i
Copyright Notice
The copyright of this thesis vests in the University of the Witwatersrand, Johannesburg, South
Africa, in accordance with the University‘s Intellectual Property Policy.
No portion of the text may be reproduced, stored in a retrieval system, or transmitted in any form
or by any means, including analogue and digital media, without prior written permission from
the University. Extracts of or quotations from this thesis may, however, be made in terms of
Sections 12 and 13 of the South African Copyright Act No. 98 of 1978 (as amended), for non-
commercial or education purposes. Full acknowledgement must be made to the author and the
University.
An electronic version of this thesis is available on the Library webpage (www.wits.ac.za/library)
under ―Research Resources‖.
For permission requests, please contact the University Legal Office or the University Research
Office (www.wits.ac.za).
LEARNER MOBILITY IN
JOHANNESBURG-SOWETO,
SOUTH AFRICA: DIMENSIONS
AND DETERMINANTS
Julia Ruth de Kadt
A thesis submitted to the Wits School of Education, Faculty of Humanities,
University of the Witwatersrand in fulfilment of the requirements for the
degree of Doctor of Philosophy.
Johannesburg, 2011
iii
Abstract and Keywords
Abstract: Many South African school children are known to travel fairly
long distances to school each day, in pursuit of the best possible educational
opportunities in a schooling system that is known to vary greatly in quality.
This thesis documents the dimensions and determinants of the daily, education-
related travel of primary school aged children in Johannesburg-Soweto, South
Africa. It uses data on a sample of 1428 children drawn from the Birth to
Twenty cohort study to provide the first population-based data on the extent of
learner mobility in contemporary urban South Africa. Learner mobility is
measured in three different ways: firstly by the straight line distance between a
child‘s home and his or her school; secondly by whether the child‘s school
falls into the same geographical area as his or her home; and thirdly by
whether the child attends his or her nearest, grade-appropriate school.
The thesis provides clear evidence for extensive mobility using all three of
these approaches to measurement. Over 25% of children were found to be
travelling more than 5km each way to school and back on a daily basis. Almost
60% of children attended a school outside of the Census 2001 Sub-Place
(roughly equivalent to a suburb) in which they lived, and fewer than 20% of
children attended the grade-appropriate school nearest to their home. Counter
to expectations, these figures were fairly stable over time, suggesting that
educational mobility does not increase substantially as children age or
transition to high school. Mobile children attended significantly more well-
resourced and well-performing schools than their non-mobile peers, and the
quality of schools attended increased with distance travelled. This substantiates
the assumption that children and families make use of educational mobility to
improve the quality of education that they are able to access.
The analyses presented in the thesis suggest that two distinct patterns of
mobility, with different determinants, are in use in the Johannesburg-Soweto
area. The first relates primarily to travel from townships to historically
advantaged schools in suburban Johannesburg, and typically requires
substantial economic investment and extensive parental involvement. The
second form of mobility operates at a more local level, and relates to children
and families making choices between a number of relatively local schools.
This form of mobility is less resource intensive. Children engaging in the first
form of mobility were more likely to attend a particularly advantaged school,
and to have a well-educated mother. By contrast, children engaged in the
second form of mobility were more likely to live in a disadvantaged area, and
come from households with moderate SES levels.
iv
The findings of this thesis provide important insights into the nature of school
choice in South Africa, which have implications for educational policy, and the
understanding of the nature of urban poverty as experienced by South African
children. They also contribute to the international school choice literature, by
providing novel information about the implications of relatively unregulated
school choice for educational inequality and segregation in the South African
context.
Keywords: Birth to Twenty; cohort data; Johannesburg; learner migration;
learner mobility; primary school; quantitative analysis; school choice; travel to
school; secondary analysis; South Africa; Soweto
v
Candidate’s Declaration
I declare that this thesis is my own unaided work. It is being submitted for the
degree of Doctor of Philosophy at the University of the Witwatersrand,
Johannesburg. It has not been submitted before for any degree or examination
at any other University.
____________________________
Julia Ruth de Kadt
11th day of May in the year 2011
vi
To my parents, in thanks for their support.
vii
Publications or Presentations Emanating from this
Research
De Kadt, J., Richter, L., Fleisch, B. & Norris, S. (2010) Measuring learner
mobility in Johannesburg-Soweto, South Africa. 2nd Isibalo Young African
Statisticians Conference, Pretoria, South Africa, December 1-3.
De Kadt, J. (2009). Learner Migration and socio-economic status in
Johannesburg, South Africa. United Kingdom Forum for International
Education and Training (UKFIET) Conference, Oxford, UK, 15-27 September.
viii
Acknowledgements
My sincere thanks to Prof. Brahm Fleisch, for supervising this thesis, and for
his ongoing enthusiasm on the topic.
Prof. Linda Richer played an essential role in enabling me to complete this
thesis. She provided me with funding for fees, travel to conferences, and a
challenging but fulfilling work environment. Despite her frantically busy
schedule, she has always found time to read drafts of my work, and to provide
guidance, mentorship, encouragement and support.
Prof. Shane Norris facilitated my access to the Birth to Twenty data, and has
since tirelessly read drafts of my work. He has been a reliable source of
valuable advice and suggestions, and his input has strengthened this thesis
immeasurably.
The whole Birth to Twenty team deserves special thanks, not just for providing
the data to me, and following up my endless queries, but also for the dedication
that has allowed the collection of such complex data over so many years.
Similarly, without the commitment of all the cohort members over many years,
this rich and valuable dataset could not exist.
The National and Gauteng Departments of Basic Education, and Statistics
South Africa, all provided data, without which this thesis would not have been
possible.
My family has been a source of strength while working on this thesis. My
father, Raphael de Kadt, provided ongoing support and encouragement, despite
my stubborn independence. My mother, Elizabeth de Kadt, has somehow
always known what I needed from her throughout this process, whether it was
warm encouragement, a delicious dinner, or a stern talking to. She has
provided support in more ways than I can count. My brothers deserve thanks
too: Daniel for help with editing, motivating, and providing a steady supply of
cute pictures of kittens and puppies; and Chris for generous gifts, dependable
advice, and lots of sushi.
My involvement in IkamvaYouth introduced me to a group of wonderful
young people, about the same age as the Birth to Twenty sample members, and
made me realise how important addressing the crises in South Africa‘s school
system is. Thank you to the organization, and all its members, for playing an
important part in my life, and for understanding when I needed to become less
involved. Special thanks to Joy Olivier & Khona Dlamini.
Thanks to Anita Craig, and her husband Denham, for allowing me to escape to
their lovely home in Cape Town from time to time, and spoiling me with food,
wine and all sorts of ideas.
ix
Colleagues at the Developmental Pathways for Health Research Unit at Wits,
from my period at the HSRC, and before that from Princeton, have been a
consistent source of encouragement and ideas. Extra special thanks to Dean
Janse van Rensburg, for sharing his wizardry with Stata and MS Access, and
providing less technical assistance in the form of friendship.
The best group of housemates I‘ve ever had also deserve a mention. Thanks to
Tawanda Makusha, Rutendo Jaison and Thabo Letsoalo for making my home a
wonderful place to be these past few years, and for always being around to
listen, talk, and de-stress. Finally, a special thank you to all my friends,
scattered all over the world, who have kept me company in one way or another
as I worked on this thesis.
x
Table of Contents
Copyright Notice ............................................................................................. i
Abstract and Keywords ................................................................................. iii
Candidate‘s Declaration ................................................................................. v
Publications or Presentations Emanating from this Research ...................... vii
Acknowledgements ..................................................................................... viii
Table of Contents............................................................................................x
List of Figures .............................................................................................. xix
List of Tables .............................................................................................. xxii
Glossary ..................................................................................................... xxix
Chapter 1: Introduction ....................................................................................... 1
1.1 Introduction .......................................................................................... 1
1.2 Defining learner mobility ..................................................................... 3
1.3 Core research questions ....................................................................... 6
1.3.1 What is the extent of learner mobility in South Africa? ............... 8
1.3.2 What are the patterns of learner mobility in South Africa,
particularly with respect to socio-economic determinants? .................. 8
1.4 Methodological approach .................................................................. 10
1.5 Rationale ............................................................................................ 11
1.5.1 Rationale for a focus on learner mobility ................................... 11
1.5.2 Rationale for a focus on school choice more broadly ................ 13
1.5.3 Rationale in terms of practical and policy implications ............. 15
1.6 Outline of the thesis ........................................................................... 17
1.7 Conclusion ......................................................................................... 18
xi
Chapter 2: Literature review ............................................................................. 20
2.1 Introduction ........................................................................................ 20
2.2 Major forms of school choice policy ................................................. 21
2.3 School choice in South Africa ........................................................... 23
2.3.1 The context of school choice in South Africa ............................ 23
2.3.2 The practice of school choice in South Africa ........................... 26
2.4 Debates in the international school choice literature ......................... 28
2.4.1 School choice and quality ........................................................... 29
2.4.2 School choice and equality ......................................................... 29
2.5 School choice and equality in South Africa ....................................... 37
2.6 Methodological approach .................................................................. 39
2.7 Conceptual framework ....................................................................... 42
2.8 Conclusion ......................................................................................... 47
Chapter 3: Methods .......................................................................................... 48
3.1 Methodological approach: quantitative secondary data analysis ....... 48
3.2 Dataset selection ................................................................................ 51
3.2.1 Birth to Twenty ........................................................................... 52
3.3 Ethical considerations ........................................................................ 54
3.4 Overview of data and variables used ................................................. 56
3.4.1 Child level variables ................................................................... 57
3.4.2 Household and maternal level variables ..................................... 59
3.4.3 Community level variables ......................................................... 63
3.4.4 School variables .......................................................................... 67
3.5 Sample selection and creation of the analytical database .................. 74
3.5.1 Sample selection ......................................................................... 74
3.5.2 Creation of the analytical dataset ............................................... 76
3.6 Operationalization of learner mobility ............................................... 79
xii
3.6.1 A distance-based definition of mobility ..................................... 79
3.6.2 An area based definition of mobility .......................................... 81
3.6.3 Mobility defined by attendance at the nearest school ................ 82
3.7 Analysis ............................................................................................. 84
3.8 Conclusion ......................................................................................... 85
Chapter 4: Sample descriptive statistics and representativity .......................... 86
4.1 Introduction ........................................................................................ 86
4.2 Sample descriptive statistics .............................................................. 86
4.2.1 Child level variables ................................................................... 87
4.2.2 Household level variables ........................................................... 88
4.2.3 Community level variables ......................................................... 90
4.3 Relationships between variables ........................................................ 91
4.4 Study sample representativity ............................................................ 91
4.4.1 How representative is Birth to Twenty? ..................................... 92
4.4.2 How representative is the study sub-sample? ............................. 99
4.4.3 Sample selection & bias: Conclusion ....................................... 107
4.5 Descriptive schools data: all Gauteng schools ................................. 108
4.5.1 School types and sectors ........................................................... 109
4.5.2 School Quintile ......................................................................... 109
4.5.3 Section 21 Status ...................................................................... 110
4.5.4 School enrolment ...................................................................... 111
4.5.5 Percentage of black learners ..................................................... 112
4.5.6 School fees ................................................................................ 117
4.5.7 Historical racial status of the school ......................................... 118
4.5.8 Matric pass rate ......................................................................... 120
4.5.9 Descriptive schools data: discussion ........................................ 123
4.6 Conclusion ....................................................................................... 124
xiii
Chapter 5: Measuring the extent of learner mobility in contemporary urban
Johannesburg-Soweto ..................................................................................... 125
5.1 Introduction ...................................................................................... 125
5.2 Distance-based operationalization of learner mobility .................... 126
5.2.1 Actual straight-line distance from home to school ................... 126
5.2.2 Binary definitions of mobility .................................................. 132
5.3 Area-based operationalization of learner mobility .......................... 135
5.4 Nearest school based operationalization of learner mobility ........... 137
5.5 Conclusion: ...................................................................................... 140
Chapter 6: Individual, family and community characteristics of mobile learners
........................................................................................................................ 142
6.1 Introduction ...................................................................................... 142
Child level characteristics ........................................................................... 143
6.2 Race ................................................................................................. 143
6.2.1 1997 .......................................................................................... 143
6.2.2 2003 .......................................................................................... 145
6.2.3 Race and mobility discussion ................................................... 147
6.3 Gender .............................................................................................. 147
6.3.1 1997 .......................................................................................... 148
6.3.2 2003 .......................................................................................... 150
6.3.3 Gender and mobility discussion ............................................... 153
6.4 Age at first school enrolment ........................................................... 153
6.4.1 1997 .......................................................................................... 154
6.4.2 2003 .......................................................................................... 155
6.4.3 Age at first enrolment and mobility discussion ........................ 156
6.5 School phase in 2003 ....................................................................... 157
6.5.1 1997 .......................................................................................... 160
xiv
6.5.2 2003 .......................................................................................... 162
6.5.3 School phase in 2003 and mobility discussion ......................... 164
6.6 Grade repetition ............................................................................... 165
6.6.1 1997 .......................................................................................... 166
6.6.2 2003 .......................................................................................... 167
6.6.3 Grade repetition and mobility discussion ................................. 169
Household level characteristics .................................................................. 169
6.7 Maternal education .......................................................................... 169
6.7.1 1997 .......................................................................................... 170
6.7.2 2003 .......................................................................................... 173
6.7.3 Maternal education and mobility discussion ............................ 175
6.8 Maternal Marital status .................................................................... 177
6.8.1 1997 .......................................................................................... 177
6.8.2 2003 .......................................................................................... 178
6.8.3 Maternal marital status and mobility discussion ...................... 180
6.9 Household SES ................................................................................ 180
6.9.1 1997 .......................................................................................... 181
6.9.2 2003 .......................................................................................... 183
6.9.3 Household SES and mobility discussion .................................. 186
Community level characteristics ................................................................ 187
6.10 Residential area poverty ............................................................... 187
6.10.1 1997 ........................................................................................ 188
6.10.2 2003 ........................................................................................ 195
6.10.3 Discussion of residential area poverty and mobility ............... 199
6.11 Conclusion .................................................................................... 201
Chapter 7: School characteristics associated with mobile learners ................ 205
7.1 Introduction ...................................................................................... 205
xv
7.2 Schools attended by study sample members, and grade-appropriate
schools closest to study sample members‘ homes ...................................... 206
7.2.1 Unweighted data ....................................................................... 206
7.2.2 Weighted data ........................................................................... 210
7.3 Which children attend which schools? ............................................ 217
7.3.1 Child level variables ................................................................. 217
7.3.2 Family & household variables: ................................................. 225
7.3.3 Community level variables: ...................................................... 231
7.3.4 Child, household and community level variables associated with
school enrolment patterns: Discussion .............................................. 236
7.4 Relationships between school attributes and mobility behaviours .. 237
7.4.1 School sector ............................................................................ 237
7.4.2 School quintile .......................................................................... 238
7.4.3 Section 21 status ....................................................................... 240
7.4.4 School Enrolment ..................................................................... 241
7.4.5 Proportion black learners .......................................................... 242
7.4.6 School fees ................................................................................ 244
7.4.7 Historical DET .......................................................................... 245
7.4.8 Matric pass rate ......................................................................... 246
7.4.9 Relationships between school attributes and mobility behaviour:
Discussion ......................................................................................... 247
7.5 Conclusion ....................................................................................... 248
Chapter 8: Changes in educational mobility over time .................................. 250
8.1 Introduction ...................................................................................... 250
8.2 Changing schools ............................................................................. 250
8.2.1 Correlates of school change during primary schooling ............ 251
8.2.2 Correlates of school change associated with the transition to high
school ................................................................................................ 255
8.3 The nature of changes in mobility ................................................... 257
8.3.1 Straight-line distance ................................................................ 257
xvi
8.3.2 Census geography ..................................................................... 265
8.3.3 Nearest school .......................................................................... 268
8.4 Correlates of type of mobility change for primary school school-
changers ...................................................................................................... 270
8.4.1 Straight-line distance ................................................................ 271
8.4.2 Census geography ..................................................................... 272
8.4.3 Nearest school .......................................................................... 275
8.4.4 Conclusion: primary school school-changers ........................... 276
8.5 Correlates of type of mobility change for children transitioning to
high school .................................................................................................. 277
8.5.1 Straight-line distance ................................................................ 277
8.5.2 Census geography ..................................................................... 278
8.5.3 Nearest school .......................................................................... 282
8.5.4 Conclusion: mobility change associated with transition to high
school ................................................................................................ 282
8.6 Conclusion ....................................................................................... 283
Chapter 9: Modelling educational mobility .................................................... 285
9.1 Introduction ...................................................................................... 285
9.2 Straight-line distance ....................................................................... 286
9.2.1 1997 .......................................................................................... 286
9.2.2 2003 .......................................................................................... 289
9.3 Census area geography .................................................................... 291
9.3.1 Sub-Place level ......................................................................... 292
9.3.2 Main Place level ....................................................................... 294
9.3.3 Census area mobility discussion ............................................... 297
9.4 Nearest school analysis .................................................................... 297
9.5 Conclusion ....................................................................................... 300
xvii
Chapter 10: Conclusion .................................................................................. 303
10.1 Introduction .................................................................................. 303
10.2 Overview of key findings ............................................................. 303
10.2.1 Developing approaches to measuring learner mobility ........... 305
10.2.2 Measuring the extent of learner mobility in Johannesburg-
Soweto ....................................................................................................... 306
10.2.3 Potential child, household and community-level determinants of
learner mobility ................................................................................. 307
10.2.4 Potential school-level determinants of learner mobility ......... 309
10.2.5 Changes in mobility as children age .......................................... 310
10.2.6 Predicting mobility ...................................................................... 311
10.2.7 Developing an evidence based conceptual framework for the
study of learner mobility ................................................................... 314
10.3 Key contributions ......................................................................... 315
10.3.1 Methodological contributions ..................................................... 315
10.3.2 Empirical contributions ............................................................... 316
10.3.3 Theoretical contributions ............................................................ 318
10.4 Contextual relevance .................................................................... 320
10.4.1 Relevance to South Africa .......................................................... 320
10.4.2 International relevance ................................................................ 322
10.5 Project limitations and future work .............................................. 323
10.5.1 Sample composition..................................................................... 323
10.5.2 Study end point and longitudinal analysis ................................ 323
10.5.3 Methodological approach ............................................................ 324
10.5.4 Future work ................................................................................... 325
10.6 Conclusion .................................................................................... 328
Appendix A: Alternative data sources considered for the thesis .................... 329
A.1 Cape Area Panel Study (CAPS) ............................................... 329
A.2 Kwa-Zulu Natal Income Dynamics Survey (KIDS)................. 330
xviii
A.3 National Income Dynamics Survey (NIDS) ............................. 331
A.4 Agincourt Health and Demographic Surveillance System
(Agincourt) ........................................................................................ 332
A.5 Africa Centre for Health and Population Studies ..................... 332
Appendix B: Letter confirming approval of ethics clearance for thesis, received
from the School of Education, University of the Witwatersrand ................... 333
Appendix C: Relationships within the study sample between variables
hypothesized to act as determinants of learner mobility ................................ 336
C.1 Race and other variables .................................................................. 336
C.2 Gender and other variables .............................................................. 337
C.3 Age at first enrolment and other variables ....................................... 337
C.4 Phase of education in 2003 and other variables ............................... 338
C.5 Grade repetition and other variables ................................................ 338
C.6 Maternal education and other variables ........................................... 338
C.7 Maternal marital status and other variables ..................................... 339
C.8 Household SES and residential area poverty levels ......................... 339
C.9 Conclusion ....................................................................................... 340
References ...................................................................................................... 341
xix
List of Figures
Figure 2.1: Theoretical framework, based on De Jong (2000) and Hanson and
Litten (1982) ......................................................................................... 44
Figure 3.1: Location of study variables within the conceptual framework
presented in Chapter 2 .......................................................................... 56
Figure 3.2: Flow chart illustrating selection of sub-sample for use in thesis. .. 76
Figure 4.1: Distribution of Gauteng schools by the proportion of their learners
who are black ...................................................................................... 113
Figure 4.2: Distribution of Gauteng schools by the proportion of their learners
who are black ...................................................................................... 114
Figure 4.3: Distribution of the proportion of black learners for public primary
and secondary schools in Gauteng ..................................................... 115
Figure 4.4: Kernel density plots of pass rates at primary and secondary schools
............................................................................................................ 121
Figure 5.1: Kernel density plots of the distribution of distances to school for
the small and large samples, curtailed at a maximum distance of 60km,
for both 1997 and 2003. ...................................................................... 128
Figure 5.2: Kernel density plot overlaid on histogram illustrating the
distribution of distances travelled by sample members in 1997. The log
transformation of the distribution is also provided. ............................ 130
Figure 5.3: Kernel density plot overlaid on histogram illustrating the
distribution of distances travelled by sample members in 2003. The log
transformation of the distribution is also provided ............................. 131
Figure 5.4: Cumulative density plot of distance between home and school, up
to 10km, laid over a histogram illustrating the density distribution of
distance ............................................................................................... 134
Figure 6.1: Kernel density plot of distance to school in 1997, on the basis of
race ..................................................................................................... 144
Figure 6.2: Kernel density plot of distance to school in 1997, on the basis of
race ..................................................................................................... 146
xx
Figure 6.3: Kernel density plot of 1997 distance from home to school, by
gender ................................................................................................. 149
Figure 6.4: Kernel density plot of 2003 distance from home to school, by
gender ................................................................................................. 151
Figure 6.5: Kernel density plot of 1997 distance from home to school, by age
at first school enrolment ..................................................................... 154
Figure 6.6: Kernel density plot of 2003 distance from home to school by age at
first school enrolment ......................................................................... 156
Figure 6.7: Kernel density plot of 1997 distance from home to school by 2003
schooling phase .................................................................................. 161
Figure 6.8: Kernel density plot of 2003 distance from home to school by phase
of education in 2003 ........................................................................... 162
Figure 6.9: Kernel density plot of 2003 distance from home to school by
gender and phase of education in 2003 .............................................. 164
Figure 6.10: Kernel density plot of 1997 distance from home to school by
grade repetition ................................................................................... 167
Figure 6.11: Kernel density plot of 2003 distance from home to school by
grade repetition ................................................................................... 168
Figure 6.12: Kernel density plot of 1997 distance from home to school by
maternal education level. .................................................................... 171
Figure 6.13: Kernel density plot of 2003 distance from home to school by
maternal education level ..................................................................... 174
Figure 6.14: Kernel density plot of 1997 distance from home to school by
maternal marital status ........................................................................ 178
Figure 6.15: Kernel density plot of 2003 distance from home to school, by
maternal marital status ........................................................................ 179
Figure 6.16: Kernel density plot of 1997 distance from home to school, by
1997 household SES quintile .............................................................. 182
Figure 6.17: Kernel density plot of 2003 distance from home to school, by
2003 household SES quintile .............................................................. 184
xxi
Figure 6.18: Kernel density plot of 1997 distance from home to school, by
SAL poverty quintile .......................................................................... 190
Figure 6.19: Kernel density plot of 1997 distance from home to school, by SP
poverty quintile ................................................................................... 192
Figure 6.20: Kernel density plot of 1997 distance from home to school, by MP
poverty quantile .................................................................................. 194
Figure 6.21: Kernel density plot of 2003 distance from home to school, by
SAL poverty quintile .......................................................................... 196
Figure 6.22: Kernel density plot of 2003 distance from home to school, by SP
poverty quintile ................................................................................... 197
Figure 6.23: Kernel density plot of 2003 distance from home to school, by MP
poverty quantile .................................................................................. 199
Figure 8.1: Kernel density plot of the change in distance from home to school
between 1997 and 2003 for all sample members with full mobility
information ......................................................................................... 259
Figure 8.2: Kernel density plot of the change in distance from home to school
between 1997 and 2003 for all sample members who changed schools
............................................................................................................ 259
Figure 8.3: Distribution of change in distance from home to school for children
travelling less far in 2003, by schooling phase ................................... 262
Figure 8.4: Distribution of change in distance from home to school for children
travelling further in 2003, by schooling phase ................................... 262
Figure 10.1: Conceptual framework revised on the basis of study findings .. 315
xxii
List of Tables
Table 3.1: Variables used in the creation of SES scores .................................. 62
Table 3.2: Sources for school-level variables used .......................................... 68
Table 4.1: Breakdown of study sample members by race ................................ 87
Table 4.2: Breakdown of study sample members by gender ............................ 87
Table 4.3: Breakdown of study sample members by age at first school
enrolment .............................................................................................. 87
Table 4.4: Breakdown of study sample members by phase of schooling in 2003
.............................................................................................................. 88
Table 4.5: Breakdown of study sample members by whether or not they have
repeated at least one grade between 1997 and 2003 ............................. 88
Table 4.6: Breakdown of study sample by highest level of maternal education
attained at the time at which the study sample member was born ....... 89
Table 4.7: Breakdown of study sample members by maternal marital status .. 89
Table 4.8: Breakdown of study sample by household SES in 1997 ................. 89
Table 4.9: Breakdown of study sample members by household SES in 2003 . 90
Table 4.10: Breakdown of study sample members by the poverty level of the
SAL in which they live ......................................................................... 90
Table 4.11: Breakdown of study sample members by the poverty level of the
SP in which they live ............................................................................ 91
Table 4.12: Breakdown of study sample members by the poverty level of the
MP in which they live ........................................................................... 91
Table 4.13: Differences between cohort members included in the study sample,
and those excluded from the study sample with regards to all available
demographic variables collected at birth ............................................ 102
Table 4.14: Differences between members of the non-attrition sample included
in and excluded from the study sub-sample, with respect to variables
collected at birth ................................................................................. 103
Table 4.15: Availability of 1997 SES data for members of the non-attrition
sample included in and excluded from the study sample ................... 105
xxiii
Table 4.16: Differences between members of the non-attrition sample included
in and excluded from the study sub-sample, with respect to SES in
1997 .................................................................................................... 106
Table 4.17: Availability of 2003 SES data for members of the non-attrition
sample included in and excluded from the study sample ................... 107
Table 4.18: Differences between members of the non-attrition sample included
in and excluded from the study sub-sample, with respect to SES in
2002/2003 ........................................................................................... 107
Table 4.19: Numbers of schools in each quintile in Gauteng province .......... 110
Table 4.20: Average number of learners for different types of schools in
Gauteng ............................................................................................... 111
Table 5.1: Comparison of findings on distances travelled from home to school
for different datasets, for 1997 and 2003. ........................................... 127
Table 5.2: Distribution of distances and log distances travelled by sample
members ............................................................................................. 130
Table 5.3: Distribution of distances and log distances travelled by sample
members ............................................................................................. 131
Table 5.4: Numbers and percentages of children classified as mobile in 1997
and 2003, for each binary definition of mobility considered. ............ 133
Table 5.5: Number and percent of children who live and attend school in the
same SP, MP and MN areas, in 1997 and 2003 ................................. 136
Table 5.6: Correlation coefficients between distance-based and area-based
measures of mobility for 1997 ............................................................ 137
Table 5.7: Correlation coefficients between distance-based and area-based
measures of mobility for 2003 ............................................................ 137
Table 5.8: Number and percentage of learners attending the school closest to
their home in 1997 and 2003, and the mean and maximum distances to
the schools nearest to sample members‘ homes ................................. 139
Table 6.1: 1997 Distance between home and school on the basis of race ...... 144
Table 6.2: 1997 mobility at different levels of census geography, by race .... 145
xxiv
Table 6.3: Children attending their nearest grade-appropriate school, by race,
for public schools only, and for all schools ........................................ 145
Table 6.4: 2003 Distance between home and school on the basis of race ...... 146
Table 6.5: 2003 Mobility across different levels of census geography, by race
............................................................................................................ 147
Table 6.6: 2003 Children attending their nearest grade-appropriate school, by
race, both for public schools only, and for all schools ....................... 147
Table 6.7: 1997 distance from home to school, by gender ............................. 148
Table 6.8: Gender breakdown of 1997 categories of distance from home to
school .................................................................................................. 149
Table 6.9: 1997 mobility across different levels of census geography, by race
............................................................................................................ 150
Table 6.10: 1997 distance from home to school, by gender ........................... 151
Table 6.11: Gender breakdown of 2003 categories of distance from home to
school .................................................................................................. 152
Table 6.12: 2003 mobility across different levels of census geography, by
gender ................................................................................................. 152
Table 6.13: 1997 distance from home to school, by age at first enrollment .. 154
Table 6.14: 2003 distance from home to school, by age at first enrolment ... 155
Table 6.15: 1997 distance from home to school, by schooling phase in 2003
............................................................................................................ 160
Table 6.16: 2003 distance from home to school, by progression to high school
by 2003 ............................................................................................... 162
Table 6.17: 2003 distance from home to school, by gender and phase of
education in 2003 ............................................................................... 163
Table 6.18: 1997 distance from home to school, by grade repetition status .. 166
Table 6.19: 2003 distance from home to school, by grade repetition status .. 168
Table 6.20: 1997 distance from home to school, by maternal education level
............................................................................................................ 170
Table 6.21: 1997 mobility across different levels of census geography, by
maternal education level ..................................................................... 172
xxv
Table 6.22: Children attending their 1997 nearest grade-appropriate public
school, by maternal education ............................................................ 173
Table 6.23: 2003 distance from home to school, by maternal education level
............................................................................................................ 173
Table 6.24: 2003 mobility across different levels of census geography, by
maternal education level ..................................................................... 175
Table 6.25: Children attending their 2003 nearest grade-appropriate school, by
maternal education .............................................................................. 175
Table 6.26: 1997 distance from home to school, by maternal marital status . 177
Table 6.27: 2003 distance from home to school, by maternal marital status . 179
Table 6.28: 1997 distance from home to school, by 1997 household SES
quintile ................................................................................................ 181
Table 6.29: 1997 mobility across different levels of census geography, by 1997
household SES .................................................................................... 183
Table 6.30: Children attending their 1997 nearest grade-appropriate school, by
1997 household SES ........................................................................... 183
Table 6.31: 2003 distance from home to school, by 2003 household SES
quintile ................................................................................................ 184
Table 6.32: 2003 mobility across different levels of census geography, by 2003
household SES .................................................................................... 185
Table 6.33: Children attending their 2003 nearest grade-appropriate school, by
2003 household SES ........................................................................... 186
Table 6.34: 1997 distance from home to school, by SAL poverty quintile .... 189
Table 6.35: 1997 distance from home to school, by SP poverty quintile ....... 191
Table 6.36: 1997 distance from home to school, by MP poverty quantile ..... 193
Table 6.37: 2003 distance from home to school, by SAL poverty quintile .... 195
Table 6.38: 2003 distance from home to school, by SP poverty quintile ....... 197
Table 6.39: 2003 distance from home to school, by MP poverty quantile ..... 198
Table 7.1: Quintiles of schools attended by and nearest to sample members‘
homes .................................................................................................. 208
xxvi
Table 7.2: Schools attended by sample members (note: schools classified as
combined are included in both columns for 2003) ............................. 212
Table 7.3: Proportion of sample members closest to and attending independent
schools ................................................................................................ 212
Table 7.4: Distribution of sample members across schools by school quintile
rating ................................................................................................... 213
Table 7.5: Proportion of schools without Section 21 status nearest to and
attended by sample members .............................................................. 213
Table 7.6: Size of schools nearest to and attended by sample members ........ 214
Table 7.7: Proportion of black learners at schools nearest to and attended by
sample members ................................................................................. 215
Table 7.8: Fees charged by schools nearest to and attended by sample members
............................................................................................................ 216
Table 7.9: Proportion of schools nearest to and attended by sample members
that were historically under the DET .................................................. 216
Table 7.10: Pass rates of schools nearest to and attended by sample members.
............................................................................................................ 217
Table 7.11: Relationship between child race (black and coloured children only)
and properties of the school he or she attends .................................... 219
Table 7.12: Relationship between child gender and properties of the school he
or she attends ...................................................................................... 220
Table 7.13: Relationship between child age at first school enrolment and
school properties ................................................................................. 222
Table 7.14: Relationship between grade repetitions between 1997 and 2003,
and school properties .......................................................................... 224
Table 7.15: Relationship between maternal education and school properties 226
Table 7.16: Relationship between maternal marital status and school properties
............................................................................................................ 228
Table 7.17: Relationship between 1997 household SES and school attributes in
both 1997 and 2003 ............................................................................ 229
xxvii
Table 7.18: Relationship between 2003 household SES and school attributes in
1997 and 2003 .................................................................................... 231
Table 7.19: Relationship between residential SAL poverty and school
attributes in 1997 and 2003 ................................................................ 232
Table 7.20: Relationship between SP poverty and school properties ............ 234
Table 7.21: Relationship between MP poverty and school properties ........... 236
Table 7.22: Relationship between sector of attended school and learner
mobility ............................................................................................... 238
Table 7.23: Relationship between quintile of attended school and learner
mobility ............................................................................................... 240
Table 7.24: Relationship between Section 21 status of attended school and
learner mobility .................................................................................. 241
Table 7.25: Relationship between size of attended school and learner mobility
............................................................................................................ 242
Table 7.26: Relationship between proportion of black students at attended
school and learner mobility ................................................................ 244
Table 7.27: Relationship between fees of attended school and learner mobility
............................................................................................................ 245
Table 7.28: Relationship between historical DET status of attended school and
learner mobility .................................................................................. 246
Table 7.29: Relationship between quintile of attended school and learner
mobility ............................................................................................... 247
Table 8.1: Proportion of sample members remaining in the same school from
1997 to 2003, and proportion changing schools (note that information
on schooling phase is only available for 1196 individuals) ............... 251
Table 8.2: Distribution of sample members with full residential and schooling
data for both 1997 and 2003 across schooling phases, and stability of
mobility behaviour .............................................................................. 257
Table 8.3: Changes in distance from home to school between 1997 for all
sample members, disaggregated by schooling phase and school change
status ................................................................................................... 258
xxviii
Table 8.4: Counts of sample members who live closer to or further from their
school in 2003 as compared to 1997, disaggregated by phase of
schooling ............................................................................................. 261
Table 8.5: Children attending schools in the same categories of distance from
their homes in 1997 and 2003, disaggregated by schooling phase ..... 263
Table 8.6: Numbers of children moving between schools in different
geographical areas, disaggregated by schooling phase ...................... 265
Table 8.7: Changes in distance from home to school experienced by children
moving between schools in different geographic areas, disaggregated
by schooling phase ............................................................................. 266
Table 8.8: Children attending school in the same area as their home, for three
different levels of geography, disaggregated by schooling phase ...... 268
Table 8.9: Children attending their nearest grade-appropriate public school in
1997, 2003 and at both points, disaggregated by schooling phase ..... 269
Table 9.1: 1997 regression results. Figures in parentheses are standard errors.
............................................................................................................ 289
Table 9.2: 2003 regression results. Figures in parentheses are standard errors.
............................................................................................................ 291
Table 9.3: 1997 & 2003 SP mobility regression results. Figures in parentheses
are standard errors. ............................................................................. 294
Table 9.4: 1997 and 2003 MP mobility regression results. Figures in
parentheses are standard errors. .......................................................... 296
Table 9.5: 1997 and 2003 nearest school attendance logistic regression results.
Figures in parentheses are standard errors. ......................................... 300
Table 9.6: Summary of variables associated with increased mobility in the
regression models presented in this chapter ....................................... 301
Table 10.1: Overview of the key findings presented in this thesis ................. 305
Table 10.2: Summarized results for the models of mobility developed in
Chapter 9. Results presented are for regressions with robust errors. . 313
Table A.1: Significance of relationships between household SES and
residential area poverty levels ............................................................ 340
xxix
Glossary
ASS – Annual Schools Survey
Bt20 – Birth to Twenty Cohort Study
DET – Department of Education and Training
DOE – Department of Education
EEA – Employment of Educators Act
EMIS – Education Management Information System
GDE – Gauteng Department of Education
GIS – Geographical Information System
MN – Municipality level
MP – Main place level
NEPA – National Education Policy Act
PCA – Principal Components Analysis
SAL – Small area level
SASA – South Africa Schools Act
SES – Socio-economic status
SGB – School Governing Body
SP – Sub-place level
SRN – School Register of Needs
1
Chapter 1: Introduction
1.1 Introduction
Much recent work has highlighted the generally poor state of education in
South Africa. Schools, particularly those serving the less privileged, tend to be
poorly performing, and the skill levels of South African children are
notoriously low (Reddy 2006; Fleisch 2008; Spaull 2011; van der Berg, Burger
et al. 2011). Simultaneously, the schooling system is also known for the highly
variable resource levels enjoyed by different schools within the public sector,
and the enormous variations in school performance that tend to accompany
this. School quality is closely connected to South African history, with most
well performing schools being those that historically educated white1 children.
This means that well performing schools are usually located in historically
white areas, and that school quality is closely related to geography (Fiske and
Ladd 2004; Woolman and Fleisch 2006; Spaull 2011; van der Berg, Burger et
al. 2011).
Within this context, and in the aftermath of Apartheid‘s Bantu education
policies, education is very highly valued by many South Africans. Completing
secondary school and reaching tertiary education are core goals held by many
young people, regardless of their backgrounds. That there is a strong
relationship between the quality of the school attended, and future
opportunities, is widely accepted. As a result, children and families are often
willing to go to great lengths to ensure the best possible educational
opportunities (Woolman and Fleisch 2006; Lombard 2007).
1 In this thesis, the four race groups defined by the Apartheid-era government will be used to
categorize individuals, due to South Africa‘s unique historical context, and the ongoing
relevance of these categories to the life experiences and educational opportunities of young
South Africans.
2
This thesis explores some of the ways in which children and families in post-
Apartheid Johannesburg-Soweto pursue the highest-quality educational
opportunities possible. In particular, it focuses on school choice, and the
associated phenomenon of learner mobility. Learner mobility is used to denote
the travel of learners to attend schools other than those closest to their homes.
Anecdotal evidence suggests that learner mobility is both widespread and
substantial in contemporary South Africa, but very little is actually known
about its dimensions, drivers and implications. Although several studies have
identified its occurrence, often at fairly substantial levels, and some have even
discussed it in relative depth (Fiske & Ladd, 2004; Nelson Mandela Children's
Fund, 2005; Paterson & Kruss, 1998; Soudien & Sayed, 2003; Woolman &
Fleisch, 2006) there has been only one attempt to construct a broader, data-
driven picture (Sekete, Shilubane, & Moila, 2001). Although this study
provides valuable baseline data for a critical period of history, it was not
conducted on a population based sample, meaning that the applicability of its
findings to a broader urban population is not clear. Given the apparently high
level of learner mobility in urban South Africa, and its potentially significant
implications for the South African educational system, updating and deepening
our understanding of this phenomenon is an urgent need.
While we know that learner mobility appears to be extensive, and that due to
the distribution of educational opportunities in contemporary South Africa it is
likely to have substantial importance to the life chances of young urban South
Africans, we know very little about the actual extent of engagement in school
choice and learner mobility. Although we have reason to believe that most
learner mobility involves travelling to historically advantaged schools, we
know very little about where children actually tend to enrol in school, relative
to where they live. Our knowledge about the determinants of school choice and
learner mobility is also very limited. This thesis uses secondary quantitative
analysis of longitudinal data from the Birth to Twenty (Bt20) study, which
tracked a sample of 3273 young people born in the highly urbanized
3
Johannesburg-Soweto area in a six week period in 1990, to address these
questions. Establishing the dimensions, patterns and correlates of learner
mobility in South Africa, and beginning to understand its implications, will fill
significant gaps in the local and international literatures on school choice, and
have important implications for the design and implementation of policies to
ensure genuinely equitable access to high quality education in South Africa.
1.2 Defining learner mobility
Before detailing the research questions addressed by this thesis, it is critical to
clearly define learner mobility. The term ―learner mobility‖ is used to refer to
the daily travel of learners to a school that is not the school nearest to their
home (Karlsson 2007). It is derived from the phrase ―learner migration‖, which
has previously been used in the South African literature on the travel patterns
of learners (Sekete, Shilubane et al. 2001; Lombard 2007). This term, however,
has not obtained widespread usage outside of this fairly limited literature. In
addition, it does not adequately differentiate between daily travel or
commuting, and genuine migration in which learners spend nights away from
their permanent family residence for the purpose of attending a specific school.
As both of these practices are believed to be widespread in contemporary
South Africa, it is essential to distinguish between them, particularly as their
implications for individual learners, the educational system and society more
broadly, along with appropriate policy responses, are likely to differ
substantially. In addition, the learners making use of them are expected to
represent different groups, particularly with regard to residential location and
family socio-economic status (SES) (Paterson and Kruss 1998). As daily travel
is the focus of this dissertation, learner mobility is used as a more appropriate
descriptor.
It should be stressed that learner mobility is only one expression of school
choice in South Africa. Other, and likely widespread, expressions of choice
4
occur when a family chooses their residence on the basis of proximity to
specific schools, when a child leaves the public sector to attend a private
school, regardless of location, or when a child lives with people other than his
or her family in order to attend a specific school. These types of choice are
excluded from the above definition of mobility for various reasons, including
that it is not possible to measure them with the data available for this project,
and that they can be understood to operate differently from the types of
movement described above in terms of the resources required and used, the
perceived benefits, the population groups most likely to be taking advantage of
them, and their implications for public policy. While all these forms of school
choice in South Africa certainly warrant further examination, they do not fall
within the scope of this thesis.
This thesis takes several different approaches to the measurement of learner
mobility, in order to capture a range of different aspects of the phenomenon.
The first approach involves measuring the straight-line distance between a
child‘s home, and his or her school. This is selected because it is theoretically
sound, capturing to some extent the level of investment required by mobility,
and because it can be accurately measured using the available data.
Additionally, and for these same reasons, it has been widely used in existing
work on learner mobility, although in some cases distance is supplemented or
replaced by travel time (Sekete, Shilubane et al. 2001; South African Human
Rights Commission 2004; South African Human Rights Commission 2006;
Pendlebury and Rudolph 2008).
The second approach involves determining whether the learner is travelling to
a school inside or outside of the area in which he or she lives. Unfortunately,
identifying in a consistent and useful way whether a school and a home address
are in fact located in the same area is very challenging in contemporary South
Africa. Educational districts, as used by the Gauteng Department of Education
(GDE) for administrative purposes, do not generally align with electoral wards,
5
census districts, or neighbourhoods. Most schools do not have a clearly defined
catchment area, unless it has been defined by the school itself. Electoral wards
and census districts have also undergone recent and substantial changes, and in
many cases bear limited resemblance to historically defined neighbourhoods.
In this context, the most feasible way to measure travel between areas proved
to be by locating both school and home within their relevant census districts,
using GIS coordinates. Although imperfect, this data does at least provide a
preliminary measure of whether a child attends school in the area in which he
or she lives.
Finally, the third approach to measuring mobility is to identify whether or not a
child is attending the grade-appropriate school nearest to his or her home. This
is used primarily as an indicator of whether or not a child and his or her family
are engaging in school choice at all. Of course, this measure is imperfect, as a
child attending the school closest to his or her home may have chosen this
school deliberately, while a child attending a school further from home may do
so for completely involuntary reasons. Nonetheless, this measure is used as the
best available indicator to provide an approximate measure of the extent of
engagement in school choice amongst children in post-Apartheid
Johannesburg-Soweto.
It is critical to note that many other definitions of learner mobility are of course
possible, and in various contexts may indeed be more appropriate. For
example, when parents and learners make school enrolment choices, they
certainly take into account a wide range of factors other than distance. Given
South Africa‘s long and racially-defined history of learner migration and
differential educational opportunities, the historical racial designations of
schools is one factor that is likely to play a significant role (Paterson and Kruss
1998; Lombard 2007). Factors such as travel time, travel cost, safety, school
reputation, and school fees, among others, are also likely to be of particular
importance (Chisholm 2004; Maile 2004; Lemon 2005).
6
The measures of learner mobility used in this thesis, and discussed above,
include both a continuous measure (distance from home to school), and binary
measures (attending a school in the area in which a child lives, and attending
the school nearest to the child‘s home). This is appealing, as mobility can be
thought of as both binary – occurring or not occurring – and as a continuum –
with the extent of mobility determined by the distance travelled to school. In
addition, combining both binary and continuous measures allows for the
application of a broader range of different analytical approaches.
1.3 Core research questions
The literature review, presented in Chapter 2, explores and defines the context
in which school choice, including learner mobility, occurs in contemporary
urban South Africa. This provides the background against which the study‘s
research questions can be addressed. Current school choice practices in South
Africa have emerged in the context of rapid, global, changes in educational
systems. Key elements of these changes have included increasing demands for
higher levels of education along with universal access to basic education, a
focus on improving the quality of education, which is often accompanied by
increased levels of testing and ICT use, and pressure to keep public sector
spending to a minimum through cost recovery and privatization (Carnoy 1999).
During the 1990s there were widespread debates around the extent to which
market forces and an increased private role in educational production could
allow these otherwise somewhat contradictory goals to be met, and
establishing school choice was viewed as a critical component of getting
educational markets to work properly (Hoxby 2002; Hoxby 2003). While these
debates have now died down, most of the pressures that spurred them have not
gone away, and some parts of the response to these pressures, such as school
choice, have become entrenched (Greene, Loveless et al. 2010). Unfortunately,
this entrenchment has not been combined with evidence that school choice is
7
an effective response to these pressures, nor, indeed that these pressures should
be accepted in the first place. Nonetheless, school choice, whether official or
unofficial, regulated or not, is now accepted – or at least tolerated – in most
countries.
In South Africa, the issue of school choice is complicated further by the
country‘s history of massive racial inequalities. Under the Bantu Education
Act, children‘s educational opportunities, along with the resources devoted to
these, were determined entirely by the colour of their skin (Fedderke, de Kadt
et al. 2000). While this policy is now a thing of the past, it has left behind a
very persistent set of geographically defined inequalities in educational
infrastructure and resources (Fiske and Ladd 2004; Fiske and Ladd 2005;
Spaull 2011; van der Berg, Burger et al. 2011). As residential desegregation is
occurring only slowly, for most black learners the only opportunity to attend a
high quality school comes through a willingness and an ability to travel in
order to reach one. While government policies do not legislate against school
choice, they also do not facilitate choice as an option for the most
disadvantaged learners, nor do they protect these learners from any harmful
effects of their inability to express choice (Maile 2004; Woolman and Fleisch
2006). The ability to express choice is therefore expected to remain closely
linked to SES, even as SES gradually dissociates from race.
School choice, and particularly learner mobility, has become entrenched and
broadly accepted in contemporary South Africa, even though relatively little is
known about its extent. In this context, it is critical to explore this phenomenon
to develop an understanding of its implications, as well as how any negative
effects can be mitigated. This thesis documents learner mobility between 1997
and 2003. This period of time is thought to be critical to the development and
establishment of new, post-Apartheid patterns of interaction with educational
opportunity. The most substantial policy changes thought to have an impact on
mobility had already been made by this period.
8
I will now move through the two major research questions addressed in this
thesis, providing critical context for each, and exploring key hypotheses.
1.3.1 What is the extent of learner mobility in South Africa?
Given good reason to believe that learner mobility, as an expression of school
choice, exists at significant levels in South Africa, the next task becomes to
measure it (Karlsson 2007). Determining the extent of learner mobility in post-
Apartheid Johannesburg-Soweto forms the first core research question in this
thesis. To date, although there is ample evidence that learner mobility is fairly
widespread, there is little concrete information about just what this means. This
thesis measures learner mobility in the Johannesburg-Soweto metropolitan
area, using the three different definitions of learner mobility described above.
The actual dimensions of mobility are both of practical and theoretical
importance. Practically, the extent of learner mobility – in each of the forms
described – has implications for educational policy and planning.
Theoretically, the extent of the phenomenon is important as it feeds into major,
ongoing debates, both domestic and international, around the implications of
school choice, particularly when unregulated, for both academic and societal
outcomes. Although some work has been done to measure the extent of learner
mobility and school choice in particular populations (Paterson and Kruss 1998;
Sekete, Shilubane et al. 2001; Msila 2005; Karlsson 2007; Lombard 2007), to
my knowledge there is no existing work which provides a measurement of the
extent of mobility at a population level in contemporary urban South Africa.
1.3.2 What are the patterns of learner mobility in South Africa,
particularly with respect to socio-economic determinants?
There appear to be two major groups of determinants of school choice. The
first set of determinants relates to the broader context in which school choice
occurs, and includes factors like policy, the nature of the schooling system, the
9
distribution of educational opportunities, and so on. This is covered in this
study‘s literature review, and has been fairly comprehensively dealt with by
the existing South African school choice literature (Woolman and Fleisch
2006).
The second set of determinants relate to the attributes of individual children,
their families, and the communities in which they live, as well as the properties
of the schools available to them. International work has documented the
importance of these determinants, but very little theoretical work, and even less
empirical work, has been done on this for South Africa. As a result, this area is
where the bulk of the analytical work conducted in this thesis is focussed.
This second research question addressed by the thesis therefore revolves
around identifying who, primarily, is involved in learner mobility, and
exploring the patterns evident in this mobility. The evidence that is available
suggests that patterns of mobility are highly mediated by race, class, gender,
age and geographic location (Paterson and Kruss 1998; Sekete, Shilubane et al.
2001; Fiske and Ladd 2004; Nelson Mandela Children's Fund 2005; Karlsson
2007). The relationship between each of these variables and mobility will be
explored in this thesis, as will the relationship between mobility and a range of
attributes of the school a child lives nearest to, and the school which that child
attends.
As the majority of documented learner mobility seems to involve children
living in historically disadvantaged urban African communities (Sekete,
Shilubane et al. 2001; Maile 2004), this thesis focuses particularly on the urban
Johannesburg-Soweto area. In line with data provided by other work (Sekete,
Shilubane et al. 2001; Chisholm 2004; Fiske and Ladd 2004), the most
substantial portion of the mobility identified is hypothesised to be learners
travelling from townships and informal settlements to schools in historically
more advantaged areas. In addition, some evidence of choice and mobility
10
within the more disadvantaged urban areas are also expected, with children
choosing to attend schools other than those nearest to their homes, and
sometimes those outside of their immediate residential area.
These forms of intra-urban learner mobility are hypothesised to be linked to
family race, SES, and maternal education. Of those involved in learner
mobility, the most advantaged learners and those with the most educated
parents are expected to travel furthest, to attend schools in the most advantaged
areas (typically, historically white schools). Somewhat less advantaged
learners, and those with somewhat less educated parents, are expected to travel
somewhat shorter distances to attend schools in more moderately advantaged
areas (typically, historically Coloured or Indian schools). Travel to attend
schools in other highly disadvantaged areas (typically, historically Black
township schools) is expected to be restricted to the least advantaged among
mobile learners.
The thesis also explores changes in the mobility behaviour of children over
time. Children are expected to become increasingly mobile as they age, with
far higher levels of mobility anticipated at the secondary school level than at
the primary school level. The final analytical component of the thesis ties
together all of the potential determinants of mobility analysed, and provides
preliminary models for the prediction of mobility behaviours in children in the
Johannesburg-Soweto area.
1.4 Methodological approach
The methodological approach taken in this thesis is quantitative secondary
analysis. This is a widely accepted research method (Bryman 2004), and is
well suited to answering the research questions outlined above. The thesis
makes use of data on a sub-sample (n=1428) of the Bt20 cohort, which is
comprised of all (n=3273) children born in a 6 week window in 1990 in the
11
Johannesburg-Soweto metropolitan area. In order to allow for the
consideration of variables relating to schools and communities in the analysis,
and the incorporation of spatial effects, the Bt20 data is combined with schools
data from the Department of Education (DOE), and census district data from
Census 2001. As detailed previously, a number of approaches to the
measurement of mobility are taken. This is followed by a range of bivariate
analyses, exploring the relationships between learner mobility and
hypothesized determinants of mobility. Change in mobility over time is also
examined. Finally, a regression approach is used to combine all relevant
variables, and generate preliminary models of learner mobility.
1.5 Rationale
1.5.1 Rationale for a focus on learner mobility
As mentioned previously, learner mobility is only one form of school choice in
South Africa. Framing this study of learner mobility as a contribution to the
international school choice literature, and as a contribution to understanding
school choice and its implications for policy and practice in South Africa,
requires answering two questions. Firstly, why only study a part of the
phenomenon, and not the entire phenomenon? Secondly, why focus on learner
mobility, as opposed to any other expression of school choice?
The argument for addressing only a single form of school choice stems
primarily from the complexity of school choice in contemporary South Africa,
combined with the current paucity of data on the phenomenon. The review of
the international, empirical literature on school choice (presented in Chapter 2)
will demonstrate that the implications of school choice depend heavily on both
the context of choice, as well as the exact nature of the choice that is available.
If multiple forms of school choice are combined in a single study, the study
will need to differentiate between these different forms of choice, and their
12
differing implications. Because we currently know so little about school choice
in South Africa, any empirical study will need to begin by identifying and
measuring each particular form of choice, before it is possible to move on to
examine implications. Providing such a thorough and deep examination of
multiple forms of choice would be ideal, but is simply beyond what is feasible
in this study.
Of the various forms of school choice in evidence in contemporary South
Africa, there are several reasons for this study to focus specifically on learner
mobility. Firstly, learner mobility appears to be extensive. Although the
literature on learner mobility itself is extremely limited, data from studies of
various other educational topics suggest that it is likely to be quite prevalent.
For example, Karlsson (2007) provides evidence that the single largest group
of commuters in contemporary South Africa is school children. Additionally,
in a representative survey of Grade 12 learners nationwide in 2001, 24.9%
were found to live more than 10km away from their school, a proportion which
excluded those at boarding schools (Cosser and du Toit 2002). Even if some of
this mobility can be explained by children in rural areas who have poor access
to schooling, it still suggests that voluntary mobility in urban areas is likely to
be prevalent.
Secondly, learner mobility is of particular theoretical interest, due to its
potentially significant role in decreasing the racial segregation of schools, but
simultaneously increasing their socio-economic segregation, with substantial
implications for educational access and equality. This is because the costs
associated with learner mobility tend to be fairly substantial, meaning that it is
likely to be shaped primarily by socio-economic status, rather than race.
Recent work suggests that although the role of race in determining educational
outcomes has been falling in post-Apartheid South Africa, the role of SES has
been becoming increasingly important (van der Berg, Burger et al. 2011).
Understanding the relationship between SES and school choice is a critical
13
precursor to understanding the nature of the shifting relationship between race,
SES and educational outcomes.
Thirdly, learner mobility is in many ways the most measurable form of non-
private school choice, and can in fact be measured, at least for samples of the
population, using data which already exists. This means that considerable
information can be obtained without requiring the time-consuming and
expensive collection of specialised data. In addition, learner mobility has
implications for a number of areas of educational policy, including school
financing and governance. The information generated by this study will allow
for a reassessment of key elements of South African educational policy, as well
as existing academic work based on the assumption of limited or no learner
mobility. It will also highlight the practical questions that need to be asked
about South Africa‘s current approach to school choice, and provide some
guidance regarding potential alternatives.
1.5.2 Rationale for a focus on school choice more broadly
As will be illustrated in Chapter 2, and as alluded to above, the South African
literature on school choice is fairly limited, and of the work that does exist,
only a very small proportion is empirical. This sparse empirical literature is in
itself a strong justification for an empirical investigation of the actual extent of
school choice in a fairly large urban population in South Africa. Additionally,
most existing empirical work tends to be either extremely local or highly
aggregated. While each of these types of work is very useful, particularly given
the limited state of existing knowledge, this does leave substantial space for
work at a more intermediate level, which allows for both the contextualization
of data, but also the identification of more generalizable patterns.
The international literature makes very clear that school choice policy and
practice are very closely connected to issues of educational quality and
educational equality. Understanding the determinants of access to high quality
14
education is of great relevance to national developmental goals of rapid and
ongoing economic growth, job creation, racial equality, and socio-economic
justice. Particularly in the context of the huge variations in educational quality
present in South African public education, it is critical to understand who has
access to good education, who does not, and how to best improve access for
this second group.
Although there is a fairly broad, and growing, literature on educational quality
in South Africa, there is as yet very little exploration of how school quality
shapes patterns of choice, and vice versa, or of the distribution of access to
high quality educational opportunities. While this project will not provide
definitive information about the implications of school choice for educational
quality or outcomes, it will provide preliminary information about the
relationship between engagement in school choice, and the quality of the
educational opportunities a child is likely to be able to access. This will
facilitate the development of data-driven hypotheses to be tested by future
work. Understanding patterns of learner mobility will also provide useful
information around questions of access to quality education, and potential
patterns of investment to address quality improvement.
Existing South African work on educational equality has tended to focus on
resource and quality differentials across schools (Chisholm 2004; Fiske and
Ladd 2004; Kanjee 2007). South Africa‘s highly unequal educational system
has been strongly shaped by the country‘s history of racial segregation, and the
race-based allocation of educational resources (Motala 1995; Fedderke, de
Kadt et al. 2000; Motala, Dieltiens et al. 2009). Racial inequalities in the
educational sector remain extensive, and rightfully receive a substantial
amount of attention, both politically and in the academic literature (Fiske and
Ladd 2004; Nkomo, McKinney et al. 2004). However, as socio-economic
inequalities within race groups begin to grow (van der Berg, Wood et al.
2002), direct attention to the implications of socio-economic status for
15
education becomes increasingly important. In focussing closely on the
correlates of learner mobility, this study will bring to light a great deal of
information about the extent to which SES and access to high quality schooling
are related, and by extension, whether it is important to begin focussing more
attention on segregation across schools on the basis of SES rather than race.
Although the South African literature on educational equality is much broader
than that on school choice, once again relatively little of it is empirical, and
concerns about the lack of empirical work at an intermediate level also hold. A
great proportion of the existing work on equality and desegregation has also
focussed on those schools generally thought of as higher quality, or more
desirable – that is, generally, former Model C and independent schools. By
contrast, schools which are perceived as low quality have been to a large extent
ignored, even though they make up a far greater proportion of schools
nationally. This project adds valuably to the literature because it will provide
reliable information about who actually attends these very numerous, lower
quality schools.
In addition to contributing to the local understanding of an important but
poorly understood phenomenon, and the local school choice literature, this
thesis has also been designed to help fill several important gaps in the
international school choice literature. Firstly, it will provide insight into the
extent to which school choice can become accepted in developing country
context, even without the existence of intentionally pro-school-choice policies.
Secondly, it will explore the implications of unplanned and largely unregulated
school choice for educational equality and segregation, in a context of
extremely high inequality.
1.5.3 Rationale in terms of practical and policy implications
The information generated in this project will assist in the assessment of
current educational policies, most of which are based on the assumption that
16
the large majority of children attend local, neighbourhood schools. For
example, if mobility levels are substantial, and mobility is related to SES, this
means that the school poverty rankings which determine the allocation of
governmental funds are likely to be inaccurate, leading to underfunding of
some learners, and overfunding of others. Similarly, high levels of learner
mobility would imply problems with current community-centred school
governance policies, in both the schools which lose learners and those which
receive them. For example, if schools in disadvantaged areas are enrolling only
the most disadvantaged of the children living in these areas, the managerial and
fund-raising capacity of their parent bodies, and by extension their School
Governing Bodies (SGBs), will probably be far more limited than would
otherwise be the case. Furthermore, if more advantaged children are attending
schools relatively far afield, their parents, who might otherwise be eager to
play an active role in school governance, will experience greater logistical
barriers to engagement in those schools. Additionally, in light of current school
fee policies, which require schools to grant fee exemptions to disadvantaged
learners, the extent of mobility is likely to have implications for the revenue
streams of both more advantaged schools enrolling relatively disadvantaged
children from further afield, and more disadvantaged schools enrolling fewer
of the local children with any ability to pay fees. Information on the extent to
which school choice is currently practiced, and by whom, will also allow for
data-driven reflection on the appropriateness of those very policies which
currently serve to shape and constrain choice.
At the level of learner outcomes, learner mobility has potential implications for
academic performance, drop-out and repetition. Acting directly on learners,
mobility can be hypothesised to have either positive implications, due to access
to enhanced educational opportunities, or negative implications, due to travel
time, sub-optimal resource allocation, and cultural differences. Mobility may
also operate on learners indirectly, through its impact on the school
environment. A question of particular interest in this regard is the extent to
17
which more advantaged children do actually leave their neighbourhood
schools, and what implications this has for the children remaining in those
schools. While these questions are not answered explicitly in this thesis, the
information generated about mobility will allow for the development of
clearer, data-driven hypotheses regarding the likely implications of learner
mobility for both individual children, and for South Africa‘s schooling system
more broadly.
1.6 Outline of the thesis
Chapter 1 has provided an overview of the topic of this thesis, along with the
specific research questions to be addressed, and preliminary hypotheses
regarding anticipated findings. In addition, it has provided a brief rationale for
the research project presented in this thesis.
In Chapter 2, the international and South African literature relating to both
school choice and learner mobility is reviewed. A particular focus is placed on
the literature documenting the relationship between school choice policies and
practices, and educational equity outcomes. The argument is made that the
study of learner mobility in South Africa can provide valuable information
about the ways in which school choice may impact on educational
opportunities, specifically in the context of a developing country in which
choice is largely unregulated.
Chapter 3 provides an overview of the methodological approach taken in the
study. Chapter 4 explores issues around sample representativity, and provides
descriptive statistics for the sample. Chapter 5 uses three different approaches
to measure learner mobility, and presents data on the extent of this mobility.
In Chapter 6 a range of bivariate analyses are conducted to explore the
relationships between mobility behaviours, and variables at the levels of
18
individual children, their families and households, and the communities in
which they live. Chapter 7 extends these analyses to document the
relationships between mobility behaviours and the attributes of the schools
which children attend, as well as the schools closest to their homes. Chapter 8
documents the ways in which mobility behaviours change over time, as
children age.
Chapter 9 combines variables considered in the previous three chapters to
generate preliminary models for each of the measures of learner mobility that
have been discussed. Distance between home and school is modelled using
OLS regression, while travel between areas, and attendance at the nearest
school are modelled using logistic regression. Finally, chapter 10 provides a
brief overview and discussion of the findings of the study, and concludes the
thesis.
1.7 Conclusion
This chapter began by providing some brief insight into the context of school
choice and learner mobility in contemporary South Africa. It then defined
learner mobility, and provided three different measures that will be used to
document its extent in post-Apartheid Johannesburg-Soweto. Firstly, the
straight-line distance between home and school will be measured. Secondly,
Census 2001 district data will be used to determine whether or not a child lives
and attends school in the same geographical area. Thirdly, whether or not a
child attends the grade-appropriate school nearest to his or her home will be
documented. The chapter then moved on to document the core research
questions to be addressed in this thesis. These relate to measuring the extent of
learner mobility in contemporary urban South Africa, and to identifying the
correlates of engagement in learner mobility at the level of the child, the
household, the community, and the school. This was followed by a brief
description of the methodological approach taken in the chapter, and a
rationale for the selection of the topic on both theoretical and practical, policy-
19
relevant grounds. The chapter then concluded by providing an outline and
overview of the chapters to follow.
20
Chapter 2: Literature review
2.1 Introduction
This chapter serves to place the key questions addressed by this thesis in the
context of the relevant academic literature, and to outline the methodological
and theoretical approaches used in addressing those questions. As learner
mobility is a particular form of school choice, this literature comprises the
international literature on school choice, and more specifically that part of this
literature which explores the relationships between choice, quality, and
equality. Due to learner mobility‘s potential implications for South African
educational equality, the literature on South African inequality and segregation
in the post-Apartheid era is also relevant. The first section of this chapter
outlines various forms of school choice found in internationally, and argues
that three key dimensions of choice policies are the extent to which they are
officially legislated, the extent to which they include provisions to protect
vulnerable groups from potential harm, and the extent to which they include
provisions to allow members of vulnerable groups to actively engage in choice.
The second section of the chapter outlines both the policy and practice relating
to school choice in post-Apartheid South Africa, and makes the argument that
the emergence of learner mobility in South Africa can best be understood as
the outcome of an unplanned, and largely unregulated, school choice system,
with very few protections for vulnerable groups.
Having described and situated South African learner mobility within the
context of school choice more broadly, the review then moves on to explore
two key debates in the international school choice literature: firstly, the debate
about whether or not school choice improves performance in educational
systems, and secondly the debate about whether school choice increases
inequality and segregation in educational systems. Following this, the literature
on educational equity and desegregation in South Africa is reviewed, with a
21
particular focus on the very limited body of work which relates these issues to
school choice. This illustrates that by providing data on the relationship
between choice, mobility and educational equality, this thesis will make a
useful contribution to the scholarly literature both locally and internationally.
The next section of the chapter places this study‘s methodological approach
within the context of the existing scholarly literature, and illustrates the
methodological contributions this thesis makes. In the final section, the
conceptual framework used as the foundation of the analyses presented in this
thesis is described, with reference to how it has been shaped by the literature,
both South African and international.
2.2 Major forms of school choice policy
Although there is a broad and varied literature on school choice, it is a highly
ideological literature, with much heated debate but fairly limited empirical
data. Much of the available work has been commissioned by interested parties,
or is driven by individuals with strong ideological positions (for a discussion of
this phenomenon with respect to the UK school choice literature, see Gorard
and Fitz (2006)). Additionally, the large majority of existing literature,
particularly empirical literature, relates to school choice as implemented in the
developed world, while work on school choice in the developing world is far
sparser. As a result, this overview of the major forms of school choice policy
draws primarily on examples from the developed world.
School choice is generally understood to be occurring when families are able
to make a decision about which school a child will be enrolled in. Globally,
public schooling systems range from those where there is almost complete
choice (such as New Zealand and the Netherlands), to those where children are
required to attend particular schools, usually on the basis of residential location
(such as Cuba, France and Japan). Most systems are located somewhere
22
between these extremes, with a global trend towards increasing levels of
choice (Goldhaber and Eide 2002; Plank and Sykes 2003).
There are some forms of school choice which are virtually impossible to
control. The most significant of these is when people select their residences on
the basis of access to particular schools (Holmes 2002). Another significant
expression occurs when people leave public schooling systems to attend
private schools, although this is obviously limited to those systems in which
private schooling is accepted (Rinne, Kivirauma et al. 2002). The extent to
which people can exercise these types of choice depends on their location, their
flexibility in terms of location, and, critically, their wealth (Hoxby 1998).
Importantly, these forms of choice interact with various different legislated
systems of choice.
Forms of choice provided by legislated systems of choice may include
provisions for choice between a number of local schools, magnet school
programmes which are accessible to any student on the basis of a lottery or
academic performance, establishment of charter schools or multiple small
schools in particular areas to provide alternatives to traditional schools, or
simply unfettered enrolment at any school, regardless of location. Each form of
choice will evidently have different implications in terms of equality of access,
and consequences for particular population groups (Teske and Schneider 2001;
Hoxby 2002).
While some school choice legislation simply provides a legal right to a certain
amount of choice between different public schools, others are designed to
counter inequalities associated with school choice, particularly when expressed
through access to private schools. Most widespread here are various voucher
systems, common in the US, but also emerging in other countries such as
Chile, which pay a varying proportion of private school fees for less
advantaged learners (Goldhaber and Eide 2002; Elacqua, Schneider et al.
23
2006). In addition, provisions to protect particular groups may also be included
in any type of school choice policy. They might, for example, require that
schools enrol a certain proportion of students from particular ethnic or
economic backgrounds, or might provide free transportation for certain
students to the school of their choice (Holmes 2002).
In summary, the key variations in school choice policies, shaping which
people, and how many of them, can engage in school choice are the extent to
which these policies:
are planned or legislated, as opposed to unplanned or unofficial;
include provisions to protect vulnerable groups from potential
harm;
and include provisions to ensure that vulnerable groups have
equitable access to choice.
I now explore school choice policy in South Africa, with particular attention to
its properties in these regards.
2.3 School choice in South Africa
2.3.1 The context of school choice in South Africa
Although the school choice literature on South Africa is quite limited,
particularly with regards to empirical work, there is some strong theoretical
literature documenting the ways in which existing policies shape school choice
in the country. School choice in South Africa is regulated primarily through the
National Education Policy Act (NEPA), the South Africa Schools Act (SASA),
and the Employment of Educators Act (EEA) (Pampallis 2003; Maile 2004;
Woolman and Fleisch 2006), although the more recent classification of a
number of schools as no-fee schools also seems likely to be important (Ahmed
and Sayed 2009).
24
While at first glance South African policy appears to constrain learners to
attend neighbourhood schools on the basis of their residence, it combines with
national history to provide both motivation and means for parents to choose
their children‘s schools. Huge variations in both empirical and perceived
school quality mean that many parents are extremely motivated to ensure that
their children attend specific schools. Policies around school financing and the
provision of teaching staff mean that schools are motivated to enrol as many
children as possible. As public funding is limited, fee-charging schools are
particularly eager to enrol large numbers of children who are able to pay fees.
Decentralisation of managerial functions to school governing body (SGB) level
means that schools do have some control over the design and enforcement of
their admissions policies, and by extension over which learners they enrol,
although this control is often de facto rather than de jure, and is subject to
some legal constraints. In addition, school choice through residential selection
continues to operate, as many advantaged schools prioritize the enrolment of
local children. The growing independent schooling sector also provides parents
with further choice. As a result, the opportunities for school choice in South
Africa are substantial, but come at a fairly marked financial cost to parents.
There is widespread evidence that significant numbers of parents are none-the-
less exercising this choice (Sekete, Shilubane et al. 2001; Maile 2004; Lemon
2005; Nelson Mandela Children's Fund 2005; Johnson 2007; Lam, Ardington
et al. 2008).
Although the policies mentioned above have played a central role in shaping
the way that school choice has developed in South Africa, it should be noted
that, perhaps with the exception of legislation around independent (private)
schooling, this role was generally not intended (Woolman and Fleisch 2006) –
instead, school choice was a largely accidental outcome of policies developed
for other reasons2. The system of school choice in South Africa can therefore
2 Although the fact that there have been no major efforts to reduce levels of choice in the
system does suggest that the existence of choice appears to suit the political and social elite.
25
be understood as one which is generally, although not entirely, unplanned and
unofficial. Due to the substantial freedom of choice that seems to exist, as well
as the considerable inequality in both South Africa‘s schooling system and
income distribution, the question of whether there are any protections in place
for vulnerable groups in the context of large-scale school choice is clearly
important. As the major determinant of the ability to exercise choice seems to
be the ability to pay higher fees and pay for additional transportation, it is
probable that the ability to exercise choice is strongly linked to socio-economic
status.
As a result, the major group at risk of an inability to engage in choice, or even
at risk of harm by being left in the most poorly performing schoos, are those of
lower socio-economic status (Pampallis 2003; Fiske and Ladd 2004). Potential
protective policies might therefore include additional support for schools
which primarily attract children from disadvantaged contexts, or the provision
of incentives to advantaged schools to enrol less advantaged learners.
However, given that there are no explicit government policies on school
choice, there are also none of these types of provisions to protect or support
vulnerable populations. Similarly, with regards to providing vulnerable groups
with access to school choice, for example by paying their school fees or
providing free transportation, there are also no policies in place3. This means
that South Africa‘s school choice policy could be considered as one that is
unplanned, unofficial, and unregulated, with few protective measures, while
simultaneously allowing quite extensive levels of choice to certain sectors of
the population.
3 Although a school fee exemption policy exists, which exempts disadvantaged learners from
the obligation to pay fees, this only applies once children have been granted admission to the
school in question. Additionally, implementation is generally acknowledged to be poor.
Veriava, F. (2005). Free to learn: A discussion paper on the School Fee Exemption policy.
Cape Town, South Africa, Children's Institute, University of Cape Town.
26
2.3.2 The practice of school choice in South Africa
Due to the limited literature on South African school choice, the description of
the practice of school choice in contemporary South Africa presented in this
section draws primarily on the theoretical literature, although empirical studies
related to school choice are cited where they exist. Currently, school choice in
South Africa appears to take four major forms: residential, private, intra-
area, and inter-area, where intra- and inter-area choice corresponds to learner
mobility as defined in this thesis.
Residential school choice occurs when parents select homes on the basis of
their proximity to particular schools. Exercising residential school choice
generally requires a relatively high level of income and parental education.
Due to the geographic distribution of good schools in South Africa, with most
good schools located in affluent, historically white areas with high property
prices, the constraints on parental ability to exercise residential choice are
likely to be particularly high. Due to the private nature of residential location
decisions, this type of school choice is also extremely difficult to measure.
Private school choice occurs when parents decide to exit the public schooling
system altogether, instead sending their child to an independent (private)
school. In South Africa, although growing, the independent schooling sector
remains relatively small, accommodating just over 3 percent of learners (du
Toit 2003; Hofmeyr and Lee 2004; Centre for Development and Enterprise
2010). While increasing numbers of independent schools offer relatively low
fees, and access appears to have expanded greatly in recent years, these
schools still serve only relatively small numbers of children (Centre for
Development and Enterprise 2010). Most high quality independent schools
charge high fees, and often select learners on the basis of academic capability.
While the sector is increasingly diverse, racially and socio-economically (du
Toit 2003; Hofmeyr and Lee 2004), choosing an independent school is still not
an option for the large majority of less-advantaged parents, due to the small
27
size of the sector. In addition, finding a space for a disadvantaged child in an
independent school is likely to require fairly substantial knowledge and effort
on the part of a parent, again making it less of an option for most
disadvantaged families.
Intra-area choice occurs when parents are able to choose between a number of
schools within their residential area, and make enrolment decisions themselves,
on the basis of any particular set of factors. This type of choice is very difficult
to measure, as there is no easy way to distinguish between learners who are
simply attending the school most accessible to their home, and those who
choosing to attend a particular school among those closest to their home for
other reasons. In addition, because most residential areas in South Africa
remain fairly homogenous, the extent to which this type of migration is likely
to matter to socio-economic segregation may be relatively limited.
Nonetheless, there is evidence from a small number of studies that parents do
distinguish between local schools, and that even within disadvantaged areas,
schools with better reputations tend to charge slightly higher fees and attract
slightly more advantaged learners (Fiske and Ladd 2004; Msila 2009).
Finally, inter-area choice occurs when parents choose a school outside of the
area of their residence. This form of choice appears to be fairly wide-spread in
South Africa, with large numbers of learners in various contexts reporting that
they attend school relatively far from home (Sekete, Shilubane et al. 2001;
Cosser and du Toit 2002; Nelson Mandela Children's Fund 2005). In some
cases, particularly in rural areas, this travel may be due to children not having
schools close to their homes, rather than choice, but in urban areas this is
typically not a concern.
In practice, the line between intra- and inter-area school choice is quite fuzzy,
particularly in the South African context where, unlike in most developed
countries, there are no consistently defined school catchment areas, and the
28
geography of school districts rarely meshes with that of residential areas.
Nonetheless, because inter-area choice generally requires parents to access
some additional information, as well as fund additional travel and potentially
higher fees, it seems probable that parents accessing inter-area choice are
likely to be somewhat more advantaged than those only able to access intra-
area choice. On the other hand, inter-area choice is also unlikely to be used by
the most advantaged families, as they tend to already be living in the areas with
the strongest schools.
This description of school choice practices in contemporary South Africa
makes clear that the expression of school choice in the country is extremely
complex and multi-layered. In some cases, multiple forms of choice may co-
occur, for example with learners travelling substantial distances to attend
independent schools, or parents sending a child to a local primary school, and a
distant high school. As this example also illustrates, multiple forms of choice
may be evident at different times during a single child‘s education. Each form
of school choice is governed by different constraints, particularly with respect
to the socio-economic attributes of those who are able to exercise them.
2.4 Debates in the international school choice literature
Two major topics of debate are evident in the international literature around
school choice. The first is the relationship between choice and educational
quality, and the second is the relationship between choice and educational
equality. Although historically these debates have been largely theoretical,
studies drawing on data are increasing in number, and these two debates will
be discussed with particular reference to this type of empirical work.
Nonetheless, the highly ideologically driven nature of much of the literature,
and the polarized nature of these debates, does make a clear and unbiased
interpretation of this literature challenging. I will explore the debates around
equality in some depth, as this is the area to which this thesis contributes most
29
directly, but will first touch briefly on the discussions around the relationship
between school choice and educational quality.
2.4.1 School choice and quality
Proponents of school choice argue strongly that choice creates markets, or at
the least, quasi-markets, in the educational arena, resulting in competition
between schools for students and resources, and by extension, for better
academic performance (Coleman 1992; Hoxby 2002; Hoxby 2003). Critics, by
contrast, argue that the market in public education is too imperfect to result in
the type of competition which would improve performance overall. Instead,
they argue, choice will lead to a growing divide between well-resourced
schools attracting good students, and poorly resourced schools attracting the
weakest learners (Levin 1991; Astin 1992). These students and their schools
are often termed ―those left behind‖ in the school choice literature, and studies
on how school choice policies affect them are particularly inconclusive (Teske
and Schneider 2001). To date, research findings on the quality implications of
school choice, both broadly and for specific populations, remain mixed. Much
of the variation evident in the quality outcomes of school choice policies
relates to differences in policy design, implementation, and evaluation
methodologies (Henig 1994; Goldhaber 2000; Teske and Schneider 2001;
Goldhaber and Eide 2002). Despite some contemporary claims of consensus
around the notion that choice improves quality, an increase in the
implementation of choice programmes, and a dramatic falloff in the quantity of
research on the topic, there in fact remains very little agreement on this issue
(Lubienski, Weitzel et al. 2009).
2.4.2 School choice and equality
The second major area of debate, closely related to the discussions around
quality, is around the implications of school choice for educational equality.
School choice policies have potential to impact equality of opportunity,
30
through shaping access to particular schools, equality in school performance,
equality for people from different backgrounds by way of approaches to
diversity, and finally, equality with regards to whether and how parents may
exercise choice in the educational realm (Godwin, Kemerer et al. 1998). While
all of these aspects of equality are important, particularly in the context of
rapid societal change, equality of opportunity is the most salient in
contemporary South Africa, and it is here that this review is focussed.
Levin (1991) argues that while school choice may provide private benefits to
individuals in the form of access to better schools, this is likely to be
accompanied by social harm. Equality of educational opportunity is closely
related to student sorting, which has long been recognized as an outcome of
any form of school choice (Hoxby 2003). Sorting along racial and socio-
economic lines, which may lead to racial or economic segregation, has been a
particular focus of the literature. Segregation between schools is a well-
documented phenomenon both in South Africa (Chisholm 2004) and
internationally, across educational systems with a range of policies towards
choice (Coleman 1992). Critics of school choice argue that choice tends to
increase segregation through two key mechanisms. Firstly, the ability and
willingness to take advantage of school choice varies along with demographic
variables; and secondly, the bases on which choices are made also vary with
demographic variables (Holmes 2002; Ladd 2003; Denessen, Driessena et al.
2005; Elacqua, Schneider et al. 2006). By contrast, proponents of school
choice tend to argue that choice has a positive effect on the equality of
educational opportunity, decreasing segregation along lines of race and class
by allowing disadvantaged students to escape from badly-performing
neighbourhood schools. They acknowledge that segregation along the lines of
performance may increase, but argue that this will simply increase incentives
for competition between schools, improving outcomes across the system
(Hoxby 2003). While some segregation is inevitable, systems must strive to
avoid segregation on the basis of race or class, and use school choice as a tool
31
to this end (Coleman 1992). Critics counter that given the clear evidence of a
close connection between class, race and academic performance, even
segregation purely on the basis of performance will increase segregation on the
basis of class and race (Astin 1992).
To date, the evidence as to whether, and how, school choice might influence
segregation levels in schools remains mixed (Goldhaber and Eide 2002;
Viteritti 2005; Godwin, Leland et al. 2006). Given that the outcomes of school
choice depend very strongly on both the context of implementation, as well as
the design of the policy, this is hardly surprising (Henig 1994; Hoxby 2003).
Different research methods have also played a role in the variable outcomes of
research into school choice and segregation. For example, in the area of
parental decision-making, the bulk of research has been conducted through
surveys and interviews. Studies of actual enrolment patterns, however, have
demonstrated that how parents claim to make enrolment decisions does not
always correspond with how they actually make enrolment decisions (Elacqua,
Schneider et al. 2006). Similarly, parental answers to questionnaires may be
markedly different to their answers during in-depth interviews (Bagley 1996).
Data constraints are another problem. For example, many studies use data only
from a single point in time, which makes it impossible for them to demonstrate
a relationship between particular polices, and changes in segregation levels
(Gorard and Fitz 2000). Finally, the deep-seated beliefs around social justice or
free market competition held by many involved in this debate has meant that
research has not always been entirely objective, making it still more difficult to
unravel the genuine implications of school choice policies for inequalities and
segregation (Gorard and Fitz 2006).
Nonetheless, it is worth presenting a brief overview of what has been found
internationally, to provide an understanding of the current state of knowledge
on the issue. In particular, these studies provide insight into how various types
of school choice interact with various types of school system structure, to
32
shape outcomes related to segregation. Particularly in the absence of much
empirical work on school choice in South Africa, exploring these international
variations is critical to understanding how South Africa‘s unplanned and
unregulated choice system is likely to play out in the context of its already
extremely segregated and highly varied schooling system.
Empirical investigations of the implications of choice policies for
segregation
While perhaps least relevant to South Africa, the evidence from small and
fairly homogenous countries, or those with a single, national policy on choice,
tends to be the most straightforward. This is particularly so when, as is the case
in many European countries, education policy and curricula are nationally
defined, and private education is minimal. For example, in 2000 in Sweden,
school admissions decisions nationwide were switched from the basis of
residential location to academic performance. This resulted in a growth of
differentiation between schools in terms of achievement, but also,
significantly, substantial growth in segregation on the basis of race, SES and
immigration status (Daun 2003; Soderstrom and Uusitalo 2005).
One of the key reasons for these types of outcomes appears to be that parents
often take the racial, religious or socio-economic composition of a school‘s
student body into account when making school choices for their children –
although they are very rarely willing to admit this. School segregation in the
Netherlands, for example, is clearly related to parental decision making
(Karsten, Felix et al. 2006). It is not only advantaged parents who take
ethnicity and class into account in decision making about schools, however. In
the Netherlands, decision making by members of the typically disadvantaged
Muslim minority appears to play an important role in shaping segregation
(Denessen, Driessena et al. 2005). Likewise, in Germany, the provision of
school choice at the primary level has been associated with ethnic segregation
33
due to the different patterns of decision making around schooling exhibited by
both German and Turkish families (Kristen 2005).
Even in these comparatively simple contexts interpretations of the data can
vary substantially. Scholars concur that the introduction of school choice
policies in New Zealand between 1989 and 1993 has had highly variable
results for different schools. But while some argue for a clear underlying theme
of increasing segregation along racial, and to a lesser extent, socio-economic,
lines between schools (Waslander and Thrupp 1995; Fiske and Ladd 2000),
others argue that this is not the case, and that segregation has actually
decreased markedly (Gorard and Fitz 2006).
The UK‘s schooling system lies somewhere between most European systems
and the American system, both in terms of the centralization of educational
decision making and planning, and in terms of the level of choice that has
traditionally been available to parents, and therefore in the complexity of the
analysis of school choice outcomes. In some ways, more so in terms of policy
than resource levels or history, the schooling system in the UK resembles the
South African system quite closely. In particular, they both combine extremely
local school management and decision making with extremely centralized
curriculum planning and policy making. The debate on the implications of
changes to school choice policy in the UK is particularly heated (Gorard and
Fitz 2006). Some evidence of a national, short-term decrease in socio-
economic segregation in response to increased school choice has been
presented (Gorard and Fitz 2000; Gorard, Fitz et al. 2001; Bradley and Taylor
2002), although the implications for racial segregation as well as long-term
effects are less clear (Bagley 1996; Noden 2000). Other scholars argue,
however, that even the evidence for a short-term decrease in socio-economic
segregation is not clear, and that segregation has actually increased with the
introduction of greater parental choice. In addition, regional variations in the
effects of school choice policy appear to have been very high, and aggregate
34
changes in segregation at the national level may in fact have very little to do
with choice policy (Gorard and Fitz 2000; Noden 2000). Evidence from the
UK also suggests that parents take race and class into account when choosing
schools. For example, white parents have been found to avoid schools with
large numbers of non-white children, although the evidence for this claim is
largely qualitative (Bagley 1996).
The large majority of available research evidence on the implications of school
choice comes from the US. While this evidence is particularly mixed, which is
to be expected given the country‘s wide range of choice policies as well as its
substantial demographic variations, it is also particularly rich. The first form of
explicit choice made available in the US was through magnet school
programmes. Originally designed as a tool to combat racial segregation, there
is some evidence of their working effectively in this regard. However, there is
also evidence that in a number of cases magnet schools have not decreased
segregation; in some cases they may have increased it, or added a new
dimension, such as SES, to existing racial segregation (Henig 1994; Goldring
and Hausman 1999; Saporito 2003).
Another type of choice programme designed to combat segregation, though in
this case with a greater focus on socio-economic segregation, are voucher
programmes. While voucher programmes were implemented starting in the
1970s, the data on their implications for segregation is extremely limited. It is
clear that there are socio-economic and demographic differences between those
parents who do and do not participate in voucher programmes, but what these
differences mean for segregation levels is not clear. While desegregation in the
private schools receiving voucher-bearing students can be expected, there are
concerns about those schools ‗left behind‘, and the students they educate
(Bridge and Blackman 1978; Capell 1981; Henig 1994; Witte and Thorn 1996;
Levin 1998; Goldhaber 1999; Hoxby 2003; Peterson, Howell et al. 2003). With
the 2002 Supreme Court decision that voucher programmes including religious
35
schools are constitutional, and the subsequent growth in voucher programmes,
it seems likely that clearer information will gradually become available.
Currently, a particularly widespread form of choice in the US is various charter
school programmes, which provide parents with a way to opt out of local
public schools. Again, those who choose to participate in this choice
programme differ from the general population. Participating parents tend to be
better educated and somewhat better off. Race is generally found to be a strong
determinant of which charter school a student will choose to attend, and there
is some evidence that charter schools have had the most appeal to black
parents, meaning that charter schools do in fact tend to be fairly highly
segregated (Weiher and Tedin 2002; Bifulco and Ladd 2006; Garcia 2008). A
final set of choice programmes has been those operating at the intra-district
level, allowing parents some choice between the different schools within a
particular school district. However, this type of policy can only be effectively
implemented in fairly densely populated, urban, areas, and again, outcomes
have been mixed, with reports of both increased and decreased segregation
(Henig 1994; Godwin, Leland et al. 2006).
Across all forms of school choice, there is clear evidence that American
parents do tend to take concerns about racial and socio-economic composition
of schools into account, even if they do not admit this explicitly (Schneider,
Marschall et al. 1998; Holmes 2002; Schneider and Buckley 2002; Saporito
2003). Furthermore, even when parents from different backgrounds vary in
their school-choice preferences in terms of variables other than race or socio-
economic composition, the end effect may remain one of segregation or sorting
(Schneider, Marschall et al. 1998).
While the US literature is both substantial and diverse, a few clear themes
stand out. The first is that in shaping outcomes, details are important, both in
policy design and in the context of implementation (Henig 1994; Hoxby 2003;
36
Greene, Loveless et al. 2010). The second is that even when choice policies are
designed to attain a particular outcome such as desegregation, their actual
effects are difficult to predict. And thirdly, we do not yet have any definitive
answers to questions about what choice means for educational equality.
Studies providing empirical data on school choice in developing countries are
relatively few and far between. In terms of racial and socio-economic
inequalities, resource levels in the educational system more broadly, and
capacity to implement policy, studies from these countries are likely to be
particularly relevant to South Africa. The structure of the schooling systems in
most of these countries, which generally rely very heavily on private
education, does however differ quite substantially from that found in South
Africa.
One of the most well-documented cases of developing world school choice is
the national voucher plan implemented in Chile in 1980, which spurred a rapid
growth in private sector educational provision (Carnoy and McEwan 2003).
There is clear evidence that those parents making use of the voucher
programme to send their children to the private religious schools which
produce the best educational outcomes, tend to be more advantaged, both
educationally and economically, than those who do not. In addition, Chilean
parents also take the socio-economic composition of a school‘s student body
into account when making school choices, and more advantaged parents are
particularly likely to enrol their children in schools with other advantaged
children, prioritizing socio-economic composition over academic performance
(Carnoy and McEwan 2003; Elacqua, Schneider et al. 2006). However, the
extent to which the voucher programme is actually responsible for the high
levels of socio-economic segregation in Chilean schools remains debated
(Narodowski and Nores 2002).
37
In China, school choice is beginning to emerge, after a long period of entirely
state-controlled schooling, but remains limited primarily to the more
advantaged members of society (Tsang 2003). Evidence of increasingly
choice-oriented schooling systems in post-Soviet countries is also beginning to
emerge, but information on their likely implications for segregation is not yet
available (Filer and Munich 2003).
While this review presents a very mixed picture, a few points are clear. Firstly,
there is potential for school choice policies to influence racial and socio-
economic segregation, in a range of different contexts, and in a number of
different ways. Secondly, the exact nature of this influence is highly dependent
on the specificities of the context of implementation, as well as on the details
of policy design. This review therefore supports the contention that examining
the implications of school choice in South Africa for racial and socio-economic
segregation is likely to provide useful information, both for those involved in
managing and improving the nation‘s education system, as well as for those
interested in understanding more clearly the interactions between context,
policy, choice and segregation.
2.5 School choice and equality in South Africa
Although the literature on school choice in South Africa is limited, there is a
fairly well developed literature which speaks more broadly to the inequalities
inherent in the country‘s educational system. This literature forms the core of
the review presented below, although work with an explicit focus on choice is
referred to whenever appropriate. In contemporary South Africa, numerous
factors shape the access of individuals to high quality educational opportunities
(Soudien, Carrim et al. 2004). For many reasons, race is the most salient of
these factors, and has received a great deal of academic attention. There is now
clear evidence that historically advantaged South African schools are
becoming increasingly heterogeneous, although to varying degrees, with
38
regards to race and language (Sekete, Shilubane et al. 2001; Maile 2004; Sujee
2004; Lemon 2005; Johnson 2007). While concerns remain about the
inclusivity of this integration, and very few schools reflect the race distribution
of the country or its regions, there has certainly been a marked change since
1994. However, while these changes in the racial composition of schools have
been well-documented, much less work has looked at school socio-economic
composition, either statically or over time. The information that is available
tends to suggest that substantial differences exist across schools in terms of
their socio-economic composition (Maile 2004; Chisholm 2005; Lemon 2005;
Reschovsky 2006; Motala 2009; Hunter 2010). Given increasing levels of
socio-economic inequality in South African society more broadly, it seems
likely that socio-economic segregation at the school level has the potential to
increase relative to racial segregation.
A major concern about the literature on school choice and equality in South
Africa is closely related to this point. While the importance of race should not
be understated, its high salience has tended to obscure, to some extent at least,
other dimensions of inclusion/exclusion and discrimination. Particularly
important in the context of school choice is SES. Given the geographically
unequal distribution of good schools in South Africa, the huge variations in
public school cost, and the strong relationship between school cost and quality,
SES is likely to be strongly related to the quality of education a learner can
access. While SES and race remain inextricably linked in South Africa, the
primary pathways through which limitations on school choice operate now
appear to have shifted away from race, and towards socio-economic status.
This makes the paucity of the literature examining the relationship between
school choice and socio-economic status particularly interesting, and alarming.
Understanding whether school choice is genuinely linked to SES, and how any
related negative implications of choice can be mitigated, requires urgent study.
39
This question is particularly critical to those historically disadvantaged
communities where populations are racially fairly homogenous, but increasing
socio-economic diversity is becoming evident. In part, this is because of the
risk that certain schools and learners will be ‗left behind‘ by school choice.
While the current system of largely unregulated choice may enable well-
performing schools to attract more resources and further improve their
performance, it may occur at the cost of schools which are struggling due to
historically low resource levels, and learners who do not have the resources to
exercise their genuine preferences for schools (Motala 2009). This division of
schools and learners is of even more concern in light of the extremely high
levels of socio-economic inequality in South Africa.
This thesis makes an important contribution to the South African literature on
school choice firstly by documenting how wide spread school choice is in the
Johannesburg-Soweto metropolitan area. Additionally, it fills theoretical and
empirical gaps by examining a range of potential contributors to school choice,
including both race and SES. Finally, by examining SES explicitly, and
including both advantaged and disadvantaged individuals, it will shed light on
how broadly school choice is available as an option for members of all racial
and socioeconomic groups. In the following section, the methodological
contributions of thesis are discussed with reference to the existing body of
work.
2.6 Methodological approach
As already mentioned, empirical work on school choice in South Africa is
extremely limited. With a few exceptions, those empirical studies which do
touch on either school choice or on educational inequality tend to belong to
one of two extremes. Either, they focus on a small sample, providing deep, rich
data, usually but not always qualitative (Soudien 2003; Msila 2005; Msila
2009; Bray, Gooskens et al. 2010; Hunter 2010), or they are quantitative and
40
extremely broad, providing only highly aggregated statistics (van der Berg,
Wood et al. 2002; Sujee 2004). Each of these types of study clearly has great
value, particularly when knowledge about a phenomenon is limited. However,
as knowledge increases, particularly in a country as diverse as South Africa, it
becomes important that a more rounded literature develops and provides
information at an intermediate level of aggregation, allowing data to become
increasingly contextualized. Very little is currently available at this level, and
even when it is, it tends to focus on either school choice or on educational
inequalities (Sekete, Shilubane et al. 2001). To my knowledge, no studies exist
which combine the examination of choice and educational inequality at an
intermediate level. This is important gap, as it prevents us from understanding,
for example, potentially substantial variations between rural, urban and peri-
urban areas, or between different parts of the country at substantially different
levels of development.
In light of this gap, there is considerable scope for an empirical analysis of the
relationship between school choice and socio-economic status, and particularly
one which makes use of a relatively large sample from a clearly specified
context. The two core questions addressed by this thesis, relating to the scope
and the correlates of learner mobility in Johannesburg-Soweto, South Africa,
seek to fill this gap. Although these questions are fundamentally empirical in
nature, addressing them requires that this thesis also tackles the methodological
and theoretical gaps evident in the literature, and proposes some novel
solutions. In addition, as will have become clear from the overview of
international literature, this thesis will also enrich the international debate
around school choice, and provide particularly valuable information for the
discussions about how school choice, inequality and segregation interact.
This study offers three further methodological contributions. Firstly, most
existing empirical work around learner mobility, whether quantitative or
qualitative in nature, comes from school-based studies (Fiske et al., 2004;
41
Sekete et al., 2001). While the school-based approach has many advantages,
particularly with regards to understanding how levels of mobility impact
school performance and functioning, it does also have drawbacks, most
notably with respect to understanding population-based levels and patterns of
mobility. This is partly because mobility appears to be highly clustered around
specific schools. Once a school first begins to enrol learners from outside the
local community, particularly when those learners are from a different race
group, members of the local community tend to begin to avoid that particular
school (Fiske et al., 2004). Focusing on particular schools, rather than on
particular populations or communities, may therefore provide either inflated or
deflated data, depending on whether the school is one that caters primarily to
mobile learners or not.
In order to overcome some of these difficulties, this thesis makes use of
population level data. As far as I am able to determine, it is the first study of
learner mobility in South Africa to do so. Unfortunately, the type of national
data necessary to generate this understanding is not available for South Africa.
The best available alternative is to draw on large-scale datasets that sample
populations in particular parts of the country, and use this information to draw
conclusions at an intermediate level. This can in turn guide future data
collection and analysis, toward a more complete understanding of the
phenomenon. In addition, the use of data at an intermediate level has
advantages of its own, such as allowing for control for geographic and other
associated variation. The reasons for the selection of the Birth to Twenty
dataset are documented in Chapter 3. Using this type of data, however,
provides an additional advantage, which is that changes in mobility can be
explored over time. Again, this thesis is the first study of which I am aware
which looks at mobility behaviour at more than one time point.
A second methodological innovation is that the project makes use of data from
a number of different sources, in order to overcome the limitations imposed by
42
each individual data source. This allows for the simultaneous examination of
school-level, child and family-level, and community-level determinants of
mobility. Again this is the first empirical South African study of which I am
aware that is able to examine this full range of potential determinates of
mobility.
Thirdly, this thesis takes a completely novel approach to the measurement of
learner mobility. Whereas previous studies have used only one approach to
measuring mobility, typically either travel distance or travel time, this study
uses three different measures of mobility, each capturing a different aspect of
the phenomenon. These are straight-line distance between home and school,
whether a child attends a school in the same area in which he or she lives, and
whether a child attends the grade-appropriate school nearest to his or her home.
This is critical in that it allows for the unpacking of different forms of choice
and mobility, as well as their varying determinants, and raises questions as to
whether learner mobility in South Africa is a unitary concept. These measures
are documented more fully in Chapter 3.
2.7 Conceptual framework
Although the international literature on school choice is fairly large, there is a
very limited body of work which actually explores, empirically, the
determinants of the choices made by individual learners and their families
(Bosetti 2004). Where empirical data does exist, it is often focused on choice
of a particular type of school, for example independent or religious, as opposed
to the choice of a particular school (Le and Miller 2003; Elacqua 2006). While
a range of variables have fairly consistently been found to be important –
particularly those relating to social class, as discussed above – there is no clear
model which is systematically used to predict engagement in choice. As a
result, I draw on work in other areas to develop the conceptual framework used
in this thesis. In particular, the framework draws on work on decision-making
43
around mobility more broadly (De Jong 2000), and the literature on the process
by which students choose the higher education institutions they apply to
(Hanson and Litten 1982). The framework that I have derived, presented in
Figure 2.1 below, therefore also serves as a contribution to the theoretical
literature around school choice in South Africa.
The framework conceives of school choice as always occurring within a
particular historical, geographical and policy context. This is guided by the
evidence presented in the literature review above that context is critical in
determining the ways in which school choice play out. Examples of relevant
aspects of historical context in contemporary South Africa include the
enormous variability of school quality, and the ways in which access to
resources, including education, are distributed across the population, on the
basis of both race and class (Fiske and Ladd 2004; Msila 2005; Msila 2009;
Spaull 2011; van der Berg, Burger et al. 2011). Geographical context includes
the variable nature of different residential areas, and the ways in which these
are physically located, as well as the geographical distribution of schools of
differing levels of quality (Hunter 2010; van der Berg, Burger et al. 2011).
Finally, examples of the ways in which policy shape school choice include the
high levels of variation in the cost of attending different public schools, the
ways in which policy allows and constrains school choice, and the extent to
which families have access to reliable information about particular schools
(Fiske and Ladd 2004; Woolman and Fleisch 2006).
44
Figure 2.1: Conceptual framework, based on De Jong (2000) and Hanson and Litten
(1982)
In this framework, variables at the level of the individual child, the family, and
the community in which the family lives, all interact in shaping each other, and
simultaneously all feed into the decision making process (De Jong 2000). In
this thesis, a range of variables at each of these levels are tested for their
relationship with mobility. The selection of variables for testing is guided by
the international and local literatures on school choice where available.
Potentially important child-level variables include gender, race, school
performance and academic aptitude, psychological adjustment, and child
location within the family (for example, number and relative ages and genders
Decision making
process
- Desired
educational
outcomes
- Investment
required for
desired
outcomes
- Investment
constraints
Residential location
Child
Household
& Family
Maternal
Historical, geographical and policy context
School
selection
- School quality
- Distance
-‘Localness’
45
of siblings). In this thesis, due to data limitations, race, gender, and a few
indicators of academic aptitude are examined. Race is suggested for inclusion
by the wide range of international literature suggesting that it is a strong
predictor of school choice behaviours (Glazerman 1998; Fiske and Ladd 2004;
Fiske and Ladd 2005; Karsten, Felix et al. 2006; West and Hind 2007). Gender
is included due to the literature documenting differential parental investment in
children‘s education on the basis of their gender (Alderman and King 1998;
Klasen 2002; Unterhalter 2005), although concerns about this in the
contemporary South African context are fairly limited. Academic aptitude is
also included as there is some reason to believe that it may also shape parental
willingness to invest in education, or contribute to school choice in other ways
(Glazerman 1998; Zietz and Joshi 2005; West and Hind 2007).
At the household level, variables such as education levels of members of the
household, the structural stability of the household, the wealth of the
household, and the household‘s residential stability are considered.
Additionally, maternal attributes are likely to be of particular importance to
schooling decisions. Relevant variables, very similar to household level
variables include the mother‘s marital status, and the stability of her
relationship status, her education level, her income, and her age. In this thesis,
the variables tested are maternal education, maternal marital status (as a proxy
for household stability), and household SES. Maternal education is included on
the basis of evidence for a relationship between maternal education and
educational choices made for children (Magnuson 2007; Andrabi, Das et al.
2009; Greenberg). Marital status is included as a proxy for household stability,
as there is evidence that both indicators are related to child outcomes and
wellbeing (Osborne and McLanahan 2007). Finally, household SES is included
due to evidence for a relationship with school choice (Glazerman 1998; Msila
2005; Andrabi, Das et al. 2009; Msila 2009; Hunter 2010).
46
At the community level, relevant variables might include the quality and
attributes of local schools, the affluence and education levels of the
community, as well as the coherence of the community itself, and the extent to
which it suffers from problems such as crime. This thesis includes a range of
measures of school quality, such as poverty quintile rating, school fees
charged, matric examination performance, and historical level of advantage,
along with the racial composition of the student body and the size of the
school, all of which have been demonstrated or hypothesized to relate to school
choice in the South African context (Paterson and Kruss 1998; Fiske and Ladd
2004; Fiske and Ladd 2005; Msila 2005; Woolman and Fleisch 2006; Lombard
2007; Msila 2009). The thesis also includes a measure of community poverty,
as this may influence the willingness of parents to send children to local
schools (Msila 2005; Lombard 2007).
All of these variables at the child, family, community levels are expected to
feed into the decision making process, in which children and families weigh
their desired outcomes in terms of school level attributes, with the investment
required, and the constraints that they face. Required investments are likely to
depend on desired outcomes, the geographical location of both the household
and the schools considered, and the socio-economic status of the household.
Constraints on investment are likely to depend on household access to human
capital, social and economic resources, the structure of the household, and the
extent to which the household prioritizes educational or other forms of
investment. This decision making process is likely to result in the identification
of a small group of schools which the child and family considers appropriate
and feasible. The school at which a child finally enrols is likely to be shaped
by some extent by school-level constraints, such as whether the school still has
space available, and also probably by some degree of chance.
47
2.8 Conclusion
This chapter has provided an overview of school choice practices and policy,
both internationally and in South Africa, and has also provided a review of the
school choice literature relevant to educational equality and segregation. It has
identified a number of gaps in the scholarly literature, which this thesis aims to
fill. At the international level, these gaps include the general lack of empirical
data relating firstly to the determinants of school choice, and secondly to the
implications of school choice for educational segregation and equality, and the
absence of an appropriate conceptual framework for the empirical investigation
of these issues. At the South African level, gaps include the absence of
information about the dimensions of learner mobility, as well as a general
shortage of methodological tools with which to explore the issue.
The chapter has also highlighted the ways in which this thesis will make
original contributions towards filling these gaps, at methodological, theoretical
and empirical levels. Methodologically, contributions will include the use of a
longitudinal, population-based dataset, at a level that provides both some
generalizability, but also fairly detailed information at the level of the
individual, the use of data from a range of different sources to explore potential
determinants of mobility at a range of different levels, and finally the use of
three different operational definitions of learner mobility to allow the
exploration of a range of different dimensions of the phenomenon.
Theoretically, the contribution will include a preliminary theoretical model of
the determinants of school choice at the child, household, community and
school levels. In addition, it will provide data and insight regarding long-
standing debates around the relationship between school choice and
educational equality and segregation. Finally, empirically, it will provide the
first population-level data documenting the scope and dimensions of learner
mobility and school choice in post-Apartheid urban South Africa, along with
preliminary data on the determinants of these phenomena.
48
Chapter 3: Methods
3.1 Methodological approach: quantitative secondary
data analysis
This project makes use of quantitative, secondary analysis of pre-existing data
to explore questions related to the extent and nature of learner mobility in
contemporary urban South Africa. Secondary analysis of pre-existing data is a
well-accepted research method with a long history of use in educational
research as well as the study of mobility (McMillan and Schumacher 2005;
Smith 2006; Fleisch and Schindler 2008). As will be described later, taking a
quantitative approach to secondary analysis is particularly well suited to
answering the questions posed in this thesis.
Secondary data analysis is an approach to research that is based on the
analysis, or in some cases the reanalysis, of pre-existing data (Bryman 2004;
McMillan and Schumacher 2005). Typically, this data, which may be
quantitative or qualitative, and may consist of primary or secondary sources,
was originally collected for a particular purpose other than the research project
under consideration. Secondary data analysis allows this data to be reused, to
answer a different set of research questions. A major strength of this
methodology is therefore the ability to make use of pre-existing data,
eliminating the need for time-consuming and expensive data collection, and
allowing for more time and effort to be dedicated to analysis. Eliminating the
need to collect data also allows time and resources for the analysis of a greater
volume of data, possibly covering a longer period of time, and greatly
improving the breadth and reliability of work. Additionally, it enables research
projects to make use of data from multiple sources, increasing the depth of
findings, or to explore a particular historical era, generating period-specific
conclusions. The volume of data available for analysis is likely to be far
greater, and potentially of higher quality, than the data that could be collected
49
during the limited time, and with the limited resources, available for most
doctoral research. Quality of data is also less of a concern, as most large, pre-
existing datasets have already gone through multiple levels of quality
checking.
Quantitative analysis typically refers to the use of statistical approaches to
deriving meaning from numerical data. Strengths of quantitative research
include its potential for extracting meaningful and non-obvious information
from large pools of data, and the efficacy with which it can be used on large
sample sizes. Weaknesses include an inherent assumption that the principles of
the scientific method apply to human phenomena, the inability to incorporate
qualitative contextual information, and a deceptive sense of accuracy generated
by the availability of numerical results (Bryman 2004). Its strength in
aggregation, which makes it so valuable in providing an overall measure of a
phenomenon, does, however, often also result in a substantial loss of individual
detail.
Taking a quantitative approach to secondary data analysis provides a research
method well-suited to the major empirical questions the project answers,
particularly in the context of an extremely limited and almost entirely
qualitative pre-existing empirical literature. Measuring the scope of learner
mobility in contemporary urban South Africa – the first major empirical task
undertaken in this thesis – is essentially a quantitative question, and requires a
quantitative approach. While we already have some information about children
travelling to particular schools, and about learner mobility within particular,
fairly constrained, communities (Sekete, Shilubane et al. 2001; Fiske and Ladd
2004; Msila 2005; Msila 2009), we don‘t currently have a broader
understanding of the scale of this mobility. Qualitative approaches have proved
informative in exploring some reasons for learner mobility, as well as
documenting the behaviours of individuals, but they cannot give us an
overview of overall levels of learner mobility in a major urban area. For this, a
50
quantitative approach to the analysis of data drawn from a fairly large sample
is required.
Answering questions about the scale of mobility also requires the use of data
collected at a population level. To date, the large majority of research on
learner mobility has explored the question either by focusing on particular
schools, or by making use of a non-representative sample, typically drawn
from a fairly geographically constrained area (Sekete, Shilubane et al. 2001;
Fiske and Ladd 2004; Msila 2005; Msila 2009; Hunter 2010). While these
approaches provide valuable data, particularly with regards to the causes and
implications of the phenomenon, learner mobility appears to be highly
clustered around particular schools, and amongst particular groups of people.
This means that any sample that is not drawn to be relatively representative of
a fairly sizeable and varied population is unlikely to provide an accurate
measure of the overall scope of learner mobility. Unfortunately, collecting data
on a relatively representative sample of a substantial population, such as that
found in major urban hubs, is an extremely complex and time-consuming
process, particularly if data is wanted for more than one point in time. Drawing
on a dataset that has already been collected offers a way to gain access to a
volume of reasonably representative, high-quality data that could not be
otherwise be obtained in the context of a PhD project.
A second major empirical question posed by this thesis relates to the patterns
and correlates of learner mobility in contemporary urban South Africa,
particularly with respect to socio-economic status. In answering this question,
using pre-existing data is particularly valuable, as it allows for access to a
wider range of variables, often over a wider interval of time, than would be
feasible to collect for a single thesis. In particular, using secondary data
provides access to data from a range of different time points. It also allows the
researcher to tap into data from a range of different sources, and combine these
to enable the exploration of dimensions of the phenomenon that might
51
otherwise not be possible. Identifying patterns in learner mobility also requires
access to data for a large and reasonably representative sample of individuals.
This provides further support for the use of quantitative secondary data
analysis for this project.
3.2 Dataset selection
Making use of quantitative secondary data analysis requires access to an
appropriate data set. Although it would be ideal to draw data from a nationally
representative sample, the type of data needed for this study (specifically data
on the school at which children are enrolled) has unfortunately not been
collected in the national census, or other nationally representative household
surveys such as the October Household Survey, the Labour Force Survey or
the Community Survey. Fortunately, South Africa has a number of other
studies tracking fairly large numbers of people of varying ages, in different
parts of the country, for different lengths of time. Making use of one of these
studies was therefore the most feasible way of obtaining the necessary data.
Deciding which of these various datasets would be most appropriate to use in
answering the questions this thesis poses was an important step in developing
the project. Key considerations were the extent of data focused on school-aged
children, and in particular the availability of residential addresses and school
enrolment information, the extent to which this data was available
longitudinally, and the extent to which the children included in the study were
representative of a fairly well-defined and large population. After substantial
consideration, the Birth to Twenty (Bt20) study, based in the Soweto-
Johannesburg area of South Africa was selected. Appendix A contains details
of each of the other datasets considered for use, and documents the reasons that
these sources were decided against in the context of this particular project. In
summary, however, there were three main reasons for the selection of Bt20.
These are explored in more detail below, but relate firstly to the availability of
52
particular variables, secondly to the availability of data for particularly
important time points both in South African history and in the educational lives
of the sample members, and thirdly to the fact that the data was made
accessible to me.
3.2.1 Birth to Twenty
The Birth to Twenty (Bt20) cohort study started in 1989 with pilot studies to
test the feasibility of a long-term follow-up study of children‘s health and
wellbeing (Yach, Cameron et al. 1991). Women were enrolled in their second
and third trimester of pregnancy through public health facilities and
interviewed regarding their health and social history and current
circumstances. Singleton children (n=3 273) born between April and June
1990 and resident for at least 6 months in the municipal area of Soweto-
Johannesburg were enrolled into the birth cohort and have been followed up 16
times between birth and 20 years of age (Richter, Norris et al. 2004; Richter,
Norris et al. 2007). During the last 7 years, young people have been seen twice
a year, at the Bt20 offices and at home. Attrition over two decades has been
comparatively low (30%), mostly occurring during children‘s infancy and
early childhood, and approximately 2 300 children and their families currently
remain in contact with the study (Norris, Richter et al. 2007). The sample is
roughly representative of the demographic parameters of South Africa with
equal numbers of male and female participants. Assessments across multiple
domains have been made of children, families, households, schools and
communities during the course of the study, including growth, development,
psychological adjustment, physiological functioning, genetics, school
performance, and sexual and reproductive health. The third generation,
children of Bt20 children, began to be born in 2004. The Bt20 research
programme, including all data collection, has received clearance by the Ethics
Committee on Human Subjects at the University of the Witwatersrand
(M010556). The Federal-Wide Assurance registration number of the
Committee is FWA00000715.
53
Of particular importance to the selection of this dataset for use in this thesis,
data on both residential address and on school enrolment was available for
each Bt20 participant at multiple time points, providing a uniquely longitudinal
record of schooling and residence, and allowing for an exploration of variation
in mobility over time. The residential data was particularly promising, as
substantial work in cleaning the data had already been conducted for another
PhD project (Ginsburg, Norris et al. 2009; Ginsburg, Richter et al. 2010), and
GIS coordinates had been collected for most residential addresses at three
recent time points. As the Bt20 study was designed as a cohort of children born
in the Johannesburg-Soweto metropolitan area, and has achieved fairly limited
attrition, this also meant that its use would permit the development of a
roughly representative understanding of the extent and nature of learner
mobility for a substantial urban area. In addition, in contrast to the Cape Town
metropolitan area, the Johannesburg-Soweto area is substantially more similar
to the rest of South Africa, and particularly urban South Africa, in terms of the
performance of the schooling system, as well as the demographic makeup of
the population. Bt20 also had the advantage of providing data on the earlier
years of schooling, rather than only focusing on adolescence and later, where
much South African research on schooling behaviour tends to be focused. A
final important consideration was that Bt20 was able to make all residential
and schooling enrolment data available to me for the purposes of this study, on
the understanding that no information that could allow for the identification of
participants would be made public. Finally, the Principal Investigator of the
project, Prof. Linda Richter, and the Project Director, Prof. Shane Norris, both
expressed interest in this project, and indicated a willingness to provide
intellectual support for the work.
Concerns about making use of the data from Bt20 included the limited amount
of previous work making use of the more detailed schooling data collected.
While data regarding the school at which each participant was enrolled in each
54
year had been collected, this had not previously been used, and was therefore
likely to require substantial cleaning. While substantial work on the residential
addresses and residential mobility of the cohort had already been conducted
(Ginsburg, Norris et al. 2009), the GIS data had also not previously been
worked with. A second concern related to the cohort nature of the database,
particularly given the historically unique period of time, marking South
Africa‘s transition to democracy, during which the participants were born and
grew up. In a society as rapidly changing as South Africa during this period, it
is possible that conclusions derived on the basis of this particular cohort might
not apply to children born and attending school slightly later. A final concern
relates to the extent to which the study sample is indeed representative of the
youth population of the Johannesburg-Soweto metropolitan area. In particular,
as will be detailed in the next section, decisions around data collection,
combined with different response rates, has led to variable levels of enrolment,
and subsequently attrition, for individuals from different racial and socio-
economic backgrounds. Minority racial groups are therefore underrepresented
in the study, as are both the most advantaged and the most disadvantaged
individuals (Norris, Richter et al. 2007; Ginsburg, Norris et al. 2009; Richter,
Panday et al. 2009). However, the study does appear to remain representative
for the middle-income African population of the area, which is the group of
greatest interest for the questions asked in this thesis. These concerns
notwithstanding, Bt20 was the most feasible and suitable dataset for use in this
thesis, and was therefore selected.
3.3 Ethical considerations
Ethical clearance for this thesis was received from the University of the
Witwatersrand. The letter of approval is attached to this thesis as Appendix B.
As the project relies only on secondary analysis of existing data, there was no
data collection conducted for this project, and there were by extension no
ethical issues related to data collection or study instruments for this study. As
55
detailed previously, all data collection for the Bt20 study received clearance by
the Ethics Committee on Human Subjects at the University of the
Witwatersrand (original Birth to Ten protocol ethics clearance: 24/1/90;
extended Birth to Twenty protocol ethics clearance: M01-05-56). Bt20 data is
owned by Birth to Twenty, which is located in the Department of Paediatrics in
the Faculty of Health at the University of Witwatersrand. Access to the data
required for the purposes of this project was provided by Birth to Twenty.
All data was made available on the basis of unique identifying numbers
attached to each individual, with names and other identifying information
removed. The sole exclusion to the removal of identifying information was
with regards to residential addresses, which were essential to the project. All
residential address data was stored securely at all times, on a password
protected computer. Care was taken to ensure that exact residential addresses
of individuals were not disclosed in writing this thesis and related work, and
that addresses were only ever presented at the level of suburb. Similarly,
graphics are presented at a level of detail which ensures that an address cannot
be identified. In all other instances, only aggregated data is presented.
The Bt20 data was supplemented by data from two additional sources. Data
from the South Africa National Census 2001 was used to provide information
regarding the demographics and socio-economic conditions of the
communities in which Bt20 participants lived and schooled. Census 2001 data
was used aggregated at the small-area level4 and higher, and did not contain
any information which might identify individuals or households. Data provided
by the South African National Department of Education (DOE, renamed the
Department of Basic Education in 2009) was also used to supplement Bt20
data. This data consisted both of publicly available data, and data made
available specifically for this project. All DOE data was provided at school
4 Detailed explanations of the various geographic levels used, including the small area level,
are provided in the section detailing the Census 2001 data used, below.
56
level, and was linked to particular schools. Non-public data was made
available on the understanding that data would not be reported in such a way as
to link it to an identifiable school.
3.4 Overview of data and variables used
In this section, I provide information on the construction of each of the
variables used during my analysis, including details of the data sources used in
their generation, and any cleaning or modifications to the data that were
required. Descriptive statistics for each variable are presented in Chapter 4. As
discussed in Chapter 2, variables relating to potential child-level, household
and maternal-level and community-level determinants of mobility are included
in this thesis. Figure 3.1, below, shows the location of each of the variables
considered within the conceptual framework used in this study.
Figure 3.1: Location of study variables within the conceptual framework presented
in Chapter 2
Community
Home location
Community poverty index
Maternal
Maternal marital status
Maternal education
Child
Race
Gender
School attended
Age at first enrolment
Grade repetition
Phase of schooling 2003
Household & Family
Household SES 1990
Household SES 1997
Household SES 2003
Change in SES, 1997 to 2003
57
3.4.1 Child level variables
Race
All children were recorded as either white, black African, coloured, or Indian,
on the basis of information provided by caregivers shortly after birth. These
groupings are not meant to denote biological categories, but represent socially
meaningful categories in the South African context, as they are the groupings
on which Apartheid-era policy was based, and which continue to shape to a
significant extent the life experiences and opportunities open to South
Africans. This variable has been extensively used in existing analyses, and did
not require any cleaning or manipulation.
Gender
All children were recorded as either male or female on the basis of information
provided by caregivers after birth. As with race, this variable has been
extensively used, and required no cleaning.
School attended
The Bt20 education data required substantial manipulation prior to use. Firstly,
as data collection waves did not mirror the academic year, it was necessary to
restructure much of this data so that school name and grade could be attached
to a particular calendar year, rather than a data collection wave. Secondly, data
was available from two different types of source: prospective data, collected
during each wave, and retrospective data collected for all calendar years to date
during study Year 14. Thirdly, to maximise the value of the school name data,
it was necessary to match each school‘s name to the correct Education
Management Information System (EMIS) number, as it is through the EMIS
number that information about a school, such as it GIS coordinates, enrolment
and resource levels, can be obtained. The school level variables derived on the
basis of EMIS numbers are discussed later, in the section on school variables.
58
This section focuses on the education-related variables obtained from the Bt20
data.
The cleaning and reorganization of Bt20 education data followed a three-step
process. Firstly, all prospective schooling data was reorganized by calendar
year. Secondly, whenever the prospective record was incomplete, retrospective
data, where available, was used to fill in these gaps. Thirdly, using information
on the school name, location, and grades, schools were matched to the
appropriate EMIS number. Due to the variable spellings used for school
names, and the existence of several pairs of schools with the same or similar
names, this process had to be done manually. All cases in which there were
inconsistencies or a lack of clarity around which school the child attended were
checked against the original data collected. When it was not possible to
identify definitively the particular school attended, the school was coded as
missing.
School attended variables were generated for two points in time, 1997 and
2003. These two points were selected respectively as the earliest point at which
all children could be expected to be enrolled in primary school, and the end of
primary schooling. Data for 1997 was drawn from the Year 7 round of
interviews, combined with retrospective data from Year 14. Over all, for 1997,
schools were identified and matched to EMIS numbers for 1241 of the 1428
study sample5 members. Data for 2003 was drawn from the Year 13 and Year
14 interview rounds, and schools were identified and linked to EMIS numbers
for 1311 of the study sample members.
Age at first enrolment in school
Age at first enrolment was calculated using the child‘s grade in 1997,
controlling for repetition. If a child started formal schooling at the earliest
5 Details for the selection of the 1428 children included in the study sample are provided in a
subsequent section of this chapter.
59
possible point, they would have been in grade 1 in 1996, and therefore, barring
repetition or failure, would be in grade 2 in 1997. These children, as well as a
few who appeared to start particularly early and were already in grade 3 by
1997, were classed as early-starters. Children in grade 1 in 1997 (with the
exception of those who had repeated a grade), as well as a handful not yet
enrolled in school by 1997, were classed as late-starters. Note that most of the
children classified as late-starters by this variable do actually start school on
time with regards to official policy, which only requires children to start school
by the year in which they turn seven. However, in order to obtain any variation
on this variable, it was necessary to look at the timing of first enrollment
within the bounds specified by policy.
Phase of schooling in 2003
Phase of schooling is a binary variable indicating whether a child was still
attending a primary school in 2003, or whether he or she had progressed to
high school already.
Grade repetition
Grade repetition is a binary indicator, coded 1 if the child repeated any grades
between 1997 and 2003, and 0 if the child did not repeat any grades.
3.4.2 Household and maternal level variables
Maternal marital status
The maternal marital status variable is based on self-reported maternal data,
and refers to 1990, the year of the child‘s birth. Mothers selected from the
options married; unmarried but living together, single, and
divorced/separated/widowed. For the purpose of this analysis, in which marital
status was used as a proxy for household stability, all options other than
married were combined into a single unmarried category.
60
Maternal education
The maternal education variable is derived from each cohort member‘s
mother‘s self-reported highest completed level of education in 1990, at the
birth of the child. Mothers selected from the following options: no formal
education; up to and including grade 5; grade 6 or 7; grade 8, 9 or 10; grade 11
or 12; and post-school education. When mothers reported post-school
education, more detailed information as to the nature of this education was
collected. As very few mothers reported no formal education (n=13 in the
study sample), this category was merged with the group of mothers completing
up to grade 5. As education beyond Grade 5 is used as the cut-off point for
determining functional literacy, this group of mothers can be classified as
functionally illiterate. Due to the relatively low numbers of mothers who had
completed any specific type of post-school education, all forms of post-school
education were combined into a single category.
1990, 1997 and 2003 household SES (raw PCA score and quintile)
During each wave of data collection, a varying number of different indicators
related to socio-economic status were collected. Using each of these indicators
independently is not really feasible. Firstly, they tend to be very highly
correlated, and secondly, the value of ownership of a particular asset, or access
to a particular service, tends to change substantially over time. For this reason,
it is more appropriate to combine those indicators appropriate to a particular
point in time into a single SES index score, and use this in analyses. Grouping
sample members into quintiles on the basis of their scores at each time point
provides a straightforward means for comparison over time.
Due to a combination of changing societal context over time, and the use of
different indicators during each data collection wave, it is challenging to
construct an SES variable comparable across different study time-points. The
best option available within the Bt20 data was to construct an SES score
drawing on asset ownership and housing quality data for three different points
61
in time (Filmer and Pritchett 2001; Rutstein and Johnson 2004; Howe,
Hargreaves et al. 2008). Data from study years 0, 7, 12, and 13 were used to
construct scores for 1990 (around the time of the child‘s birth), 1997 (around
the time of enrolment in primary school), and 2003 (around the time of
completion of primary school). See Table 3.1 below for details of the variables
used in composing each SES index score.
Time point 1990 1997 2003
Variables used
in score
creation
All variables collected
during pregnancy or
the first two years of
the child’s life (1989-
1992):
Home type (1=house
or flat; 0=anything
else)
Home ownership
(1=owned;
0=anything else)
Water type (1=indoor
running water;
0=anything else)
Water use (1=sole
use; 0=shared)
Toilet type (1=indoor
flush; 0=anything
else)
Toilet use (1=sole
use; 0=shared)
Electricity in the
home (1=available;
0=not available)
TV ownership
(1=owned; 0=not
owned)
Car ownership
(1=owned; 0=not
owned)
Fridge ownership
(1=owned; 0=not
owned)
Washing machine
ownership (1=owned;
0=not owned)
All variables
collectd in study
year 7 (1997-1998):
Home type
(1=house or flat;
0=anything else)
Home ownership
(1=owned & fully
paid; 0=anything
else)
Water type
(1=running indoor;
0=other)
Toilet type
(1=indoor flush;
0=other)
Radio ownership
(1=owned; 0=not
owned)
Car ownership
(1=owned; 0=not
owned)
Washing machine
ownership
(1=owned; 0=not
owned)
VCR ownership
(1=owned; 0=not
owned)
Microwave
ownership
(1=owned; 0=not
owned)
Variables collected
in study years 12
(2002-2003) or 13
(2003-2004):
Home type (Year
13; 1=house or flat;
0=anything else)
Water type (Year
13; 1=hot and cold
indoor running
water; 0=anything
else)
Toilet type (Year
13; 1=indoor flush
toilet; 0=anything
else)
Electricity in the
home (Year 12;
1=yes; 0=no)
TV ownership
(Year 12; 1=yes;
0=no)
Radio ownership
(Year 12; 1=yes;
0=no)
Motor vehicle
ownership (Year
12; 1=yes; 0=no)
Fridge ownership
(Year 12; 1=yes;
0=no)
Washing machine
ownership (Year
12; 1=yes; 0=no)
Telephone
ownership (Year
62
12; 1=yes; 0=no)
VCR ownership
(Year 12; 1=yes;
0=no)
Microwave
ownership (Year
12; 1=yes; 0=no)
MNet ownership
(Year 12; 1=yes;
0=no)
Satellite TV
ownership (Year
12; 1=yes; 0=no)
Cellphone
ownership (Year
12; 1=yes; 0=no)
Table 3.1: Variables used in the creation of SES scores
After the relevant variables for each time point had been identified, all non-
binary variables were recoded to become binary variables (see Table 3.1 for
the final coding scheme used). A process of manual imputation was then
conducted for the 1990 and 2003 data, to reduce the number of missing values.
Imputation was done on the principle that if one variable predicted the value of
another correctly for 75% or more of the cases without missing data, it could
be used to impute the other variable where it was missing. As data for 1997
was either uniformly present, or uniformly missing for all variables, it was not
possible to impute any values in this year.
Once the amount of missing data had been minimized as far as possible,
principal components analysis (PCA) was run for each year, to determine the
appropriate weighting for each component variable (Vyas and Kumaranayake
2006). The literature is somewhat divided as to whether it is appropriate to use
PCA on binary data, as has been done here. While some studies have shown it
to perform as well as, or better than, alternatives (Filmer and Pritchett 2001;
Howe, Hargreaves et al. 2008), others have found that its performance is
suboptimal (Kolenikov and Angeles 2008). Nonetheless, it remains the
accepted standard approach used for working with binary SES data (Filmer and
63
Pritchett 2001; Rutstein and Johnson 2004; Vyas and Kumaranayake 2006),
and as such, has been used here. Depending on the nature of analysis for which
the SES data is used in this thesis, either the raw scores generated by the PCA
process are used, or the appropriate sample is divided into quintiles, which are
then used.
Change in household SES from 1997 to 2003
An additional variables, change in SES over time, was constructed for analyses
exploring changes in mobility behaviour over time. This was constructed by
taking the sample member‘s SES quintile in 2003, and subtracting the sample
member‘s SES quintile in 1997.
3.4.3 Community level variables
Home location
Address information was collected for all cohort members during each wave of
data collection, as this information was critical to maintaining contact. GIS
coordinates for home addresses, however, were not captured until Year 13 of
the study, when home visits were initiated. During Year 13, 15 and 16 home
visits, data collectors stood outside of the participant‘s home, and used a
mobile GPS device to record the coordinates at that location. These
coordinates were then either manually captured on the home visit instrument,
or downloaded into a study database at a later point. Unfortunately, while
working with this data, it became evident that between a third and a half of the
coordinates captured during each data collection wave were incorrect. These
problems were traced to data collectors not resetting the GPS device correctly
between uses, and therefore not always collecting the correct coordinates for
each location. Unfortunately this meant that the GIS coordinates available
could not be used without being checked manually for accuracy. Additionally,
due to limited available street-name and number data for the Soweto area, and
64
the vague nature of many addresses provided, it was not possible to generate
accurate coordinates for the majority of addresses using software or maps.
Due to time constraints, a decision was made to pursue the correct GIS
coordinates only for those children who had not moved home between 1996
and 2004, as this meant that only one set of coordinates would be required per
child. Details of sample generation, and implications of the decision to limit
the study sample on the basis of residential mobility are discussed in the
section on sample selection and bias below. Of the initial BTT cohort of 3273,
66% (n=2158) completed a residential history questionnaire in 2005 or 2006.
These individuals comprise the cumulative non-attrition cases. Data from the
questionnaire was used to generate a longitudinal dataset containing all address
and residential movement information for these sample members, from birth
through to 2004 (Ginsburg, Norris et al. 2009). This dataset was used to
distinguish between those individuals who had and had not changed residence
during the period 1996-2004. 1470 individuals reported stable residential
addresses during this period, and these formed the basis for the study sample.
Very few of these children had moved between 2004 and the end of Year 16
data collection, meaning that 3 different sets of GIS data, corresponding to
three different study waves, were available in most cases.
The residential GIS coordinates for these remaining children were manually
checked for accuracy using Google Earth, and fortunately, for the majority of
participants, at least one of the sets of coordinates collected did appear to be
correct, and could be used. For the remaining children with traceable
addresses, GIS data was generated using Google Earth where possible, or
otherwise re-collected through an additional visit to the address. Reliable GIS
coordinates for residential addresses could not be obtained for 27. These
instances of missing information were largely due to street name changes,
house numbering systems, and redevelopment of areas, and were distributed
throughout the greater Soweto area.
65
Once residential GIS coordinates had been obtained for sample members, a
number of additional community-level variables were generated for each set of
coordinates using data from Census 2001 – the most recent national census,
and that most relevant to the time period under consideration. This data was
used for two purposes. Firstly, Census 2001 geographical area boundary data
was used to delineate the boundaries of residential areas, suburbs and
municipalities. These boundaries were then used to identify the area in which
each child lived. This allowed for the later examination of whether children
were living in and schooling in the same areas, or not. Secondly, I also made
use of Census 2001 data to provide contextual information, in the form of a
poverty index, for the areas in which learners live, and the areas in which they
attend school.
Census geographical area delineation
Census geography was used to define area boundaries as census geography
generally – although not always – corresponds fairly closely to local
perceptions of area boundaries. They are sensitive to the ways in which socio-
economic factors and history, along with geographical features, have shaped
perceptions of areas. In addition, the Census 2001 provides 4 different levels of
geography, which allow for the exploration of mobility and context at a range
of different levels.
The smallest level used here is the Small Area Level (SAL). Each SAL
typically corresponds to between one and three enumerator areas, and contains
approximately 200 households. This level of geography is the lowest level at
which census data is released. Given the relatively small size of most SALs,
most ‗areas‘, as typically perceived, tend to contain several of them. The next
level of geography is Sub-Place name level (SP). This level typically
corresponds with residential suburbs, or small but distinct areas of a city.
Examples might be Pimville or Diepsloot. The third level, Main-Place name
66
(MP) corresponds roughly to small cities or towns, or large but distinct areas
within a large city, for example Soweto. Finally, the largest level of geography
used in these analyses is the Municipal level (MN), which represents entire
municipalities or districts. Each SAL is entirely contained within a particular
SP, each SP within a particular MP, and each MP within a particular MN.
The GIS coordinates home address for each sample member was linked to the
SAL, SP, MP and MN in which the address falls, using gvSIG software.
Similarly, the GIS coordinates for the school attended by each child, in both
1997 and 2003, were also linked to the SAL, SP, MP and MN within which
they fall. Once these linkages have been made, it is possible to test whether
children live and attend school in the same area or not. It is also possible to use
these linkages to obtain information about the area in which the home or
school is situated, as is discussed in the next section.
Community poverty rating
To explore the nature of the areas in which the BT20 families live, small area
level data from Census 2001 was used. For each area of geography (SAL, SP,
MP and MN), a PCA was conducted using a range of variables related to
affluence, and obtained from the Census 2001 SAL dataset. Variables used
were the percent of the working age population employed; average household
income; the percent of households living in informal dwellings; percent of
adults who had no secondary schooling; the percent of the area‘s population
who were black Africans; and the percent of households who did not have
access to services such as running water, electricity, hygienic toilets, refuse
removals and landline telephones. The results of the PCA were used to
generate a poverty level score for each area. This score has been used in its raw
form, and has also been used to divide sample members into quintiles on the
basis of the poverty of the area in which they live or attend school, where
appropriate.
67
3.4.4 School variables
In addition to the child, family and community-level variables above, which
are hypothesised to shape and constrain decision making around school
choices, attributes of potential and selected schools are also expected to play a
role in shaping decisions. In addition, data on the properties of the school a
child attends also provides valuable information on the quality of education
that child is likely to receive. As a result, the variables documented above are
supplemented with additional data on Gauteng province schools, obtained from
the South African National Department of Basic Education (formerly the
Department of Education). The bulk of the data comes from the Educational
Management Information System (EMIS) 2008, 2009 and master schools lists
for Gauteng, and the 2002 and 2003 Annual Schools Survey (ASS). This data
includes the GIS coordinates of schools, historical classification of schools,
school poverty quintile rating, enrolment data, and data on school fees.
Although there are some important concerns about the quality of data coming
from the Department of Education, particularly given that most data is self-
reported by schools, it is by far the best available source of school information.
This self-reported school-level data is supplemented with matric pass rates for
secondary schools in 2002, which were obtained from the Department of Basic
Education.
Although this thesis examines schooling patterns from 1997, the schools data
used is for 2002 or later. This is for a few different reasons. Firstly, data
available for the post-Apartheid period prior to 2002 is primarily through the
School Register of Needs (SRN), conducted in 1996 and 2000. Despite several
requests, I was unable to obtain a copy of the school-level data collected in the
SRN. Secondly, even had access been possible, issues of compatibility
between SRN and ASS data would have limited the utility of earlier data
(Yamauchi 2004). Finally, there is reasonable evidence to believe that the
variables of primary interest in this thesis are unlikely to have changed
substantially for individual schools over the period under consideration. These
68
reasons, along with the data source used, are discussed in the following
paragraphs. Data sources are also summarized in Table 3.2 below.
Variable Source Year
School location
(street address & GIS
coordinates)
Gauteng Schools Master List 2008 (The list includes
information on schools
operating in 2002, but which
subsequently closed)
School sector (Public
or private)
Gauteng Schools Master List 2008
Section 21 Status Gauteng Schools Master List 2008
Phase (Primary,
Secondary or
Combined)
Gauteng Schools Master List 2008
School resource
levels:
School fees charged
Annual Schools Survey 2002
School resource
levels:
Poverty Quintile
rating
Gauteng Schools List 2008
Historical
Department of
Education
Gauteng Schools List 2008
School enrolment Annual Schools Survey 2002
Racial composition of
the student body
Annual Schools Survey 2002
Matric pass rates by
school
Department of Basic Education 2002
Table 3.2: Sources for school-level variables used
School location, school sector, school phase, and Section 21 status
School location, school sector (whether the school is public or private), school
phase (primary, secondary or combined) and Section 21 status (whether or not
the school is allowed to manage its own finances) are all typically fairly stable,
and the data for these variables was obtained from the Gauteng schools master
list for 2008. Using this data source meant that the variables were available not
just for schools operating in 2008, but also for those schools which had closed
previously. Although it would have been ideal to obtain these variables, with
the exception of school location, from an earlier source, limited access to data
69
meant that this was not possible. Given the stability of school locations, the
2008 data was given preference over earlier sources due to the generally
increasing accuracy of GIS data over time.
School quintile rating
Data on school resource levels proved somewhat complicated to obtain. The
resource variable which is clearest and most widely used for South African
schools is probably the school quintile rating. The quintile system, which ranks
schools in categories from 1-5, depending on their resource levels, was first
instituted in 1998 by the Department of Education‘s Norms and Standards for
School Funding. At this point, schools were assigned to quintiles on a
provincial basis. This was amended in 2006, after which schools were assigned
to quintiles at the national level. This meant that many of the poorer school in
the wealthier provinces were moved to higher quintiles at this point (Pampallis
2008). In using school quintiles as a measure of resource levels in 2002, it was
necessary to decide between the old, provincial quintiles, and the newer,
national quintiles. Unfortunately it did not prove possible to obtain the quintile
ratings under the older provincial system, and so the ratings used here are those
for 2008, and therefore developed under the national system. Although this is
not ideal, it should be noted that despite changes in the quintile system, the
relative resource levels and of schools with respect to each other, and related to
this, their performance, has remained relatively constant (Fiske and Ladd 2004;
Fiske and Ladd 2005; Fleisch 2008). That is, those schools which were most
advantaged in 2002 are, more or less, the same schools that were most
advantaged in 2008. Therefore, although the actual quintiles assigned to
schools may have changed somewhat between 2002 and 2008, the rating of
schools relative to each other is unlikely to have changed substantially.
Using the 2008 poverty quintile ratings for the 1997 time point, when cohort
members were entering primary school is even more problematic than its use
for the 2003 time point. Between 1996 and 2000, substantial infrastructural
70
investments were made in historically under-resourced schools, particularly in
upgrading the basic infrastructure and facilities at schools, for example,
sanitation, telecommunications, water and power (Department of Education
2000). However, most of this investment was focussed on rural areas, and
therefore was of less relevance with regards to schools in the Gauteng
province. More importantly, the redistribution of resources occurred largely
between provinces. By contrast, within provinces, even if the overall levels of
resources changed, the relative proportions going to schools which historically
served different race group remained fairly similar (Fiske and Ladd 2005).
Historically advantaged schools continued to receive higher levels of state
funding than historically disadvantaged schools, and this discrepancy was
further exacerbated by the ability of parents at historically advantaged schools
to supplement state funding through higher school fees (Fiske and Ladd 2005).
Overall, then, those schools which were most advantaged in 2002 were also
those that were most advantaged 1997, while those with the fewest resources in
2002 were those which had always received the fewest resources.
School fees
Due to the age of the data, as well as various concerns that have been raised
about the quintile system more broadly (Kanjee and Chudgar 2009; Kanjee and
Chudgar 2009), this thesis also makes use of two additional measures of school
resources. Firstly, the school fees charged by schools in 2002, as reported by
schools in the ASS 2002, is used as an indicator of the school‘s access to
resources. It is widely accepted that schools with higher resource levels charge
higher fees (Fiske and Ladd 2004; Pampallis 2008). This data is also valuable
in that it dates to the time at which the children in this study were actually
attending the schools in question.
Historical department of education
Secondly, as much of the performance of South African schools today
continues to be explained by the educational department under which they fell
71
during the Apartheid era, a variable indicating whether the school was operated
by the Department of Education and Training (DET) during the Apartheid era
was used. The DET was responsible for the running of urban schools serving
black children, and DET schools received far fewer resources than all other
urban schools, and typically continue to be under-resourced and poorly
performing to the present time (Fiske and Ladd 2004; Fiske and Ladd 2005;
Fleisch 2008). This variable was obtained from the 2008 schools master list.
Not all currently disadvantaged schools operated under the DET during
Apartheid, and any schools which have been opened since the end of Apartheid
will obviously not fall under historical DET status. Nonetheless, given the
large number of education departments in place during the apartheid era, the
binary variable used here to indicate the historical status of the school is the
most feasible available option, and contains valuable information about the
history of a school. As it is a historical variable, the use of the 2008 data in its
composition is not problematic.
School enrollment
The next group of variables used related to school size and the composition of
the student body. For school size, the enrolment reported by the school in the
ASS 2002 was used. As this data is self-reported by schools, concerns have
been raised that the figures may be inflated, particularly for less well-managed
schools. However, this is the best available data, and as such, is used here.
With regards to figures around school size, Fiske and Ladd (2005) and The
Department of Education (2000) note some fluctuations in enrolment levels,
but these do not appear to be very substantial. Importantly, school choice had
already been possible for a number of years by 1997, suggesting that any initial
surge in changing school enrolments post-Apartheid had probably already
largely stabilised. Nonetheless, the data for 2002 was the earliest that could be
obtained, and is therefore used as a proxy for 1997.
72
Racial composition of student body
The variable used to describe the racial composition of the student body of the
school was also derived from ASS 2002 data. For this variable, the reported
number of black African students enrolled at a school was divided by the
schools total enrolment, to obtain the proportion of the student body who were
black African. Again, the use of 2002 data for the 1997 time point raises some
concerns, as discussed above.
Matric pass rate
The final variable examined here is the Matric pass rate – the proportion of a
school‘s students who write the national school-leaving examinations (the
Matric examinations) and pass – which is used here as a proxy for school
performance. This data was obtained for 2002 from the Gauteng Department of
Education, and is available for the majority of public secondary schools in
Gauteng. Private schools can choose between writing independent matric
exams, or writing those that public sector schools write. Results are only
available here for those schools which chose to write the public sector exams.
It should be noted that well performing private schools typically choose the
independent option, and so the results presented here for private schools are
likely to be extremely biased. Unfortunately, no measure of the academic
performance of primary schools is currently available for South Africa. In
order to generate a proxy of the likely performance of primary schools in
Gauteng, each primary school was matched to its nearest secondary school
using gvSIG software. The primary school was then provided with the matric
pass rate of its nearest secondary school6. Obviously this method is highly
imperfect7, but given the geographical clustering of school performance in
6 Thank you to my examiners for suggesting that a three-year average of matric pass rate be
used in future work, as this will help to counter concerns about the high annual variability in
performance of many historically disadvantaged secondary schools.
7 Particular concerns about this measure relate to the fact that within any given context, there
will always be a few schools which perform particularly well or particularly poorly. This
approach has no way of differentiating these schools from those performing at more expected
levels, which means that proxy pass rates generated for some schools will be inaccurate.
73
South Africa, it is at least plausible, and given the available data it is the only
real option.
School name and EMIS number
Once all variables to be used in the analyses had been sourced, the next
challenge was to create a list of the schools operating in the Gauteng province
during the 1997-2003 period, and combine the variables with these schools
appropriately. The best available base list of schools obtained was the 2008
Gauteng master schools list. This includes all registered schools in the
province, including those which had previously been registered, and had
subsequently closed. This list was therefore used as the base for the list of
schools used in analyses. All special schools, as well as non-school institutions
(FET colleges, pre-primary schools, exam centres, administrative offices, and
so on) were dropped from this list, leaving 2604 institutions, both open and
closed. All other variables were then merged with this list.
Aggregated school quality variables
Due to the strong correlations identified between many of the school related
variables, and particularly those relating to school quality, it was necessary to
combine these variables into an index for use in model generation. This was
done by the use of PCA on all school attribute variables that were consistently
found to be significantly related to mobility (school quintile, school fees,
school enrolment, percent black learners, school sector, historical DET status,
and pass rate). This process was repeated for the school attended by each child
in 1997 and 2003, as well as for the nearest grade-appropriate school to the
child‘s home in 1997 and 2003. In all cases, the eigenvalues of the first two
components of the PCA were both greater than 1, and were therefore both
retained.
74
3.5 Sample selection and creation of the analytical
database
As indicated previously, when conducting secondary data analysis, it is critical
to develop a clear understanding of the datasets being used, and the way that
the methodological decisions involved in creating the data may impact on data
analysis and findings. In this section, I begin by providing an overview of how
the study sample was composed, and detail the construction of the study‘s
analytical database.
A major concern with regards to the study sample is the extent to which it is
representative firstly of the Bt20 cohort in general, and secondly of the youth
population of urban Johannesburg-Soweto more broadly. These issues are
explored extensively in Chapter 4, which includes a review of the way in
which the Bt20 study was designed and conducted, and how this has
influenced sample composition. It also explores how the study population,
participant attrition, and composition of the sub-sample used in this thesis are
likely to have impacted on the outcomes of the analysis, and on the extent to
which findings are likely to be representative of the population of
contemporary Johannesburg-Soweto youth. In the current chapter, however,
the focus remains on the way in which the study sample and the analytical
databases were constructed.
3.5.1 Sample selection
As discussed previously, due to unanticipated challenges in preparing the GIS
data for use, the study sample was limited to the 1470 cohort members who did
not change address between 1996 and 2004. Subsequently, 28 additional
children who were either not attending school in 2003, were enrolled in a
special school at any time from 1996 to 2003, were enrolled in a school outside
of the Gauteng province, or were known to be boarding at a school within the
Gauteng province, were also excluded from the study sample. Children with
75
special educational needs were excluded as they and their families are unlikely
to experience the same degree, if any, of choice around where they will be
educated, due to their particular educational needs. Children attending
boarding schools or extremely distant schools were excluded, as they are not
travelling on a daily basis. An additional 14 cases of children who had changed
address between 1996 and 2004, or who were spending substantial time a
different address from their home, we also identified and removed from the
sample. This left a sample of 1428 individuals, which formed the sample on
which all subsequent analysis for this thesis was conducted. Members of all
race groups were retained in the study sample. However, due the very small
numbers of Indian and white participants, no results are presented for these
groups. See Figure 3.2, below, for a flow chart illustrating this process of
sample composition.
76
Figure 3.2: Flow chart illustrating selection of sub-sample for use in thesis.
3.5.2 Creation of the analytical dataset
Selection of time points
1997 was selected as the initial year within the analysis as it was the earliest
point for which reliable schooling information was available for the majority of
cohort members. It also reflected the first point in time by which all cohort
members, then aged 6-7, could safely be expected to be enrolled in formal
education. 2003 was selected as the point closest to the end of primary
schooling for the majority of the sample. In addition, it was selected as the year
Cases removed due to non-standard
schooling or residence (n=42)
Final study sub-
sample
(n=1428)
No residential
change 1996-2004
(n=1470)
Cases removed due to residential mobility
(n=730)
=1115
Non-attrition sample
2006, (full residential
information)
(n=2158)
Cases lost to attrition (n=1115)
Initial Bt20 Cohort
(n=3273)
77
for which the most reliable address data was available and during which GIS
coordinates were first collected, but this became irrelevant once the problems
with the GIS coordinates had been discovered.
Selection of these two time points, 1997 and 2003, allows this thesis to explore
both ends of the primary school experience, as well as to address questions
about both changes and consistency in schooling experience over time. One
drawback of the use of these time points, however, is that by 2003, roughly a
third of the sample had already completed primary school, and are enrolled in
secondary schooling. This progression to secondary schooling is strongly
related to socio-economic status, age at initial school enrolment, and academic
performance. As these variables are expected to relate to learner mobility, and
as mobility behaviour is also expected to differ between primary and secondary
school children, this makes it particularly challenging to draw definitive
conclusions about the cause of changing mobility at the end of primary
schooling. This concern is discussed where appropriate in the results chapters.
Details of the 1997 dataset
Of all the schooling data encountered during this project, that for 1997 was,
expectedly, the most problematic, for a few different reasons. The prospective
data was not as reliable as in subsequent years, as in 1997 the children were
still too young to report their own school names. School names were therefore
generally provided by caregivers, who were not always aware of which school
the child attended, or knew of the school only by an informal name. Secondly,
data capturing for 1997 schooling data was only done in 2009, and was done
using a drop down menu filled with school names as of 2009. This induced a
number of data capture errors, particularly when several schools had similar
names, as the data capturers would simply choose the first name that appeared
to match on the drop-down menu. It also introduced problems when schools
had closed or changed their names between 1997 and 2009, as the school name
would not appear on the drop down list. In these cases, data capturers were
78
also inclined to simply choose the most similar name on the drop down menu –
which would be incorrect. For this reason, all cases in which the school name
provided for the child was similar to other contemporary or historical primary
schools in the Gauteng area were double checked, and corrections made
whenever possible. However, it is quite probable that there do remain a small
number of incorrect school attributions in the data. The major reasons for
concern about the quality of the retrospective data relates to the substantial
length of time between 1997, and 2005-2006 when the retrospective schooling
data was collected, and the fact that if children had changed schools frequently,
they may have struggled to remember accurately which school they attended in
1997, when they were only 6 to 7 years old.
As a result, two different home and schooling datasets were created for 1997.
The first dataset was created by combining the 1997 address and GIS data with
the prospective schooling data for that year. While this schooling data was
highly accurate, it was only available for 760 of the sample individuals, just
over 50%. Given this extremely high level of missing information, a second
dataset was constructed using the less reliable, but more comprehensive
retrospective data to fill in as many gaps as possible. This second dataset
contained schooling information for 1244 sample members, with EMIS data
missing for only 184 individuals. Given the complementary nature of these two
datasets, with one being far more complete, but the data in the other being far
more reliable, initial analyses were conducted using both datasets. As both
datasets provided similar results, the more comprehensive one was used for the
majority of analysis presented in this thesis.
Details of the 2003 dataset
For the same reasons as described above, two home and schooling datasets
were also constructed for 2003. The dataset using only the more reliable
prospective data contained information for only 760 cases, again just over 50%
of the sample. When the retrospective data was integrated, schooling
79
information became available for 1310 sample members, with data missing for
only 118 individuals. Given that the retrospective schooling data was collected
in 2005-2006, the relatively lower number of cases with missing data is as
expected. The accuracy of the retrospective data is also expected to be greater
than for 1997, more closely resembling the prospective data collected in 2003.
Again, analyses using both sets of data provided similar results, and the more
comprehensive data including retrospective information is therefore used.
3.6 Operationalization of learner mobility
Given the poorly developed state of theory around the concept of learner
mobility, particularly within the South African context, as discussed in
Chapters 1 and 2, I make use of data from a range of different sources to
explore a number of alternative ways in which learner mobility might be
operationalized. I begin by using GIS coordinates to calculate the straight line
distance between each participant‘s home and school. I then explore definitions
shaped by movement between different areas as defined by census geography.
Finally I explore whether a child attends his or her nearest grade-appropriate
public school.
3.6.1 A distance-based definition of mobility
As discussed previously, a number of distance based definitions of what
constitutes learner migration or mobility have been advanced. Typically, these
work on the principle of assessing what constitutes excessive travel, on the
basis of assumptions around the age of the child, the safety of the area, the
availability of safe and affordable public transportation, and possibly other
context-dependent concerns. Internationally, the literature suggests a range of
maximum distances, ranging up to 10km. In the South African context, the
maximum distance that a child should need to travel has tended to be fixed
from between 2.5km up to this maximum of 10km. In current official policy, a
80
school‘s catchment area is defined as the area within a 3km radius of the
school, suggesting that this is felt to be, at the policy level, the maximum
distance a child should travel (Martin 2010).
In working with a distance as an indicator of mobility, it is possible to either
create a binary variable, using a particular distance as an indicator of whether
or not mobility is occurring, or to work with distance as a continuous measure
of the extent to which mobility is occurring. The binary, all-or-nothing
indicator approach is the one typically implied by existing distance-based
definitions of mobility, where mobility either occurs, or does not, on the basis
of a specified cut-off point, with no middle ground. While a binary approach to
measurement is therefore most useful from a policy assessment point of view,
mobility can also be understood as something which always occurs, but at
variable levels. As distance travelled has a close relationship to the resources
required to be dedicated to that travel, a continuous definition of mobility can
be seen as more closely related to reality as experienced by individuals
engaged in mobility. As these two different approaches to measurement are
complementary, this analysis makes use of both of them.
A second concern when working with distance-based definitions of mobility
relates to the way in which the distance from home to school should be
determined. Straight-line measurements are both methodologically most
simple, and are what tends to be used in mobility-related policy and
assessment. However, it does have shortcomings, the most obvious being that
children do not travel to school in a straight line, but make use of roads,
footpaths, and transportation networks. Furthermore, the relationship between
straight-line distance, and actual distance, is likely to be quite variable.
Calculating actual distance travelled for this sample would have required the
collection of substantial historical route data, and was not practicable. A more
practical alternative is using GIS software to calculate the shortest feasible rout
between two points, drawing on road network information. Unfortunately, road
81
network data for the Johannesburg-Soweto area was prohibitively expensive,
which meant that this alternative could not be pursued for this project. As road
network and similar data becomes increasingly available, however, this is an
avenue that could be valuably explored.
For the purposes of this project, however, all distances are straight line
distances. These are calculated using the Haversine formula applied to the GIS
coordinates of the child‘s home and the child‘s school (Sinnott 1984). The
Haversine formula is calculated as follows, where R is the radius of the earth in
kms:
dlon = lon2 - lon1
dlat = lat2 - lat1
a = sin^2(dlat/2) + cos(lat1) * cos(lat2) * sin^2(dlon/2)
c = 2 * arcsin(min(1,sqrt(a)))
d = R * c
3.6.2 An area based definition of mobility
The second approach to the operationalization of learner mobility draws on
tests of whether the learner attends school in the same area, using a range of
definitions, in which he or she lives. This is motivated by the concern that in
some cases, a child‘s ‗local‘ school may not be the closest school, but the
school that is located in the same community in which a child lives.
Additionally, in some cases, barriers such as rivers, hills, busy roads or train
tracks may mean that a child is cut off from the school that is closest to him or
her on the basis of straight-line distance. In these cases, it would also be more
natural for a child to attend a school that is slightly further away, but is located
in the same geographic community.
The various definitions of area which are used in this thesis are those
developed for Census 2001, and discussed previously. For each of these levels,
82
SAL, SP, MP and MN, a binary indicator was created for each of 1997 and
2003, coded one if the child lived and attended school in the same area, and
zero if the child did not live and attend school in the same area. If the child is
mobile at the smallest level of geography, the SAL level, all other tests will
also categorize a child as mobile. Similarly, if a child is mobile at the SP level,
the MP and MN tests will also categorize him or her as mobile.
3.6.3 Mobility defined by attendance at the nearest school
The final operationalization used for mobility is based on whether or not the
child is attending his or her nearest grade-appropriate school. The two
approaches to measuring mobility discussed so far tend to focus on those
learners who are travelling particularly substantial distances. While travelling a
substantial distance is an important indicator of the amount of effort and
money invested in school choice, and certainly identifies the forms of school
choice that are most salient in a gradually integrating post-apartheid South
Africa, it does not reveal much about individuals who may only be able to
participate in school choice at a relatively local level. One way to measure
mobility without losing these individuals is to determine whether a child is
attending the age-appropriate school nearest to their home, or whether they are
choosing to travel slightly further to attend a different school. This is a
particularly important aspect of mobility and school choice to explore in an
area such as Soweto, where the density of schools is extremely high, with most
children living within easy walking distance of more than one school.
Obviously, the figures obtained using this operationalization will be an
imprecise reflection of school choice. One particular concern is around those
children who are explicitly choosing to attend the school closest to their home,
and not simply attending it because it is closest. The extent of this phenomenon
is unfortunately not measurable with the existing data, and if it is substantial,
would mean that levels of school choice are actually higher than is reflected in
the available data. A second concern is around the possibility that there may be
83
children who would like to attend their nearest school, but are unable to do so,
perhaps because the closest school is over-enrolled, the child is discouraged by
higher school fees at this school, or the child is (illegally) refused admission
due to poor academic performance. In these cases, this definition would
suggest that the child is engaging in school choice and mobility, when in fact
the child is actively being prevented from exercising their choice. Thirdly, it is
possible that for reasons of geography, the closest school on the basis of a
distance calculation is not actually the closest school for a child on foot. In this
instance, it is possible that children who are simply attending the nearest
school on the basis of available roads are being misclassified as engaging in
school choice and mobility. Nonetheless, this nearest-school analysis should
provide a more accurate reflection of the extent of school choice when local
school choice is included, than any other definitions explored this far.
In order to measure the proportion of children attending their nearest age-
appropriate school, a spatial join was conducted in gvSIG to identify, for each
child, the relevant school nearest to their home. This was done twice, firstly for
all schools, both public and private, and secondly using only public schools.
For 1997, only primary, intermediate and combined schools were included in
the analysis. For 2003, both the nearest primary phase school and the nearest
secondary phase school was calculated. Then, on the basis of the child‘s grade,
the most appropriate of these two schools was selected. Once the nearest
grade-appropriate school for each child, at each time point, was obtained, this
school was then compared to the school actually attended by the child at that
time point. A binary indicator was created and coded one when the child did
attend his or her nearest grade-appropriate school, and zero when the child did
not attend that school.
84
3.7 Analysis
Data management was conducted using Microsoft Access. GIS analysis was
done with gvSIG, and statistical analysis with Stata Standard Edition 11. The
analytical procedures followed to generate each set of findings are described in
the relevant analytical chapters. In general, however, as most variables were
non-normally distributed, non-parametric analyses were used. In particular,
analyses involving only categorical variables were conducted with chi-square
tests, unless the numbers in any category dropped below 5, in which case a
Fisher exact test was used. Analyses involving both categorical and continuous
variables were conducted using Mann-Whitney Wilcoxon rank-sum or
Kruskal-Wallis tests, and analysis involving only continuous variables was
conducted using Spearman rank correlation. For multivariate analysis, a
multiple regression approach was used. Distributions are displayed as kernel
density plots, which show a smoothed curve based on the estimated non-
parametric probability densities of a variable‘s distribution (Scott 1992; van
der Berg, Wood et al. 2002).
Chapter 4, the first results chapter, presents a range of descriptive data,
covering sample representativeness and bias, and properties of schools
attended by sample members in relation to the universe of schools available to
them. Chapter 5 documents the overall extent of mobility using each of the
three different operationalizations detailed earlier in this chapter. In Chapter 6,
the relationship between mobility behaviour and characteristics of learners,
their families and households, and their communities is explored. Chapter 7
details the relationship between learner mobility and properties of the schools
that children attend, and the schools that are closest to their homes. Chapter 8
explores the extent to which mobility behaviours are subject to change over
time. Finally, Chapter 9 uses the findings of all previous chapters to generate
preliminary models of learner mobility. Chapter 10 summarizes and discusses
the findings, and serves as a conclusion to the thesis.
85
3.8 Conclusion
This chapter has outlined the methodological approach taken in this thesis, that
of quantitative secondary data analysis, as well as the rationale that lay behind
this choice. It has documented the range of datasets that were considered for
use in the study, and justified the selection of the Bt20 longitudinal cohort
study as the most appropriate. The composition of this cohort, particularly with
respect to demographic variables, SES, and residential mobility were
discussed. Existing data was used to demonstrate that while the overall Bt20
cohort does under-represent the most advantaged and disadvantaged children
in Soweto-Johannesburg, and particularly those in minority racial groups, it
remains highly representative of black low and middle-income township
residents – the group of primary interest to this thesis. The creation of the sub-
sample of the Bt20 cohort used in this thesis was explained, and the tests for
sample representativity and bias, which are will be presented in Chapter 4,
were outlined. Finally, the three different approaches to the operationalization
of learner mobility that are used in this thesis were presented, and approaches
to data analysis were described.
86
Chapter 4: Sample descriptive
statistics and representativity
4.1 Introduction
This chapter begins by providing descriptive statistics for the sample with
respect to the variables hypothesised to have a relationship to engagement in
learner mobility. It then moves on to ask whether the study sample is
representative firstly of the Bt20 cohort as a whole, and secondly of the youth
population of Johannesburg-Soweto metropolitan area more broadly. These
questions around representativity are answered both through a range of
statistical tests, and through a discussion of the process by which the Bt20
cohort and the study sample were created. Finally, to provide some context for
the discussions of school choice to follow in subsequent chapters, some basic
descriptive statistics for schools in the Gauteng province (within which the
study area falls) are presented.
4.2 Sample descriptive statistics
This section presents descriptive statistics for key child, household and
community variables for which there are theoretical grounds to anticipate a
relationship with educational mobility. At the child level, race, gender, age at
first school enrolment, school phase in 2003, and grade repetition between
1997 and 2003 are explored. At the household level, maternal education,
maternal marital status, and household SES in both 1997 and 2003 are
considered. Finally, at the community level, the poverty of the area in which
the child lives is documented, for three different levels of geography. The
relationship between these variables and mobility behaviour will be tested in
Chapter 6.
87
4.2.1 Child level variables
Race
As noted previously, the study sample is mostly black. While coloured children
are reasonably well represented, white and Indian children are under-
represented, and their numbers are also extremely small. The exact breakdown
of study sample members across race groups is presented in Table 4.1 below.
Race Black African White Coloured Indian
Number of
children
(n=1428)
1,145
(80.18%)
41
(2.87%)
192
(13.45%)
50
(3.50%)
Table 4.1: Breakdown of study sample members by race
Gender
The study sample is approximately evenly split between males and females
(see Table 4.2 below).
Gender Male Female
Number of children (n=1428) 711 (49.79%) 717 (50.21%)
Table 4.2: Breakdown of study sample members by gender
Age at first school enrolment
Overall, a very slight majority of sample members enrolled in school for the
first time either early or on time for their age, while a slight minority enrolled
late (see Table 4.3 below). The extent of late enrolment, at over 47%, is
striking. However, as noted in Chapter 3, although the late-starters being their
schooling a year later than their peers, the majority of them do not start their
schooling outside of the two-year window for enrolment specified by policy.
Age at first enrolment Earlier Later
Number of children (n=1275) 673 (52.78%) 602 (47.22%)
Table 4.3: Breakdown of study sample members by age at first school enrolment
88
Schooling phase in 2003
Table 4.4, below, illustrates that by 2003, just under one third of children had
progressed to high school, while just over two thirds remained in primary
school. A sample member who started their primary schooling on time, and
who had not repeated a grade, would be expected to have reached high school
by 2003, whereas those who started late, or who had repeated a grade, would
typically not be expected to have reached high school.
Schooling phase (03) Primary High
Number of children (n=1330) 897 (67.44%) 433 (32.56%)
Table 4.4: Breakdown of study sample members by phase of schooling in 2003
Grade repetition
As shown in Table 4.5 below, slightly more than one third of children repeated
a grade between 1997 and 2003, while the remainder did not. This figure is
similar, though slightly higher than that reported for other work on the Bt20
cohort (Fleisch and Schindler 2009).
Grade repetition between 1997 and
2003
No repetitions One or more repetitions
Number of children (n=1240) 778 (62.74%) 462 (37.26%)
Table 4.5: Breakdown of study sample members by whether or not they have
repeated at least one grade between 1997 and 2003
4.2.2 Household level variables
Maternal education
The distribution of maternal educational levels is shown in Table 4.6 below.
The largest proportion of mothers have completed some secondary school,
while relatively few are have grade 5 education or less, which is equivalent to
functional illiteracy. The proportion with post-school education is also low.
89
Maternal
Highest
Completed
Education
Level
Grade 5 or
below
Grade 6 or
7
Grade 8, 9
or 10
Grade 11 or
12
Post-school
education
Number of
children
(n=1305)
86
(6.59%)
94
(7.20%)
610
(46.74%)
399
(30.57%)
116
(8.89%)
Table 4.6: Breakdown of study sample by highest level of maternal education
attained at the time at which the study sample member was born
Maternal marital status
Slightly over one third of mothers were married at the time of the birth of study
sample member, while just less than two thirds were unmarried (see Table 4.7
below).
Maternal marital status in 1990 Married Unmarried
Number of children (n=1418) 506 (35.68%) 912 (64.32%)
Table 4.7: Breakdown of study sample members by maternal marital status
Household SES: 1997
The grouping of households into quintiles on the basis of SES in 1997 is
shown in Table 4.8 below. Due to several clusters of households with similar
scores, it was not possible to create completely even quintiles.
Household
SES quintile
1997
1
(poorest)
2 3 4 5
(wealthiest)
Number of
children
(n=1205)
254
(21.08%)
240
(19.92%)
233
(19.34%)
246
(20.41%)
232
(19.25%)
Table 4.8: Breakdown of study sample by household SES in 1997
90
Household SES: 2003
Similarly, the household SES quintiles for 2003 are also not completely even,
as evident in Table 4.9 below. Additionally, the small proportion of sample
member for whom SES data is available for 2003 should be noted.
Household
SES quintile
2003
1
(poorest)
2 3 4 5
(wealthiest)
Number of
children
(n=887)
181
(20.41%)
174
(19.62%)
179
(20.18%)
177
(19.95%)
176
(19.84%)
Table 4.9: Breakdown of study sample members by household SES in 2003
4.2.3 Community level variables
Small Area Level poverty
An index of community poverty was calculated for each level of census
geography, as described in Chapter 3. Households were broken down into
quintiles on the basis of poverty level of the SAL in which they lived. The
distribution of households across the quintiles of SAL poverty level is
illustrated in Table 4.10 below.
SAL poverty
quintile
1
(lowest
poverty)
2 3 4 5
(highest
poverty)
Number of
children
(n=1399)
280
(20.01%)
280
(20.01%)
280
(20.01%)
281
(20.09%)
278
(19.87%)
Table 4.10: Breakdown of study sample members by the poverty level of the SAL in
which they live
Sub Place poverty
Table 4.11, below, shows the distribution of households across quintiles based
on the poverty level of the SP in which they are located.
91
SP poverty
quintile
1
(lowest
poverty)
2 3 4 5
(highest
poverty)
Number of
children
(n=1399)
284
(20.30%)
287
(20.51%)
288
(20.59%)
273
(19.51%)
267
(19.09%)
Table 4.11: Breakdown of study sample members by the poverty level of the SP in
which they live
Main Place poverty
Due to the small number of MPs represented in the data, with most sample
members concentrated in just a few MPs, attempts to create poverty quintiles
based on this level of geography were unsuccessful. The creation of tertiles
was slightly more successful, and is illustrated in Table 4.12 below, although
substantial clustering is still evident.
MP poverty tertile 1 (lowest
poverty)
2 3 (highest poverty)
Number of children
(n=1399)
491
(35.07%)
776
(55.43%)
133
(9.50%)
Table 4.12: Breakdown of study sample members by the poverty level of the MP in
which they live
4.3 Relationships between variables
Tests were conducted to explore the relationships between each of the
variables discussed above within the study sample. All relationships operated
in the expected directions, and are documented in Appendix C.
4.4 Study sample representativity
Understanding the representativity of the study sample consists of two
different elements. Firstly the representativity of the full Bt20 cohort with
regards to the broader population of similarly aged children in the
92
Johannesburg-Soweto area needs to be understood. This requires a discussion
of the initial sample composition, as well as of subsequent sample attrition.
Secondly, the representativity of the study sample, with regards to the full,
non-attrition Bt20 cohort must be explored. This section begins by describing
Bt20 cohort composition and attrition, and implications for cohort
representativity. This is followed by series of tests to determine whether the
decision to limit the study sample to only those children who did not change
residential address between 1996 and 2004 introduced any additional
representativity concerns.
4.4.1 How representative is Birth to Twenty?
Cohort composition
The Bt20 study enrolled and collected longitudinal data on a birth cohort of
3273 singleton children, born to mothers resident in the Johannesburg-Soweto
area between April 28 and June 8, 1990. These children have been followed up
16 times to date, meaning that data is available for them at a range of points
throughout their schooling. Along with data regarding home situation,
caregivers, and a range of health and psychometric measures, Bt20 data
relevant to schooling includes each child‘s school‘s name, grade, repetition,
drop out, and academic performance as captured on standardized tests and by
the children‘s school reports. Although the quality and depth of the data
collected by Bt20 is high, and attrition is generally low, concerns about how
representative the Bt20 cohort is of Johannesburg-Soweto children and youth,
particularly over time, remain important. For this study, this is particularly
relevant when the cohort may differ from the broader population with respect
to variables such as SES that are expected to have a substantial influence on
schooling choices and learner mobility.
The Bt20 cohort was intended to consist of all the singleton children born in
the Johannesburg-Soweto area between 23 April and 8 June, 1990. For various
93
reasons, however, not all eligible children were actually enrolled. A subsequent
study undertaken to identify these ‗missing‘ children – children eligible to be
in the cohort but who had either never been identified by the study, or had been
identified but had not enrolled– revealed that non-enrolment was largely for
two reasons. Firstly, for practical and resource-related reasons, study
recruitment was concentrated in public sector health facilities. This meant that
members of the more advantaged groups, who typically used private facilities,
were less likely to come into contact with study recruiters, and were therefore
less likely to be enrolled (Richter, Norris et al. 2004). Secondly, a number of
individuals who had initially been identified by study recruiters either declined
to enrol, or agreed to study participation but could not subsequently be traced
for full enrolment. Reasons for non-enrolment included mobility, often
combined with incorrect or incomplete address data, incorrect or incomplete
recording of caregiver‘s names, particularly when multiple names were in use,
and participant concerns about invasion of privacy or about participation being
overly time-consuming. The majority of eligible children who did not enrol
again came from relatively affluent backgrounds compared to the rest of the
cohort, although some were also relatively disadvantaged (Richter, Norris et al.
2004; Richter, Panday et al. 2009).
As a result of these two sets of factors, the initial Bt20 cohort underrepresented
white and Indian children, along with more affluent children more generally,
but was largely representative of the predominantly black African population
of similarly aged children living in the area in 1990. This under-representation
of more advantaged children, particularly white and Indian, may result in an
under-representation of those children who live very close to high quality
schools, and therefore do not experience any pressure to travel to attend
schools further afield. It may also, however, result in an under-representation
of those children who are able to travel particularly great distances in order to
access the most desirable schools. It is therefore unclear how, in aggregate, the
under-representation of more affluent children is likely to influence the
94
outcomes of this study. Given, however, that the population of primary interest
to this study is the largely black African lower-middle-class population of
Johannesburg-Soweto, and that the cohort was representative in this regard,
these concerns are not of substantial significance. The relatively small size of
the affluent population of Johannesburg-Soweto, as well as of the white and
Indian populations, also means that any impact that their under-representation
is likely to have on study outcomes will be fairly minor.
A second concern relating to overall Bt20 cohort composition relates to the
fact that, with the passage of time, the population of the greater Johannesburg-
Soweto area has changed. This means that regardless of how representative the
cohort was when it was launched in 1990, over time it is likely to have become
less representative of the population of same-aged children actually living in
the area. This is, of course, a concern with any cohort study, but while it is not
specific to Bt20, Bt20 might be particularly seriously affected due the
historical era which the study covers. The post-Apartheid era, during which the
Bt20 participants grew up and attended school, has been characterized by
substantial changes in the residential patterns of people, including children and
youth. In part, this has been a response to the demise of Apartheid‘s strictly
enforced segregationist residential rules, in which black South Africans were
only permitted to live in urban areas if they were employed there. The
perception that people living in urban areas, and particularly in Gauteng, are
better off than those in rural areas, particularly with regards to access to
services and economic resources, combined with the repeal of Apartheid-era
segregationist policies, has triggered a substantial influx of new residents to the
Soweto-Johannesburg area during the past 20 years (Richter, Norris et al.
2006).
A sub-study of Bt20, the 2002 Children‘s School Survey, collected data on all
children born between April 23 and June 8 1990 enrolled at 81% of the
primary schools in the greater Johannesburg-Soweto area (Richter, Norris et al.
95
2006; Richter, Panday et al. 2009). With a 92% response rate, detailed
demographic data was collected on 5367 children. Almost half of these
children had not been born in the greater Johannesburg-Soweto area, indicating
a substantial level of in-migration amongst children of this age. Analysis of
this data revealed significant differences on a number of key indicators
between those children who had been born in the greater Johannesburg-Soweto
area and those who had moved there at a later point. Generally, in-migrating
children appeared to be living in more adverse circumstances and were at
greater risk of poverty than their peers who had been born in the area. They
were less likely to live in formal housing, and their parents were less likely to
own their dwellings. They had lower levels of ownership of most household
assets, and poorer access to basic household services such as running water,
electricity, sanitation, and refuse removal. Parents of in-migrating children
were more likely to be unemployed, and a higher proportion of those who were
employed were in unskilled employment. The in-migration of children
appeared to be closely connected to school attendance, with the majority of
children migrating into the area doing so before commencing their schooling.
In-migrating children were also more likely to have started their schooling late,
although no evidence of any impact of this delayed start on academic
performance in 2002 was found (Richter, Norris et al. 2006). While we don‘t
have access to information about out-migration, it is likely that the effects of
this on the population have been captured fairly well by attrition from the
cohort, which is discussed in the next section.
From the information collected from the Children‘s School Survey, however, it
is clear that the children migrating into the Johannesburg-Soweto area do tend
to differ in terms of their home environments from those children born in the
area, and represented in the Bt20 cohort. Specifically, cohort members are
likely to be more advantaged, and therefore may be more likely to have access
to the necessary resources to participate in school choice and learner mobility.
As such, we may again expect to see higher levels of learner mobility in the
96
cohort than would be found in the overall population of same-aged children in
Johannesburg-Soweto. Given, however, that we know very little about how
income shapes learner mobility, it is not possible to estimate the scale of this
effect. It appears, nonetheless, that our cohort remains largely representative of
the lower-middle-income black people who make up the bulk of the population
of Johannesburg-Soweto.
Finally, as is the case with any cohort study, it cannot be assumed that the
findings for children born in 1990 can necessarily be extrapolated to children
born at other points in time. For the same reasons as those outlined above,
relating to the political and social transformations that South Africa has
undergone during the lives of the cohort members, it is also likely that Bt20 is
particularly sensitive to these sorts of changes. Nonetheless, given the
strengths of the cohort, and the absence of alternative sources of data, it
remains the best source from which to derive findings relevant to primary
school children in contemporary South Africa. Additionally, it seems likely
that the period during which Bt20 cohort members were attending primary
school was the period during which new, post-Apartheid patterns of school
enrolment were defined and stabilized. As a result, the data presented
documents a critical period in the evolution of South Africa‘s post-Apartheid
schooling system, and can probably, with some caution, be extrapolated to
more recent points in time.
Bt20 cohort attrition
Changes in the composition of the cohort over time, predominantly due to the
non-random attrition of participants, will also affect the extent to which it is
representative of the population that it was designed to represent. Attrition in
Bt20, while low for a study of this length and magnitude, is known to be
related to certain variables such as race and socio-economic status (Richter,
Norris et al. 2004; Richter, Norris et al. 2007; Ginsburg, Norris et al. 2009). As
these variables are expected to be related to learner mobility and school choice
97
more broadly, it is important to explore the nature of this attrition, along with
the ways in which it might influence study findings.
As mentioned previously, the Bt20 study covers a particularly eventful era in
South African history, during which the Apartheid system of controls and
regulations governing where people could live, work and be educated was
dismantled, with huge implications for population distribution in South Africa.
Along with in-migration, as discussed above, the Gauteng province was also
affected by out-migration, particularly amongst women and young children,
driven by political tension, and overstretched public services. High levels of
circular mobility between urban and rural areas, along with mobility within
urban areas, became features of the Gauteng province. Migration out of the
study area, and as well as mobility within it, caused attrition of cohort
members from the study. Attrition has also been caused by maternal or child
death, child abandonment or adoption, and study fatigue (Richter, Norris et al.
2004).
Even in times and areas of relative stability, the maintenance of a longitudinal
sample is difficult, and the internationally accepted norm for sample attrition is
between 10 and 20% per annum. In the early post-Apartheid South African
context, the Bt20 study was greatly challenged to find ways to maintain the
birth cohort, and minimize attrition (Richter, Norris et al. 2004). Using a
combination of approaches at both the community and individual level, Bt20
succeeded in keeping sample attrition at an extremely low level, averaging
below 3% per year. At the community level, efforts included cooperation with
a Community Advisory Board, the use of local fieldworkers, and strict
adherence to ethical guidelines including confidentiality, to build a strong
relationship with participant communities, and by extension, trust. The
provision of some limited social and health services to cohort members also
encouraged them to maintain contact. At the individual level, participants were
98
contacted regularly by post, telephone and in person, and were followed up
extensively if contact was lost.
While attrition has been greatly limited by the efforts described above, it has
nonetheless been non-random in nature. Most notably, attrition has been
substantially higher among white participants, as well as participants with
higher socio-economic status, exacerbating the existing under-enrolment of
these groups (Richter, Norris et al. 2004). By contrast to more advantaged
groups, retention amongst more vulnerable members of the sample has been
extremely high, with black African mothers and their children being
particularly likely to remain in the study.
Ginsburg et al. (2009) present a number of mobility-related analyses
comparing all cases lost to the study prior to 2005, and those cases remaining
in the cohort. The attrition cases had experienced significantly higher
frequency of residential movement, with 81.3% of them having moved at least
once by 2005, and 13.3% moving within any of the documented intervals. By
contrast, amongst children remaining in the cohort, only 55.5% had
experienced any residential movement. The attrition group also contained
significantly more white participants, those born in private hospitals, and those
residing in the inner city or suburbs, as opposed to the townships. Mothers in
the attrition group were more likely to have been married at the time of child‘s
birth, and to have either no formal education, or to have attained some level of
post-school education. As suggested by the maternal education information,
attrition children were also more likely to live in particularly highly or poorly
resourced households. These finding echo expectations that the most
advantaged children, particularly white children and those living in the most
affluent areas are likely to be underrepresented, as are the most disadvantaged
children. As neither of these two groups are likely to be participating
substantially in learner mobility (the disadvantaged due to inability, the
advantaged due to living close to good schools), we may again expect that
99
figures on mobility are likely to be somewhat inflated for urban populations as
a whole. However, as attrition amongst the predominantly African, township
dwelling majority of the cohort has remained low, we can expect results fairly
representative of this particular population, which is fortunately the group of
most interest for this particular project.
4.4.2 How representative is the study sub-sample?
As documented in Chapter 3, unanticipated problems with the residential GIS
coordinate data, requiring a substantial and time-consuming cleaning process,
meant that it was infeasible to use the full, non-attrition Bt20 cohort for this
thesis, as had originally been planned. Of the initial Bt20 cohort of 3273, 66%
(n=2158) completed a residential history questionnaire in 2005 or 2006
(Ginsburg, Norris et al. 2009), and forms the cumulative non-attrition cohort as
of 2006. Of this group, 1470 individuals reported no changes in residential
address during this period. Once children attending special schools or boarding
schools, those attending schools outside of the Gauteng province, and those
who were not attending school at all were removed from this group, along with
a small number of children who were resident at multiple addresses during the
study period, this left 1428 individuals. The data on these 1428 individuals
forms the basis for all analysis presented in this thesis. While not ideal, the
decision to focus on this sub-sample of cohort members was made to maximize
the available sample size in light of problems with the residential GIS data
which required a lengthy cleaning procedure. However, as residential stability
between 1996 and 2004 appears likely to be closely related to SES and other
variables which may influence school choice, it is important to measure and
document the differences between this non-random sub-sample and the full,
non-attrition cohort, and to think about how this is likely to influence study
findings.
Given that residential mobility levels are greatest amongst the most advantaged
and most disadvantaged sectors of the population (Ginsburg, Norris et al.
100
2009), a sub-sample constructed primarily on the basis of mobility behaviours
is clearly unlikely to be representative of the full sample from which it is
drawn. Ginsburg et al. (2009) provide some valuable data on the differences
between Bt20 participants who had experienced some residential mobility prior
to 2005, and those who had not. While these figures are valuable as indicators,
it should be noted that the sub-sample used in this thesis excludes only those
children who experienced mobility between 1996 and 2004, and not those who
moved earlier or later. The highest levels of mobility, however, were seen in
the earliest years, with the commencement of primary schooling typically
having a stabilizing effect on children‘s residence. For this reason, the
differences between residentially mobile and non-mobile children presented in
Ginsburg et al (2009) are likely to be somewhat more substantial than the
differences between those children included and excluded from the sample
used in this thesis. Additionally, sample construction seems less likely to have
influenced representativeness for black African children with mid-range SES
levels, the population of primary interest in this study.
In order to better understand the nature of the sub-sample used in this thesis,
and in particular the ways in which included cases may differ from the
excluded, I compare both groups of cases. Firstly, I compare the members of
the study sub-sample (n=1428) to all cohort members not included in the sub-
sample (n=1845). This provides an indication of how different this sub-sample
is from the full Bt20 cohort, including all those individuals lost to contact, as
well as those excluded from this sub-sample for any other reasons. Secondly, I
focus on the group of cohort members not lost to attrition (n=2158), and
contrast those included in my sub-sample (n=1428) with those excluded from
my sub-sample (n=730). This provides an indication of how different my
sample is from those members of the full Bt20 non-attrition cohort who were
not included in this study.
101
Thesis sub-sample compared to all excluded cohort members
For SES, and each available demographic variable, the group of individuals
included in the study sub-sample was compared to the group of all cohort
members excluded from the sub-sample for any reason. Chi-squared tests were
conducted to determine whether the distribution of individuals was
significantly different across groups for each of the variables. As evident in
Table 4.13, below, the included and excluded cohort members differed to a
statistically significant degree on all tested variables, with the exception of
gender. Importantly, and as hypothesized, the different distributions support
the contention that the study sample appears to under-represent those at either
extreme of the socioeconomic scale. Under-representation of the historically
more advantaged race groups, those born in private hospitals, and outside of
the Soweto-DiepMeadow area all suggest that the most advantaged are likely
to be underrepresented, but can provide little information about the most
disadvantaged section of the cohort. However, an examination of the SES data,
as well as the data on maternal educational level suggests that the most
disadvantaged are also underrepresented. This is clearest with regards to the
SES variable, which shows that the study sample is biased towards the 3
middle quintiles, while the group of excluded individuals contains a higher
proportion of individuals falling into the first (poorest) and fifth (wealthiest)
quintiles. As discussed previously, the exclusion of the most advantaged and
disadvantaged, who for various reasons are hypothesized to be least likely to
engage in learner mobility, may lead to somewhat inflated findings regarding
levels of learner mobility. However, given that the study sample does appear to
be reasonably representative of the black, township-based middle-class, the
group which is of greatest interest in this examination of school choice
behaviours, this is not anticipated to be likely to be a major problem.
102
Variable Value Included # (%
of included)
Excluded # (%
of excluded)
Chi-squared
results
Child gender Male 711 (49.79%) 880 (47.70%) χ2(1) =1.412
not significant
n=3273
Female 717 (50.21%) 965 (52.30%)
Race Black 1145 (80.18%) 1423 (77.13%) χ2(3) =55.307
p<0.001
n=3273
Coloured 192 (13.45%) 191 (10.35%)
Indian 50 (3.50%) 65 (3.52%)
White 41 (2.87%) 166 (9.00%)
Maternal
age
18 or younger 162 (11.35%) 180 (9.76%) χ2(2) =20.733
p<0.001
n=3271
19-34 1084 (75.96%) 1511 (81.94%)
34 or older 181 (12.68%) 153 (8.30%)
Place of
birth
Soweto/DiepMeadow 1134 (79.41%) 1295 (70.19%) χ2(3) = 135.345
p<0.001
n=3273
Historically Indian/Coloured
area
221 (15.48%) 211 (11.44%)
Inner city JHB 5 (0.35%) 64 (3.47%)
Suburban JHB 68 (19.83%) 275 (14.91%)
Hospital of
birth
Public 1255 (87.89%) 1576 (85.47%) χ2(1) = 4.038
p<0.05
n=3272
Private 173 (12.11%) 268 (14.53%)
Maternal
marital
status
Married 506 (35.68%) 696 (37.97%) χ2(3) = 42.236
p<0.001
n=3251
Cohabiting 53 (3.74%) 160 (8.73%)
Separated/Divorced/Widow
ed
28 (1.97%) 19 (1.04%)
Single 831 (58.60%) 958 (52.26%)
Maternal
highest
educational
level
No formal education 13 (1.00%) 34 (2.09%) χ2(3) = 23.812
p<0.001
n=2932
Primary schooling 167 (12.80%) 241 (14.81%)
Secondary schooling 1009 (77.32%) 1140 (70.07%)
Post-school education 116 (8.89%) 212 (13.03%)
SES at birth Quintile 1 (most poor) 216 (18.56%) 393 (27.14%) χ2(4) = 36.803
p<0.001
n=2612
Quintile 2 215 (18.47%) 223 (15.40%)
Quintile 3 286 (24.57%) 284 (19.61%)
Quintile 4 238 (20.45%) 251 (17.33%)
Quintile 5 (least poor) 209 (17.96%) 297 (20.51%)
Table 4.13: Differences between cohort members included in the study sample,
and those excluded from the study sample with regards to all available
demographic variables collected at birth
Thesis sub-sample compared to other non-attrition cases: historical
data
This second analysis explores the extent to which the study sub-sample differs
from the group of cohort members from which it is drawn; that is, the full non-
attrition sample. This captures the way in which those cases excluded because
103
the children moved home between 1996 and 2004 differ from those who did
not move during this period. The same 1990 data was used for this set of
analyses as in the section above. In particular, the SES estimates and quintile
allocations for each case were not re-calculated, but were used as generated in
the previous set of analyses, on the basis of the data from the full cohort. The
results are presented in Table 4.14, below.
Variable Value Included # (%
of included)
Excluded # (%
of excluded)
Chi-squared
results
Child gender Male 711 (49.79%) 342 (46.85%) χ2(1) = 1.672
not significant
n=2158
Female 717 (50.21%) 388 (53.15%)
Race Black 1145 (80.18%) 601 (82.33%) χ2(3) = 3.442
not significant
n=2158
Coloured 192 (13.45%) 86 (11.78%)
Indian 50 (3.50%) 18 (2.47%)
White 41 (2.87%) 25 (3.42%)
Maternal
age
18 or younger 162 (11.35%) 92 (12.62%) χ2(2) = 19.940
p<0.001
n=2156
19-34 1084 (75.96%) 590 (80.93%)
34 or older 181 (12.68%) 47 (6.45%)
Place of
birth
Soweto/DiepMeadow 1134 (79.41%) 584 (80.00%) χ2(3) = 9.950
p<0.05
n=2158
Historically
Indian/Coloured area
221 (15.48%) 90 (12.33%)
Inner city JHB 5 (0.35%) 7 (0.96%)
Suburban JHB 68 (4.76%) 49 (6.71%)
Hospital of
birth
Public 1255 (87.89%) 642 (88.07%) χ2(1) = 0.015
not significant
n=2157
Private 173 (12.11%) 87 (11.93%)
Maternal
marital
status
Married 506 (35.68%) 247 (34.12%) χ2(3) = 3.144
not significant
n=2142
Cohabiting 53 (3.74%) 31 (4.28%)
Separated/Divorced/
Widowed
28 (1.97%) 8 (1.10%)
Single 831 (58.60%) 438 (60.50%)
Maternal
highest
educational
level
No formal education 13 (1.00%) 4 (0.60%) χ2(3) = 7.886
p<0.05
n=1971
Primary schooling 167 (12.80%) 64 (9.61%)
Secondary schooling 1009 (77.32%) 521 (78.23%)
Post-school education 116 (8.89%) 77 (11.56%)
SES Quintile 1 (most poor) 223 (19.16%) 135 (22.06%) χ2(4) = 6.8454
not significant
n=1776
Quintile 2 253 (21.75%) 104 (16.99%)
Quintile 3 241 (20.70%) 137 (22.39%)
Quintile 4 238 (20.45%) 122 (19.93%)
Quintile 5 (least poor) 209 (17.96%) 114 (18.63%)
Table 4.14: Differences between members of the non-attrition sample included in
and excluded from the study sub-sample, with respect to variables collected at
birth
104
The results of this second set of analyses suggest that the study sub-sample,
selected on the basis of not having changed residence between 1996 and 2004,
while differing from the full non-attrition sample in some regards (maternal
age, place of birth, and maternal education), is not significantly different in
others. Additionally, for those variables which are significantly different, the
levels of significance are lower, with only maternal age remaining significant
at p<0.001. These results suggest that children with mothers under the age of
35 were significantly more likely to be excluded from the study sub-sample
than those with older mothers. Children with mothers with particularly high
levels of education were also significantly more likely to be excluded from the
sub-sample, as were children born in the typically more affluent suburban
areas of Johannesburg. By contrast, children born in the historically Indian and
coloured areas were particularly likely to be included in the sub-sample.
While these figures do suggest that more advantaged children may be
somewhat under-represented in the study sub-sample, compared to in the non-
attrition sample, the lack of any statistical significance on this variable
suggests that any genuine differences are likely to be fairly minor. The absence
of any significant difference on race, hospital of birth, and maternal marital
status is also encouraging. It therefore seems reasonable to conclude that while
the composition of the thesis sub-sample selection may under-represent both
the most advantaged and disadvantaged children, this effect is not as
substantial as that anticipated on the basis of Ginsburg et al. (2009), and is
certainly less severe than that caused by sample attrition.
Thesis sub-sample compared to other non-attrition cases: 1997 and
2003 data
A second question with regards to study sample representativity is whether,
despite initial similarities, those included in and excluded from the study
sample have changed over time in systematically different ways. Ideally, one
105
would also want to ask this question with regards to emergent differences
between the study sample and the full cohort, but this is not feasible as, due to
attrition, data for later time points is not available for all cohort members. For
this reason, this second exploration of potential sample bias is restricted to
testing for differences between the study sample and those non-attrition sample
members excluded from it. SES data is available for both 1997 and 2003, and
this is used to test whether the study sub-sample differs from the non-attrition
sample in this regard at each of these time points.
Socio-economic status in 1997
Full SES data for 1997 was available for 1758 of the 2158 cases in the non-
attrition sample, and this is the data that was used for the following analyses. It
is worth noting that cases removed from the study sub-sample due to
residential mobility were substantially more likely to be missing SES data
(χ2(1)= 23.828; pr<0.001) than those that were retained in the study sub-sample
(see Table 4.15 below). While this makes sense, in that children who were
mobile were probably harder to locate during any particular round of data
collection, the implications of this difference in levels of missing data for the
validity of the following analysis are not clear.
Included:
n (% of included)
Excluded: n (% of excluded)
Yr7 SES data available 1205 (84.38%) 553 (75.75%)
Yr7 SES data missing 223 (15.62%) 177 (24.25%)
Table 4.15: Availability of 1997 SES data for members of the non-attrition sample
included in and excluded from the study sample
As described in Chapter 3, PCA was used to estimate SES scores for each
individual using asset ownership data (see Table 3.1). These scores were then
used to generate poverty quintiles. Distribution across the quintiles differed
significantly between those included in the study sample, and those excluded
from the study sample on the basis of residential mobility (see Table 4.16
106
below). Substantially more cases in the very lowest quintile were removed in
generating the study sample, while a relatively lower proportion of cases in the
other quintiles were removed. This suggests that the study sample includes a
somewhat higher proportion of children living in middle-class and affluent
families in 1997 than the full non-attrition sample. Reasons why this might be
the case are not obvious, but may relate to differences in the timing of mobility
for families with different levels of SES, as only children who moved during
their primary school years were excluded.
Variable Value Included # (%) Excluded # (%) Chi-squared
results
SES quintile 1 (poorest) 254 (21.08%) 160 (28.93%) χ2(4) = 14.587
p<0.01
n=1758
2 204 (16.93%) 87 (15.73%)
3 261 (21.66%) 97 (17.54%)
4 236 (19.59%) 108 (19.53%)
5 (least poor) 250 (20.75%) 101 (18.26%)
Table 4.16: Differences between members of the non-attrition sample included in
and excluded from the study sub-sample, with respect to SES in 1997
Socio-economic status in 2003
As described in Chapter 3, SES for 2003 was estimated using an assets index
collected during study year 12, and housing quality data collected during study
year 13 (see Table 3.1). Following manual imputation of missing values,
complete data was available for 1296 individuals, or approximately 60% of
non-attrition cases. PCA was used to estimate an SES variable for each sample
member. These were additionally used to categorize individuals into 5 poverty
quintiles. The extent to which data was missing for cases included in the study
sample, and for the non-attrition cases excluded, were compared (see Table
4.17 below). Once again, a substantially higher proportion of those excluded
from the study sample on the basis of residential mobility were missing SES
data (χ2(1) = 7.462; pr<0.01).
107
Included in study sample:
n (% of included)
Excluded from study sample: n
(% of excluded)
2003 SES data
available
887 (62.11%) 409 (56.03%)
2003 SES data
missing
541 (37.89%) 321 (43.97%)
Table 4.17: Availability of 2003 SES data for members of the non-attrition sample
included in and excluded from the study sample
A chi-squared analysis was conducted to test whether the distribution of cases
across the SES quintiles was different for those participants included in the
study sample, and those excluded (see Table 4.18 below). The test revealed no
significant differences between these distributions. On the basis of these 2003
SES scores then, sample selection does not appear to have created any
additional sample bias in favour of children with mid-range or high SES
scores. This combines with the analyses of SES scores at other points in time
to suggest that while sample selection on the basis of residential mobility may
reduce the representation of the most disadvantaged and the most advantaged
participants, this effect is relatively minor, and is weakest for the most
contemporary data.
Variable Value Included # (%) Excluded # (%) Chi-squared
results
SES quintile 1 (poorest) 181 (20.41%) 79 (19.32%) χ2(4) = 1.1916
not significant
n=1296
2 173 (19.50%) 86 (21.03%)
3 180 (20.29%) 80 (19.56%)
4 181 (20.41%) 78 (19.07%)
5 (least poor) 172 (19.39%) 86 (21.03%)
Table 4.18: Differences between members of the non-attrition sample included in
and excluded from the study sub-sample, with respect to SES in 2002/2003
4.4.3 Sample selection & bias: Conclusion
In summary, it seems likely that cohort enrolment, attrition over time, and the
non-random selection of the sub-sample used for this thesis are likely to have
each played a small role in contributing to a somewhat biased sample. In
108
particular, there seems to be some indication of an over-representation of black
children, and t hose from middle-income families. By contrast, there is some
under-representation of members of minority racial groups, as well as children
whose families are amongst the richest or poorest 20% of the population.
Overall, however, the extent of this under-representation does not appear to be
extreme, particularly when the length of time for which data is available is
considered. However, caution should of course be used in generalizing the
findings of this study to the broader population, and particularly to children at
extreme ends of the socio-economic continuum. In particular, it is possible that
the level of mobility found in this cohort may be slightly higher than that found
in the population more broadly, as it seems possible that children from
medium-income families may be the most likely to engage in learner mobility.
4.5 Descriptive schools data: all Gauteng schools
The final section in this chapter presents a range of descriptive statistics for the
schools found in the Gauteng province of South Africa. As detailed in Chapter
3, the data presented here comes from a variety of time points between 2002
and 2008, but covers all registered schools, public and independent, known to
have operated in the Gauteng province in the post-Apartheid era. This
information is presented to provide an overview of the nature of the
educational opportunities available to children growing up in the
Johannesburg-Soweto area, which is essential to understanding the findings
presented in subsequent chapters. Properties of schools with regards to each of
the school level variables considered are described, and bivariate relationships
with other school properties are described. Additional data relating solely to
that subsample of schools attended by study sample members is presented in
Chapter 7, as is information relating different school attributes to mobility
behaviours.
109
4.5.1 School types and sectors
The final schools dataset used for the analyses in this thesis contains data on
2604 schools, both public and independent, known to have operated in the
Gauteng province in the post-Apartheid era. Of these, 1570 (60.29%) are
primary schools, covering grades 1-7, and 289 (11.10%) are combined schools,
running all the way from grade 1 through to grade 12. There are 656 (40.90%)
secondary schools, covering grades 8-12, along with small numbers of
intermediate schools (n=73; 4.55%), and finishing schools (n=16; 0.61%).
Just below 20% of the schools in the dataset are independent schools. The
majority of the combined (n=229, 79.24%) and finishing schools (n=15;
93.75%) are independent. Amongst public schools, 1406 (67.27%) are primary
schools, 552 (26.41%) are secondary schools, and only 132 (6.32%) are other
school types. This fact, that Gauteng contains a substantially larger number of
public primary schools than public high schools will be revisited in subsequent
chapters, as it relates to changes in mobility behaviour as children move from
primary to high school. Of particular note is the fact that, on average, a child‘s
nearest secondary school will be somewhat further away from his or her home
than his or her nearest primary school. This means that, all else held constant, a
child should be expected to travel somewhat further to school on enrolling at a
high school. The smaller number of high schools also means that the range of
schools which children are choosing between is more limited, reducing the
extent of choice available to children.
4.5.2 School Quintile
The quintile rating system8 applies only to public schools. As discussed in
Chapter 3, it rates schools from 1 (being the poorest) to 5 (the most affluent),
primarily on the basis of the community within which the school is located
8 It should be noted that despite its name, the quintile rating system does not divide either
schools or learners evenly into five different groups. Very little information is publicly
available as to how exactly the poverty quintile ratings for schools were arrived at, or why the
quintiles are so variable in size. Available information is summarized in Chapters 2 and 3.
110
(Kanjee and Chudgar 2009). Within Gauteng province, schools are not very
evenly distributed between the different quintiles, and the majority of schools
are in quintiles 3 or above, which is in line with Gauteng being a primarily
urban, and comparatively affluent province. Secondary schools appear to be
somewhat more likely to be in higher quintiles than primary schools. The
distribution of Gauteng public schools across the poverty quintiles is shown in
Table 4.19 below.
Quintile Primary schools (% of
public primary
schools)
High school (% of
public high schools)
Total (% of all public
schools)
1 n=176 (13.02%) n=42 (8.11%) n=218 (11.66%)
2 n=118 (8.73%) n=43 (8.30%) n=161 ( 8.61%)
3 n=433 (32.03%) n=145 (27.99%) n=578 (30.91%)
4 n=355 (26.26%) n=155 (29.92%) n=510 (27.27)
5 n=270 (19.97%) n=133 (25.68%) n=403 (21.55%)
Table 4.19: Numbers of schools in each quintile in Gauteng province
4.5.3 Section 21 Status
Section 21 status is also only relevant to public schools, and is based on 2008
data. Any public school can apply to operate as a Section 21 school, which
places responsibility for the management of school finances at the school level,
substantially increasing autonomy. If schools are not Section 21, their finances
are operated by their provincial Department of Education, which is typically
much less efficient, and can be substantially more expensive. As a result,
schools typically pursue Section 21 status whenever they have any managerial
capacity, although in some cases applications for Section 21 status are turned
down. The large majority of the Gauteng schools on which data is available,
87%, have Section 21 status. High schools appear less likely to have Section
21 status (84%) than primary schools (90%), and a chi square test confirms
that this is a significant difference (χ2(1)=13.6702, Pr=0.000). Predictably, there
is a strong relationship between quintile rating and Section 21 status, with
111
more affluent schools being significantly more likely to manage their own
finances (χ2(4)=35.1538, Pr=0.000).
4.5.4 School enrolment
Due to the substantially larger number of public primary schools in Gauteng, it
seems likely that they would tend to be smaller than high schools, even though
primary schools cover 7 years of schooling, as opposed to 5 years covered by
high schools. The available schools data bears this out, with the mean
enrolment for primary schools in 2002 being 687, compared with 979 for
secondary schools (see Table 4.20 below). Combined schools had the smallest
mean size of any school type, probably because only a small proportion of
them were public. Independent schools were typically smaller than public
schools, with the mean size for a independent primary school at just 301
learners, while the mean for public primary schools was 711 learners.
Similarly, independent secondary schools were a mean size of 344, whereas
public secondary schools had a mean enrolment of 1067. The mean size overall
was 525 learners, rising to 776 for public combined schools, and falling to 448
for independent combined schools.
School Type Number in Gauteng Average number of learners
Combined 297 620
Intermediate 76 782
Primary 1807 744
Secondary 702 1003
Table 4.20: Average number of learners for different types of schools in Gauteng
Although school enrolment does vary significantly by quintile rating,
according to a Kruskal-Wallis test (Pr=0.00), the nature of this relationship is
not entirely clear. Quintile 2 schools are on average the largest, quintile 1
schools the smallest, and the average enrolments of schools in quintile 3, 4 and
5 are between these two extremes. This pattern holds when all schools are
examined and when primary schools only are examined, but becomes less
112
extreme when only secondary schools are examined. Overall, there is no
evidence that enrolment rises or falls strictly in line with quintile rating. An
explanation for the particularly small size of quintile 1 schools may be their
predominantly rural locations.
The relationship between school size and Section 21 is clearer, with schools
with Section 21 status tending to be substantially larger than those without.
This may again relate to many of those schools without Section 21 status being
located in rural areas, and therefore having lower enrolments. On the other
hand, it may also be the case that those few schools without Section 21 status
are a particularly poorly performing subset of schools, and therefore
particularly unattractive to learners, leading to lower enrolment.
4.5.5 Percentage of black learners
Although the mean proportion of black African learners in Gauteng schools
was just over 73%, much in line with the population of the province as a
whole, this figure obscures the actual distribution of black learners across the
province‘s schools. When the data is broken down, it becomes clear that over
50% of schools have an enrolment that is over 99% black, while a full 10% of
schools have fewer than 5% black learners. Schools that are meaningfully
integrated, and representative of the racial composition of the province‘s
population as a whole, are extremely rare. These figures are in line with those
presented by Sujee (2004). Figure 4.1 below illustrates the distribution of the
proportion of black learners across all Gauteng province schools.
113
Figure 4.1: Distribution of Gauteng schools by the proportion of their learners who
are black
Exploring the racial distribution of learners by schooling sector, public or
independent, reveals that overall, independent schools have a lower proportion
of black students. For public schools, the mean proportion of black students is
77%, while for independent schools it falls to 55%. In addition, a Wilcoxon
rank-sum test indicates that the distributions of the proportion of black learners
across schools is significantly different (P=0.000) across independent and
public schools. Public schools are substantially more likely than independent
schools to be 100% black, while independent schools are much more likely
than public schools to have very low proportions of black children. Figure 4.2
below illustrates the different ways in which black learners are distributed
across public and independent schools.
114
Figure 4.2: Distribution of Gauteng schools by the proportion of their learners who
are black
These figures highlight that there remains some accuracy to the perception of
many independent schools as ‗white‘ institutions, even as there is evidence of
the development of a substantial sub-group of independent schools which serve
an entirely black student body (Centre for Development and Enterprise 2010).
It seems likely that the disaggregation of independent schools into two groups,
one entirely black, and one almost entirely white, is likely to fall largely along
the same lines as the division of independent schools into two groups on the
basis of performance – one excellent, and the other extremely poor (although
the Centre for Development and Enterprise report referred to above does
contest this hypothesis).
Within the public sector, the distribution of black children across schools
differs between primary and secondary schools, with primary schools being
more likely to be entirely black than secondary schools, while secondary
schools are more likely to be almost entirely white than primary schools (see
Figure 4.3 below). A Wilcoxon rank-sum test finds that the distributions of the
115
proportion of black learners across primary and secondary schools are
statistically significantly different (P=0.0112), and an examination of the
differences in distributions reveals that while for primary schools the mean
proportion of black learners is 78%, this falls to 71% for secondary schools.
The differences in the racial composition of student bodies at primary and
secondary schools seem likely to be an artifact of the high number of relatively
small primary schools and the smaller number of relatively large secondary
schools found in historically black areas. Unfortunately it is not possible to test
this here. It is also important to point out that the data reflected here is only for
one year (2002), during a period of extremely rapid change in South African
society and schools, and that the distributions may well have since shifted.
Figure 4.3: Distribution of the proportion of black learners for public primary and
secondary schools in Gauteng
The relationships between the proportion of black children in a school, and that
school‘s quintile rating and Section 21 status are more straightforward than
those relationships discussed previously, and are highly statistically significant
(Kruskal-Wallis test, Pr=0.000). Almost all schools in quintiles 1-3 are entirely
116
black African. In quintile 4, the mean proportion of black children falls to
78%, and in quintile 5, the most advantaged schools, the mean proportion of
black children is only 31%. Equally predictably, amongst schools without
Section 21 status, almost all are entirely black. By contrast, amongst those with
Section 21 status, the distribution is very similar to the distribution across all
public schools as described previously.
The evidence around the relationship between proportion of black learners and
total school size is mixed. When all schools are considered, there is a very
weak positive correlation, which does not reach statistical significance.
However, when independent schools are removed from the sample and only
public schools are considered, a statistically significant – but not extremely
strong – negative correlation emerges. A scatter plot suggests that schools
which are predominantly black have a much wider range of sizes than schools
which are predominantly white, which tend to be middle-sized. Interestingly,
the negative correlation between school size and percent black learners is
particularly strong amongst public-sector primary schools – that is, public
primary schools with a higher proportion of black learners tend to be smaller
than those with a lower proportion of black learners. By contrast, among public
secondary schools, there is evidence of a (more weakly) statistically significant
positive correlation between school size and percent of learners that are black,
suggesting that secondary schools with a higher proportion of black learners
are larger than those with a smaller proportion. This tends to suggest that
historically disadvantaged primary schools are likely to be particularly small,
while historically disadvantaged secondary schools are likely to be particularly
large. This will have implications for the average distance from a child‘s home,
in a historically disadvantaged area, to his or her nearest schools – the distance
to the nearest primary school is likely to be substantially shorter than the
distance to the nearest secondary school. In addition, the child is likely to have
access to a larger number of local primary schools than secondary schools.
117
4.5.6 School fees
Of all the school data considered, the data on school fees is one of the most
problematic, and is probably the least reliable. That said, school fees serve as
an extremely useful measure of how accessible a school is, and are worth
exploring even in the context of imperfect data. Overall, the recorded school
fees for 2002 range from R0 to R9900. It seems likely that some schools,
especially in the independent sector, were charging higher fees, but for some
reason the captured figure was capped at R9900. In the full sample, the mean
school fee charged was R1117, however, this obscured an extremely skewed
distribution, as revealed by the median school fee of R120. When the schools
are separated on the basis of public and independent, it is evident that while a
small number of independent schools charged very low or no fees, the majority
charged substantial fees, with the median figure being R2500. By contrast,
when looking only at public schools, the median falls to R100, and the
maximum to R8600.
Within the public sector, school fees also vary substantially by phase, with
secondary schools tending to be substantially more expensive than primary
schools. For public primary schools, fees range from R0 to R6500, with a
mean of R683, and a median of R70. For public secondary schools, the range is
from R0 to R8600, with a mean of R1302, and a median of R200. Somewhat
predictably, school fees also vary significantly by school quintile rating, with
more affluent schools charging substantially higher fees, although there is
some anomalous data for quintile 1 schools, which probably reflects poor
reporting by those schools (Kruskal-Wallis test, Pr=0.0001). Along similar
lines, fees are significantly higher (Wilcoxon rank sum test) at those schools
with Section 21 status, as opposed to those without.
There is a statistically significant positive relationship between school fees,
and school size (Spearman correlation, Pr=0.0009). Much of this is probably
explained by the higher fees typically charged by secondary schools, which are
118
also larger than primary schools. When only public primary schools are
examined, however, the relationship between fees charged and enrolment
becomes even stronger. By contrast, within the group of public secondary
schools, fees tend to fall as enrolment increases. These divergent patterns are
consistent with previous data showing that smaller primary schools typically
have more black learners, and are therefore likely to be historically
disadvantaged and less affluent, while in the secondary phase it is the larger
schools that typically have more black learners.
Also in line with this data, direct analysis of the relationship between school
fees and the proportion of a school‘s learners that are black reveals an
extremely strong relationship (Spearman correlation, Pr=0.0000). As schools
become increasingly black, school fees fall substantially. This pattern is
consistent across all schools, public and independent, and within the groups of
primary and secondary schools as well. The relationship is stronger within the
public sector than the independent sector, however, suggesting that a
proportion of black learners attending independent schools may well be buying
out of the public sector. Of all the relationships described so far, this negative
relationship between school fees and black enrolment is by far the strongest. In
the South African context where economic disadvantage and race are so
strongly conflated, this is not in the least surprising.
4.5.7 Historical racial status of the school
Of the Gauteng schools for which Apartheid-era department data is available,
just under 60% fell under the DET. Although the majority of these schools
remain in the public sphere, there are a small proportion of them – about 10%
– that have subsequently become independent schools. When only public
schools are considered, a significantly higher (χ2(1)=4.1224, Pr<0.042)
proportion of primary schools are historically DET schools than the proportion
of secondary schools. However, when independent schools are included, this
distinction disappears. It is also noticeable that the majority of public schools
119
which are intermediate or combined are historically DET schools. Together,
this suggests that fewer DET schools have become public secondary schools
than would be expected, and that this may be because a number have moved
into the independent sector, while others remain categorized as combined or
intermediate schools.
Predictably, historical DET status is strongly linked to quintile (χ2(4)=961.5693,
Pr=0.000), with very few Quintile 4 or 5 schools having historically operated
under the DET. Surprising, at first glance, is the fact that by far the largest
proportion of historically DET schools are found in Quintile 3. However, given
that the DET was an urban department, and that schools in the more rural areas
of Gauteng were typically run by other departments, this does in fact make
sense. Clearly, historical DET status cannot be used purely as a proxy for low
resource levels at a school, as few of the quintile 1 and 2 schools were operated
by the DET. However, it can probably operate as a useful proxy for poor
schools located in urban contexts. Schools historically operated under the DET
were substantially less likely to have obtained Section 21 status by 2002,
compared to schools operated under other departments (χ2(1)=42.5644,
Pr=0.000).
Public schools historically operated by the DET typically had lower
enrolments in 2002 than other schools (Wilcoxon rank-sum test, Pr=0.0000).
When these schools are broken down into primary and secondary schools,
however, it become clear that historically DET operated high schools are larger
than other high schools (Wilcoxon rank-sum test, Pr=0.0000), while it is only
the primary schools that actually tend to be smaller (Wilcoxon rank-sum test,
Pr=0.0000).
Historical DET status also provided a strong predictor of the 2002 proportion
of a school‘s learners who were black (Wilcoxon rank-sum, Pr=0.0000). For
ex-DET schools, the mean proportion of black students was 99%, while for
120
others it was 39%. This was not substantially different when further
disaggregated by phase. In line with all data presented previously, historical
DET status was also a strong predictor of school fees, with ex-DET schools
charging substantially lower fees than others (Wilcoxon rank-sum, Pr=0.0000).
4.5.8 Matric pass rate
As detailed in Chapter 3, due to the absence of school performance data for
primary schools, the matric pass rate of the nearest secondary school was used
to approximate performance for primary schools. While highly suboptimal,
particularly given recognition that the matric pass rate is often a dubious
indicator of school performance in secondary schools, this was the best
available option, and as a result is presented here. Figure 4.4 below provides
the kernel density plots for the matric pass rates at secondary schools, and the
rates extrapolated onto primary schools as discussed in Chapter 3. Overall,
both distributions are fairly similar. The one aberration is the much higher
proportion of primary schools with pass rates around the 70% level. This is
probably related to the trend of a larger number of smaller primary schools in
historically black urban areas, where the typical pass rates of high schools are
around the 70% level. The similarity of the distributions of matric pass rates
applied to primary and secondary schools is also evident when details of the
two distributions are examined. Both have means in the low seventies,
although the median score in the secondary school distribution is substantially
higher (78%) than that for the primary school distribution (72%). Standard
deviations and scores at more extreme percentiles are also very similar.
121
Figure 4.4: Kernel density plots of pass rates at primary and secondary schools
For various reasons, when matric pass rate is explored by school status, public
or independent, the data is not very consistent. At the primary school level, the
distinction does not make sense, as the imputed performance bears no
relationship to whether the school is independent or public, but only to the
performance of the secondary school closest to it. At the secondary level, the
data is more meaningful, but, as mentioned in Chapter 3, performance data for
the more highly performing secondary schools is largely missing. As a result,
although performance varies significantly on the basis of whether a school is
public or independent (Wilcoxon rank sum test, Pr=0.0951), with independent
schools performing more poorly, this is due to a bias in the data and cannot be
taken at face value. Due to concerns about the validity of the data for
independent schools, the remaining analyses will be conducted on public
schools only, unless otherwise specified.
Examining the pass rates of public schools on the basis of their poverty quintile
rating reveals, unsurprisingly, a strongly significant difference (Kruskal-Wallis
122
test, Pr=0.0001). Schools in the lowest quintiles (that is, the poorest schools)
perform substantially more poorly than those in the higher quintiles,
particularly at the extremes. The mean pass rate for quintile 1 schools is 64%,
for quintile 2 schools it is 69%, for quintile 3 schools a slightly inconsistent
65%, for quintile 4 schools 72%, and for quintile 5 schools 92%. When
primary and secondary schools are examined in isolation, this pattern does not
change substantially.
A strongly significant relationship was also identified between section 21
status in public schools and matric pass rates, with section 21 schools
performing substantially better than those without section 21 status (Wilcoxon
rank-sum test, Pr=0.0001). The mean pass rate was 66% for schools without
section 21 status, and 74% for those with this status. The difference between
schools with and without section 21 status is greater at the secondary school
level.
There is also a statistically significant relationship between pass rate and the
size of a school, with larger schools tending to perform somewhat better
(Spearman correlation, Pr=0.0005). However, when schools are broken down
into primary and secondary, this relationship changes. At the primary school
level, the positive relationship between school size and performance persists,
whereas at the secondary level this relationship reverses, with smaller schools
out-performing larger schools. This almost certainly relates to the high
numbers of comparatively smaller primary schools found in less affluent urban
areas, and the tendency for the secondary schools in these same areas to be
larger than their counterparts in more advantaged areas.
The strongest and most statistically significant relationship identified with pass
rate is that with the proportion of a school‘s learners who are black (Spearman
correlation, Pr=0.0000). As the proportion of black learners in a school rises,
performance falls. The relationship is strongest for public sector secondary
123
schools, and somewhat weaker at the primary school level. The relationship
between school fees and pass rates is almost as strong, with performance rising
substantially in the schools charging the highest fees (Spearman correlation,
Pr=0.0000). Again this relationship is weaker – though still strong – at the
primary school level.
The final analysis undertaken explored pass rate by whether the school was
historically a DET school or not. Predictably, performance between DET and
non-DET schools was significantly different, with DET schools substantially
underperforming all other schools (Wilcoxon rank-sum test, Pr=0.0000). This
was consistent across public and independent schools, as well as primary and
secondary schools.
4.5.9 Descriptive schools data: discussion
In summary, this descriptive schools data suggests that schools in Gauteng are
strongly clustered, with a relatively small group of highly performing, well-
resourced and expensive schools, and a much larger group of less well-
performing, and more poorly resourced schools. Resource levels, school fees,
racial composition of the student body, and school-level academic performance
all remain closely related, even in the context of post-Apartheid South Africa.
Additionally, these analyses provide evidence that the distribution of primary
and high schools with respect to each other is somewhat different in different
areas of the Gauteng province. In particular, there are substantially higher
numbers of public primary schools than public high schools. In less
advantaged urban areas (typically township areas), primary schools tend to be
fairly small, and rather densely distributed. In these same areas, high schools
tend to be particularly large, and far more sparsely distributed. In more
advantaged areas, and amongst independent schools, primary and high schools
are far more similar in terms of size, as well as in the numbers of schools
available at each phase.
124
4.6 Conclusion
This chapter began by providing an overview of the study sample with regards
to the key explanatory variables examined in this thesis. It then moved on to
address issues related to sample representativity and bias. By comparing the
study sub-sample to both the full Bt20 cohort, as well as the 2005/6 non-
attrition cohort, it was possible to establish that while the study sub-sample
does under-represent children at the most advantaged and disadvantaged
extremes of the population, this was largely the result of sample attrition,
rather than the a result of the method of sub-sample construction. While
caution should therefore be taken in applying the findings of the study to the
full urban population of Johannesburg-Soweto, the sub-sample appeared to
remain largely representative of the black African majority with mid-range
levels of SES, who are the group of primary interest with regards to the
questions asked in this thesis.
Finally, the third component of the chapter presented descriptive statistics for
the schools in the Gauteng province. It provided evidence that schools across
the province are far from comparable, varying widely in terms of their student
bodies, their fees, their access to resources, and their performance. Historically
advantaged schools, typically located in historically advantaged areas, continue
to outperform historically disadvantaged schools. All findings were very much
in line with other analyses of the South African educational system (Fiske and
Ladd 2004; Sujee 2004). In addition, this section provided evidence that in
historically disadvantaged urban areas, a large number of relatively small
primary schools are found, along with a fairly small number of much larger
secondary schools. By comparison, in more advantaged areas, as well as in the
independent sector, primary and secondary schools are much closer in size, and
more evenly distributed. This has clear implications for the range of schools
between which children in different areas are choosing, as well as the distances
they are likely to need to travel from home to school.
125
Chapter 5: Measuring the extent
of learner mobility in
contemporary urban
Johannesburg-Soweto
5.1 Introduction
This chapter answers the first major empirical question asked in this thesis, by
presenting data on the extent of leaner mobility in contemporary
Johannesburg-Soweto. As discussed in previous chapters, due to the limited
levels of knowledge and theory about learner mobility, particularly in the
South African context but in the international literature as well, the best
approach to measuring learner mobility is not immediately clear. While the
majority of existing studies have looked at travel distance or time (Sekete,
Shilubane et al. 2001), there is reason to believe that other approaches to
measuring mobility, such as whether home and school are in the same area
(Msila 2005; Karlsson 2007; Hunter 2010), or whether children attend their
nearest school (Msila 2009), may also be important. Therefore, this chapter
uses the three different operationalizations of learner mobility discussed
previously to measure learner mobility amongst members of the study sample
in both 1997 and 2003. Firstly, straight line distance between home and school
is used as an indicator of distance. Secondly, whether or not the child lives and
attends school in the same ‗area‘ is used as an indicator of whether a child
attends a local school. Thirdly, whether or not a child attends his or her nearest
grade-appropriate school is used as an indicator of choice.
126
5.2 Distance-based operationalization of learner
mobility
In this section, two different approaches to the use of the straight line distance
between a child‘s home and school in measuring mobility are presented. The
first approach simply looks at the distance between home and school, while the
second is to use this distance to create binary indicators coded one if a child is
travelling further than a particular distance, and zero if he or she is not. This
allows us to answer two different question – firstly, how far children are
travelling, and secondly, how many children are actually mobile.
5.2.1 Actual straight-line distance from home to school
Comparing datasets
As discussed in Chapter 3, two different variables for the school attended were
created for each time point. The first variable for each time point was created
using purely prospective schooling data, collected at that particular point in
time. The second variable was based on the first, but used additional
retrospective data collected at a later point to fill in gaps. For both timepoints,
the prospective dataset therefore has a much higher number of missing cases,
but is likely to have greater accuracy than the retrospective data, which makes
use of recollection at a later point. For this reason, initial explorations of
distance travelled to school focused on comparing these two different datasets
for each point in time, to establish whether or not they provided satisfactorily
similar results. The figures obtained using the different datasets are presented
in Table 5.1 below.
Sample No. of
Observations
Mean
(km)
Standard
Deviation
(km)
Minimum
(km)
Maximum
(km)
1997, small
sample
746 5.623 10.798 .007 79.867
1997, small
sample
742 5.249 9.544 .007 57.766
127
constrained
at 60km
1997, large
sample
1221 5.901 10.955 .007 105.038
1997, large
sample
constrained
at 60km
1214 5.493 9.524 .007 57.766
2003, small
sample
745 5.479 8.725 .045 57.551
2003, large
sample
1285 5.625 9.768 .045 105.948
2003, large
sample
constrained
1281
5.355 8.462 .045 57.551
Table 5.1: Comparison of findings on distances travelled from home to school for
different datasets, for 1997 and 2003.
It is immediately apparent that although the minimum and mean distances from
home to school using the prospective and retrospective variables are extremely
similar, there is a substantial difference in the maximum distances, with much
higher maximums presented in the retrospective data. For both time points,
however, this is due to a small number of cases in the retrospective variable
that travelled further than the maximum distance reported in the prospective
dataset. Re-examination of the raw data in these cases did not provide any
definitive information as to whether these distances were correct, but in most
cases it seemed implausible that a child would travel that distance to attend the
school identified, suggesting that perhaps an incorrect school name had been
provided, or that the child was in fact not actually resident at the reported home
address.
Given that a few extreme outliers would bias the results of any further analysis,
it was decided to constrain the data by recoding as missing any reported
distances of over 60km between home and school. This removed 7 cases from
the 1997 data, and 4 cases from the 2003 data. Once these outliers had been
removed, the figures and distributions for the prospective and retrospective
128
data became substantially more similar, as is evident in Table 5.1 above, and
Figure 5.1 below.
Figure 5.1: Kernel density plots of the distribution of distances to school for the
small and large samples, curtailed at a maximum distance of 60km, for both 1997
and 2003.
Examination of the properties of both distributions for 1997, constrained at
60km, reveals that they are extremely similar. The same holds true for both
distributions for 2003. Conducting an unmatched t-test on the two different
samples for each time point also fails to reject the hypothesis that the means
are the same in both cases, further supporting the argument that the
distributions are indeed similar. Given the similarity in each year between the
small sample and the large sample when the most extreme cases have been
removed, all subsequent analysis makes use of these constrained larger
samples, unless otherwise specified.
1997 distance from home to school
The distance data for 1997 show a very high concentration of learners at the
lowest levels of mobility (see Figure 5.2 and Table 5.2 below). 25% of learners
are travelling less than half a kilometre to school, and almost half travel less
than a full kilometre. As the distance of travel increases above 1km, however,
the distribution begins to spread out substantially. The 75th percentile is
reached at just below 6km, and the remaining 25% of the sample forms a long,
2003 1997
129
thin tail reaching out to 60km. In practical terms, this means that although
almost half of learners attend a school that is extremely close to their home,
there are also almost 25% of learners who travel over 6km – a fairly substantial
distance for a 7 year old, and one that almost certainly indicates that these
children or their families are making use of school choice, and are investing
some financial resources into this choice, at the very least in terms of paying
for transportation. These children are also likely to be travelling to schools in
communities that differ substantially from those in which the live, particularly
with regards to community affluence, resource levels, and historical racial
designation.
Given the shape of the distribution of distances travelled, taking a log
transformation provides a useful way to compress the tail, and makes the
distribution somewhat more normal. Key data regarding the distributions of the
distance and transformed distance are provided in Table 5.2 below, and the
kernel density plots are provided in Figure 5.2. The log transformation is
particularly interesting in that it pulls together the cases spread out over the tail
of the untransformed distribution. The second peak, around 3, is the effect of
concentrating all these cases, and demonstrates that despite their low density,
they do actually form a significant proportion of the distribution when
considered together. Although the log transformation still fails standard tests
for normality (Shapiro-Wilk in Stata 11), it is far closer to a normal
distribution than the original data.
130
Figure 5.2: Kernel density plot overlaid on histogram illustrating the distribution of
distances travelled by sample members in 1997. The log transformation of the
distribution is also provided.
1997 25% 50% 75% 95% Mean
Distance
(km)
0.471km 1.032km 5.825km 24.686km 5.493km
Natural log
of distance
-0.753 0.031 1.762 3.206 0.454
Table 5.2: Distribution of distances and log distances travelled by sample members
2003 distance from home to school
The distribution of the 2003 data resembles the distribution of the 1997 data
very closely, even though the children are substantially older at this time point
(13 years, as opposed to 7 years), with a number already enrolled in high
school (see Figure 5.3 and Table 5.3 below). The distribution is however
slightly more compressed, indicated by the lower mean even as the values at
the percentile levels are generally slightly higher. This may relate to the better
quality of the data, which has cleared out spurious cases from the tail of the
distribution, or it may relate to actual differences in the behaviour of children
or the distribution of the relevant schools.
The 2003 data is analysed more closely, controlling for schooling phase
(primary or secondary) in Chapter 6, and a detailed comparison of the data for
1997 and 2003 is presented in Chapter 8. For the moment, however, the key
points to note are that around 50% of children are attending schools less than
131
1.25km away from their home, but also that over a quarter of 13 year old
children are travelling more than 6.5km to get to school on a daily basis. This
is very similar to the data for 1997, which is unanticipated as it was
hypothesized that mobility would increase substantially as children aged.
Instead, it suggests that a fairly similar (although still high) proportion of
children and their families are participating and investing resources in school
choice in 2003 as was the case in 1997.
Figure 5.3: Kernel density plot overlaid on histogram illustrating the distribution of
distances travelled by sample members in 2003. The log transformation of the
distribution is also provided
2003 25% 50% 75% 95% Mean
Distance
(km)
0.578km 1.243km 6.879km 23.362km 5.355km
Natural log
of distance
-0.549 0.218 1.928 3.151 0.584
Table 5.3: Distribution of distances and log distances travelled by sample members
Again, taking the log transformation of the distances travelled is helpful in
compressing the distribution, and revealing the extent to which cases are
concentrated in the tail end of the distribution. Once again, although it still fails
tests for normal distribution, the transformed distribution is closer to a normal
curve.
132
Actual straight line distance to school: Conclusion
Overall, calculation of the distances between children‘s homes and schools
reveals two key points. Firstly, the average (one-way) distance from home to
school, regardless of which year or sample size is examined, is somewhere
between 5 and 6km. Secondly, the distances are distributed as an
approximately normal curve centred somewhere between 2 and 3km, with an
extremely long and narrow tail to the right representing the roughly 25% of
children who appear to live particularly far away from their schools.
5.2.2 Binary definitions of mobility
While examining the actual distances between children‘s homes and their
schools provides more detailed information about how far children are
travelling, the use of binary definitions of learner mobility can facilitate the
development of policy around learner mobility and school catchment areas, as
well as the assessment of the implementation of existing policies. While some
information is lost in moving from a continuous measure to a binary definition,
analysis and interpretation are also simplified. As discussed in Chapter 3,
various cut-off points for the binary definition of learner mobility suggest
themselves on the basis of the existing literature and information on the topic,
and the analysis presented here makes use of a number of them. 3km is used as
this is the maximum distance a learner can travel and still be considered to
attend a local school in South African policy (Martin 2010). It is also probably
the maximum distance that a young child can be expected to walk to school.
5km and 10km cutoffs are also used, as they are frequently encountered in the
local and international literature (Sekete, Shilubane et al. 2001; South African
Human Rights Commission 2004). Working with these definitions, and
exploring cumulative density plots suggested that various other definitions,
particularly around 1, 1.5, 2 and 2.5km would also provide useful information.
In all instances, the variable is defined by coding all children travelling up to
and including the cut off distance as 0 (not mobile), and all those travelling
more than the cut off distance as 1 (mobile).
133
As the data for 1997 and 2003 were again extremely similar, both time points
are discussed together here, and data for both are presented in Table 5.4,
below, which provides the numbers and percentages of children who are
classified as mobile and non-mobile for each of the various binary definitions
of mobility considered. The first important outcome of exploring the various
binary definitions of mobility as proposed above is that there is remarkably
little difference between the proportions of children defined as mobile across
the different definitions, once a distance of 2.5km has been exceeded. After
this point, the most striking shift occurs in the interval from 3 to 5 km, where
the proportion of children classified as mobile falls from 33.53% to 27.59%
(1997) and 33.96% to 28.96% (2003) with an increase in travel distance of a
full 2km. While the absolute decreases in mobility from the 5km to the 10km
definition are greater, this is spread over an interval of 5km. Given the
steepness of the distribution curve prior to 2.5km, the relative flatness of the
curve at the 2.5km to 3km interval is somewhat surprising, and indicates that
this interval may have some significance.
Mobility definition 1997: Number (%) mobile 2003: Number mobile (%)
Travel more than 1 km 613 (50.49%) 727 (56.75%)
Travel more than 1.5 km 505 (41.60%) 574 (44.81%)
Travel more than 2km 451 (37.15%) 503 (39.27%)
Travel more than 2.5 km 418 (34.43%) 458 (35.75%)
Travel more than 3km 407 (33.53%) 435 (33.96%)
Travel more than 5km 335 (27.59%) 371 (28.96%)
Travel more than 10km 226 (18.62%) 239 (18.66%)
Table 5.4: Numbers and percentages of children classified as mobile in 1997 and
2003, for each binary definition of mobility considered.
Looking at a plot of the distribution of distances (see Figure 5.4 below)
substantiates this indication that something important is happening around the
2-3km interval. The initial, parabolic distribution ends here, and the long flat
tail of the distribution seems to begin. Similarly, the slope of the cumulative
density function shifts from steep to flat during this interval. That this shift in
134
distributions occurs in the interval between 2 and 3km is fairly compelling for
both empirical and theoretical reasons. Empirically, it corresponds roughly to
the distance that that a young, school-age child could reasonably be expected
to walk to school on a regular basis. Theoretically, the 3km endpoint of this
interval corresponds to the South African definition of a local school as being
within 3km of a child‘s home.
Figure 5.4: Cumulative density plot of distance between home and school, up to
10km, laid over a histogram illustrating the density distribution of distance
The second important outcome of these analyses is that they do indicate that
for any of the proposed definitions of mobility, particularly those that are
guided by the learner mobility literature, a substantial proportion of children
are actually travelling substantial distances on a daily basis. In particular,
roughly one third of children are travelling more than 3km. This is pretty clear
indication that they are not attending local schools, particularly in an area such
as Soweto where there is an extremely high density of public schools. It also
suggests – at least in an urban area – that at least a third of children are making
use of transportation, whether public or independent, to access schooling. This
entails a substantial additional level of family investment in the schooling of
these children.
2003 1997
135
5.3 Area-based operationalization of learner mobility
As described in Chapter 3, the second approach to the operationalization of
learner mobility draws on various levels of geographic areas as defined by
Census 2001. The smallest area is the Small Area Level (SAL), followed by
the Sub-Place (SP), the Main Place (MP), and finally the largest area, the
Municipality (MN). Preliminary investigations of the mobility at the SAL level
revealed that, due to the small size of SALs, very few children (less than 7%)
attended school in the same SAL in which they lived. This makes sense, as
very few urban schools are small enough to serve a community of only 200
households, suggesting that SALs are likely to be sharing schools, and as a
result, data for mobility at the SAL level is not presented here. Once again, due
to the strong similarity between the data for 1997 and 2003, findings for both
time points are presented together.
The numbers and proportions of children who are mobile for each of SP, MP
and MN are presented in Table 5.5 below, for both 1997 and 2003. The SP
level of analysis shows that just over 40% of children attended school in the
same SP as they lived in in 1997, and just below 37% in 2003. Given that SP
geography is roughly equivalent to residential suburbs, this suggests that
around 40% of children are attending a local school within their suburb, while
the other 60% are travelling to schools outside of their suburb. At the MP
level, which corresponds to major areas of the city (for example Soweto,
Meadowlands, Johannesburg, and so on), the proportion of children attending
school within the MP where they live rises to over 70% for both 1997 and
2003. Interestingly, the proportion of children travelling across MP boundaries
is very similar to the proportion travelling over 5km. Finally, at the MN level,
very few children are travelling to a different MN area for school. Given the
size of MN areas, and the fact that each MN includes a wide range of schools
in terms of performance and cost, this is unsurprising.
136
1997: number (%) not mobile (i.e.
school and home in same area)
2003: number (%) not mobile (i.e. school
and home in same area)
SP 494 (40.63%) 473 (36.90%)
MP 882 (72.53%) 901 (70.28%)
MN 1157 (95.15%) 1,236 (96.41%)
Table 5.5: Number and percent of children who live and attend school in the same
SP, MP and MN areas, in 1997 and 2003
The figure for mobility at the MP level is particularly significant, because the
boundaries at the MP level of geography correspond most closely to the
historical boundaries between areas designated for different race groups. This
is critical, because the historical racial designation of a school remains one of
the strongest predictors of school performance in contemporary South Africa,
and is likely to be one of the major determinants of school choice.
Additionally, historical racial group is also a strong predictor of the cost of
attending a school. For this reason, those children crossing MP boundaries can
be roughly equated to the group that are choosing to attend schools that were
historically restricted to white, Indian or coloured children. Those children
crossing SP, but not MP, boundaries, can by contrast be roughly equated to
those children exercising some degree of school choice but without travelling
to areas that were historically designated for other racial groups. This group of
just under 30% of children should be roughly equivalent to those who are
exercising school choice within historically disadvantaged areas, without
leaving those areas.
Determining the correlations between the different possible measures of
mobility reveals that there is a very strong overlap between the MP definition,
and the distance based definition using travel greater than 5km (see Tables 5.6
and 5.7 below). This substantiates the notion that both of these measures are
identifying roughly the same group of children, those travelling fairly
substantial distances to attend historically more advantaged schools, and that
these measures are therefore likely to be of particular significance.
137
1997 Travel >2.5km Travel >3km Travel >5km Travel>10km
SAL 0.1963 0.1924 0.1672 0.1296
SP 0.5897 0.5847 0.5114 0.3962
MP 0.7650 0.7700 0.8036 0.7131
MN 0.3063 0.3043 0.3334 0.4241
Table 5.6: Correlation coefficients between distance-based and area-based
measures of mobility for 1997
2003 Travel >2.5km Travel >3km Travel >5km Travel>10km
SAL 0.1503 0.1445 0.1287 0.0965
SP 0.5674 0.5486 0.4885 0.3664
MP 0.7850 0.8010 0.8212 0.6980
MN 0.2558 0.2482 0.2614 0.3440
Table 5.7: Correlation coefficients between distance-based and area-based
measures of mobility for 2003
5.4 Nearest school based operationalization of learner
mobility
The final approach to measuring learner mobility involves determining
whether or not children are enrolled at their nearest grade-appropriate school.
As discussed in Chapter 3, while this is not a perfect indicator of engagement
in school choice, the proportion of children not attending their nearest school is
expected to provide a fair approximation of the proportion of children
engaging in choice. Again, due to substantial similarities over time, the data for
1997 and 2003 is presented together.
The first key finding using this approach to measuring mobility is that less than
20% of children are actually attending the grade-appropriate school nearest to
their homes in both 1997 and 2003 (see Table 5.8 below). This figure is
surprisingly low, and suggests that over 80% of children are travelling further
than strictly necessary in order to attend school. One possible reason that
children might not be attending their nearest grade-appropriate school could be
that the school in question is an independent (private) school. Due to this, two
138
sets of figures are presented, one including only public schools, and one
including independent schools as well. As is clear from the data in Table 5.8,
this makes very little difference to the results.
However, when the 2003 data is disaggregated by schooling phase – that is,
when children enrolled in primary school are separated from those enrolled in
secondary school – an interesting pattern is revealed. Despite hypotheses that
mobility should be higher amongst high school children, a substantially higher
proportion of these children are attending their nearest school (just under
22%). The overall proportion of children attending the nearest school remains
the same because the proportion of primary school children attending the
nearest primary school actually falls fairly markedly to just over 15%. While
the higher proportion of high school children attending the closest school may
just be due to a smaller number of available high schools (see Chapter 4), the
lower proportion of primary school children attending their closest school at
age 13 is more intriguing. One potential explanation is that children who are
attending schools further afield perform more poorly, making them more likely
to still be in primary school at age 13. An alternative, and somewhat more
plausible explanation may be that when children fail a grade, their parents are
more likely to try sending them to different schools, which may be further
from their homes. These hypotheses, and others, will be explored in the
subsequent chapters.
Number (%) of learners
attending the school
closest to their home
Mean distance
to nearest
school
Maximum
distance to
nearest school
1997 public schools
only
219 (17.92%) 0.417km 3.142km
1997 public and
independent
schools
217 (17.76%) 0.398 km 2.767km
2003, public
schools only
235 (18.58%) 0.489km 4.230km
2003, public
schools only;
141 (16.49%) 0.428km 3.142km
139
primary school
learners only
2003, public
schools only; high
school learners
only
94 (22.33%) 0.616km 4.230km
2003, public and
independent
schools
222 (17.40%) 0.466km 4.230km
2003, public and
independent
schools; primary
school learners
only
131 (15.32%) 0.410km 2.767km
2003, public and
independent
schools; high
school learners
only
91 (21.62%) 0.583km 4.230km
Table 5.8: Number and percentage of learners attending the school closest to their
home in 1997 and 2003, and the mean and maximum distances to the schools
nearest to sample members’ homes
The data on the distance from children‘s homes to their nearest schools
provides an additional interesting finding: the mean distance a child needs to
travel to attend their nearest primary phase school is just approximately 400m,
and less than 5% of children need to travel over 1km. When contrasted to the
actual distances children are travelling – previous calculations indicated over
50% of children travelling over 1km – this highlights the extent to which
travel, even of moderate levels, appears to be due to children attending schools
further from home than is strictly necessary.
In 2003, however, not all children are still within easy walking distance of a
grade-appropriate school. Although 95% of children in 2003 have to travel less
than 1.15km to reach their nearest school, there are a small number of children
who have to travel over 3km. This is probably primarily due to the fact,
discussed in Chapter 4, that there are substantially fewer high schools in the
Johannesburg-Soweto area, due to their typically having a somewhat larger
140
size. Disaggregating the children by schooling phase supports this hypothesis,
as the data generated for the primary school children remains similar to that
generated in 1997. Nonetheless, even when children are in high school, the
average distance travelled remains substantially greater than the distance a
child would need to travel to access his or her nearest school.
5.5 Conclusion:
This chapter has explored three different approaches to defining and measuring
learner mobility, and provided data about the extent of learner mobility in
Johannesburg-Soweto on the basis of each of these definitions. Each definition
is likely to prove particularly valuable for certain purposes, and in certain
contexts. Using a distance-based measure provides both a binary and a
continuous measure of mobility. The distance-based binary measure is of the
form that is typically used in school choice related policy, and is therefore
particularly valuable in assessments of the appropriateness or applicability of
policy. The continuous measure of distance is particularly useful in exploring
the actual extent of mobility and what it entails for particular learners in terms
of the investments they are required to make, both financially and in terms of
time. In addition, it allows for the examination of the distribution of the
distances travelled by the entire sample, and, as it is the measure that has been
most commonly used in the existing literature, it also allows for comparison
with previous findings.
The definition of mobility based on census geography is particularly useful in
that it makes use of generally accepted geographical areas to explore the extent
to which mobility is occurring within and between these areas. This is helpful
in identifying whether learners are travelling between areas historically
designated for different race groups, and thereby significantly enhancing the
quality of education they are likely to receive. Additionally, it is, and thus
identifying those learners who are likely to be making the most substantial
141
economic investments in their education. Finally, the definition based on
whether or not the learner is attending the age-appropriate school closest to his
or her home is useful in highlighting the extent to which even learners with
relatively low levels of mobility may be engaging in more travel than strictly
necessary or anticipated, and may also be engaging in school choice,
particularly within the historically disadvantaged areas.
This chapter has made two key contributions to the literature. Firstly, in
providing three different approaches to the conceptualization and measurement
of learner mobility, it has significantly enhanced the methodological tools
available to the study of this practice. Secondly, it has, for the first time,
provided population-based data on the extent of learner mobility in
contemporary urban South Africa. In so doing, it has identified preliminary
evidence to suggest that there may in fact be two patterns of school choice and
mobility in operation in Johannesburg-Soweto. Firstly, there is a group of
approximately 25% of the sample who are engaged in substantial travel from
home to school on a daily basis, and who seem likely to be making significant
investments in this mobility. Secondly, and somewhat less expectedly, there is
also evidence that a large proportion of children who are not travelling
substantial distances to school are still engaging in mobility and school choice.
Even though they are attending schools relatively close to home, they are not
attending the nearest grade-appropriate school to their home, and are often
travelling to schools that are not located in the same residential areas as their
homes. These patterns, and their importance to understanding the implications
of learner mobility to educational access and equality, are explored in greater
detail in the subsequent chapters.
142
Chapter 6: Individual, family and
community characteristics of
mobile learners
6.1 Introduction
The previous chapter illustrated the extent of learner mobility in post-
Apartheid Johannesburg-Soweto. Even using the most stringent definitions of
mobility, the numbers of children engaged in mobility are substantial. Having
developed an understanding of the extent to which learner mobility is taking
place amongst school-age children in Johannesburg-Soweto, along with a
clearer idea of exactly what this mobility entails, two further questions arise.
The first relates to which children in particular are most likely to be engaging
in learner mobility. The second relates to the characteristics of schools these
children are choosing to attend, and the types of schools they are travelling to
avoid. This chapter addresses the first of these questions, while the second will
be addressed in Chapter 7.
A child‘s educational mobility is expected to be closely related to the level of
investment the child‘s family makes in his or her education. For this reason, it
is useful to explore variation in both a family‘s access to resources to invest in
education, and in variables which might be associated with a family‘s
propensity to invest in education. This chapter explores the relationship to
mobility of a range of variables at the levels of the individual child, the child‘s
household, and his or her community. The child characteristics considered are
race, gender, age at first enrolment in school, grade repetition and schooling
phase. Family characteristics examined are maternal education, maternal
marital status, and household SES, in both 1997 and 2003. At the community
level, SAL, SP and MP poverty levels are examined.
143
Child level characteristics
6.2 Race
Given the strong relationship between race, access to resources, and area of
residence in South Africa, it is anticipated that race is related to educational
mobility behaviours (Fiske and Ladd 2004; Fiske and Ladd 2005). While white
children are most likely to have access to the resources required to engage in
educational mobility, they are also least likely to need to engage in it to access
good schools, given that most highly performing schools are located in
historically white areas. Black children, by contrast, are likely to have the
greatest incentives to engage in mobility, typically living in the areas with
poorest schools, but are simultaneously least likely to have access to the
necessary resources. Indian and coloured children are likely to fall somewhere
in between the black and white children in terms of both incentives and ability
to engage in mobility. As the numbers of white and Indian children present in
the study sub-sample (28 and 25, respectively) are extremely small, findings
for these groups are unlikely to be broadly representative, and are therefore not
presented here. Discussion will be limited to the behaviour of black and
coloured children.
6.2.1 1997
Straight-line distance
Examining the distance between home and school on the basis of race reveals a
strong relationship with race. As is evident in Table 6.1, black children tend to
travel substantially further to school than coloured children (Wilcoxon rank-
sum, Pr=0.0000). A kernel density plot (see Figure 6.1 below), illustrates just
how different the distances from home to school are for black and coloured
children. The kernel density plot for the coloured children is far more
concentrated at very low levels of travel for coloured children than for their
black peers.
144
Race Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Black
African
1002 5.773km 9.589km 0.492km 1.138km 6.904km
Coloured 159 3.935km 9.533km 0.360km 0.561km 1.834km
Table 6.1: 1997 Distance between home and school on the basis of race
Figure 6.1: Kernel density plot of distance to school in 1997, on the basis of race
Census geography
An area-based approach to measuring mobility finds similar patterns (see
Table 6.2 below). At all levels of geography other than MN, coloured children
are substantially more likely to live and attend school in the same area than
black children.
Black (n=1003) Coloured(n=160) χ2
School and home in
same SAL
48
(4.79%
29
(18.31%)
χ2(1) = 39.7146,
Pr=0.000
School and home in 373 105 χ2(1) = 46.0932,
145
same SP (37.19%) (65.63%) Pr=0.000
School and home in
same MP
696
(69.39%)
141
(88.13%)
χ2(1) = 24.0038,
Pr=0.000
School and home in
same MN
957
(95.41%)
150
(93.75%)
χ2(1) = 0.8334,
N.S.
Table 6.2: 1997 mobility at different levels of census geography, by race
Nearest school
Mobility analyses exploring whether or not children attended their nearest
grade-appropriate schools again provided similar results, with coloured
children being substantially more likely to attend their nearest school than
black children (see Table 6.3). The analysis was conducted firstly using only
public schools, and secondly using both public and independent schools, and in
both cases similar figures were obtained. The finding that black children are
the least likely to attend their nearest school makes sense given that they are
likely to live in the poorest areas (as described in Appendix 3), and their
nearest school is therefore more likely to be particularly poorly performing.
Black (n=1009) Coloured
(n=155)
χ2
Child attends nearest school
(public or independent)
146
(14.47%)
60
(38.71%)
χ2(1) = 54.2007,
Pr=0.000
Child attends nearest school
(public only)
143
(14.17%)
62
(38.75%)
χ2(1) = 57.6862,
Pr=0.000
Table 6.3: Children attending their nearest grade-appropriate school, by race, for
public schools only, and for all schools
6.2.2 2003
Straight-line distance
As evident in Table 6.4, the distances from home to school in 2003 are slightly
different from those in 1997, with an increase in the difference between the
mean distances travelled by black and coloured children (Wilcoxon rank-sum,
Pr=0.0000). In Figure 6.2, the extent to which coloured children are more
likely to live very close to their school than black children is highly evident.
146
Race Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Black
African
1065 5.834km 8.617km 0.635km 1.380km 8.090km
Coloured 163 2.625km 7.121km 0.402km 0.668km 1.315km
Table 6.4: 2003 Distance between home and school on the basis of race
Figure 6.2: Kernel density plot of distance to school in 1997, on the basis of race
Census Geography
The 2003 area-based analysis (see Table 6.5 below) provides very similar
results to the 1997 analysis. Once again, coloured children are much more like
to attend school in the same SAL, SP and MP areas in which they live than
black children.
Black (n=1066) Coloured (n=163) χ2
School and home in
same SAL
28
(2.63%)
16
(9.82%)
χ2(1) = 72.4870
Pr=0.000
School and home in
same SP
350
(32.83%)
110
(67.48)
χ2(1) = 72.4870
Pr=0.000
147
School and home in
same MP
702
(65.95%)
150
(92.02%)
χ2(1) = 45.5357
Pr=0.000
School and home in
same MN
1026
(96.25%)
159
(97.55%)
χ2(1) = 0.6904
N.S.
Table 6.5: 2003 Mobility across different levels of census geography, by race
Nearest school
Coloured children remain significantly more likely to attend their nearest
grade-appropriate school than black children in 2003 (see Table 6.6 below).
Black (n=1009) Coloured (n=155) χ2
Child attends nearest
school (public or
independent)
154
(14.50%)
61
(39.35%)
χ2(1)= 57.4420
Pr=0.000
Child attends nearest
school (public only)
161
(15.16%)
66
(40.99%)
χ2(1)= 61.7256
Pr=0.000
Table 6.6: 2003 Children attending their nearest grade-appropriate school, by race,
both for public schools only, and for all schools
6.2.3 Race and mobility discussion
There is strong evidence that both in 1997 and 2003, race is closely related to
mobility behaviour, regardless of the way in which mobility is measured.
Overall, black children appear to be substantially more engaged in all forms of
learner mobility than coloured children. There are a range of possible
explanations for this, including that coloured children live in areas with better
schools, that the coloured community is more cohesive and prefer to keep their
children at local schools, or that coloured families are less likely to want to
make substantial investments in their children‘s education for various reasons.
6.3 Gender
It is possible that families approach the education of girls and boys differently.
Certainly, in contemporary South Africa, girls are known to remain in formal
148
education longer, and also tend to outperform boys (Unterhalter 2005; Fleisch
and Schindler 2009). During the Apartheid era, both policy and practice
favoured different approaches to education on the basis of gender (Fiske and
Ladd 2004), and some legacy of this might be expected to persist, particularly
around the levels of investment in the education of children of different
genders. To determine whether mobility, and by extension, educational
investment, differs on the basis of gender, male and female populations were
compared, using different definitions of mobility.
6.3.1 1997
Straight-line distance
Examining the distribution of distance by gender does suggest girls, on
average, travel slightly further than boys (see Table 6.7 and Figure 6.3 below).
Closer examination of the data, however, seems to suggest that this difference,
particularly evident in the means, may be caused primarily by a cluster of girls
travelling fairly substantial distances, particularly between 20 and 60km,
pulling the overall mean for girls (along with the standard deviation and
percentile breaks) upwards. A Wilcoxon rank-sum (Mann-Whitney) test fails
to find any significant difference in the distribution of distance from home to
school on the basis of gender.
Gender Number
of
children
Mean
distance to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Boys 592 4.976 km 8.782 km 0.466 km 0.941 km 4.997 km
Girls 622 5.985 km 10.163 km 0.475 km 1.086 km 6.928 km
Table 6.7: 1997 distance from home to school, by gender
149
Figure 6.3: Kernel density plot of 1997 distance from home to school, by gender
If distance from home to school is grouped into categories, a chi-square test
does, however, reveal a significant difference between boys and girls (see
Table 6.8 below). Overall, while distributions are fairly similar, boys are
somewhat more likely to be travelling extremely short distances, and girls are
somewhat more likely to be travelling distances over 20km.
Up to
1km
1km-
2.5km
2.5km-
5km
5km-
10km
10km-
20km
Over
20km
χ2(5)
=12.156,
Pr=0.033
Boys
(n=592)
306
(51.69%)
88
(14.86%)
50
(8.45%)
47
(7.94%)
66
(11.15%)
35
(5.91%)
Girls
(n=622)
295
(47.43%)
107
(17.20%)
33
(5.31%)
62
(9.97%)
68
(10.93%)
57
(9.16%)
Table 6.8: Gender breakdown of 1997 categories of distance from home to school
Census geography
Girls were significantly more likely to attend a school in the same SAL in
which they lived than boys (see Table 6.9 below). At all other levels of
150
geography, however, there was no evidence that girls and boys behaved
differently.
Boys (n=592) Girls (n=622) χ2
School and home in
same SAL
33 (5.56%) 50 (8.03%) χ2(1) =2.893
Pr = 0.089
School and home in
same SP
242 (40.81%) 252 (40.45%) χ2(1) =0.0163
Not significant
School and home in
same MP
439 (74.03%) 443 (71.11%) χ2(1) =1.303
Not significant
School and home in
same MN
570 (96.12%) 587 (94.22%) χ2(1) =2.376
Not significant
Table 6.9: 1997 mobility across different levels of census geography, by race
Nearest School
Chi-squared tests provided no indication for any gender differences in the
likelihood of children attending their nearest school, regardless of whether
independent schools were included in the analysis or not.
6.3.2 2003
Straight-line distance
Overall, patterns of mobility by gender in 2003 remained largely consistent
with those identified in 1997, although the mean distance travelled to school by
girls did decrease slightly (see Table 6.10 ). Nonetheless, girls still continue to
travel, on average, almost a kilometre further than boys. Interestingly, the
standard deviation on the distances travelled by girls has fallen quite
substantially, approaching fairly closely the standard deviation on the distances
travelled by boys. For both genders, the percentile distances have increased
slightly, with the effect more noticeable for girls, particularly from the 75th
percentile up. In sum, this suggests that distribution of travel distances for girls
may have spread out slightly towards the tail end (representing greater
distances), with proportionally fewer girls continuing to travel particularly
short distances, although this effect is not large enough to be evident on the
151
kernel density plot of the distributions of distance travelled by gender (see
Figure 6.4 below). A Wilcoxon rank-sum test indicates that girls travel further
than boys in 2003 (Pr= 0.0489).
Gender Number of
observations
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Boys 631 4.953469 8.224416 0.522km 1.175km 5.524km
Girls 650 5.745164 8.67542 0.619km 1.303km 7.446km
Table 6.10: 1997 distance from home to school, by gender
Figure 6.4: Kernel density plot of 2003 distance from home to school, by gender
In 2003, the chi-square analysis of distance grouped into categories no longer
reveals any significant difference between boys and girls, although a higher
percentage of girls continue to travel particularly great distances (see Table
6.11 below).
152
Up to
1km
1km-
2.5km
2.5km-
5km
5km-
10km
10km-
20km
Over
20km
χ2(5)
=8.428;
Not
significant
Boys
(n=631)
290
(45.96%)
123
(19.49%)
47
(7.45%)
65
(10.30%)
66
(10.46%)
40
(6.34%)
Girls
(n=650)
264
(40.62%)
146
(22.46%)
40
(6.15%)
67
(10.31%)
71
(10.92%)
62
(9.54%)
Table 6.11: Gender breakdown of 2003 categories of distance from home to school
Census geography
Examining mobility by census area in 2003 reveals that the gender difference
previously evident at the SAL has disappeared (see Table 6.12). A gender
difference at the level of SP level has emerged, although in the opposite
direction, with boys being more likely to attend school in the same SP where
they live. There is no evidence of any gender difference at the MP or MN
levels.
Boys (n=631) Girls (n=651) χ2
School and home in
same SAL
24
(3.80%)
26
(3.99%)
χ2(1) =0.031
Not significant
School and home in
same SP
247
(39.14%)
226
(34.72%)
χ2(1) =2.699
Pr =.100
School and home in
same MP
452
(71.63%)
449
(68.97%)
χ2(1) =1.087
Not significant
School and home in
same MN
609
(96.51%)
627
(96.31%)
χ2(1) =0.037
Not significant
Table 6.12: 2003 mobility across different levels of census geography, by gender
Nearest school
As with 1997, there is no evidence in that children of either gender were more
likely to attend their nearest school in 2003, regardless of whether independent
schools are included in the analysis or not.
153
6.3.3 Gender and mobility discussion
There was some evidence that girls travelled, on average, further than boys in
both 1997 and 2003, but this was extremely sensitive to the way in which
mobility was measured. If girls are travelling further, one possible explanation
is that parents value the education of girl children more highly, and therefore
tend to invest more in their education. An alternative explanation may be that
girl children are more likely to travel with parents, and attend school close to a
parents‘ work place, perhaps due to safety concerns. Overall, however, the data
presented here does not provide conclusive evidence for any substantial
difference in mobility on the basis of gender.
6.4 Age at first school enrolment
In South Africa, children have a legal window of two years during which to
start their schooling. It is possible that the point during this window at which
children begin their formal schooling relates to the level of interest or
commitment that parents feel towards their child‘s schooling, with more
committed parents enrolling children earlier. By contrast, it may also relate to a
parent‘s ability to fulfil care-giving responsibilities, in which case parents with
fewer resources may be more likely to pursue the earliest possible enrolment of
their children to reduce their care-giving burden. It may also relate to different
enrolment and application policies applied in different schools, with more
selective schools preferring to enrol older and more independent children. It is
therefore conceivable that the distance a child travels to school is connected to
their age at first school enrolment, although the expected direction of this
relationship is not evident.
154
6.4.1 1997
Straight-line distance
Table 6.13 and Figure 6.5 show that children who start school at a later age
travel significantly further than those who start at an earlier age (Wilcoxon
rank-sum test; Pr= 0.0004).
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Earlier
starters
646 4.704km 9.216km 0.449km 0.880km 4.569km
Later
starters
557 6.482km 9.866km 0.498km 1.221km 9.044km
Table 6.13: 1997 distance from home to school, by age at first enrollment
Figure 6.5: Kernel density plot of 1997 distance from home to school, by age at first
school enrolment
Census geography
Although there was no relationship between starting school late and travelling
between SALs or MNs for schooling, children who started school early were
155
significantly more likely to school within their residential SP (χ2(1)=10.8425;
Pr=0.001), as well as in their residential MP (χ2(1)=9.8651; Pr=0.002).
Nearest school
There was no significant relationship between age at first enrolment and
whether or not a child attended their closest grade-appropriate school,
regardless of whether independent schools are included or excluded.
6.4.2 2003
Straight-line distance
In 2003, although late starters still travel further on average than early starters,
this difference is no longer statistically significant (Wilcoxon rank-sum test).
Additionally, although the distance for the 75th percentile of late starters is still
substantially higher than for early starters, at the 25th and 50th percentile, the
early starters are actually travelling further. The distribution for late starters is
therefore wider, but slightly flatter, than that for early starters (see Table 6.14
and Figure 6.6 below). One potential explanation for this change over time is
that by 2003, more of the children who started school early are enrolled in high
school, and that this requires them to travel somewhat further. This hypothesis
is explored in the next section of this chapter, on the relationship between
schooling phase in 2003 and mobility behaviours.
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Earlier
starters
644 5.122km 8.143km 0.615km 1.288km 6.187km
Later
starters
563 5.750km 8.867km 0.567km 1.181km 8.077km
Table 6.14: 2003 distance from home to school, by age at first enrolment
156
Figure 6.6: Kernel density plot of 2003 distance from home to school by age at first
school enrolment
Census geography analysis
By 2003, there is no longer any evidence for differential mobility at any level
of census geography on the basis of whether a child started school early or late.
Nearest school
There is a weakly significant relationship between whether a child starts school
late, and whether he or she attends his or her nearest grade-appropriate school
in 2003, but only when both public and independent schools are considered
(χ2(1)= 2.9904, Pr = 0.084), with children starting late being less likely to be
attending their nearest school in 2003. When only public schools are
considered, there is no significant relationship.
6.4.3 Age at first enrolment and mobility discussion
In 1997, there is a relationship between mobility and whether a child starts
school earlier or later, with children starting later being more likely to travel
157
further. By 2003, this relationship has however largely disappeared. This may
relate to the fact that a proportion of those who started school early have begun
attending secondary schools, while those who started late are almost all still in
primary schools. The relationship between mobility and schooling phase is
explored further in the next section of this chapter.
There is no clear and obvious explanation for why, in 1997, children who
started school late are likely to have a greater distance between their homes
and their schools. It may be the case that parents who plan to send their
children to schools further afield, necessitating independent travel, as well as in
many cases the ability to adapt to a different cultural environment, are waiting
until children are slightly older before enrolling them in school. It may also
relate to enrolment practices at more advantaged schools, were children are in
some instances required to take entrance tests or undergo interviews. A third
potential explanation is that less affluent parents, who are not able to send
children to schools far from home, may also not be able to afford pre-school or
child care for their children, and therefore prefer to send them to school as
early as possible. More affluent parents, by contrast, may be less pressed to
enrol children in primary school, preferring to ensure that children are
genuinely school-ready.
6.5 School phase in 2003
As discussed in Chapter 4, there is evidence that those children who have
reached secondary school in 2003 differ systematically from those that have
not on the basis of race, gender, age at first enrolment, grade repetition,
maternal education, and 1997 SES. All of these variables have been
hypothesized to have some relationship to mobility, and as a result, children
who have reached high school in 2003 may be exhibiting different mobility
behaviours to other children simply because of this. On the other hand, high
158
school status may in itself have implications for mobility behaviour for a range
of reasons.
As described in Chapter 4, the Gauteng province contains a fairly small
number of large high schools, and a much larger number of much smaller
primary schools. Due to this, all else being constant, we would expect to see
high school children travelling slightly further to school on average, and we
would also expect a higher proportion of high school children to attend their
nearest school. The descriptive mobility data presented in Chapter 5 suggests
that this is indeed the case.
An additional reason to anticipate changes in mobility behaviour between
primary and secondary schooling is that the costs and benefits of mobility may
change, in turn changing preferences around the selection of schools. For
example, children of high school age can travel greater distances, on their own,
more safely than younger children, and may also be able to walk further,
lowering the cost of attending a more distant school. A factor in favour of
stability in mobility behaviours between the primary and secondary schooling
years, however, is any element of path dependence. For example, secondary
schools may give preference to children from local primary schools, and those
primary schools might likewise encourage children to enrol in local high
schools. The difficulty of moving between schools in different areas may be
much higher for older children.
Preferences may also be shaped through different criteria at the secondary
school level. For example, the academic performance of a secondary school is
more immediately salient than the performance of a primary school, largely
due to the availability of some information about the matric exam pass rate.
The greater availability of evidence with regards to school academic
performance may affect the importance attributed to school academic
performance in school selection. At the high school level, children may also be
159
far more actively involved in the selection of their school, and may well be
driven by different priorities than those used by their parents in selecting a
primary school. By contrast, however, secondary schooling is typically more
expensive than primary schooling, and this may influence parents to maintain
or even increase their role in school selection. Finally, secondary schooling
may simply be attributed greater value than primary schooling, changing the
level of investment which families and children are willing to make in
education. It is not immediately evident, however, in which direction these
potential changes in school preferences should influence mobility when
aggregated.
Differences between the mobility behaviours in 2003 of children who have
reached high school, and those who haven‘t, could be attributed either to their
being different with regards to variables associated with mobility, or
alternatively simply to the fact that they have reached high school. It is also
possible that any difference is due to a combination of these factors. Due to the
nature of the sample used in this study, in which only a relatively small, non-
random group of children has reached high school by 2003, the relative
contributions of individual and household variables on mobility cannot be
separated out from any independent effect of high school status. Really
untangling the extent to which schooling phase shapes mobility independent of
socio-economic and other individual and family-level variables could be
approached either through the use of data for years beyond 2003 for the current
sample, or through the use of a broader, or differently structured sample.
However, the available data does provide one way of obtaining some insight
into this issue, by looking at whether the two groups of children had similar
mobility behaviours in 1997 or not. If they behaved similarly in 1997, it seems
likely that more of the difference in 2003 can be attributed to schooling phase.
By contrast, if behaviour was already different in 1997, this suggests a more
160
important role for individual, family and community variables associated with
mobility. This data is presented below.
6.5.1 1997
Straight-line distance
Children who were still in primary school in 2003, and those who had reached
secondary school in 2003, had very similar distances from home to school in
1997, as evident in Table 6.15 and Figure 6.7. A Wilcoxon rank-sum test
confirmed that distance from home to school in 1997 is not statistically
significantly different for the two groups.
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Primary
school in
2003
780
(65.60%)
5.729km 9.350km 0.469km 1.132km 6.739km
High
school in
2003
409
(34.40%)
5.285km 10.073km 0.466km 0.910km 5.414km
Table 6.15: 1997 distance from home to school, by schooling phase in 2003
161
Figure 6.7: Kernel density plot of 1997 distance from home to school by 2003
schooling phase
Census Geography
Children who have reached high school by 2003 are slightly more likely to
attend a school in 1997 that is in the same SAL as their home (χ2(1)=4.3214,
Pr=0.038), and the same SP as their home (χ2(1)=7.6029, Pr=0.006). There is,
however, no evidence of any differences in mobility on the basis of 2003
schooling phase at either the MP or MN levels.
Nearest School
There is no evidence that schooling phase in 2003 is associated with a child‘s
likelihood of attending his or her nearest grade-appropriate school in 1997,
regardless of whether independent schools are included in the analysis or not.
162
6.5.2 2003
Straight-line distance
As evident in Table 6.16 and Figure 6.8 below, children attending high school
travel significantly further to school than those still in primary school in 2003
(Wilcoxon rank-sum test; Pr=0.0263).
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
High
school in
2003
418 5.791km 9.216km 0.712km 1.421km 7.367km
Primary
school in
2003
853 5.164km 8.085km 0.533km 1.175km 6.475km
Table 6.16: 2003 distance from home to school, by progression to high school by
2003
Figure 6.8: Kernel density plot of 2003 distance from home to school by phase of
education in 2003
163
Given the fairly strong relationships between gender, age at first enrollment,
and schooling phase in 2003 (see Appendix 3 for details), it is worth exploring
the combined interactions of these variables with distance from home to
school. Table 6.17 and Figure 6.9, below, illustrate that distance from home to
school is different for each group when broken down by both gender and
schooling phase. In all cases, boys travel less far than girls, with primary
school boys travelling the shortest distances of all. High school girls are
travelling further than any other group, including primary school girls. Both
girls and boys at high school level are less likely to be travelling short
distances to school, and it is only at the particularly high distances that the
distributions actually differ on the basis of gender. At shorter distances (as is
evident in Table 6.18), the distributions for high school girls and boys are
fairly similar, as are the distributions for primary school girls and boys. A
Wilcoxon rank-sum test indicates a weakly significant positive relationship
between high school status and distance travelled for girls (Pr= 0.0689), but
not for boys. There is, however, no evidence that girls travel significantly
further than boys at either the primary or high school level.
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Primary
school
boys
459 4.778km 7.620km 0.509km 1.144km 5.816km
High
school
boys
166 5.564km 9.805km 0.655km 1.260km 5.524km
Primary
school
girls
394 5.614km 8.583km 0.567km 1.183km 7.425km
High
school
girls
252 5.941km 8.823km 0.758km 1.481km 8.211km
Table 6.17: 2003 distance from home to school, by gender and phase of education
in 2003
164
Figure 6.9: Kernel density plot of 2003 distance from home to school by gender and
phase of education in 2003
Census geography
There is no evidence for any significant relationship between mobility at any
level of census geography in 2003, and whether a child is in primary or high
school.
Nearest school
Children who are in high school in 2003 are significantly more likely to be
attending their nearest grade-appropriate school, whether independent schools
are included (χ2(1)= 7.9497, Pr = 0.005) or excluded (χ
2
(1)= 6.3958, Pr = 0.011).
6.5.3 School phase in 2003 and mobility discussion
The results presented in this section have suggested that those children who
had reached secondary school by 2003, and those who had not, behaved fairly
similarly with respect to educational mobility in 1997. In 2003, by contrast,
those children who have reached secondary school do travel further than those
165
still in primary school. This makes sense as high schools tend to be larger and
therefore less densely distributed, meaning that the nearest high school will on
average be slightly further from a child‘s home than the nearest primary
school. School distribution is also likely to explain why children in high school
are more likely to be attending their nearest school – there are simply fewer
options, particularly when a child faces financial constraints. The fact that
these differences only emerge in 2003, once these children have reached high
school, also supports the hypothesis that the changes in behaviour are linked
more to high school status itself, than to the individual, family and community
attributes of the learners in question.
While it would be ideal to explore these patterns further, the relatively small
number of children in high school by 2003, and the non-random constitution of
this group, raises problems. These questions could, however, be usefully
explored in future work making use of data from subsequent years, or a
broader sample. The data presented in this section also suggest that gender-
based differences in mobility may increase at the secondary school level.
Again, however, additional data will be required before these relationships can
be more conclusively tested.
6.6 Grade repetition
Once enrolled in formal schooling, children have different experiences with
grade progression. While many children do pass smoothly through the grades,
a fairly substantial number are forced to repeat a particular grade, or even a
number of grades (Fleisch and Schindler 2009). This section explores whether
grade repetition between 1997 and 2003 has any relationship to distance
travelled to attend school. This relationship is explored both in 1997, prior to
repetition, and in 2003, after repetition.
166
Grade repetition may be thought of as an indicator of a child‘s inherent
academic capabilities. In this case, it is possible that it might influence parental
decisions on investment in schooling in both 1997 and 2003. Parents might
choose to invest less in an academically less gifted child, or they might choose
to invest more in the hopes of ensuring that child‘s success. To the extent that
grade repetition reflects inherent academic capacity, the direction in which it
should be expected to impact mobility is not clear. However, grade repetition
may also simply reflect the quality of the school which a child attends (Lam,
Ardington et al. 2008). In this case, grade repetition would be expected to be
negatively associated with mobility.
6.6.1 1997
Straight-line distance
Children who repeated a grade between 1997 and 2003 are significantly more
likely to have shorter distances between home and school than those who did
not repeat any grades (see Table 6.18 and Figure 6.10 below), as confirmed by
a Wilcoxon rank-sum test (Pr=0.0000).
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Repeaters
440 4.082km 7.997km 0.398km 0.823km 3.751km
Non-
repeaters
730 6.516km 10.364km 0.506km 1.195km 8.152km
Table 6.18: 1997 distance from home to school, by grade repetition status
167
Figure 6.10: Kernel density plot of 1997 distance from home to school by grade
repetition
Census geography
Children who repeated a grade at least once between 1997 and 2003 were less
likely to attend schools outside of the SP (χ2(1)= 5.1148; Pr = 0.024) and MP
(χ2(1)= 12.6441; Pr = 0.000) in which they lived. There was no relationship
between repetition and mobility at the SAL and MN levels.
Nearest school
There is no significant relationship between grade repetition and whether or
not a child attends their closest grade-appropriate school in 1997, regardless of
whether independent schools are included or excluded.
6.6.2 2003
Straight-line distance
The relationship between grade repetition and distance travelled in 2003 is
similar to that with distance in 1997. Again, as illustrated in Table 6.19 and
168
Figure 6.11 below, children who have repeated grades have a shorter distance
from home to school (Wilcoxon rank-sum test; Pr=0.0000).
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Repeaters 439 4.249km 7.236km 0.464km 1.013km 4.640km
Non-
repeaters
750 6.145km 9.121km 0.664km 1.447km 8.583km
Table 6.19: 2003 distance from home to school, by grade repetition status
Figure 6.11: Kernel density plot of 2003 distance from home to school by grade
repetition
Census geography
In 2003, children who have repeated grades are again significantly more likely
to attend a school in their residential SP (χ2(1)= 8.3371, Pr = 0.004) or MP
(χ2(1)= 8.3578, Pr = 0.004). There remains no evidence for any relationship at
the SAL or MN levels.
169
Nearest school
There is no evidence for any significant relationship between grade repetition,
and whether or not a child attends his or her nearest school, public or
independent, in 2003.
6.6.3 Grade repetition and mobility discussion
Although there is a strong and consistent relationship between repetition and
mobility, the available data says nothing about the causal direction of this
relationship – does mobility shape repetition, or the reverse? Because most
sample members live in relatively disadvantaged areas, mobility is usually
associated with attendance at more advantaged schools, which tend to have
lower levels of repetition. By extension, attending a local, less-advantaged
school is likely to be associated with higher levels of repetition. For this
reason, it seems more likely that mobility predicts repetition, than the reverse.
A similar explanation seems plausible for the relationship between repetition
and mobility at SP and MP levels of census geography. Of course, repetition is
not determined purely by the school a child attends, and there is likely to also
be an interaction effect operating, with children whose family circumstances
favour better academic performance also being more likely to travel. The next
set of analyses presented will explore the relationship between family
attributes often associated with academic performance, and mobility.
Household level characteristics
6.7 Maternal education
Maternal education is anticipated to have a positive relationship with mobility,
as more educated mothers are expected to place a higher premium on
educational investment, and to have access to more resources to invest in their
children‘s education.
170
6.7.1 1997
Straight-line distance
As expected, mean distance travelled to school increases with maternal
education level (see Table 6.20 and Figure 6.12 below). The mean figures for
children with mothers with lower levels of education, up to grade 7, are
skewed upwards by a few children who are travelling extremely substantial
distances. This effect is evident in the high standard deviations and the very
low distances at the 25th, 50th and 75th percentile.
Maternal
Education
level
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Up to
Grade 5
64 4.591km 13.075km 0.353km 0.708km 1.456km
Grade 6 or
7
78 5.572km 11.411km 0.468km 1.121km 3.304km
Grade 8, 9
or 10
530 4.207km 8.353km 0.434km 0.81km 3.646km
Grade 11
or 12
349 6.828km 9.400km 0.608km 1.491km 9.769km
Post-
school
education
92 7.224km 9.392km 0.642km 2.148km 11.879km
Table 6.20: 1997 distance from home to school, by maternal education level
Figure 6.12, below, provides a kernel density plot of the distances travelled by
children to school, grouped by maternal education. The graph illustrates very
clearly the variable nature of mobility by educational level, with children with
functionally illiterate mothers being most likely to travel very short distances.
As maternal education increases, however, a smaller proportion of children can
be seen to travel very short distances, and the distributions gradually spread
out, becoming slightly more normal. There is, however, substantial overlap
between those children whose mothers have either grade 6 or 7, and those
whose mothers have either grade 8, 9 or 10. A Kruskal-Wallis test indicates
171
that distance from home to school varies significantly on the basis of maternal
education (Pr=0.0001).
Figure 6.12: Kernel density plot of 1997 distance from home to school by maternal
education level.
Census geography
Using chi-squared tests, whether or not a child attended school in the same MP
as their home was the only geographical measure significantly linked to the
distribution of maternal education for all levels of maternal education (see
Table 6.21 below). When the children were divided on the basis of whether
their mothers had completed primary education or not, significantly different
levels of mobility at the SAL, SP and MP levels were identified.
172
Grade 5 and
lower vs.
Grade 6 or
higher
(functionally
illiterate vs.
functionally
literate)
Grade 7 and
lower vs.
Grade 8 or
higher
(primary
school vs.
higher than
primary school
education)
Grade 10 and
lower vs.
Grade 11 or
higher
Grade 12 and
lower vs. any
post-school
education
(no post-
school
education vs.
any post-
school
education)
School in
same SAL as
home
χ2(1)= 1.902
Not significant
χ2(1)= 4.202
Pr = 0.040
χ2(1)= 0.815
Not significant
χ2(1)= 0.153
Not significant
School in
same SP as
home
χ2(1)= 2.578
Not significant
χ2(1)= 15.201
Pr = 0.000
χ2(1)= 2.1208
Not significant
χ2(1)= 1.734
Not significant
School in
same MP as
home
χ2(1)= 6.409
Pr = 0.011
χ2(1)= 41.05
Pr = 0.000
χ2(1)= 9.363
Pr = 0.002
χ2(1)= 8.216
Pr = 0.004
School in
same MN as
home
χ2(1)= 0.299
Not significant
χ2(1)= 0.081
Not significant
χ2(1)= 3.13
Pr = 0.077
χ2(1)= 0.693
Not significant
Table 6.21: 1997 mobility across different levels of census geography, by maternal
education level
Nearest school
A chi-square test indicated that children whose mothers were functionally
illiterate were significantly more likely to attend their nearest public school
than children whose mothers had higher levels of education (see Table 6.22
below). However, this relationship did become less linear at the highest levels
of maternal education. When independent schools were included in the
analysis, figures were largely similar, although somewhat more statistically
significant (χ2(4) = 12.707, Pr=0.013).
173
Maternal
education
Up to
grade 5
Grade 6
or 7
Grade 8, 9
or 10
Grade 11
or 12
Post-
school
education
χ2 test
results
Child attends
nearest public
school (n=198)
19
(28.79%)
13
(16.67%)
103
(19.36%)
48
(13.64%)
15
(16.13%)
χ2(4)=
10.7994
Pr = 0.029
Table 6.22: Children attending their 1997 nearest grade-appropriate public school,
by maternal education
6.7.2 2003
Straight-line distance
As was the case in 1997, distance from home to school in 2003 is also closely
connected to maternal education (Kruskal-Wallis test, Pr=0.0001). As evident
in Table 6.23 below, the gap between children whose mothers have completed
up to Grade 10, and those who have completed Grade 11 or higher appears to
have grown, while the distributions for children whose mothers have
completed Grade 11 or 12, and those whose mothers have some post-school
education, have become more similar. This is even more evident in Figure 6.13
below, in which the distributions of distance travelled are shown on the basis
of maternal education.
Maternal
Education
level
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Up to
Grade 5
73 3.417km 8.443km 0.397km 0.778km 1.532km
Grade 6 or
7
83 3.021km 5.826km 0.398km 0.788km 2.332km
Grade 8, 9
or 10
560 3.855km 7.372km 0.506km 1.002km 2.692km
Grade 11
or 12
364 7.792km 9.403km 0.835km 2.870km 12.175km
Post-school
education
97 7.811km 9.189km 0.938km 4.410km 11.858km
Table 6.23: 2003 distance from home to school, by maternal education level
174
Figure 6.13: Kernel density plot of 2003 distance from home to school by maternal
education level
Census geography
Again, as in 1997, chi-squared tests indicated that children with more educated
mothers were more likely to attend a school outside of the area in which they
lived for all levels of census geography other than MN (see Table 6.24 below).
The 2003 data generally seems to indicate a stronger relationship between
maternal education level, and whether or not children live and school within
the same geographic area than was evident in 1997.
Grade 5 and
lower vs.
Grade 6 or
higher
(functionally
illiterate vs.
functionally
literate)
Grade 7 and
lower vs.
Grade 8 or
higher
(primary
school vs.
higher than
primary school
education)
Grade 10 and
lower vs.
Grade 11 or
higher
Grade 12 and
lower vs. any
post-school
education
(no post-
school
education vs.
any post-
school
education)
School in χ2(1)= 7.417 χ
2
(1)= 5.215 χ
2
(1)= 3.579 χ
2
(1)= 1.765
175
same SAL as
home
Pr = 0.006 Pr = 0.022 Pr = 0.059 Not significant
School in
same SP as
home
χ2(1)= 3.925
Pr = 0.048
χ2(1)= 52.025
Pr = 0.000
χ2(1)= 12.829
Pr = 0.000
χ2(1)= 9.414
Pr = 0.002
School in
same MP as
home
χ2(1)= 4.135
Pr = 0.042
χ2(1)= 78.232
Pr = 0.000
χ2(1)= 17.673
Pr = 0.000
χ2(1)= 10.817
Pr = 0.001
School in
same MN as
home
χ2(1)= 0.215
Not significant
χ2(1)= 2.229
Not significant
χ2(1)= 0.686
Not significant
χ2(1)= 0.592
Not significant
Table 6.24: 2003 mobility across different levels of census geography, by maternal
education level
Nearest school
The distribution of levels of maternal education is significantly different
between children attending their closest grade-appropriate school, public or
independent, and those travelling further afield (Table 6.25). Children whose
mothers have completed schooling up to grade 7 appear to be substantially
more likely to attend their nearest public school than children whose mothers
have higher levels of education. As with all previous analyses of maternal
education, this relationship again appears to be stronger in 2003 than 1997.
Mother’s
education
level
Up to
grade 5
Grade 6
or 7
Grade 8,
9 or 10
Grade 11
or 12
Post-
school
education
χ2 test
results
Child
attends
nearest
public
school
(n=213)
20
(27.03%)
26
(31.71%)
111
(19.89%)
43
(11.85%)
13
(13.54%)
χ2(4)=
26.285
Pr=0.000
Table 6.25: Children attending their 2003 nearest grade-appropriate school, by
maternal education
6.7.3 Maternal education and mobility discussion
In 1997, at the lowest levels of maternal education, there is a very high level of
variability around the distance travelled. This effect seems largely to have
176
disappeared by 2003, and may relate to poor quality data around these
children‘s education and residence, resulting in inaccurate mobility data, or
alternatively, lack of residential stability resulting in substantial travel.
Overall, however, there is evidence that as maternal education increases, so too
does educational mobility. The effect of maternal education on mobility also
appears to increase over time, strengthening as children reach the end of
primary school and start to enrol in secondary school. This relationship
between mobility and maternal education is not strictly linear, and children
whose mothers have attained varying levels of education between grade 6 and
10 all appear to behave quite similarly rather than as distinct groups.
Additionally, the relationship between distance and maternal education appears
to stop holding at the very highest levels of maternal education. One possible
explanation for this is that children with the most highly educated mothers are
more likely to live in affluent areas, and by extension, close to good schools.
An interesting feature of the maternal education analyses is the fairly strong
inverse relationship between maternal education, particularly at intermediate
levels, and the likelihood that children will attend their nearest grade-
appropriate school. This raises questions around the relationship between
maternal education and the decision to engage in school choice, even if the
resources available allow only for choice between fairly local schools. Of
course, with both this relationship between maternal education and local school
choice, and with the broader relationship between maternal education and
distance travelled, it is critical to explore the role of SES. As SES is highly
correlated with maternal education, it is important to attempt to separate the
roles of resource availability and maternal education in shaping schooling
decisions. This is explored both in the section on household SES later in this
chapter, and in Chapter 9.
177
6.8 Maternal Marital status
The next potential determinant of learner mobility to be examined is maternal
marital status at the time of the child‘s birth. This provides an indicator of the
home environment into which a child is born, with married mothers typically
being associated with a more stable, socio-economically advantaged home
environment than non-married mothers. It is therefore plausible to expect that
the children of married mothers may be more likely to travel further to attend
school.
6.8.1 1997
Straight-line distance
As shown in Table 6.26 and Figure 6.14 below, children of married mothers
have, on average, a slightly greater distance from home to school than the
children of unmarried mothers (Wilcoxon rank-sum test; Pr= 0.0090).
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Married
mothers
410 6.284km 10.102km 0.510km 1.280km 8.102km
Unmarried
mothers
797 5.125km 9.228km 0.448km 0.934km 5.298km
Table 6.26: 1997 distance from home to school, by maternal marital status
178
Figure 6.14: Kernel density plot of 1997 distance from home to school by maternal
marital status
Census geography
There was no evidence of a significant relationship between maternal marital
status at birth, and a child‘s mobility at any level of census geography in 1997.
Nearest school
There was no evidence of a significant relationship between maternal marital
status at birth and whether a child attended his or her nearest grade-appropriate
school in 1997.
6.8.2 2003
Straight-line distance
Although it is evident from Table 6.27 and Figure 6.15, below, that in 2003
children with married mothers continued to travel slightly further than children
of unmarried mothers, this difference is no longer statistically significant
(Wilcoxon rank-sum test).
179
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
Married
mothers
426 5.511km 8.235km 0.639km 1.264km 7.446km
Unmarried
mothers
848 5.292km 8.603km 0.554km 1.238km 6.327km
Table 6.27: 2003 distance from home to school, by maternal marital status
Figure 6.15: Kernel density plot of 2003 distance from home to school, by maternal
marital status
Census geography
There was no evidence for a significant relationship between maternal marital
status at birth and a child‘s mobility at any level of census geography in 2003.
180
Nearest school
There was no evidence for a significant relationship between maternal marital
status at birth and whether a child attends his or her nearest grade-appropriate
school in 2003.
6.8.3 Maternal marital status and mobility discussion
Although there does seem to be a weak relationship between distance from
home to school and maternal marital status in 1997, this effect seems to have
disappeared by 2003. There is no evidence that maternal marital status is
related to any other measures of mobility. This may suggest that to the extent
that maternal marital status does influence schooling choices, this effect is
strongest when the children are young, and have fairly limited independence. It
may also simply reflect the stronger relationship between marital status at birth
and 1997 SES, as opposed to 2003 SES. Indeed, the extent to which any effect
of maternal marital status on schooling choices is operating through the
relationship between marital status and SES also merits further investigation.
This is covered in Chapter 9.
6.9 Household SES
Household SES is likely to play a core role in shaping school choice. Access to
resources determines how much a family can afford to spend on travel to
school, as well as how much they can afford to contribute to school fees and
related expenses. It may also be highly correlated with determinants of the
value the family places on education, such as parental education levels.
Generally, it is anticipated that as SES increases, so too will, on average, the
distance travelled to school.
181
6.9.1 1997
Straight-line distance
Examining the means and distributions of distance from home to school on the
basis of 1997 SES (see Table 6.28 and Figure 6.16 below) confirms the
hypothesis that children from wealthier families do tend to travel further to
school. Interestingly, the pattern for quintile 1 is quite distinct, while those for
quintiles 2 and 3 are quite similar, as are those for quintiles 4 and 5. There is
also some non-linearity in the relationship between SES and distance,
particularly in quintiles 2 and 3.
SES
Quintile
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1 (most
poor)
232 3.896km 10.559km 0.370km 0.694km 1.380km
2 217 4.789km 8.988km 0.449km 0.842km 4.230km
3 217 4.460km 8.226km 0.465km 0.946km 3.995km
4 225 6.256km 9.430km 0.483km 1.445km 8.654km
5 (least
poor)
187 6.993km 8.865km 0.611km 2.416km 11.549km
Table 6.28: 1997 distance from home to school, by 1997 household SES quintile
182
Figure 6.16: Kernel density plot of 1997 distance from home to school, by 1997
household SES quintile
Census Geography
As is evident in Table 6.29 below, more children become mobile at both the SP
and MP levels of census geography, as SES increases. At the SAL level, the
proportion of children who are mobile is extremely high for all SES groups,
and the chi-squared test does not provide evidence that mobility varies on the
basis of SES. At the MN level so few children are mobile that there is no
difference in behaviour on the basis of SES. Due to the very small numbers of
children involved the MN level is not shown in the table below.
183
Quintile 1 (most
poor)
2 3 4 5 (least
poor)
χ2 test
results
School in
same SAL as
home
23
(9.87%)
11
(5.07%)
16
(7.37%)
17
(7.56%)
13
(6.95%)
χ2(4)= 3.8529
N.S.
School in
same SP as
home
126
(54.08%)
89
(41.01%)
94
(43.32%)
79
(35.11%)
64
(34.22%)
χ2(4)=
23.2285
Pr=0.000
School in
same MP as
home
200
(85.84%)
164
(75.58%)
167
(76.96%)
143
(63.56%)
122
(65.24%)
χ2(4)=
38.2041
Pr=0.000
Table 6.29: 1997 mobility across different levels of census geography, by 1997
household SES
Nearest school
Table 6.30, below, illustrates the proportions of children from each quintile
who attend their nearest school. Although a higher proportion of children from
quintiles 1 and 5 do appear to be attending their nearest school in both
analyses, this is not statistically significant. The U-shaped nature of the
relationship may explain the absence of significant result.
Quintile 1 (most
poor)
2 3 4 5 (least
poor)
χ2 test
results
Attends
nearest
public school
51
(21.89%)
31
(14.22%)
37
(16.97%)
40
(17.70%)
38
(20.32%)
χ2(4)=5.2706
N.S.
Attends
nearest
public or
independent
school
51
(21.89%)
33
(15.21%)
37
(16.97%)
38
(16.81%)
38
(20.65%) χ2(4)=4.6710
N.S.
Table 6.30: Children attending their 1997 nearest grade-appropriate school, by
1997 household SES
6.9.2 2003
Straight-line distance
The relationship between distance from home to school and household SES in
2003 remains very similar to that for 1997 (see Table 6.31 and Figure 6.17
184
below). However, the distances travelled have increased, particularly for
children in the higher quintiles. This may in part relate to the higher likelihood
that more advantaged children have reached high school by 2003. A Kruskal-
Wallis test confirms that distance from home to school varies significantly on
the basis of 2003 SES (Pr=0.0001).
SES
Quintile
Number
of
children
Mean
distance
to school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1 (most
poor)
170 1.422km 2.480km 0.420km 0.749km 1.395km
2 169 4.717km 9.098km 0.560km 1.019km 4.267km
3 172 5.106km 8.476km 0.457km 1.185km 5.145km
4 170 7.664km 9.548km 0.637km 2.443km 12.466km
5 (least
poor)
162 9.501km 9.433km 1.315km 6.392km 14.994km
Table 6.31: 2003 distance from home to school, by 2003 household SES quintile
Figure 6.17: Kernel density plot of 2003 distance from home to school, by 2003
household SES quintile
185
Census Geography
In 2003 there is a fairly straightforward relationship between SES and mobility
at the SP and MP levels (see Table 6.32 below), with children from higher
income families more likely to be mobile. At the SAL level, there is again no
relationship between mobility and SES. At the MN level, by contrast, in 2003
there is a significant relationship (χ2(4)=7.8098, Pr=0.099), with more wealthy
children being more likely to travel between MNs for schooling. However, the
actual numbers of children travelling between MNs remains very small (n=31),
and this level is therefore not included in the table below.
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 χ2 test
results
School in
same SAL as
home
5
(2.94%)
4
(2.37%)
5
(2.91%)
9
(5.29%)
4
(2.47%)
χ2(4)=3.1465
N.S
School in
same SP as
home
95
(55.88%)
63
(37.28%)
61
(35.47%)
46
(27.06%)
33
(20.37%)
χ2(4)=52.6549
Pr=0.000
School in
same MP as
home
154
(90.59%)
128
(75.74%)
123
(71.51%)
95
(55.88%)
75
(46.30%)
χ2(4)=92.3332
Pr=0.000
Table 6.32: 2003 mobility across different levels of census geography, by 2003
household SES
Nearest school
In 2003 there is evidence that the poorest children are significantly more likely
to attend the school nearest to their home (see Table 6.33 below).
186
Quintile
1
Quintile
2
Quintile
3
Quintile
4
Quintile
5
χ2 test
results
Attends
nearest
public
school
41
(24.26%)
26
(15.38%)
28
(16.28%)
27
(15.88%)
18
(11.04%)
χ2(4)=11.0515
Pr=0.026
Attends
nearest
public or
independent
school
40
(23.67%)
25
(14.79%)
26
(15.20%)
25
(14.71%)
17
(10.49%)
χ2(4)=11.6124
Pr=0.020
Table 6.33: Children attending their 2003 nearest grade-appropriate school, by
2003 household SES
6.9.3 Household SES and mobility discussion
In both 1997 and 2003, SES is a strong predictor of the distance between a
child‘s home and school. The relationship becomes slightly more direct in
2003, when children are older, and at least for wealthier children appears to
become stronger. There is also a fairly strong relationship between SES and
mobility at both the SP and MP levels, although not at the SAL or the MN
levels. In 2003, poorer children are more likely to attend their nearest grade-
appropriate schools than their wealthier peers.
As noted in Chapter 4, SES is strongly correlated with maternal education, and
the similarities in the relationships between maternal education and distance
and SES and distance are therefore unsurprising. One of the most interesting
features of this set of analyses is the absence of a significant relationship
between SES and whether or not a child attends their closest school in 1997.
Given the significant nature of the relationship between maternal education
and enrolment at the closest school, this is unexpected. It suggests one way in
which the relationship between mobility and maternal education may differ
from the relationship between mobility and SES. One potential explanation for
these results is that more educated mothers are more likely to engage in school
choice. To the extent that they have access to additional resources (which is
reflected in the SES data), they may choose to access more advantaged schools
187
outside of their immediate area. This is evident in the relationship between
SES and distance travelled. However, to the extent that these mothers do not
have access to additional resources, and have lower SES, they may be
constrained to choose from schools closer to home. However, they do continue
to exercise choice, which is demonstrated by the lower levels at which their
children attend the nearest schools, even when they aren‘t able to travel far.
Whether this is indeed the explanation, and if so, why this effect disappears by
2003, is not clear, and is explored further in Chapter 9.
Community level characteristics
6.10 Residential area poverty
The area in which a child and his or her family live is also likely to influence
school choice, and particularly whether travel over a substantial distance is a
seen as a potentially beneficial option. An area‘s affluence is likely to be
related to, but not identical with, the affluence of any particular household
within this area. A wealthier family is more likely to be able to afford to live in
an affluent area, but within well-off areas, for a range of reasons such as
historical accident, an employer providing somewhere to live, or the rapid
emergence of an informal settlement, a number of comparatively and
absolutely disadvantaged families can typically be found. Similarly, some
wealthy families can typically be found in even the most disadvantaged
communities, choosing to live there for various reasons which might be
historical, social or even economic (lower expenditure on rent, purchasing a
home, or rates and property taxes may allow for greater expenditure on
consumer goods or investments such as education).
An area‘s relative affluence is expected to play a role in the educational
choices available to families, and the costs of those choices. This is particularly
188
true in South Africa, where the affluence of areas is closely connected to their
historical racial designation, which simultaneously also shaped the quality of
the schooling available in those areas. Simply put, historically white areas are
typically more affluent than other areas, and also offer higher quality
educational opportunities locally. A child living in a particularly affluent
community is therefore not likely to need to travel substantial distances to
access a good school. By contrast, a child living in a poor and historically
disadvantaged community is likely to need to travel a great distance. We can
therefore expect that a family‘s SES, which shapes ability to travel, and the
nature of the area in which they live, which shapes the extent to which travel is
beneficial, will operate together in determining levels of learner mobility. As a
result, the more affluent children living in less affluent areas are those expected
to travel the greatest distances.
The relationship between the poverty level of the area in which a child lives,
and the distance that they are likely to travel to school seems likely to be
considerably more complex than any of the other relationships explored thus
far. The nature of the relationship seems likely to depend strongly on the level
of geography which is being considered – SAL, SP or MP. For this reason, the
results will be explored level by level.
6.10.1 1997
Small Area Level (SAL)
Straight-line distance
In the two lowest poverty quintiles of SAL (that is, the wealthiest areas),
children typically travel fairly substantial distances to school (see Table 6.34
and Figure 6.18 below). Children in quintile 3 travel substantially shorter
distances than children in any other quintile. The distances travelled by
children in quintiles 4 and 5 (the poorest areas), are between these two
extremes. Given this non-linear relationship between SAL poverty and
189
distance travelled, it is understandable that the correlation between SAL area
poverty and distance from home to school is fairly low, although negative
overall (-0.0591 for raw scores). Overall, children in wealthier SALs with
lower poverty levels travel somewhat further than others, as confirmed by both
a Kruskal-Wallis test (Pr=0.0001). However, this relationship is clearly not
linear, and particularly worth notice is that the children living in the very
poorest SALs are actually typically travelling further than those in only
moderately poor SALs.
Area
Poverty
Quintile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
213 6.568km 9.206km 0.596km 1.912km 9.348km
2 248 6.957km 10.526km 0.492km 1.345km 9.264km
3 256 3.790km 7.433km 0.422km 0.758km 2.076km
4 259 4.451km 8.453km 0.398km 0.824km 4.285km
5 (poorest) 237 5.966km 11.334km 0.523km 1.177km 5.925km
Table 6.34: 1997 distance from home to school, by SAL poverty quintile
190
Figure 6.18: Kernel density plot of 1997 distance from home to school, by SAL
poverty quintile
Census Geography
There is no evidence for a relationship between the poverty level of the SAL in
which a child lives, and the child‘s likelihood of a attending a school in that
same SAL. Children living in the very poorest and most affluent SAL quintile
are however more likely to be travelling to a school outside their home SP than
children in intermediate SAL quintiles. Children in SAL poverty quintile 3 are
most likely to attend a school in the same MP in which they live
(χ2(4)=14.2914, Pr = 0.006). Finally, children living in the poorest SAL quintile
are most likely to travel to a school outside of the MN in which their home is
(χ2(4)=8.0681, Pr = 0.089). However, given the extremely low levels of MN
mobility, it is not clear that these figures are particularly meaningful.
Nearest school
There is an almost linear relationship between the poverty of the SAL and
whether children attend their nearest school, public or independent. Children in
191
the most affluent SALs are most likely to attend their nearest school, while
children in the poorest SALs are least likely to attend their nearest schools
(χ2(4)= 17.7483, Pr=0.001 for public schools only; χ
2
(4)= 13.3720, Pr=0.01 for
public and independent schools).
Sub Place (SP)
Straight-line distance
Children in quintile 2 and quintile 5 (highest poverty) SP areas have the
greatest distance from home to school, with children in quintiles 3 and 4
travelling particularly short distances (see Table 6.35 and Figure 6.19 below).
Accordingly, the correlation between SP area poverty and distance travelled is
weaker than that for SAL, although it remains negative (-0.0263 for raw
scores). Although children in wealthier SPs are continuing, on average, to
travel slightly further than children in poorer SPs, this relationship is somewhat
weaker than was the case for the SALs (Kruskal-Wallis test, Pr=0.0001).
Area
Poverty
Quintile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
218 5.762km 8.820km 0.495km 1.524km 6.698km
2 247 7.274km 10.490km 0.496km 1.500km 11.237km
3 267 3.504km 6.166km 0.424km 0.843km 3.915km
4 248 3.694km 6.860km 0.413km 0.743km 2.142km
5 (poorest) 233 7.542km 13.246km 0.633km 1.404km 7.608km
Table 6.35: 1997 distance from home to school, by SP poverty quintile
192
Figure 6.19: Kernel density plot of 1997 distance from home to school, by SP
poverty quintile
Census Geography
Children living in poorer SP areas are more likely to be travelling to a school
outside of the SAL in which they live than their peers in more advantaged
areas (χ2(4)=11.1595, Pr = 0.025). Children living in poorer SP areas are also
more likely to be travelling to a school outside of the SP in which they live
(χ2(4)=28.1873, Pr = 0.000), but this relationship is not strictly linear. Children
in quintile 5 (poorest areas), and quintile 2 seem to be far more likely than any
other groups to be mobile at the SP level. Children in SP quintile 5 and 2 are
also substantially more likely to be mobile at the MP level than their peers
(χ2(4)=25.9282, Pr = 0.000). A similar relationship also holds with MN mobility
(χ2(4)= 17.9573, Pr = 0.001), although only a very small number of children are
travelling at this level. Overall, although the relationships are not strictly
linear, children living in the very poorest SPs are more mobile at each level of
census geography than their peers in more affluent areas.
193
Nearest school analysis
Children living in wealthier SP areas are more likely to attend their nearest
school than children living in SP areas with higher poverty (Kruskal-Wallis
test; Pr=0.0001 regardless of whether independent schools are included).
Main Place
As discussed in Chapter 3, due to extreme clustering only 3 poverty quantiles
are used at the MP level, and even these are very uneven, meaning that the
findings presented in the section should be treated with some caution.
Straight-line distance
Children living in the poorest MP areas live the furthest distance away from
their schools (see Table 6.36 and Figure 6.20 below; Kruskal-Wallis test,
Pr=0.0155). Although this is a different result from that found at the SAL and
SP level, it is not clear whether it is a feature of the larger area size considered,
or a function of grouping children into three quantiles as opposed to five
quintiles.
MP Area
Poverty
Quantile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
394 5.349km 9.299km 0.447km 0.909km 5.303km
2 705 5.453km 9.580km 0.455km 1.040km 5.816km
3 (poorest) 115 6.233km 9.983km 0.699km 1.554km 7.608km
Table 6.36: 1997 distance from home to school, by MP poverty quantile
194
Figure 6.20: Kernel density plot of 1997 distance from home to school, by MP
poverty quantile
Census geography
Children living in the most affluent MP areas are most likely to attend school
in the SAL in which they live (χ2(2)=16.2439, Pr = 0.000). These children are
also more likely to attend school in the SP in which they live (χ2(2)= 24.4440,
Pr = 0.000), and also in the MP in which they live (χ2(2)= 16.6547, Pr = 0.000).
At the MN level, however, children in MP poverty quintile 2 are substantially
more likely to travel between MNs than their peers living in more affluent or
poorer MPs (χ2(2)= 27.0329, Pr = 0.000).
Nearest School analysis
Chi-square tests indicate that children living in more advantaged MPs are
significantly more likely to be attending their nearest schools (χ2(2)=34.0264, Pr
= 0.000 for public schools only; χ2(2)= 28.5264, Pr = 0.000 for public and
independent schools).
195
6.10.2 2003
Small Area Level (SAL)
Straight-line distance
In 2003, it is the quintile of children living in the poorest SALs that travel the
shortest distances to school, although children in SAL poverty quintile 3 only
travel slightly further (see Table 6.37 and Figure 6.21 below). The change in
the relative travel of children living in the poorest SALs may relate to the
opening of new schools, these children being less likely to have reached high
school, or more accurate data for these children. The correlation between SAL
poverty and distance travelled remains weakly negative (-0.0721 for raw
scores), and a Kruskal-Wallis test indicates that distance to school varies
significantly with SAL poverty level (Pr= 0.0053).
Area
Poverty
Quintile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
223 6.006km 8.949km 0.664km 1.720km 7.410km
2 252 6.439km 9.058km 0.534km 1.629km 10.246km
3 270 4.910km 8.693km 0.464km 0.873km 5.387km
4 271 5.268km 7.604km 0.631km 1.214km 7.796km
5 (poorest) 264 4.304km 7.946km 0.669km 1.247km 3.276km
Table 6.37: 2003 distance from home to school, by SAL poverty quintile
196
Figure 6.21: Kernel density plot of 2003 distance from home to school, by SAL
poverty quintile
Census geography
Children living in more affluent SAL areas are more likely to attend a school
in the same SAL as their home than children living in poorer SAL areas (χ2(4)=
11.1184, Pr = 0.025). There is no significant relationship between SAL poverty
and mobility at the SP, MP or MN levels, however.
Nearest school
Children living in more affluent SALs are significantly more likely to attend
their nearest school (χ2(4)= 16.2545, Pr = 0.003 for public schools only, and
χ2(4)= 13.7166, Pr = 0.008 for public and independent schools).
Sub Place (SP)
Straight-line distance
As evident in Table 6.38 and Figure 6.22, children living in poorer SP areas
travel shorter distances to school than their peers living in more affluent SP
197
areas (Kruskal-Wallis test; Pr=0.0133). A weakly negative correlation between
SP poverty and distance travelled remains in place (-0.0455 for raw scores).
Area
Poverty
Quintile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
223
5.314km
8.712km 0.535km 1.263km 5.391km
2 264 6.345km 8.666km 0.635km 1.742km 9.920km
3 283 5.636km 9.370km 0.522km 1.020km 7.425km
4 258 4.393km 6.912km 0.572km 1.010km 4.667km
5 (poorest) 252 5.011km 8.340km 0.678km 1.421km 5.916km
Table 6.38: 2003 distance from home to school, by SP poverty quintile
Figure 6.22: Kernel density plot of 2003 distance from home to school, by SP
poverty quintile
Census geography
Again, children living in more affluent SP areas are more likely to attend
school within the same SAL in which they live than their peers in poorer SP
areas (χ2(4)= 8.0401, Pr = 0.090). A similar pattern is also found for mobility at
198
the SP (χ2(4)= 22.6349, Pr = 0.000), MP (χ
2
(4)= 25.5792, Pr = 0.000) and MN
(χ2(4)= 8.4396, Pr = 0.077) levels.
Nearest school
Children living in the most affluent SPs are again more likely to attend their
nearest school than those in poorer SPs (χ2(4)= 21.7449, Pr = 0.000 for public
schools only; χ2(4)=19.7176 Pr = 0.001 for public and independent schools).
Main Place (MP)
Straight-line distance
At the MP level, the 1997 finding that the children living in the poorest MPs
tend to travel further than children in more affluent MPs is replicated (see
Table 6.39 and Figure 6.23 below). Children in the wealthiest MPs travel the
least far, children in mid-range MPs travel somewhat further, and the children
in the highest-poverty MPs travel furthest of all (Kruskal-Wallis test,
Pr=0.053). There is a positive correlation between MP poverty score and
distance travelled (0.0824), indicating that as MP poverty increases, so too
does distance travelled. The fact that the 1997 finding is replicated here
suggests that MP area poverty might have a different relationship with mobility
than poverty at smaller area levels.
MP Area
Poverty
Quantile
Number
of
children
Mean
distance
to
school
Standard
deviation
25th
percentile
distance
50th
percentile
distance
75th
percentile
distance
1
(wealthiest)
413 4.809km 8.737km 0.500km 1.106km 4.300km
2 743 5.707km 8.504km 0.635km 1.293km 7.377km
3 (poorest) 125 5.070km 7.145km 0.653km 1.572km 7.583km
Table 6.39: 2003 distance from home to school, by MP poverty quantile
199
Figure 6.23: Kernel density plot of 2003 distance from home to school, by MP
poverty quantile
Census geography
Children living in more affluent MP areas are more likely than their peers in
poorer MP areas to attend a school in the same SAL (χ2(2)= 13.3806, Pr =
0.001), SP (χ2(2)= 20.7043, Pr = 0.000), MP (χ
2
(2)= 27.7652, Pr = 0.000) and
MN (χ2(2)= 13.5678, Pr = 0.001) as their homes in 2003.
Nearest school
Children living in more affluent MPs are significantly more likely to attend
their nearest school in 2003 than their peers living in less affluent MP areas
(χ2(2)= 39.5873 Pr = 0.000 for public schools only; χ
2
(2)= 37.8819, Pr = 0.000
for public and independent schools).
6.10.3 Discussion of residential area poverty and mobility
Two main streams of findings emerge from all the above analyses of
residential area poverty. Firstly, there is a fairly complex relationship between
200
the area a child lives in, and his or her education-related mobility The exact
nature of this relationship is not clear from the analyses documented above,
although the overall trend seems to be that on average children living in
relatively advantaged areas, as well as those in the most disadvantaged areas,
tend to travel furthest. Secondly, however, there is a strong and straightforward
relationship between the area a child lives in, and the likelihood that he or she
will attend the nearest grade appropriate school, whether public or
independent.
The patterns evident in the first set of findings seem likely to relate to complex
interactions between neighbourhood poverty and a range of household and
individual characteristics. Overall, though, with the exception of some of the
nearest-school analyses, children living in wealthier areas do seem to be more
mobile than children living in poorer areas. As these children are, on average,
more affluent than the children from poorer areas, it makes sense that their
parents or families are more likely to have the resources (economic and social),
to engage in mobility. However, children living in poorer areas, with poorer
local schools, almost certainly have far stronger incentives to engage in
mobility. The patterns of mobility revealed by the analyses presented here
seem most likely to represent this interaction between incentives and capacities
for mobility.
The second set of findings is simpler to explain, and almost certainly relates to
the higher quality of schools in more affluent areas, regardless of the
geographic level at which these areas are defined. Children in more affluent
areas therefore almost certainly have less of an incentive to choose to attend a
school other than the one closest to their home. Two alternative explanations
should, however, also be considered. Firstly, in affluent areas, which are
generally more spread out, a child is more likely to be faced with having only
one school within easy walking distance of his or her home, and may,
therefore, be constrained to attend this particular school in a way that children
201
in less affluent areas with a denser distribution of schools are not constrained.
However, given that there is a fairly strong relationship between household
SES and the affluence of the area in which a child lives, it seems likely that
this sort of constraint should not affect many children living in more
advantaged areas. Secondly, and particularly in 1997 when the children are
younger, is the fact that in affluent areas most schools are English or Afrikaans
medium, while in less affluent areas, they may be operating in any one of
several African languages that are heavily represented in the Johannesburg-
Soweto area. While parents may consider education in English or Afrikaans
even if these are not the home language, they may not be willing to accept
education in an African language other than the child‘s home language. This
may, to some extent, be governing decision making around which school
children will attend, and whether the nearest school to the home is an option.
The finding that children in less affluent areas are less likely to attend their
nearest school – even if they still don‘t travel very far – is of particular
importance. It appears to indicate that children and families living in poor areas
still exercise school choice as far as they are able to, given the resource
constraints that they face.
6.11 Conclusion
This chapter has explored the relationship between a range of individual,
household and community level variables, and learner mobility in South
Africa. At the individual level, race, gender, age at first enrolment, and grade
repetition were examined. Clear evidence was presented that race is strongly
linked to mobility behaviour, and particularly that coloured children are less
likely to engage in mobility than black children. Although girls travel slightly
further than boys, there was not sufficient evidence to demonstrate a consistent
relationship between gender and mobility.
202
Interestingly, and contrary to expectations, children who started school in
1997, as opposed to 1996, tended to be substantially more mobile in 1997.
While the reasons for this are unclear, the most plausible relates to whether or
not parents can afford preschool or other child care if a child is not attending
school. This effect had largely worn off by 2003, perhaps in part due to the
effect of late starters being more likely to have reached secondary school in
2003. There was a strong relationship between mobility and repetition, with
repetition being much less likely for children travelling further. This is almost
certainly a reflection of household SES, which was higher with greater travel,
and the fact that for most children, travelling further is associated with
attending a higher quality school.
At the household level, maternal education, maternal marital status, and
household SES were explored. A strong, positive relationship between
mobility and maternal education was identified. In particular, children with
mothers who had an education up to and including grade 10 had a very
different distribution of mobility from those children with mother who had
completed grade 11 or higher. Although much of the relationship between
maternal education and learner mobility is likely to be mediated by SES, there
was also some evidence for an independent effect of maternal education on
school choice, particularly in 1997. In particular, children with more educated
mothers were substantially less likely to attend their local schools, suggesting
that more educated mothers were more actively engaged in school choice.
There was, however, very limited evidence for a relationship between maternal
marital status and mobility behaviour.
As anticipated, a particularly strong relationship between mobility and socio-
economic status was identified. Although this relationship was not strictly
linear, children from more affluent households generally tended to live further
away from their schools. This relationship appears to become stronger over
time, with SES being more closely related to mobility in 2003.
203
At the community level, measures of residential area poverty were explored at
three different levels of census geography. The relationship between these
measures and mobility behaviour is relatively complex, which is probably due
to the interaction of area and household SES. While children living in poorer
areas are likely to have a greater incentive to travel to school, they are less
likely to have the resources to do so. By contrast, children in more affluent
areas are less likely to need to travel to attend a school of their preference, but
are much more likely to have the ability to do so. Overall, however, the
children living in relatively advantaged areas, and those living in the most
disadvantaged areas appear to be most likely to travel substantial distances,
and to travel outside of their residential area to attend school. A more clear
relationship exists between area poverty and whether or not a child attends
their nearest grade-appropriate school – children living in more affluent areas
are substantially more likely to attend the school nearest to their home.
The data presented in this chapter substantiates the notion that there are two
different forms of school choice operating in the South African educational
market. Firstly, families choose to send children to relatively distant but
historically advantaged schools whenever possible. This is evident in the
substantially greater distances typically travelled by the more affluent members
of the sample, and in the extent to which the relationships between mobility
and its correlates seem likely to be shaped by SES.
By contrast, however, there is also evidence that less-advantaged children and
their families are also engaged in choice, even if this does not typically involve
substantial travel to historically advantaged schools. The prevalence of choice
at a more local level is evident in the data on which children attend their
nearest school. Unlike the distance travelled to school, this appears to be
inversely associated to family wealth. By contrast, maternal education and area
poverty play a more significant role in shaping this form of choice. This
204
suggests that when sending children to historically advantaged schools is not
an option, engaged parents, and particularly those in less affluent areas,
attempt to provide their children with the best possible education by choosing
from amongst local schools.
In the next chapter, Chapter 7, the relationship between mobility and school
attributes will be explored. This will be followed in Chapter 8 by an
investigation of the trajectories of individual learners with respect to mobility
over time. This will provide insight as to whether mobility in 1997 can be used
to predict mobility in 2003, as well as to potential determinants and correlates
of changes in mobility patterns over time. Chapter 9 will then combine the data
presented in the current chapter with the findings of Chapters 7 and 8 to
develop a preliminary, partial model of the determinants of various forms of
learner mobility.
205
Chapter 7: School characteristics
associated with mobile learners
7.1 Introduction
Along with the individual, household and community level variables discussed
so far, a child‘s mobility is also likely to be related to the attributes of the
schools available to the child, both locally and further afield. School attributes,
such as resource levels, educational quality, cost, and racial composition are all
likely to play various roles in attracting or deterring potential learners. It
therefore makes sense to explore key variables of both the schools local to a
child, and the schools a child attends, in attempting to understand the mobility
of any particular child.
This chapter consists of three sections. The first section explores the attributes
of those schools attended by sample members from 1997-2003, along with the
attributes of those schools closest to the homes of sample members. Data is
presented first in unweighted form, and secondly, weighted by the number of
children enrolled in each school. The second section explores the distribution
of sample members across the schools attended, and provides data on which
children attended schools with which properties. Finally, the third section
explores the relationships between patterns of travel, and the schools that
children attended.
206
7.2 Schools attended by study sample members, and
grade-appropriate schools closest to study sample
members’ homes
7.2.1 Unweighted data
Obviously, the schools attended by the members of the sample used in this
study are only a subset of all of the schools in the Gauteng province described
in Chapter 4. They are not likely to be an extremely representative subset
either, due to the urban location of sample members, and the under-
representation of the extremely poor and the extremely wealthy, as discussed
in Chapter 4. For these reasons, it is useful to generate some descriptive data
specific to the group of schools attended by sample members. For similar
reasons, descriptive data around the subset of schools which comprise the
nearest grade-appropriate school for sample members will also be generated. It
must be stressed that the data provided in this section is not weighted for the
number of cohort members enrolled at each school. Rather, all schools
attended by cohort members are weighted equally here – whether they are
attended by just one cohort member, or many. Data weighted by the number of
sample members enrolled at each school will be explored in the subsequent
section.
Overall, in 1997, the 1428 members of the study were attending a total of 365
different registered schools. However, as data is missing for a few individuals,
and a small number of children also attended unregistered schools, this figure
is probably slightly low. In 2003, this had risen to 465 schools. The figure is
higher because both primary schools (310) and secondary schools (155) are
included at this point in time. 378 different primary schools are identified as
being the nearest primary school to members of the study sample, while 212
secondary schools are identified as being the nearest secondary school to
members of the study sample. These figures are consistent with the larger
number of relatively smaller primary schools found in the Gauteng province,
207
and the smaller number of relatively larger secondary schools. From these
figures, it is also evident that the number of schools actually attended is, for
both phases, smaller than the number of schools that would be attended if all
children simply attended their nearest school.
School sector
10.54% of the schools attended by sample members during the study period
were independent schools. By contrast, roughly 20% of the registered schools
in Gauteng were independent. The schools nearest to the homes of sample
members represent far more closely the distribution of independent schools
found in the full list of registered schools. Roughly 10% of the primary schools
closest to a sample member‘s home were independent, as were 17% of
secondary schools.
Quintile
Very few of the primary and secondary schools closest to sample members‘
homes fall into quintiles 1 (poorest), 2 and 5 (most advantaged). As evident in
Table 7.1, the majority of the schools closest to sample members‘ homes are in
either quintile 3 or quintile 4 – in both cases a higher proportion than expected
given the proportion of quintile 3 and 4 schools found in Gauteng province as a
whole. The proportion of schools actually attended by cohort members that fall
into quintiles 1-3 is lower than would be expected on the basis of the schools
nearest to sample members‘ homes. By contrast, the proportion of attended
schools falling into quintiles 4 and 5 is somewhat higher than would be
expected.
208
School
Quintile
Proportion
of Gauteng
schools
Proportion of
primary
schools
closest to
sample
members’
homes
Proportion
of
secondary
schools
closest to
sample
members’
homes
Proportion
of schools
attended by
sample
members in
1997
Proportion
of schools
attended by
sample
members in
2003
1 11.66% 3.08% 2.37% 2.85% 1.46%
2 8.61% 8.31% 11.83% 5.38% 5.84%
3 30.91% 44.62% 35.50% 37.66% 36.50%
4 27.27% 35.08% 37.28% 38.61% 41.36%
5 21.55% 8.92% 13.02% 15.51% 14.84%
Table 7.1: Quintiles of schools attended by and nearest to sample members’ homes
Section 21 status
Just over 86% of public schools in Gauteng province have Section 21 status,
while just under 89% of the schools closest to sample members‘ homes have
this status. By contrast, around 95% of schools actually attended by sample
members having Section 21 status.
Enrolment
In line with the data presented for all Gauteng, the primary schools closest to
the homes of sample members are, on average, substantially smaller (mean:
626; median: 558) than the secondary schools closest to their homes (mean:
971; median: 1049). By contrast, the mean size of the schools attended 1997 is
663, while for primary schools attended in 2003 it rises to 673 – in both cases
higher than might be expected on the basis of the schools nearest to sample
members‘ homes. The mean size of secondary schools attended by sample
members is 958, which is lower than would be expected on the basis of the
data on the secondary schools nearest to sample members‘ homes.
209
Percentage of black learners
The average proportion of black learners across all Gauteng schools is 73%.
The average for the primary schools closest to sample members‘ homes is
83%, and for secondary schools it is 76%. In both cases, the median proportion
of black students is almost or exactly 100%, indicating that around or over half
of the schools found closest to participants‘ homes are entirely black. In 1997,
the average percentage of black learners at the schools attended by sample
members was 79%. For primary schools attended in 2003, this figure rises to
81%. By contrast, the corresponding figure for the secondary schools attended
in 2003 is 72%. In all cases, these figures are lower than would be expected on
the basis of the schools closest to sample member homes.
School fees
The school fees charged by both the primary and secondary schools closest to
sample members‘ homes are substantially lower than those figures for all
Gauteng area schools, with the primary schools charging an average of R504,
and the secondary schools and average of R790. However, the fees at the
schools which sample members actually attended are substantially higher –
although with extremely large standard deviations. The primary schools
attended in 1997 charging an average of R910 (minimum R0, maximum
R9510, median R80). Primary schools attended in 2003 charged an average of
R769 (minimum R0, maximum R7000, median R100). Secondary schools
attended in 2003 charged an average of R1470 (minimum R0, maximum
R9650, median R400).
Historical racial status of school
70% of the primary schools closest to participant homes were historically DET
schools, as were 63% of secondary schools. The schools actually attended by
sample members, however, are substantially less likely to be historically DET
schools. In 1997, 56% of schools attended were DET schools, and in 2003, this
210
figure was lower still at 54%. These enrolment patterns suggest that children
are tending to avoid DET schools.
Matric pass rate
Given that the pass rate data attributed to primary schools was imputed on the
basis of the performance of the nearest secondary school, it makes sense that
the figures for the groups of primary and secondary schools closest to sample
members‘ homes will be almost identical. The mean pass rate for these
secondary schools is 70%, while the mean imputed pass rate for these primary
schools is 68%. The figures for the schools actually attended are quite similar,
with the mean pass rate for schools attended in 1997 at 70%. Primary schools
attended in 2003 also have a mean pass rate of 70%, while for secondary
schools it is 72%.
Discussion of unweighted data
The data presented thus far suggests that, as would be expected, the schools
closest to study sample members‘ homes tend to be slightly more
disadvantaged than the average schools in the Gauteng province. The schools
actually attended, however, are somewhat more advantaged than this. In the
following section, the schools data weighted for attendance is explored to
determine whether this pattern still holds.
7.2.2 Weighted data
In this section, the focus shifts from using schools to using sample members as
the unit of analysis. It will describe the school environments experienced by
different proportions of sample members by presenting school data weighted
by the number of sample members attending a particular school.
Although the 1428 sample members attended a total of 365 different schools in
1997, and 465 different schools in 2003, they were by no means evenly
211
distributed across these schools, as illustrated in Table 7.2 below. In 1997, 132
schools had only one sample member attending, while 11 different schools had
10 or more sample members attending. The single most-attended school had 20
sample members enrolled. The largest proportion of sample members, 14.02%,
attended schools with 6 sample members enrolled. By contrast, relatively even
proportions of the sample live closest to each of the school in the set of schools
closest to any sample member‘s home.
In 2003, with the sample members divided between both primary and
secondary schools, the distribution of enrolment shifted towards smaller
numbers of children at a range of different schools. This trend is particularly
evident amongst children attending secondary schools in 2003, with only
7.67% of secondary school children attending schools with ten or more sample
members. By contrast, almost 70% of the sample lives closest to one of 57
secondary schools which are also closest to at least 9 other sample members.
This probably relates to the typically larger size, and sparser distribution, of
secondary schools. This data suggests that as learners move into secondary
schooling we should expect to see larger numbers of children enrolled at each
of a smaller number of schools. It is interesting to note that the available data,
however, appears to reflect an opposite trend. However, this may simply be
due to the relatively small numbers of children who have reached secondary
school in 2003, and data for subsequent years is needed before any firm
conclusions can be drawn.
Sample
members per
school
Nearest
primary
Nearest
secondary
1997
attended
2003
attended,
primary
2003
attended,
secondary
% at schools
with 1-4
sample
members
38.59%
(272
schools)
13.24%
(119
schools)
40.21%
(264
schools)
56.03%
(280
schools)
44.19%
(110 schools)
% at schools
with 5-9
sample
members
34.37%
(76
schools)
17.22%
(35 schools)
48.19%
(90 schools)
38.31%
(65
schools)
48.13%
(40 schools)
212
% at schools
with over 10
sample
members
27.03%
(28
schools)
69.54%
(57 schools)
11.6%
(11 schools)
5.16%
(6 schools)
7.67%
(5 schools)
Table 7.2: Schools attended by sample members (note: schools classified as
combined are included in both columns for 2003)
Sector
The proportion of children living nearest to and attending independent schools
is lower than suggested by the unweighted data presented earlier. Overall, in
1997, 7.09% of sample members were attending independent schools, as
illustrated in Table 7.3 below. A slightly higher proportion of children had
independent schools as their nearest primary (7.60%) and secondary schools
(8.23%). The relatively low proportion of sample members attending
independent schools may relate to a number of different factors, including the
typically smaller size of independent schools, the poor quality of some
independent schools in disadvantaged areas, and the relatively low
representation in this study sample of the extremely affluent children most
likely to attend the more highly performing independent schools.
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended
school 03 –
primary
Attended
school 03 –
secondary
% of sample at
independent
schools
7.60% 8.23% 7.09% 7.00% 7.67%
Table 7.3: Proportion of sample members closest to and attending independent
schools
Quintiles
The proportion of children attending schools in quintiles 1, 2 and 5 is lower
than would be expected on the basis of the proportions of schools of each
quintile in Gauteng, while the proportion of children in quintile 3 and 4
schools, by contrast, is higher than expected. When contrasting quintile ratings
of schools attended with nearest schools (see Table 7.4 below), it is again clear
213
that the proportion of children attending schools in quintiles 1 and 2 is lower
than would be expected on the basis of the schools nearest to sample members‘
homes. This time, however, the proportion attending quintile 3 schools is also
lower, while the proportions attending schools in quintiles 4 and 5 are higher.
School
quintile
Nearest
primary
school
Nearest
secondary
school
Sample
members
by quintile,
1997
Sample
members by
quintile, primary
school only 2003
Sample members
by quintile,
secondary school
only 2003
1 3.35% 1.93% 1.43% 1.37% 0.00%
2 6.69% 4.59% 3.84% 3.98% 4.03%
3 50.76% 53.51% 46.47% 42.61% 38.79%
4 35.46% 35.86% 39.68% 42.61% 48.36%
5 3.75% 4.11% 8.58% 9.44% 8.82%
Table 7.4: Distribution of sample members across schools by school quintile rating
Section 21 Status
In both 1997 and 2003, only slightly over 4% of learners were attending
schools without Section 21 status (see Table 7.5 below). These figures are
lower than the proportion of learners whose nearest primary school did not
have Section 21 status (6.82%), and substantially lower than the proportion
whose nearest secondary school did not have that status (17.42%).
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended
school 03 –
primary
Attended
school 03 –
secondary
% of sample at
schools without
Section 21 status
6.82% 17.42% 4.34% 4.53% 4.20%
Table 7.5: Proportion of schools without Section 21 status nearest to and attended
by sample members
Total school enrolment
As larger schools enrol more children, it is expected that a higher proportion of
our sample members will be attending larger schools. As a result, the school
size experienced our average learner should be higher than the size of the
214
average school available to him or her. This is indeed reflected in the data. The
mean primary school size experienced by sample members in 1997 was 675,
and 720 in 2003, both of which are higher than the average size (663) of
attended primary schools. The mean school size experienced by sample
members attending secondary school in 2003 is 1060 learners, compared to a
mean attended secondary school size of 971. A similar argument applies to the
schools nearest to children‘s homes, and this is also reflected in the data. The
average sample member‘s closest primary school has just over 647 children,
compared to an average of 626 learners for all the primary schools nearest to
sample members‘ homes. Similarly, the average sample member‘s nearest
secondary school enrols 1016 children, compared to an average school size of
971.
As is evident in Table 7.6 below, the mean school size experienced by a
sample member in 1997 is a little larger than would be expected on the basis of
the mean size of the closest primary school. The figure for those children still
in primary school in 2003 is even higher. The mean size of secondary schools
attended in 2003 is also a little higher than would be expected on the basis of
the mean size of the secondary schools closest to the children‘s homes.
Nearest
primary
school size
Nearest
secondary
school size
Attended
school 1997
Attended
school 03 –
primary
Attended
school 03 –
secondary
Mean school
size
647.17 1016.13 674.75 719.72 1060.35
Median
school size
563 1090 641 694 1131.5
Table 7.6: Size of schools nearest to and attended by sample members
Proportion Black
For the average child in the sample, both the nearest primary and secondary
schools were 86% black, while for the median child both schools were 100%
black, as illustrated in Table 7.7 below. These figures are higher than would be
215
predicted from the data for all Gauteng schools, as well as from the figures for
those schools that sample members actually attended. These figures do,
however, make sense, as the majority of sample members lived in majority
black areas, and majority black schools were more likely to have multiple
sample members enrolled, particularly at the secondary school level.
The figures for the schools actually attended in 1997 and in 2003 are
substantially lower than those for the nearest schools, suggesting that on
average, children are attending schools with a lower proportion of black
learners than those nearest to their homes. This difference is particularly
marked for those children who have reached secondary school by 2003.
However, as this is a non-representative sub-section of the sample, it is not
possible to determine whether this larger difference is attributable to their
being in secondary school, or to their being more advantaged or academically
more promising than their peers still in primary school.
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended
school 03 –
primary
Attended
school 03 –
secondary
Mean % black
learners
86.29 85.78 81.98 81.07 77.25
Median % black
learners
100 100 100 100 99.75
Table 7.7: Proportion of black learners at schools nearest to and attended by
sample members
School Fees
Table 7.8 below illustrates the fees charged at the primary and secondary
schools nearest to, and attended by, the average sample member, for both 1997
and 2003. The fees charged at the schools attended by sample members are
higher than would be predicted on the basis of the fees charged by the schools
nearest to their homes. This divergence is particularly notable at the secondary
school level, although again it is not clear whether this is due to the nature of
216
those children in secondary school in 2003, or whether it simply relates to the
fact of their attending secondary schools.
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended
school 2003 –
primary
Attended
school 2003 –
secondary
Mean school
fees
R255 R358 R564 R654 R812
Median
school fees
R50 R100 R60 R100 R150
Table 7.8: Fees charged by schools nearest to and attended by sample members
Historical racial status of school
Table 7.9, below, illustrates that a smaller proportion of sample members
attended ex-DET schools than would be predicted on the basis of the schools
closest to their homes. The proportion attending DET schools falls between
1997 and 2003, and is lowest for those children attending secondary schools in
2003.
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended school
2003 – primary
Attended school
2003 – secondary
% of sample at
ex-DET schools
76.72% 76.23% 63.01 58.39% 54.50%
Table 7.9: Proportion of schools nearest to and attended by sample members that
were historically under the DET
Matric Pass Rate
Once again, the average sample member attends a school with better pass rates
than would be predicted on the basis of the school closest to their home (see
Table 7.10 below). Also, similarly to other variables explored, this difference
becomes more marked at the secondary school level, although, as indicated
previously, the reasons for this increase cannot be determined from the
available data.
217
Nearest
primary
Nearest
secondary
Attended
school 1997
Attended school
2003 – primary
Attended school
03 – secondary
Mean pass
rate
64% 64% 66% 67% 70%
Median pass
rate
63% 63% 66% 66% 69%
Table 7.10: Pass rates of schools nearest to and attended by sample members.
Discussion of weighted data
A few key themes emerge from the data presented here. Firstly, it is clear that
across all variables, on average, a sample member attends a school that is more
advantaged than would be predicted on the basis of the school nearest to his or
her home. Secondly, across all variables, this difference is greater at the
secondary school level, although the reasons for this are not clear. Thirdly,
those children still attending primary school in 2003 are typically attending
schools that more advantaged than those that were attended in 1997. These
findings suggest that to the extent that school choice is being used by sample
members, it is being used to enhance the educational opportunities available to
sample members.
7.3 Which children attend which schools?
This section explores the distribution of sample members across schools with
different properties on the basis of child, household and community attributes.
Tests are conducted using data for both 1997 and 2003.
7.3.1 Child level variables
Race
Table 7.11, below, shows that schools attended by children of different races
vary significantly on the basis of the majority of the school attribute variables
examined, with black children typically more likely to be attending lower
quality schools.
218
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector n.s. Pr=0.008
(Black children
more likely to
attend public
schools)
n.s. Pr=0.013
(Black children
more likely to
attend public
schools)
Quintile Pr=0.0000
(Black children
more likely to
attend lower
quintile
schools)
Pr=0.0000
(Black children
more likely to
attend lower
quintile
schools)
Pr=0.0000
(Black children
more likely to
attend lower
quintile
schools)
Pr=0.0000
(Black children
more likely to
attend lower
quintile
schools)
Section 21 Pr=0.026
(Black children
less likely to
attend Section
21 schools)
n.s. Pr=0.070
(Black children
less likely to
attend Section
21 schools)
Pr=0.004
(Black children
more likely to
attend Section
21 schools)
School size Pr=0.0000
(Black children
are more likely
to be enrolled
in smaller
schools)
Pr=0.0000
(Black children
are more likely
to be enrolled
in smaller
schools)
Pr=0.0000
(Black children
are more likely
to be enrolled
in smaller
schools)
Pr=0.0000
(Black children
are more likely
to be enrolled
in smaller
schools)
Proportion
black
Pr=0.0000
(Black children
likely to attend
schools with a
higher
proportion of
black learners)
Pr=0.0000
(Black children
likely to attend
schools with a
higher
proportion of
black learners)
Pr=0.0000
(Black children
likely to
attend schools
with a higher
proportion of
black learners)
Pr=0.0000
(Black children
likely to attend
schools with a
higher
proportion of
black learners)
School fees Pr=0.0000
(Black children
more likely to
attend schools
with lower
fees)
Pr=0.0000
(Black children
more likely to
attend schools
with lower
fees)
Pr=0.0000
(Black children
more likely to
attend schools
with lower
fees)
Pr=0.0000
(Black children
more likely to
attend schools
with lower
fees)
Historical DET
status
Pr=0.000
(Black children
were more
likely to attend
a historically
DET school)
Pr=0.000
(Black children
were more
likely to attend
a historically
DET school)
Pr=0.000
(Black children
were more
likely to
attend a
historically
DET school)
Pr=0.000
(Black children
were more
likely to attend
a historically
DET school)
Matric Pass Pr=0.0299 n.s. n.s. n.s.
219
rate (Black children
more likely to
attend schools
with higher
pass rates)
Table 7.11: Relationship between child race (black and coloured children only) and
properties of the school he or she attends
Gender
As illustrated in Table 7.12 below, only two school properties are significantly
related to child gender in 1997: boys are more likely to attend schools with a
higher proportion of black learners, as well as schools that are former DET
schools. In 2003, a number of significant relationships are evident, but these
appear to relate largely to the higher proportion of girls attending secondary
school by 2003, and all disappear when school phase is controlled for. The one
exception to this pattern is school sector, with girls being significantly more
likely to attend independent secondary schools.
Overall, there is weak evidence that during the early years of schooling, girls
may be more likely to attend more integrated schools than boys, and less likely
to attend former DET schools. These relationships do not survive in 2003,
when school phase is controlled for. Girls are, however, substantially more
likely to have reached secondary school in 2003, which may be obscuring
relationships between gender and enrolment patterns at the secondary school
level .
1997 2003 (all data) 2003 (primary
schools only)
2003 (secondary
schools only)
Sector n.s. n.s. n.s. Pr = 0.030 (Girls
more likely to
attend
independent
schools)
Quintile n.s. Pr=0.0468
(Girls more
likely to attend
n.s. n.s.
220
higher quintile
schools)
Section 21 n.s. n.s. n.s. n.s.
School size n.s. Pr=0.0173
(Girls more
likely to attend
larger schools)
n.s. n.s.
Proportion
black
Pr=0.0677
(Girls more
likely to
attend schools
with lower
proportion of
black learners)
Pr=0.0562
(Girls more
likely to attend
schools with
lower
proportion of
black learners)
n.s. n.s.
School fees n.s. Pr=0.0293
(Girls more
likely to attend
schools with
higher fees)
n.s. n.s.
Historical
DET status
Pr=0.054
(Girls less
likely to
attend former
DET schools)
Pr=0.060 (Girls
less likely to
attend former
DET schools)
n.s. n.s.
Matric Pass
rate
n.s. n.s. n.s. n.s.
Table 7.12: Relationship between child gender and properties of the school he or
she attends
Age at enrolment
The 1997 analysis shows strong evidence for a relationship between the age of
school enrolment and the type of school attended, with children starting school
at a later age attending more advantaged schools (see Table 7.13 below). This
is in line with the finding in Chapter 6 that children who start school later are
more likely to live further away from their schools. Given the extremely strong
relationship between age at school enrolment and the child‘s phase of
schooling in 2003, results for 2003 should be treated with some caution,
although they are largely in line with the 1997 results. Overall, however, this
data suggests that children who start school at a later age tend to access more
highly performing schools.
221
Age at first
school
enrolment
1997 2003 (all data) 2003 (primary
schools only)
2003 (secondary
schools only)
Sector Pr=0.001
(children
starting
school late
less likely to
attend public
schools)
n.s. Pr= 0.067
(children
starting school
late less likely
to attend
public
schools)
Pr=0.000
(children starting
school late less
likely to attend
public schools)
Quintile Pr=0.000
(children
starting
school late
more likely to
attend
quintile 5
schools)
Pr=0.011
(children
starting school
late more
likely to
attend quintile
5 schools)
Pr=0.077
(children
starting school
late more
likely to
attend
quintile 5
schools)
n.s.
Section 21 n.s. n.s. n.s. Pr=0.008
(children starting
school late less
likely to attend
Section 21
schools)
School size n.s. Pr=0.0000
(children
starting school
late likely to
attend smaller
schools)
n.s. Pr=0.0621(childr
en starting
school late likely
to attend smaller
schools)
Proportion
black
Pr=0.0019
(children
starting
school late
likely to
attend
schools with a
lower
proportion of
black
learners)
n.s. Pr=0.0779
(children
starting school
late likely to
attend schools
with a lower
proportion of
black learners)
n.s.
School fees Pr= 0.0000
(children
starting
school late
likely to
Pr=0.0979
(children
starting school
late likely to
attend schools
Pr=0.0151
(children
starting school
late likely to
attend schools
n.s.
222
attend
schools with
higher fees)
with higher
fees)
with higher
fees)
Historical
DET status
P=0.000
(children
starting
school late
less likely to
attend former
DET schools)
Pr=0.058
(children
starting school
late less likely
to attend
former DET
schools)
n.s. n.s.
Matric Pass
rate
Pr= 0.0007
(children
starting
school late
more likely to
attend
schools with
higher pass
rates)
n.s. n.s. n.s.
Table 7.13: Relationship between child age at first school enrolment and school
properties
Repetitions
The analyses related to this grade repetition should be interpreted somewhat
differently to the other variables presented, as repetition seems to be shaped
more by the properties of the school a child attends than by the attributes of the
child him or herself. In addition, while repetition precedes 2003 schooling
choices, it occurs only after 1997 schooling choices have already been made,
meaning that it cannot be influencing school choice for that year (although a
child‘s academic capability might be influencing choice). However, it is quite
likely that school choice in 1997 influences repetition in subsequent years. In
addition, grade repetition strongly influences whether or not a child has
reached secondary school by 2003. This means that finding for 2003,
particularly for the secondary school level, should be treated with some
caution.
223
Although the causality behind the relationships documented in Table 7.14,
below, is not clear, there are highly significant relationships between grade
repetition and most indicators of school quality, other than Section 21 status.
All relationships operate in the expected direction, with children who have
repeated a grade tending to attend less advantaged and more poorly performing
schools than those who have never repeated a grade. The relationship between
repetition and school size is likely to be due to the tendency for primary
schools in poorer areas to be somewhat smaller than those in more affluent
areas, as mentioned previously.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.009
(Grade
repeaters
more likely to
attend public
schools)
Pr=0.013
(Grade
repeaters
more likely to
attend public
schools)
Pr=0.012
(Grade
repeaters
more likely to
attend public
schools)
n.s.
Quintile Pr=0.000
(Grade
repeaters
more likely to
attend schools
in quintiles 1,
2 or 3)
Pr=0.000
(Grade
repeaters
more likely to
attend schools
in quintiles 1,
2 or 3)
Pr=0.000
(Grade
repeaters
more likely to
attend schools
in quintiles 1,
2 or 3)
n.s.
Section 21 n.s. n.s. n.s. n.s.
School size Pr= 0.0000
(Grade
repeaters
more likely to
attend smaller
schools)
Pr= 0.0000
(Grade
repeaters
more likely to
attend smaller
schools)
Pr= 0.0284
(Grade
repeaters
more likely to
attend smaller
schools)
n.s.
Proportion
black
Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with a
higher
proportion of
black learners)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with a
higher
proportion of
black learners)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with a
higher
proportion of
black learners)
Pr= 0.0208
(Grade
repeaters
more likely to
attend a
school with a
higher
proportion of
black learners)
224
School fees Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with
lower fees)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with
lower fees)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
school with
lower fees)
n.s.
Historical DET
status
Pr= 0.000
(Grade
repeaters
more likely to
attend a
former DET
school)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
former DET
school)
Pr= 0.000
(Grade
repeaters
more likely to
attend a
former DET
school)
Pr= 0.014
(Grade
repeaters
more likely to
attend a
former DET
school)
Matric Pass
rate
Pr=0.0149
(Grade
repeaters
more likely to
attend a
school with a
higher pass
rate)
Pr=0.0010
(Grade
repeaters
more likely to
attend a
school with a
higher pass
rate)
Pr=0.0091
(Grade
repeaters
more likely to
attend a
school with a
higher pass
rate)
n.s.
Table 7.14: Relationship between grade repetitions between 1997 and 2003, and
school properties
Child level variables associated with school enrolment patterns:
Discussion
The data presented here suggests that race remains closely connected to the
attributes of the school a child attends, with Black children particularly likely
to attend historically disadvantaged schools. The data does not provide strong
evidence for any consistent relationship between gender and schooling
enrolment patterns, with the exception that girls are substantially more likely
than boys to have reached secondary schooling by 2003. It does, however,
indicate that children enrolling in school for the first time at a later age are
significantly more likely to attend more advantaged schools than children who
enrol earlier. Finally, there is also evidence that children who have repeated a
grade attend less advantaged schools than children who have never repeated a
grade, although with the available data it is not possible to disentangle the
225
causal roles of school quality and a learner‘s inherent academic capabilities in
this relationship.
7.3.2 Family & household variables:
Maternal education
Table 7.15, below, provides evidence for highly significant relationships
between maternal education and almost all of the school attributes examined.
In all cases where a relationship between a school property and expected
school quality exists, children with more educated mothers are more likely to
attend higher quality schools. The one exception to this is Section 21 status,
which is unrelated to maternal education. Additionally, in 1997 there is no
relationship between maternal education and the imputed pass rate of primary
schools, but this probably relates to the imputation process used.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr= 0.000
(children of
more
educated
mothers less
likely to attend
public schools)
Pr= 0.000
(children of
more
educated
mothers less
likely to attend
public schools)
Pr= 0.000
(children of
more
educated
mothers less
likely to attend
public schools)
n.s.
Quintile Pr=0.0001
(children of
more
educated
mothers likely
to attend
higher quintile
schools)
Pr=0.0001
(children of
more
educated
mothers likely
to attend
higher quintile
schools)
Pr=0.0001
(children of
more
educated
mothers likely
to attend
higher quintile
schools)
Pr=0.0002
(children of
more
educated
mothers likely
to attend
higher quintile
schools)
Section 21 n.s. n.s. n.s. n.s.
School size n.s. Pr= 0.0490
(children of
more
educated
mothers likely
to attend
n.s. Pr= 0.0490
(children of
more
educated
mothers likely
to attend
226
larger schools) smaller
schools)
Proportion
black
Pr=0.0001
(children of
more
educated
mothers likely
to attend
school with
lower
proportion of
black learners)
Pr=0.0001
(children of
more
educated
mothers likely
to attend
school with
lower
proportion of
black learners)
Pr=0.0001
(children of
more
educated
mothers likely
to attend
school with
lower
proportion of
black learners)
Pr=0.0320
(children of
more
educated
mothers likely
to attend
school with
lower
proportion of
black learners)
School fees Pr= 0.0001
(children of
more
educated
mothers likely
to attend
school with
higher fees)
Pr= 0.0001
(children of
more
educated
mothers likely
to attend
school with
higher fees)
Pr= 0.0001
(children of
more
educated
mothers likely
to attend
school with
higher fees)
Pr= 0.0001
(children of
more
educated
mothers likely
to attend
school with
higher fees)
Historical DET
status
Pr= 0.000
(children of
more
educated
mothers less
likely to attend
a former DET
school)
Pr= 0.000
(children of
more
educated
mothers less
likely to attend
a former DET
school)
Pr= 0.000
(children of
more
educated
mothers less
likely to attend
a former DET
school)
Pr= 0.033
(children of
more
educated
mothers less
likely to attend
a former DET
school)
Matric Pass
rate
n.s. Pr=0.0002
(children of
more
educated
mothers likely
to attend a
school with a
higher pass
rate)
Pr=0.0098
(children of
more
educated
mothers likely
to attend a
school with a
higher pass
rate)
Pr=0.0015
(children of
more
educated
mothers likely
to attend a
school with a
higher pass
rate)
Table 7.15: Relationship between maternal education and school properties
Maternal marital status
As reflected in Table 7.16 below, maternal marital status is significantly
related to most school attributes, with the children of married mothers typically
being more likely to attend more advantaged schools.
227
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr= 0.001
(children of
married
mothers less
likely to attend
a public
school)
Pr= 0.005
(children of
married
mothers less
likely to attend
a public
school)
Pr= 0.028
(children of
married
mothers less
likely to
attend a public
school)
Pr= 0.094
(children of
married
mothers less
likely to attend
a public
school)
Quintile Pr=0.019
(children of
married
mothers more
likely to attend
a quintile 4 or
5 school)
Pr=0.000
(children of
married
mothers more
likely to attend
a quintile 4 or
5 school)
Pr=0.000
(children of
married
mothers more
likely to
attend a
quintile 4 or 5
school)
n.s.
Section 21 n.s. n.s. n.s. Pr=0.048
(children of
married
mothers less
likely to attend
a Section 21
school)
School size Pr=0.0009
(children of
married
mothers likely
to attend
larger schools)
Pr=0.0179
(children of
married
mothers likely
to attend
larger schools)
Pr=0.0041
(children of
married
mothers likely
to attend
larger schools)
n.s.
Proportion
black
Pr=0.0000
(children of
married
mothers likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0000
(children of
married
mothers likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0000
(children of
married
mothers likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0000
(children of
married
mothers likely
to attend
schools with a
lower
proportion of
black learners)
School fees Pr=0.0000
(children of
married
mothers likely
to attend
schools with
higher fees)
Pr=0.0000
(children of
married
mothers likely
to attend
schools with
higher fees)
Pr=0.0000
(children of
married
mothers likely
to attend
schools with
higher fees)
n.s.
228
Historical DET
status
Pr=0.0000
(children of
married
mothers less
likely to attend
former DET
schools)
Pr=0.0000
(children of
married
mothers less
likely to attend
former DET
schools)
Pr=0.0000
(children of
married
mothers less
likely to
attend former
DET schools)
Pr=0.0000
(children of
married
mothers less
likely to attend
former DET
schools)
Matric Pass
rate
Pr=0.0191
(children of
married
mothers likely
to attend
schools with
higher pass
rates)
Pr=0.0033
(children of
married
mothers likely
to attend
schools with
higher pass
rates)
Pr=0.0315
(children of
married
mothers likely
to attend
schools with
higher pass
rates)
Pr=0.0690
(children of
married
mothers likely
to attend
schools with
higher pass
rates)
Table 7.16: Relationship between maternal marital status and school properties
Household SES: 1997
All relationships were explored using both the derived SES quintiles, as well as
the raw SES scores. As the results were largely identical, only the results of the
tests based on the quintiles are documented in Table 7.17 below. These results
indicate the existence of strongly significant relationships between 1997
household SES, and all indicators of school quality, in both 1997 and 2003,
with the exception of Section 21 status. The highly significant nature of all of
these relationships is expected given existing evidence for a close relationship
between SES and educational opportunity in urban South Africa.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.003
(Higher SES
children less
likely to attend
public schools)
Quintile Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
229
Section 21 n.s. n.s. n.s. n.s.
School size Pr=0.0001
(Higher SES
children likely
to attend
larger schools)
Pr=0.0026
(Higher SES
children likely
to attend
larger schools)
Pr=0.0001
(Higher SES
children likely
to attend
larger schools)
n.s.
Proportion
black
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
School fees Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Historical DET
status
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Matric Pass
rate
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher pass
rate)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher pass
rate)
Pr=0.0190
(Higher SES
children likely
to attend a
school with
higher pass
rate)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher pass
rate)
Table 7.17: Relationship between 1997 household SES and school attributes in both
1997 and 2003
Household SES: 2003
Again, tests were conducted using both 2003 SES quintiles and raw scores, and
as the results were largely identical, only the results of the tests conducted
using the quintiles are presented in Table 7.18 below. These results were
extremely similar to those for 1997 SES, which is anticipated as the two SES
scores are strongly related. Once again, the only school attribute which was not
related to SES was a child‘s school had Section 21 status.
230
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.000
(Higher SES
children less
likely to attend
public schools)
Pr=0.015
(Higher SES
children less
likely to attend
public schools)
Quintile Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Pr=0.0001
(Higher SES
children likely
to attend
higher quintile
schools)
Section 21 n.s. n.s. n.s. n.s.
School size Pr=0.0001
(Higher SES
children likely
to attend
larger schools)
n.s. Pr=0.0148
(Higher SES
children likely
to attend
larger schools)
n.s.
Proportion
black
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
Pr=0.0001
(Higher SES
children likely
to attend
schools with a
lower
proportion of
black learners)
School fees Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher fees)
Historical DET
status
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Pr=0.000
(Higher SES
children less
likely to attend
a former DET
school)
Matric Pass
rate
Pr=0.0044
(Higher SES
children likely
to attend a
school with
higher pass
Pr=0.0001
(Higher SES
children likely
to attend a
school with
higher pass
Pr=0.0015
(Higher SES
children likely
to attend a
school with
higher pass
Pr=0.0002
(Higher SES
children likely
to attend a
school with
higher pass
231
rate) rate) rate) rate)
Table 7.18: Relationship between 2003 household SES and school attributes in 1997
and 2003
Household level variables associated with school enrolment
patterns: Discussion
Significant relationships were identified between each of the household level
variables examined, and all school attributes with the exception of Section 21
status. In all cases, relationships operated in the expected direction. Children of
more educated mothers and married mothers were more likely to be attending
more advantaged schools, as were children living in more advantaged
households in either 1997 or 2003. This provides strong evidence that more
advantaged children tend to have access to better educational opportunities in
contemporary urban South African than their less advantaged peers.
7.3.3 Community level variables:
Residential Area Poverty
Small Area Level
Again, as tests on residential area poverty quintiles and raw scores produced
largely identical results, only the results of analyses conducted using the
quintiles are presented in Table 7.19 below. Overall the results are fairly clear,
with children living in poorer areas more likely to attend less advantaged
schools.
SAL Poverty 1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.000
(Children living
in poorer SALs
more likely to
attend a public
school)
Pr=0.000
(Children living
in poorer SALs
more likely to
attend a public
school)
Pr=0.000
(Children living
in poorer SALs
more likely to
attend a public
school)
n.s.
Quintile Pr=0.0001 Pr=0.0001 Pr=0.0001 Pr=0.0001
232
(Children living
in poorer SALs
likely to attend
lower quintile
schools)
(Children living
in poorer SALs
likely to attend
lower quintile
schools)
(Children living
in poorer SALs
likely to attend
lower quintile
schools)
(Children living
in poorer SALs
likely to attend
lower quintile
schools)
Section 21 n.s. n.s. n.s. Pr=0.092
(Children living
in richer SALs
less likely to
attend Section
21 schools)
School size Pr=0.0001
(Children living
in richer SALs
likely to attend
larger schools)
Pr=0.0001
(Children living
in richer SALs
likely to attend
larger schools)
Pr=0.0001
(Children living
in richer SALs
likely to attend
larger schools)
Pr=0.0232
(Children living
in richer SALs
likely to attend
larger schools)
Proportion
black
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with a
higher
proportion of
black learners)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with a
higher
proportion of
black learners)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with a
higher
proportion of
black learners)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with a
higher
proportion of
black learners)
School fees Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower fees)
Historical DET
status
Pr=0.000
(Children living
in poorer SALs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SALs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SALs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SALs
more likely to
attend former
DET schools)
Matric Pass
rate
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower pass
rates)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower pass
rates)
Pr=0.0001
(Children living
in poorer SALs
likely to attend
schools with
lower pass
rates)
Pr=0.0012
(Children living
in poorer SALs
likely to attend
schools with
lower pass
rates)
Table 7.19: Relationship between residential SAL poverty and school attributes in
1997 and 2003
233
Sub-place level
The results for the analyses based on SP poverty levels (see Table 7.20 below)
were similar to those conducted for the SAL poverty levels, with children in
poor areas typically attending less advantaged schools. However, for matric
pass rate, the relationship shifted from a linear relationship in which children
living in more affluent areas attended schools with higher pass rates, to a non-
linear relationship in which children in both the wealthiest and poorest areas
attended schools with higher pass rates than children in moderate-poverty
areas. Potential explanations for this shift are not clear.
SP Poverty 1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.000
(Children living
in poorer SPs
more likely to
attend a public
school)
Pr=0.000
(Children living
in poorer SPs
more likely to
attend a public
school)
Pr=0.000
(Children living
in poorer SPs
more likely to
attend a public
school)
n.s.
Quintile Pr=0.0001
(Children living
in poorer SPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
lower quintile
schools)
Section 21 Pr=0.003
(Children living
mid-range SPs
least likely to
attend Section
21 schools)
Pr=0.000
(Children living
mid-range SPs
least likely to
attend Section
21 schools)
Pr=0.007
(Children living
mid-range SPs
least likely to
attend Section
21 schools)
Pr=0.000
(Children living
in richer SPs
least likely to
attend Section
21 schools)
School size Pr=0.0001
(Children living
in richer SPs
likely to attend
larger schools)
Pr=0.0001
(Children living
in richer SPs
likely to attend
larger schools)
Pr=0.0001
(Children living
in richer SPs
likely to attend
larger schools)
Pr=0.0068
(Children living
in richer SPs
likely to attend
larger schools)
Proportion
black
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with a
higher
proportion of
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with a
higher
proportion of
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with a
higher
proportion of
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with a
higher
proportion of
234
black learners) black learners) black learners) black learners)
School fees Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with
lower fees)
Pr=0.0001
(Children living
in poorer SPs
likely to attend
schools with
lower fees)
Historical DET
status
Pr=0.000
(Children living
in poorer SPs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SPs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SPs
more likely to
attend former
DET schools)
Pr=0.000
(Children living
in poorer SPs
more likely to
attend former
DET schools)
Matric Pass
rate
Pr=0.0001
(Children living
in mid-range
SPs likely to
attend schools
with lower
pass rates)
Pr=0.0001
(Children living
in mid-range
SPs likely to
attend schools
with lower pass
rates)
Pr=0.0001
(Children living
in mid-range
SPs likely to
attend schools
with lower
pass rates)
Pr=0.0001
(Children living
in mid-range
SPs likely to
attend schools
with lower
pass rates)
Table 7.20: Relationship between SP poverty and school properties
Main Place level
As evident in Table 7.21, the relationships between residential MP poverty
levels and school attributes remains fairly consistent with those identified at
the SAL and SP levels, although the statistical significance of some of these
relationships has decreased. Overall, children living in wealthier MPs are more
likely to attend more advantaged schools than their peers in poorer MPs. Once
again, however, children living in the mid-range MPs likely to attend the
schools with the poorest pass rates – although this relationship disappears
when only those children who have reached secondary school by 2003 are
examined. The reasons for the persistence of this pattern remain unclear.
235
MP Poverty 1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Sector Pr=0.061
(Children living
in wealthier
MPs less likely
to attend a
public school)
Pr=0.062
(Children living
in wealthier
MPs less likely
to attend a
public school)
Pr=0.010
(Children living
in wealthier
MPs less likely
to attend a
public school)
n.s.
Quintile Pr=0.0001
(Children living
in poorer MPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer MPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer MPs
likely to attend
lower quintile
schools)
Pr=0.0001
(Children living
in poorer MPs
likely to attend
lower quintile
schools)
Section 21 Pr=0.002
(Children living
in poorer MPs
less likely to
attend Section
21 schools)
n.s. Pr=0.001
(Children living
in poorer MPs
less likely to
attend Section
21 schools)
Pr=0.022
(Children living
in poorer MPs
less likely to
attend Section
21 schools)
School size Pr=0.0001
(Children living
in wealthier
MPs likely to
attend larger
schools)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend larger
schools)
Pr=0.0002
(Children living
in wealthier
MPs likely to
attend larger
schools)
n.s.
Proportion
black
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with a lower
proportion of
black learners)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with a lower
proportion of
black learners)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with a lower
proportion of
black learners)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with a lower
proportion of
black learners)
School fees Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with higher
fees)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with higher
fees)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with higher
fees)
Pr=0.0001
(Children living
in wealthier
MPs likely to
attend schools
with higher
fees)
Historical DET
status
Pr=0.000
(Children living
in wealthier
SPs less likely
to attend
Pr=0.000
(Children living
in wealthier
SPs less likely
to attend
Pr=0.000
(Children living
in wealthier
SPs less likely
to attend
Pr=0.000
(Children living
in wealthier
SPs less likely
to attend
236
former DET
schools)
former DET
schools)
former DET
schools)
former DET
schools)
Matric Pass
rate
Pr=0.0001
(Children living
in mid-range
MPs likely to
attend schools
with lower
pass rates)
Pr=0.0028
(Children living
in mid-range
MPs likely to
attend schools
with lower
pass rates)
Pr=0.0008
(Children living
in mid-range
MPs likely to
attend schools
with lower
pass rates)
n.s.
Table 7.21: Relationship between MP poverty and school properties
Community level variables associated with school enrolment
patterns: Discussion
At the community level, the relationships between residential SAL poverty and
school attributes were clearest, with children living in poorer SALs more likely
to attend less advantaged schools. At the SP and MP levels, this pattern held
overall, although strange results emerged with respect to the school‘s Section
21 status and matric pass rates. Explanations for these unexpected results are
not clear.
7.3.4 Child, household and community level variables
associated with school enrolment patterns: Discussion
The analyses presented in this section present clear evidence for a relationship
between child, household and community attributes associated with advantage,
and school-level variables associated with more highly performing schools.
Socio-economic status and maternal education appear to be particularly
strongly related to attending higher quality schools, while living in high
poverty areas is, predictably, inversely associated with attending high quality
schools. At the level of the individual child, there is no evidence for a
relationship between gender and school quality or other school properties, once
schooling phase has been controlled for. However, children who start school
late for their age are much more highly represented among those attending
more advantaged and highly performing schools. The analyses around Section
21 status given unexpected and often contradictory results, suggesting that
237
there may be reason for concern around the accuracy of this variable. Overall,
these results highlight the extent to which access to high quality education in
contemporary urban South Africa is determined by a child‘s home
circumstances, and the area in which he or she lives.
7.4 Relationships between school attributes and mobility
behaviours
The final section of this chapter tests for and explores relationships between
various school attributes, and the mobility of learners enrolled in those schools.
Children attending higher quality schools are expected to be more strongly
engaged in mobility.
7.4.1 School sector
Table 7.22, below, illustrates the relationships between the sector (public or
independent) of the school a child attends, and his or her mobility behaviour.
There is strong evidence that children attending independent schools are
significantly more mobile than those attending public schools. This holds for
all definitions of mobility.
School sector 1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.000
(Children at
independent
schools travel
significantly
further to
school than
children at
public schools)
Pr=0.000
(Children at
independent
schools travel
significantly
further to
school than
children at
public schools)
Pr=0.000
(Children at
independent
schools travel
significantly
further to
school than
children at
public schools)
Pr=0.000
(Children at
independent
schools travel
significantly
further to
school than
children at
public schools)
Movement
between areas
SAL: Pr=0.052
(Children at
public schools
SAL: n.s.
SAL: n.s. SAL: n.s.
238
are more likely
to attend
school in their
home SAL)
SP: Pr=0.000
(Children at
public schools
are more likely
to attend
school in their
home SP)
SP: Pr=0.000
(Children at
public schools
are more likely
to attend
school in their
home SP)
SP: Pr=0.000
(Children at
public schools
are more likely
to attend
school in their
home SP)
SP: Pr=0.001
(Children at
public schools
are more likely
to attend
school in their
home SP)
MP: Pr=0.000
(Children at
public schools
are more likely
to attend
school in their
home MP)
MP: Pr=0.000
(Children t
public schools
are more likely
to attend
school in their
home MP)
MP: Pr=0.000
(Children at
public schools
are more likely
to attend
school in their
home MP)
MP: Pr=0.004
(Children at
public schools
are more likely
to attend
school in their
home MP)
Nearest
school
Pr=0.000
(Children at
public schools
are more likely
to attend their
nearest
school)
Pr=0.000
(Children at
public schools
are more likely
to attend their
nearest school)
Pr=0.000
(Children at
public schools
are more likely
to attend their
nearest
school)
Pr=0.000
(Children at
public schools
are more likely
to attend their
nearest
school)
Table 7.22: Relationship between sector of attended school and learner mobility
7.4.2 School quintile
As indicated by the data in Table 7.23, children attending high quintile schools
are more likely to engage in learner mobility than children attending lower
quintile schools. However, in some tests, particularly relating to movement
between areas, there is evidence that the children attending the very lowest
quintile schools are as likely to engage in mobility as children at the attending
the highest quintile schools. This has two potential explanations: firstly, it may
relate to poor quality data on those children attending the most disadvantaged
schools, or secondly, it may be that the children attending the most
disadvantaged schools are living in areas with particularly few educational
opportunities, requiring them to travel further. The second hypothesis is
239
supported by the fact that higher mobility amongst children attending most
disadvantaged schools largely disappears at higher levels of geography.
School
quintile
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0001
(Children at
higher quintile
schools are
likely to travel
further than
children at
lower quintile
schools)
Pr=0.0001
(Children at
higher quintile
schools are
likely to travel
further than
children at
lower quintile
schools)
Pr=0.0001
(Children at
higher quintile
schools are
likely to travel
further than
children at
lower quintile
schools)
Pr=0.0001
(Children at
higher quintile
schools are
likely to travel
further than
children at
lower quintile
schools)
Movement
between areas
SAL: Pr=0.042
(Children at
mid-quintile
schools are
more likely to
attend a
school in their
home SAL)
SAL: n.s. SAL: n.s. SAL: n.s.
SP: Pr=0.000
(Children at
mid-quintile
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.000
(Children at
mid-quintile
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.000
(Children at
mid-quintile
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.000
(Children at
mid-quintile
schools are
more likely to
attend a
school in their
home SP)
MP: Pr=0.000
(Children at
low and mid-
quintile
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
low and mid-
quintile
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
low and mid-
quintile
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
low and mid-
quintile
schools are
more likely to
attend a
school in their
home MP)
Nearest
school
Pr=0.000
(Children at
low and mid-
quintile
schools more
likely to attend
Pr=0.000
(Children at
low and mid-
quintile
schools more
likely to attend
Pr=0.000
(Children at
low and mid-
quintile
schools more
likely to attend
Pr=0.000
(Children at
low and mid-
quintile
schools more
likely to attend
240
their nearest
school)
their nearest
school)
their nearest
school)
their nearest
school)
Table 7.23: Relationship between quintile of attended school and learner mobility
7.4.3 Section 21 status
The evidence for a relationship between the Section 21 status of a child‘s
school and his or her mobility is somewhat weaker, as evident in Table 7.24
below. While there is evidence of a significant relationship between Section 21
status and mobility for all definitions of mobility for those children attending
secondary school in 2003, the evidence for primary school children is less
clear. There does appear to be a weakly significant relationship between
Section 21 status and distance travelled, but there is no evidence that Section
21 status is related to movement between areas or to whether a child attends
his or her nearest school. In all cases where there is evidence for a relationship
between Section 21 status and mobility, attending a school with Section 21
status is associated with increased mobility.
School Section
21 status
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0214
(Children at
schools
without
Section 21
status travel
less far to
school than
children at
Section 21
schools)
Pr=0.0540
(Children at
schools
without
Section 21
status travel
less far to
school than
children at
Section 21
schools)
n.s. Pr=0.0716
(Children at
schools
without
Section 21
status travel
less far to
school than
children at
Section 21
schools)
Movement
between areas
SAL: n.s. SAL: n.s. SAL: n.s. SAL: n.s.
SP: n.s. SP: n.s. SP: n.s. SP: Pr=0.001
(Children at
schools
without
Section 21
status more
241
likely to attend
school in their
home SP)
MP: n.s. MP: n.s. MP: n.s. MP: Pr=0.014
(Children at
schools
without
Section 21
status more
likely to attend
school in their
home MP)
Nearest
school
n.s. n.s. n.s. Pr=0.038
(Children at
Section 21
schools less
likely to attend
their nearest
school)
Table 7.24: Relationship between Section 21 status of attended school and learner
mobility
7.4.4 School Enrolment
The direction of the relationship between a child‘s mobility and the size of his
or her school is heavily moderated by schooling phase (see Table 7.25 below).
At the secondary school level, as distance increases, the size of the school
attended tends to fall. By contrast, at the primary level, as distance increases,
school size also tends to rise. This is likely to relate to the tendency, described
in Chapter 4, for more advantaged areas to have primary and secondary
schools that are roughly equivalent in size, while less advantaged areas tend to
have a few particularly large secondary schools, and a large number of far
smaller primary schools.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0000 (As
children travel
further to
Pr=0.0010 (As
children travel
further to
Pr=0.0000 (As
children travel
further to
Pr=0.0005 (As
children travel
further to
242
school, school
size increases)
school, school
size increases)
school, school
size increases)
school, school
size falls)
Movement
between areas
SAL: n.s. SAL: n.s. SAL: n.s. SAL: n.s.
SP: Pr=0.0000
(Children at a
school outside
of their home
SP attend
larger schools)
SP: n.s. SP: Pr=0.0008
(Children at a
school outside
of their home
SP attend
larger schools)
SP: Pr=0.0024
(Children at a
school outside
of their home
SP attend
larger schools)
MP: Pr=0.0000
(Children at a
school outside
of their home
MP attend
larger schools)
MP: Pr=0.0226
(Children at a
school outside
of their home
MP attend
larger schools)
MP: Pr=0.0000
(Children at a
school outside
of their home
MP attend
larger schools)
MP: Pr=0.0008
(Children at a
school outside
of their home
MP attend
smaller
schools)
Nearest
school
n.s. Pr=0.0000
(Children at
their nearest
school are
likely to be
attending a
larger school)
n.s. Pr=0.0005
(Children at
their nearest
school are
likely to be
attending a
larger school)
Table 7.25: Relationship between size of attended school and learner mobility
7.4.5 Proportion black learners
Predictably, given that the majority of sample members live in predominantly
black areas, those attending school close to their homes are likely to attend
schools with a higher proportion of black learners than those who are travelling
to schools further afield (see Table 7.26 below). There are, however, two
exceptions to this pattern.
Firstly, when looking at primary school children who attend school within the
same SAL as their home, they are likely to attend a school with a lower
proportion of black learners than those children travelling outside of the SAL
in which their home is. This is likely to be due to the extremely small numbers
of children attending school in their home SAL, and the earlier finding children
243
living in more affluent areas were more likely to attend school in their home
SAL.
Secondly, there is no evidence for a relationship between whether or not
children attend their nearest school, and the proportion of black learners in that
school. This absence of a relationship is likely to be because quite a few of
those children not attending their nearest school are still attending a school in
the same, relatively disadvantaged area in which they live.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0000
(Children at
schools with a
lower
proportion of
black learners
travel further
to school)
Pr=0.0000
(Children at
schools with a
lower
proportion of
black learners
travel further
to school)
Pr=0.0000
(Children at
schools with a
lower
proportion of
black learners
travel further
to school)
Pr=0.0000
(Children at
schools with a
lower
proportion of
black learners
travel further
to school)
Movement
between areas
SAL: Pr=0.0433
(Children at
schools with a
lower
proportion of
black children
more likely to
attend a
school in their
home SAL.
SAL: n.s. SAL: Pr=0.0133
(Children at
schools with a
lower
proportion of
black children
more likely to
attend a
school in their
home SAL.
SAL: n.s.
Pr=0.0000
(Children at
schools with a
higher
proportion of
black learners
are more likely
to attend
school in their
home SP)
Pr=0.0000
(Children at
schools with a
higher
proportion of
black learners
are more likely
to attend
school in their
home SP)
Pr=0.0000
(Children at
schools with a
higher
proportion of
black learners
are more likely
to attend
school in their
home SP)
Pr=0.0031
(Children at
schools with a
higher
proportion of
black learners
are more likely
to attend
school in their
home SP)
Pr=0.0000
(Children at
schools with a
Pr=0.0000
(Children at
schools with a
Pr=0.0000
(Children at
schools with a
Pr=0.0000
(Children at
schools with a
244
higher
proportion of
black learners
are more likely
to attend
school in their
home MP)
higher
proportion of
black learners
are more likely
to attend
school in their
home MP)
higher
proportion of
black learners
are more likely
to attend
school in their
home MP)
higher
proportion of
black learners
are more likely
to attend
school in their
home MP)
Nearest
school
n.s. n.s. n.s. n.s.
Table 7.26: Relationship between proportion of black students at attended school
and learner mobility
7.4.6 School fees
Table 7.27 below provides evidence for a significant positive relationship
between the fees charged by a child‘s school, and the mobility of that child.
One finding, that children attending school inside their home SAL are likely to
attend schools with lower fees, slightly inconsistent with the finding that these
same children are likely to attend a school with a lower proportion of black
learners. One plausible explanation is the nature of the distribution of the
values of school fees, with only a very few very high values present.
School fees 1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0000
(Children at
schools with
higher fees
travel further)
Pr=0.0000
(Children at
schools with
higher fees
travel further)
Pr=0.0000
(Children at
schools with
higher fees
travel further)
Pr=0.0000
(Children at
schools with
higher fees
travel further)
Movement
between areas
SAL: Pr=0.0917
(Children at
schools with
lower fees are
more likely to
attend school
in their home
SAL)
SAL: n.s. SAL: n.s. SAL: n.s.
SP: Pr=0.0000
(Children at
schools with
SP: Pr=0.0000
(Children at
schools with
SP: Pr=0.0000
(Children at
schools with
SP: Pr=0.0000
(Children at
schools with
245
lower fees are
more likely to
attend school
in their home
SP)
lower fees are
more likely to
attend school
in their home
SP)
lower fees are
more likely to
attend school
in their home
SP)
lower fees are
more likely to
attend school
in their home
SP)
MP: Pr=0.0000
(Children at
schools with
lower fees are
more likely to
attend school
in their home
MP)
MP: Pr=0.0000
(Children at
schools with
lower fees are
more likely to
attend school
in their home
MP)
MP: Pr=0.0000
(Children at
schools with
lower fees are
more likely to
attend school
in their home
MP)
MP: Pr=0.0000
(Children at
schools with
lower fees are
more likely to
attend school
in their home
MP)
Nearest
school
Pr=0.0192
(Children at
schools with
lower fees are
more likely to
attend their
nearest school)
Pr=0.0002
(Children at
schools with
lower fees are
more likely to
attend their
nearest
school)
Pr=0.0004
(Children at
schools with
lower fees are
more likely to
attend their
nearest
school)
Pr=0.0130
(Children at
schools with
lower fees are
more likely to
attend their
nearest
school)
Table 7.27: Relationship between fees of attended school and learner mobility
7.4.7 Historical DET
Table 7.28, below, presents evidence for a significant negative relationship
between the historical DET status of the school a child attends, and that child‘s
mobility. However, this relationship does not hold when mobility is defined at
the SAL level, or by attendance at the nearest school. As discussed with
reference to the findings on the proportion of black students enrolled at a
school, this is likely to relate to the fact that many of the children engaging in
these two forms of mobility are still attending schools that are no less
disadvantaged than the school nearest to their home, or within their home SAL.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0000
(Children at
former DET
schools travel
Pr=0.0000
(Children at
former DET
schools travel
Pr=0.0000
(Children at
former DET
schools travel
Pr=0.0000
(Children at
former DET
schools travel
246
less far) less far) less far) less far)
Movement
between areas
SAL: n.s. SAL: n.s. SAL: n.s. SAL: n.s.
SP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home SP)
SP: Pr=0.007
(Children at
former DET
schools are
more likely to
attend a
school in their
home SP)
MP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home MP)
MP: Pr=0.000
(Children at
former DET
schools are
more likely to
attend a
school in their
home MP)
Nearest
school
n.s. n.s. n.s. n.s.
Table 7.28: Relationship between historical DET status of attended school and
learner mobility
7.4.8 Matric pass rate
There is strong evidence for a positive relationship between mobility and
school performance, as documented in Table 7.29 below.
1997 2003 (all data) 2003 (primary
schools only)
2003
(secondary
schools only)
Distance
travelled
Pr=0.0000
(Children at
schools with
higher pass
rates travel
further)
Pr=0.0000
(Children at
schools with
higher pass
rates travel
further)
Pr=0.0000
(Children at
schools with
higher pass
rates travel
further)
Pr=0.0000
(Children at
schools with
higher pass
rates travel
further)
Movement
between areas
SAL: Pr=0.0033
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
SAL: Pr=0.0216
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
SAL: n.s. SAL: Pr=0.0046
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
247
their home
SAL)
their home
SAL)
their home
SAL)
SP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home SP)
SP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home SP)
SP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home SP)
SP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home SP)
MP: Pr=0.0010
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home
MP)
MP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home
MP)
MP: Pr=0.0048
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home
MP)
MP: Pr=0.0000
(Children at
schools with
higher pass
rates are less
likely to attend
a school in
their home
MP)
Nearest
school
Pr=0.0047
(Children at
schools with
higher pass
rates are less
likely to be
attending their
nearest
school)
Pr=0.0004
(Children at
schools with
higher pass
rates are less
likely to be
attending their
nearest
school)
Pr=0.0811
(Children at
schools with
higher pass
rates are less
likely to be
attending their
nearest
school)
Pr=0.0002
(Children at
schools with
higher pass
rates are less
likely to be
attending their
nearest
school)
Table 7.29: Relationship between quintile of attended school and learner mobility
7.4.9 Relationships between school attributes and mobility
behaviour: Discussion
Overall, the results presented in this section provide support for the hypothesis
that learner mobility is related to the pursuit of higher quality education. Using
a number of different school variables associated with educational quality, and
a number of different approaches to the measurement of learner mobility, the
analyses described above found positive, largely consistent and highly
significant relationships between engagement in learner mobility, and the
quality of the school a child attends. The one definition of mobility which did
produce inconsistent results was that based on whether or not a child attended
248
school in the same SAL in which he or she lived. One possible explanation is
simply that SAL mobility is not a good measure of learner mobility, as it does
not differentiate sufficiently between children. This is supported by the very
small numbers of children who do attend school in the SAL in which they live.
7.5 Conclusion
This chapter began by contrasting those schools closest to the homes of study
sample members with the schools actually attended by study sample members.
This provided clear evidence that, on average, the schools that children attend
are somewhat more advantaged than the schools that are closest to their homes.
This suggests that children and families are actively engaging in school choice
and mobility to pursue higher quality educational opportunities than would
otherwise be accessible to them. It does, however, also raise questions around
who is attending the most poorly performing of schools. One possibility is that
these are predominantly filled by in-migrants from other areas. Alternatively,
these schools may simply be extremely under-enrolled.
Secondly, the chapter explored the distribution of children with different
individual, household and community attributes across schools. This provided
strong evidence that more advantaged children typically attend more
advantaged schools. This suggests that even though parents and children are
able to use learner mobility as a tool to access higher quality educational
opportunities, even these enhanced educational opportunities remain strongly
related to the affluence of a family. This ties in well with the hypothesis that
two different forms of mobility exist in Johannesburg-Soweto, and that these
require the investment of different levels of resources, but also providing
access to different levels of schooling quality. The more resource intensive
form of mobility requires substantial investment in travel, school fees and
associated costs, but allows children to access historically advantaged schools.
Given the resource requirements associated with this form of mobility, only
249
children from relatively advantaged families are able to engage in it. The
second form of mobility, which requires less in the way of resources and is
therefore open to a wider group of children, is the one in which choices are
made between a number of fairly local schools. While this may still mean that
children attend a better school than the school closest to their home, it does not
give them access to the most advantaged schools of all.
Finally, in testing the relationships between learner mobility and the attributes
of the school a child attends, the chapter has also provided evidence that
learner mobility is associated with enrolment at those schools expected to
provide higher educational quality. This again suggests that children and
families are using mobility, in at least two different forms, to gain access to
higher quality educational opportunities than would otherwise be accessible to
them.
250
Chapter 8: Changes in
educational mobility over time
8.1 Introduction
This chapter documents changes in the educational mobility of children over
time. It begins by looking at those sample members who have changed schools
between 1997 and 2003. School change is a prerequisite to mobility change in
this sample of children with constant residential addresses. Correlates of school
change at the individual, household, community and school levels are
documented. Tests are also conducted to determine whether children who
change schools behave differently with regards to mobility than those who do
not change schools. Secondly, the chapter explores the nature of the changes in
mobility resulting from school change. These results are then disaggregated by
whether the school change is a change between two primary schools, or a
change from a primary to a high school. The implications of school change and
changes in mobility over time are then discussed.
8.2 Changing schools
Given that residential addresses for the study sample are by definition constant
over the period under examination, if children are enrolled in the same school
in both 1997 and 2003, their mobility will also be constant. When children
move between schools, however, their mobility will change. This means that
school change provides a window onto mobility change.
Of the 1210 children for whom full schooling information is available in both
1997 and 2003, 373 (30.93%) were attending the same school at both points in
time, whereas 833 (69.07%) were attending different schools (see Table 8.1
below). Of course, a number of these school changes relate to the 433 children
251
who have moved from primary school to high school by 2003. When only
children still in primary school by 2003 are examined, 357 (45.5%) are
enrolled in the same school at both points in time, while 427 (54.39%) have
changed schools. While this is a lower level of change than that found in the
full study sample, it indicates that the majority of children who did not move
between schooling phases still changed schools at least once between 1997 and
2003.
Remained in same
school from 1997-2003
Changed schools
between 1997 and 2003
All children with full
schooling data (n=1210)
373 (30.93%) 833 (69.07%)
Children in primary school in
2003 (n=781)
357 (45.71%) 424 (54.39%)
Children in high school in
2003 (n=415)
14 (3.37%) 401 (96.63%)
Table 8.1: Proportion of sample members remaining in the same school from 1997
to 2003, and proportion changing schools (note that information on schooling
phase is only available for 1196 individuals)
8.2.1 Correlates of school change during primary schooling
Individual, household and community correlates
A range of tests was conducted to determine whether, amongst the group of
children still enrolled in primary school in 2003, those who had changed
schools at least once differed systematically from those who had never
changed schools. A chi-square test indicates that black children were
significantly more likely to change schools than coloured children (χ2(1)=
17.9177, Pr=0.000). There was, however, no evidence for a relationship
between school changing and gender, age at first enrolment, or grade
repetition.
Maternal education was also significantly related to school change
(χ2(4)=13.6864, Pr=0.008), with children of mothers with some secondary
schooling the most likely to change schools. There was no evidence for a
252
relationship between school change and maternal marital status, or, more
surprisingly, household SES in either 1997 or 2003, or change in household
SES between 1997 and 2003.
There was, however, strong evidence for a relationship between school change
and the poverty of the area in which a child lives, whether measured as the
SAL (χ2(4)=25.7324, Pr=0.000), SP (χ
2
(4)=29.4967, Pr=0.000) or MP
(χ2(2)=18.0766, Pr=0.000) level. At the SAL level, children living in poorer
areas more likely to change schools. At the SP and MP levels, by contrast,
while children in poor areas are still more likely to change schools than
children in affluent areas, it is the children living in mid-level areas who are
most likely to change schools.
Overall, contrary to expectations, the data does not suggest that disadvantaged
children are any more likely to change schools during primary schooling than
their more advantaged peers. In fact, children with more poorly educated
mothers appear to be somewhat protected against school change during this
period.
School level correlates
With the exception of the few children enrolled in quintile 1 schools (the least
advantaged schools), who are very unlikely to change schools during primary
schooling, there is a negative relationship between school quintile and school
change, with school change decreasing as quintile increases, in both 1997
(χ2(4)=18.3413, Pr=0.001) and 2003 (χ
2
(4)=8.5163, Pr=0.074). There is no
evidence for a relationship between school change and Section 21 status in
1997, but a weakly significant relationship is found in 2003 (χ2(1)=3.1192,
Pr=0.077).
Wilcoxon rank-sum tests indicate that children attending smaller schools in
1997 (Pr=0.0000) or in 2003 (Pr=0.0043) are more likely to change schools.
253
Children attending schools with a higher proportion of black learners in 1997
(Pr=0.0000) or 2003 (Pr=0.0000) are also more likely to change schools. There
is also a significant positive relationship between school change and the school
fees in 1997 (Pr=0.0002) and 2003 (Pr=0.0000). Children attending former
DET schools in either 1997 (χ2(1)=42.5454, Pr=0.000) or 2003 (χ
2
(1)=19.6727,
Pr=0.000) were also significantly more likely to change schools. Finally,
children attending schools with lower imputed pass rates in 1997 (Pr=0.0713)
or 2003 (Pr=0.0185) were also more likely to change schools.
Interestingly, all of the school properties associated with increased levels of
school change are also typically associated with schools located in township
areas. It appears, therefore, that a large proportion of school change occurring
during the primary school years is between primary schools located in the
townships. As relationships in 2003 are typically weaker than in 1997, this
suggests that to the extent that children are changing between schools in
different areas, they are tending to leave township schools in favour of schools
in other, presumably more advantaged areas.
Mobility related correlates
Mobility in 1997
Amongst children who remained in primary school throughout the study
period, school-changers travelled average marginally further than non-
changers in 1997, although this difference was not statistically significant.
There was no evidence for a relationship between school change and whether
he or she attended a school in the same SAL, SP or MP as his or her home in
1997. Finally, there was also no evidence for a relationship between school
change and whether a child attended his or her nearest school in 1997,
regardless of whether private schools were included in the analysis. Overall,
therefore, data about a child‘s mobility in 1997 is unlikely to serve as a useful
predictor of whether he or she is likely to change primary schools between
1997 and 2003.
254
Mobility in 2003
A Wilcoxon rank-sum test indicated that there was no significant relationship
between school change and distance from home to school in 2003. Children
attending a school outside of their home SAL in 2003 were, however,
significantly more likely to have changed schools (χ2(1)=14.5194, Pr=0.000).
This relationship is inverted at the SP level (χ2(1)=3.714, Pr=0.054), with
children attending a school in their home SP more likely to have changed
schools. There is no evidence for a relationship between children attending
schools in their home MP and school change. Finally, children who change
schools are significantly less likely to be attending their nearest school in 2003,
regardless of whether only public schools (χ2(1)=8.0904, Pr=0.004), or both
public and private schools (χ2(1)=10.0275, Pr=0.002) are considered.
Overall, while a child‘s mobility in 2003 is a better predictor of school change
than mobility in 1997, there is still not much evidence for a relationship
between school change and mobility. The strongest finding is that children
who attended their nearest school in 2003 were less likely to change schools,
which is consistent with earlier findings that the children most likely to attend
their nearest schools were those living in the affluent areas, and who would
therefore be unlikely to have much incentive to change schools.
Discussion
Black children, and children with mothers with some secondary schooling
appear to be the most likely to change school during the primary schooling
period. Children living in areas with high or intermediate poverty levels are
also more likely to change schools than their peers living in more affluent
areas. While children changing schools seem to be more likely to attend
township schools, there is little evidence to suggest that their patterns of
mobility in either 1997 or 2003 differ from children who do not change
255
schools. Overall, change between primary schools does not seem to be strongly
related to mobility.
8.2.2 Correlates of school change associated with the transition
to high school
The transition to high school typically requires that a child change schools. The
correlates of school change associated with the transition to high school will
therefore simply echo the correlates of high school status itself. Furthermore,
as school change and transition to high school are not independent events for
this group of children, but are different aspects of the same phenomenon, it
does not make sense to try to determine the role of each, independently, in
shaping mobility. Rather, it is appropriate to see both the transition to grade 8
and the school change that typically accompanies it as a single event. The data
detailing the correlates of reaching high school by 2003, presented in
Appendix C, Chapter 5 and Chapter 7, is briefly reviewed in the following
sections.
Individual, household and community correlates
As documented in Appendix C, children who have reached high school by
2003 differ systematically from children still in primary school at this point.
They are more likely to be coloured, to be girls, and to have started school
early, and less likely to have ever repeated a grade. Their mothers are likely to
be more highly educated, with an attainment of at least Grade 11, and they are
more likely to have lived in comparatively advantaged homes in 1997. There
is, however, no evidence that they differ from their peers who are still in
primary school in 2003 with regards to the poverty levels of the areas in which
they live.
256
School level correlates
As documented in Chapter 7, secondary schools attended by study sample
members differ systematically from the primary schools attended. Secondary
schools are more likely to be in quintile 4, and less likely to be in quintile 3 or
quintile 5 than those attended at the primary level. They also tend to have a
substantially larger number of learners enrolled, and a slightly lower
proportion of black learners. Fees are higher, while the schools are slightly less
likely to be former DET schools, and have slightly higher pass rates.
Mobility related correlates
Mobility in 1997
As detailed in Chapter 5, there is no evidence that group of children who have
reached high school in 2003 differ from their peers with regards to the distance
from home to school in 1997, or their likelihood of attending the school closest
to their home in 1997. Those children who transition to high school by 2003
were, however, more likely to attend a school in the same SAL or SP as their
home in 1997. At the MP and MN levels, however, no differences are evident.
Mobility in 2003
Children at the high school level in 2003 do, however, live significantly further
from their schools than their peers still at primary school level do. However,
there is no evidence that patterns of travel between different areas, as defined
by census geography, change with schooling phase. Children at the high school
level, however, are more likely to be attending their nearest school. These
changes are thought to be related to the different density distributions of
primary and high schools in Johannesburg-Soweto.
Discussion
As is clear from the data reviewed above, children who have reached high
school by 2003 differ from their peers who have not, with regards to a number
of variables associated with mobility, as well as differing with regards to
257
certain aspects of mobility itself. School changes associated with the transition
to high school seem to be more strongly related to changes in mobility than
school changes associated with movement between two primary schools.
However, this appears to be due primarily to the different distributions of
primary and high schools in urban Gauteng province.
8.3 The nature of changes in mobility
8.3.1 Straight-line distance
Of the 1210 children with full schooling data for 1997 and 2003, 1177 also
have full residential data, allowing the changes in their mobility over time to
be calculated. These children are distributed across primary and high schools
as illustrated in Table 8.2 below. Just over two thirds of them experience a
change in distance from home to school between 1997 and 2003.
No change in mobility
between 1997 and 2003
Change in mobility
between 1997 and 2003
All children with full
schooling and residential
data (n=1177)
n=368 (31.27%) n=809 (68.73%)
Children in primary school
in 2003 (n=763)
n=353 (46.26%) n=410 (53.74%)
Children in high school in
2003 (n=404)
n=13 (3.21%) n=391 (96.78%)
Table 8.2: Distribution of sample members with full residential and schooling data
for both 1997 and 2003 across schooling phases, and stability of mobility behaviour
There is a small overall decrease in distance from home to school between
1997 and 2003 (see Table 8.3 below). This, however, obscures a very broad
distribution of changes in distance, and the 31% of children who experience no
change in distance at all. When only those children who have experienced a
change in mobility are examined (n=807), the mean decrease in travel distance
becomes larger, and the distribution as a whole becomes more spread out. The
258
distribution of distance change for both the full sample and for school-changers
only is illustrated in Figures 8.1 and 8.2 below. It is evident from these figures
that although changes in distance range from extreme values at each end (both
very large increases and decreases), the majority of the sample is concentrated
around very moderate levels of change, even once all zero values have been
removed.
Mean change in
distance from
1997 to 2003
Standard
Deviation
25% Median 75%
Full Sample (n=1177) -0.167km 9.723km -0.178km 0.000km 0.530km
School changers only
(n=807)
-0.244km 11.743km -0.754km 0.121km 1.111km
Children moving
between primary
and high school only
(n=404)
0.543km 12.439km -0.488km 0.137km 1.420km
Children moving
between primary
and high school and
changing schools
only (n=390)
0.560km 12.660km -0.529km 0.211km 1.519km
All primary school
children (n=763)
-0.568km 7.951km 0.000km 0.000km 0.181km
Children moving
between different
primary schools
(n=409)
-1.060km 10.841km -1.314km 0.070km 0.795km
Table 8.3: Changes in distance from home to school between 1997 for all sample
members, disaggregated by schooling phase and school change status
259
Figure 8.1: Kernel density plot of the change in distance from home to school
between 1997 and 2003 for all sample members with full mobility information
Figure 8.2: Kernel density plot of the change in distance from home to school
between 1997 and 2003 for all sample members who changed schools
260
When change in distance is disaggregated by whether it is due to transition
from primary to high school, or to movement between two primary schools,
both distributions remain centred around zero, but otherwise become quite
distinct (see Table 8.3 above). The mean change in distance for children
moving between primary schools remains negative, but the mean change in
distance for children moving to a high school becomes positive, at just over
half a kilometre. A Wilcoxon rank-sum test indicates that children moving
between primary schools, and those moving from primary to high schools,
represent different populations in terms of the distribution of change in
distance from home to school (Pr=0.0211). It is evident however, that both of
these distributions, and particularly their means, are quite heavily influenced
by particularly large values – as can be seen from the values of the 25th and
75th percentiles.
For this reason, it is also useful to examine counts of children travelling higher
and lower distances in 2003 than in 1997 (Table 8.4 below). These counts
indicate that in both the group of children moving between primary schools,
and those transitioning to high school, roughly the same proportions are
travelling further in 2003. Similarly, the proportions of children travelling less
far in the two groups are also much the same. A chi-square test confirms that
there is no evidence that the distribution of children travelling both further and
less far differs for these two groups.
Given that the Wilcoxon test referred to earlier found a significant difference
between these groups with regards to change in distance experienced, but the
proportion of children increasing and decreasing their travel distance in both
groups is the same, the difference between these two groups must be found in
the extent of the increase or decrease in distance travelled. As evident in Figure
8.3 below, amongst children who travel further in 2003, the children moving
between primary schools are far more concentrated around very small
increases than the children moving from primary to high school. By contrast,
261
when looking at those children travelling less far in 2003, this time it is the
children moving from primary to high school that are far more concentrated
around the very low decreases (see Figure 8.4 below). Wilcoxon rank-sum
tests confirm that for both the group travelling further (Pr=0.0146), and those
travelling less far (Pr=0.0373), the distributions of the change in distance are
significantly different.
Overall, this evidence suggests that although both groups of children are
approximately equally likely to travel either further or less far, the average
decrease experienced by primary school children is larger. By contrast, the
average increase in distance experience by high school children is larger.
Mean change in
distance from
1997 to 2003
Number (%)
travelling further
in 2003
Number (%)
travelling less far
in 2003
School changers only
(n=807) (1 child changed
between two schools
located in the same place)
-0.244km 448 (55.51%) 358 (44.36%)
Children moving between
different primary schools
(n=409)
-1.060km 221 (54.03%) 187 (45.72%)
Children moving between
primary and high school
only (n=390)
0.560km 223 (56.92%) 168 (43.08%)
Table 8.4: Counts of sample members who live closer to or further from their
school in 2003 as compared to 1997, disaggregated by phase of schooling
262
Figure 8.3: Distribution of change in distance from home to school for children
travelling less far in 2003, by schooling phase
Figure 8.4: Distribution of change in distance from home to school for children
travelling further in 2003, by schooling phase
263
Categories of distance
Another way to think about changes in mobility over time is to group children
according to their distance between home and school at each point in time, and
explore the numbers who remain in the same category, and those who change.
Particularly salient categories are those travelling under 2.5km – that is, those
who are able to walk to school; those travelling 2.5 to 5km – these can be
thought of as children attending ‗local schools‘; and those travelling over 5km
– that is, those attending non-local schools. Table 8.5, below, shows that the
proportion of children found in the various distance categories is fairly
consistent over time.
Group Number (%) travelling less than
2.5km
Number (%) travelling 2.5-
5km
Number (%) travelling over 5km
Time point 1997 2003 Both 1997 2003 Both 1997 2003 Both
Full Sample
1997: n=1214
2003: n=1281
Both: n=1177
796
(65.57%)
823
(64.25%)
655
(55.65%)
83
(6.84%)
87
(6.79%)
40
(3.40%)
335
(27.59%)
371
(28.96%)
244
(20.73%)
School
changers
only
1997: n=810
2003: n=812
Both: n=807
539
(66.54%)
529
(65.15%)
424
(53.54%)
51
(6.30%)
50
(6.16%)
12
(1.49%)
220
(27.16%)
233
(28.69%)
133
(16.48%)
Children
moving
between two
primary
schools
1997: n=410
2003: n=413
both: n=409
263
(64.15%)
272
(65.86%)
207
(50.61%)
29
(7.07%)
25
(6.05%)
6
(1.47%)
118
(28.78%)
116
(28.09%)
68
(16.63%)
Children
moving
between
primary and
high school
1997: n=392
2003: n=391
both: n=390
269
(68.62%)
250
(63.94%)
210
(53.85%)
21
(5.36%)
25
(6.39%)
6
(1.54%)
102
(26.02%)
116
(29.67%)
65
(16.67%)
Table 8.5: Children attending schools in the same categories of distance from their
homes in 1997 and 2003, disaggregated by schooling phase
264
However, there is less consistency over the actual individuals who are found in
each category over time. Chi-square tests conducted on all children who
changed schools (χ2(4)= 185.9723, Pr=0.000), those moving between different
primary schools (χ2(4)=83.7371, Pr=0.000), and those moving between primary
and high schools (χ2(4)=103.0342, Pr = 0.000), all found highly significant
differences in distribution across these mobility categories between the two
points in time.
Amongst children who changed schools between 1997 and 2003, of the 644
children who were travelling less than 2.5km at any point in time, 424, or
65.84%, were travelling less than 2.5km at both points in time. Of the 89
children travelling between 2.5 and 5km at either point in time, 12, or 13.48%,
were found in this distance category at both points in time. Finally, of the 320
children travelling over 5km at either point in time, 133, or 41.56%, fell into
this category for both points in time. This data suggests that while the overall
proportion of children found in the different distance categories is fairly
constant over time, a fairly high proportion of individuals are actually moving
between categories. There is no evidence of a significant difference in
movement between distance categories for those children moving between
primary schools, and those transitioning to high school. This is particularly
interesting, as a proportion of children moving to high school are expected to
be doing so because they have completed their primary schooling. By contrast,
change during primary schooling is more likely to be related to dissatisfaction
with current schooling arrangements, providing more of a reason for a child to
change the distance he or she travels. Nonetheless, both groups seem to be
equally likely to move between different categories.
Overall, just under 56% of the sample falls into the 0-2.5km at both points in
time. This means that 44% of the children in the sample are attending a school
over 2.5km away from their home during some part of their schooling. 20% of
265
the sample attends schools over 5km away from their home at both points in
time, meaning that a full fifth of the sample is consistently educated over 5kms
away from their home. Clearly the assumption that children attend schools
within walking distance of their homes does not hold for a fairly sizeable
proportion of the current sample at various points in their schooling, and fairly
consistently for almost 25% of the sample who never attend a school within a
2.5km radius of their homes.
8.3.2 Census geography
This section explores changes in mobility over time with regards to whether a
child‘s school falls into the same area as his or her home. Table 8.6, below,
illustrates the numbers of children experiencing mobility changes between
1997 and 2003, when this approach is used.
No. (%) attending
schools in different
SALs in 1997 & 2003
No. (%) attending
schools in different
SPs in 1997 & 2003
No. (%) attending
schools in different
MPs in 1997 & 2003
Full Sample with
school location data
(n=1206)
782 (64.84%) 561 (46.52%) 244 (20.23%)
School changers only
(n=831)
779 (93.74%) 560 (67.39%) 244 (29.36%)
Children moving
between two primary
schools (n=423)
398 (94.09%) 282 (66.67%) 128 (30.26%)
Children moving
between primary and
high school (n=400)
373 (93.25%) 275 (68.75%) 116 (29.00%)
Table 8.6: Numbers of children moving between schools in different geographical
areas, disaggregated by schooling phase
What is particularly striking about Table 8.6 is that the proportion of children
experiencing changes in the area in which they attend school again appears to
be very similar for children moving between primary schools, and those
transitioning to secondary schools. Chi-square tests find no significant
266
difference in the distribution of mobility at the SAL, SP or MP level on the
basis of whether a child is moving between primary schools or to a high
school.
Exploring the changes in distances travelled for each of these groups of mobile
children, however, reveals interesting results (see Table 8.7 below). Primary
school children who are moving between schools in two different geographical
areas appear to be doing so primarily by decreasing their distance travelled,
while high school children appear to be doing so by increasing their distance
travelled. This is particularly significant for the SP level. It may be that any
relationship at the MP level is being partially obscured by the relatively small
numbers moving between different MPs, as well as the typically larger
distance that needs to be travelled to bring someone across an MP boundary.
Mean
change in
distance
Median
change in
distance
No. (%)
travelling
further in 2003
χ2 for children
travelling
further
Children
changing
from
school in
one SP to
another
High
school
(n=267)
0.830km 0.584km 163 (61.05%)
χ2(1)=7.2998
Pr = 0.007 Primary
schools
(n=269)
-1.683km -0.716km 133 (49.44%)
Children
changing
from
school in
one MP to
another
High
school
(n=110)
0.820km 5.329km 66 (60.00%)
χ2(1)=3.1536
Pr = 0.076 Primary
schools
(n=122)
-2.814km -1.096km 59 (48.36%)
Table 8.7: Changes in distance from home to school experienced by children
moving between schools in different geographic areas, disaggregated by schooling
phase (Note: Sample size is not the same as in Table 8.6 as only children for whom
full schooling and residential data is available are included in Table 8.7)
The final question explored in this section is whether there is change over time
with regards to the proportions of children attending school in the same area in
which they live, and whether these are the same children in 1997 and 2003.
267
Table 8.8, below, shows the numbers of children attending schools in the same
area in which they live. At the SAL level, very few children attend school in
the same area as their home at either point in time, and the proportion of
children who change schools but continue to attend school in the same SAL in
which they live is extremely small. Given the small size of the SALs, this is
unsurprising. At the SP level, a higher proportion of children, roughly a quarter
of the sample, are found attending school in the same SP in which they live at
both points of time, even when only those children who have changed schools
are considered. Finally, at the MP level, over half the sample is attending
school in the MP in which they live in both 1997 and 2003. Chi-square tests
indicate that the differences in the distributions of sample members attending
school in their home SP or MP is highly significant, regardless of the group
considered. Results at the SAL level are less clear, which is probably related to
the extremely small numbers of children attending school in their home SAL.
Number (%) attending
school in same SAL as
home
Number (%) attending school in
same SP as home
Number (%) attending school in
same MP as home
Time
point
1997 2003 Both 1997 2003 Both 1997 2003 Both
Full
Sample
(n=1180)
80
(6.78%)
47
(3.98%)
33
(2.80
%)
479
(40.59%)
437
(37.03%)
335
(28.39%)
851
(72.12%)
829
(70.25%)
748
(63.39%)
School
changers
only
(n=810)
56
(6.91%)
23
(2.84%)
9
(1.11
%)
350
(43.21%)
309
(38.15%)
207
(25.56%)
585
(72.22%)
563
(69.51%)
482
(59.51%)
Children
moving
between
primary
schools
(n=411)
20
(4.87%)
5
(1.22%)
1
(0.24
%)
165
(40.15%)
168
(40.88%)
105
(25.55%)
286
(65.59%)
289
(70.32%)
239
(58.15%)
268
Children
moving
between
primary
and high
school
(n=391)
36
(9.21%)
18
(4.60%)
8
(2.05
%)
180
(46.04%)
137
(35.04%)
98
(25.06%)
292
(74.68%)
267
(68.29%)
236
(60.36%)
Table 8.8: Children attending school in the same area as their home, for three
different levels of geography, disaggregated by schooling phase
In line with earlier findings, there is again little evidence for an impact of
schooling phase on the proportions of children attending schools within the
area in which their home is located. However, there is evidence that the
members of these groups change quite substantially over time, particularly at
smaller levels of geography. The finding that only about a quarter of the
sample is educated within the SP in which they live at both points in time is
particularly clear evidence for the limited proportion of children who are
consistently educated at local schools. Similarly, less than two thirds of
children attend schools within their home MP at both points in time.
8.3.3 Nearest school
A final approach to understanding changes in mobility is to look at the
proportion of children attending their nearest school in 1997 and 2003. Table
8.9, below, shows the numbers of sample members attending their nearest
public school in 1997, in 2003, and at both points in time. Although results are
not presented below, the same tests were conducted including private schools,
and in all instances produced extremely similar results. For all time points, as
has been discussed previously, the numbers of children attending their nearest
public school are extremely low, and the numbers who attended their nearest
schools at both points in time are lower still.
269
No. (%)
attending
nearest
school in 1997
No. (%)
attending
nearest school
in 2003
No. (%)
attending
nearest school
in 1997 & 2003
Chi-squared
test for
similarity in
1997 and 2003
Full Sample
(n=1178)
211 (17.91%) 214 (18.17%) 100 (8.49%)
χ2(1)= 147.6961
Pr=0.000
School changers
only (n=808)
136 (16.83%) 141 (17.45%) 28 (3.47%)
χ2(1)= 1.1176
Pr=0.290
Children moving
between
primary schools
(n=414)
67 (16.18%) 51 (12.32%) 0 (0.00%)
χ2(1)=11.2308
Pr=0.001
Children moving
between primary
and high school
(n=394)
69 (17.51%) 90 (22.84%) 28 (7.11%)
χ2(1)= 14.9315
Pr=0.000
Table 8.9: Children attending their nearest grade-appropriate public school in 1997,
2003 and at both points, disaggregated by schooling phase
When only children moving between two primary schools are considered, there
is a highly significant decrease in the proportion of children attending their
nearest school between 1997 and 2003. By contrast, amongst children
transitioning to secondary school, there is a significant increase in the
proportion attending their nearest school.
There are several potential explanations for this pattern. Firstly, obviously, if a
child changes schools, but not home location or phase of schooling, he or she
cannot possibly attend the nearest school at both points in time. By contrast, if
the child changes schooling phase, he or she is able to attend the nearest school
at both points in time. This is one reason to expect a higher proportion of
children attending their nearest school to be found amongst those transitioning
to high school. Secondly, as mentioned previously, high schools tend to be
larger, and more widely spaced, than primary schools. As a result, a child
choosing between high schools has a smaller number of options available to
him or her. The chances of attending the nearest school are therefore higher.
The different sizes and distributions of primary and high schools also explains
why the finding that children transitioning to high school are simultaneously
270
more likely to experience an increase in travel distance and also increased
likelihood of attending their nearest school, is not inconsistent.
Of the 354 children in the full sample who attended their nearest school at
either point in time, 100, or 28.25%, attended their nearest school at both time
points. Of the 250 school changers who attended their nearest school at either
point in time, 28, or 11.20%, attended their nearest school at both points in
time. Finally, of the 131 children transitioning to high school who attended
their nearest school at either point in time, 28, or 21.37%, attended their
nearest school at both points in time. These figures echo the findings presented
earlier in this chapter that children who change school, whether at the primary
or high school level, are less likely to be attending their nearest school in 2003
than children who do not change school.
8.4 Correlates of type of mobility change for primary
school school-changers
The previous section has documented the differences between mobility
changes related to the transition to high school, and those related to children
moving between two primary schools. This section now focuses on those
children changing between primary schools, and explores the correlates
associated with different patterns of mobility change. Results presented below
are based on the 409 sample members known to have changed primary schools
between 1997 and 2003. Of course, given that we only have primary school
change data for the non-random sub-group of the study sample who have not
progressed to high school by 2003, it is not appropriate to assume that findings
will be representative of all children who move between schools during their
primary schooling. However, findings will provide a preliminary idea of
correlates of changing mobility, which will assist with hypothesis and theory
development.
271
8.4.1 Straight-line distance
Individual level variables
There was no evidence for any relationship between race or gender and the
nature of mobility change associated with primary school change. Children
with a later first school enrolment were more likely to decrease distance from
home to school when changing primary schools (Wilcoxon rank-sum test;
Pr=0.0089). Children who have never experienced a grade repetition were also
more likely to decrease distance from home to school (Wilcoxon rank-sum
test; Pr=0.0092).
Household level variables
Children with mothers with very little education (functionally illiterate), and
children whose mothers have completed grade 10 or higher are significantly
more likely to experience an increase in distance from home to school than
children whose mothers have intermediate levels of education (χ2(1)= 8.4919,
Pr=0.075). Children whose mothers have completed grade 10 or higher also
experience a mean increase in distance, compared to a mean decrease in
distance for all other groups. There was no evidence for any difference in
mobility change on the basis of maternal marital status, SES in either 1997 or
2003, or change in SES between these two years.
Community level variables
Children living in the wealthiest and poorest SAL areas were most likely to
experience a substantial decrease in distance travelled (Wilcoxon rank-sum
test; Pr=0.0577). By contrast, children in areas with moderate poverty levels
were more likely to experience a small increase in travel. The same pattern,
although with a lower significance level (Pr=0.0752), was found at the SP
level. There was, however, no evidence for a relationship between MP area
poverty and change in distance from home to school.
272
8.4.2 Census geography
Two ways of measuring change in mobility through the use of census data
were presented earlier in this chapter. The first was to explore whether a child
moved between schools in two different levels of geography between 1997 and
2003. The second was to explore whether a child moved either into, or out of, a
school in the same geographic area in which he or she lived. In the first part of
this section, children who moved between primary schools in different
geographic areas will be compared to primary school changers who did not
move between areas. In the second part of this section, children who moved
into a school in the same area as their home will be compared to those children
who moved out of a school in the same area as their home.
Children moving between schools in different areas
SAL
Of the 423 children still in primary school in 2003 who changed school
between 1997 and 2003, only 25 moved between two schools in the same SAL,
while the remaining 398 moved between schools in different SALs. There was
no evidence that these two groups of children differed in any way with regards
to race, gender, age at first enrolment in school, and whether or not they had
ever experienced grade repetition. There was also no evidence for any
difference with regards to any of the household level variables explored,
namely maternal education, maternal marital status, household SES in 1997,
household SES in 2003, or change in household SES between 1997 and 2003.
Finally, although there is no evidence that the groups differ with respect to
SAL poverty, there is evidence that they differ in both SP area poverty (χ2(4)=
10.8582, Pr=0.021) and MP area poverty (χ2(2)=5.2592, Pr=0.057). Those
children who move between schools in different SALs tend to live in
somewhat more affluent SP and MP areas.
273
SP
Of the 423 school-changers still in primary school in 2003, 282 (66.67%)
moved between schools in different SPs, while 141 (33.33%) moved between
two schools in the same SP. As was the case at the SAL level, there is no
evidence that children moving between schools in the same SP differed
significantly from those moving between schools in different SPs with regards
to any of the individual level variables considered (race, gender, age at first
enrolment, and grade repetition). Children moving between schools in two
different SPs did, however, have more highly educated mothers (χ2(4)=10.6102,
Pr=0.031), although there was no significant difference with respect to
maternal marital status. Those children who moved between schools in
different SPs were likely to have a higher SES in both 1997 (χ2(4)=20.2271,
Pr=0.000) and 2003 (χ2(4)=13.7227, Pr=0.008). There was no evidence that the
two groups of children differed with respect to the poverty levels of the SALs
or SPs in which they lived. Children moving between schools in different SP
areas were, however, more likely to live in an extremely advantaged or
disadvantaged MP (χ2(2)=14.5783, Pr=0.001). In summary, movement between
schools in different SPs was associated with higher maternal education and
higher household SES, and with living in either a particularly advantaged, or
disadvantaged MP.
MP
128 (30.26%) of the school changers still in primary school in 2003 moved
between schools in two different MPs, while the remaining 295 (69.74%)
moved between two schools in the same MP. There was no evidence that these
two groups of children differed with respect to race, gender, age at first
enrolment, grade repetition, maternal education, or maternal marital status.
They did, however, differ significantly with respect to household SES in 1997
(χ2(4)=13.2473, Pr=0.010) and in 2003 (χ
2
(4)=22.0211, Pr=0.000), although not
with respect to change in SES between these two points. In both 1997 and
2003, those children moving between schools in two different MPs were more
274
likely to be from more affluent homes. There was no evidence that the two
groups of children differed with respect to the poverty area of the SALs in
which they lived, but children moving between schools in different MPs were
more likely to be living in particularly disadvantaged SP (χ2(4)=10.3958,
Pr=0.034) and MP (χ2(2)=11.2848, Pr=0.004) areas.
Children moving into, and out of, schools in the same area as their
home
SAL
At the SAL level, the number of children attending school in the same area as
their home is extremely small (n=20 in 1997; n=5 in 2003). Due to this small
sample size, no further analyses will be conducted on these groups.
SP
At the SP level, the numbers are somewhat higher, with 60 children attending
primary school in the same SP as their home in 1997, but not 2003, and 63
doing so in 2003, but not 1997. There is no evidence that these two groups of
children differ with respect to race or gender. The group of children attending
local schools in 2003, but not 1997, is however significantly more likely to
have first enrolled in school at an older age (χ2(1)=4.2981, Pr=0.038). They are
also less likely to have ever repeated a grade (χ2(1)=7.9301, Pr=0.005). There is
no evidence that the two groups of children differ with respect to maternal
education, maternal marital status, or household SES in either 1997 or 2003.
There is, however, evidence for a weakly significant relationship with the
change in household SES between 1997 and 2003 (χ2(6)=11.4982, Pr=0.063,
with children whose SES has fallen between 1997 and 2003 being more likely
to move to a school in the same SP as their home. A weakly significant
relationship was also found with both SAL (χ2(4)=8.3838, Pr=0.076) and SP
(χ2(4)=9.6659, Pr=0.046) poverty levels. In both cases, children in areas with
intermediate poverty levels were more likely to be moving away from schools
located in the same area as their home, while children in areas with either very
275
high or low poverty levels were more likely to be moving into schools in the
same area as their home. There was no evidence of any relationship between
mobility change and the poverty level of the MP in which the child lived.
MP
At the MP level, of the 427 children still in primary school in 2003 who
changed schools at least once, 47 had been attending a school in their MP in
1997, but by 2003 no longer did so. A further 50 children who had not been
attending a school in their MP in 1997 were doing so by 2003. There was no
evidence that these two groups of children differed with respect to either race
or gender. Children moving into a school in the same MP as their home were,
however, more likely to have started school later (χ2(1)=9.9730, Pr=0.002), and
never have repeated a grade (χ2(1)=5.8091, Pr=0.016), than those children
moving away from schools in the same MP as their home.
The two groups of children did not differ with respect to any of the household
level variables considered (maternal education, maternal marital status,
household SES in either 1997 or 2003, and the change in SES between 1997
and 2003). Children moving out of schools in their home SAL (χ2(4)=8.1504,
Pr=0.090) and SP (χ2(4)=12.7700, Pr=0.012) areas were more likely to be living
in areas with intermediate levels of poverty, while those children moving into
schools in the same area as their home were more likely to be living in areas
with either very high or very low poverty levels. There was, however, no
evidence for any difference in MP poverty.
8.4.3 Nearest school
The final approach to measuring changes in mobility is to look at whether
children move into, or out of, their nearest grade-appropriate school. Between
1997 and 2003, 67 primary school school-changers moved out of their nearest
primary school into another primary school further from home, while 51
276
moved from a primary school further away into their nearest school. There was
no evidence that these two groups of children differed systematically with
regards to race, gender, age at first enrolment or grade repetition. Children
moving out of their nearest school tending to have mothers with higher levels
of education than those moving into their nearest schools (χ2(4)=10.2479,
Pr=0.034). There was no evidence of a relationship with maternal marital
status, household SES in either 1997 or 2003, change in household SES
between 1997 and 2003, or poverty at the SAL, SP or MP area levels.
8.4.4 Conclusion: primary school school-changers
Children who enrolled in school late for their age, and those who had never
repeated a grade were more likely to experience a decrease in distance from
home to school. Children living in the most advantaged and disadvantaged
SAL and SP areas were also more likely to experience a decrease in distance
from home to school. By contrast, children moving to schools further from
home between 1997 and 2003 tended to have mothers with particularly low, or
particularly high, levels of education.
Children moving between schools in different areas appeared to differ from
their peers moving between two schools in the same area primarily with
respect to maternal education, household SES, and the poverty level of the area
in which they lived. Moving between schools in different areas was typically
associated with relative advantage, although in some cases, there was a non-
linear relationship in which both the most and least advantaged children were
particularly likely to move between areas. Children moving into schools in the
same area as their homes differed from those moving away from these schools
in that they tended to have been older at their first enrolment in school, were
less likely to have repeated grades, and lived in areas with either particularly
low or high levels of poverty.
277
Children moving out of their nearest primary schools tended to have mothers
with higher levels of education than children moving into their nearest primary
schools, but otherwise the two groups did not differ significantly.
It is not clear that analyses presented here have generated any consistent
findings. This may be due to a non-representative sample, and fairly small
sample size in some of the analyses, or may simply indicate that mobility
change is not significantly related to any of the variables considered here.
However, extreme levels of SES, later first enrolment in school, and grade
repetition, were the variables most prominent in these analyses, which suggests
that they are likely to have stronger relationship to schooling change than the
other variables considered. Unfortunately, the nature of this relationship is not
clear at this point.
8.5 Correlates of type of mobility change for children
transitioning to high school
In this section, the correlates of different types of mobility associated with the
transition to high school are documented. Unless otherwise indicated, analyses
presented here refer to the 390 children in the sample known to have moved
from a primary to a high school between 1997 and 2003. Children enrolled at
combined schools are excluded from these analyses. Again, the non-random
nature of the sub-sample considered here must be emphasized, and findings
should be used primarily as a basis for the development of hypotheses
requiring further testing.
8.5.1 Straight-line distance
Individual level variables
A Kruskal-Wallis test found a significant relationship between change in
mobility and ethnicity (Pr=0.0365), with black children experiencing greater
increases in distance from home to school than coloured children. There was
278
no evidence of any relationship between change in distance from home to
school and child gender, age at first school enrolment, or grade repetition.
Household level variables
There was no evidence for a relationship between change in distance and
maternal education, but children of unmarried mothers were more likely to
experience larger increases in distance from home to school (Wilcoxon
ranksum, Pr=0.0028). Children in the most advantaged and disadvantaged SES
quintiles in 1997 were most likely to experience the largest fall in distance
from home to school (Kruskal Wallis, Pr=0.0040). Those in the middle quintile
in 2003, by contrast, were those experiencing the largest fall in distance
(Kruskal Wallis, Pr=0.0247). There was also no evidence for any relationship
between change in SES from 1997 to 2003 and change in distance from home
to school.
Community level variables
There was no evidence for any relationship between area poverty, at the SAL,
SP or MP level, and change in distance between home and school.
8.5.2 Census geography
Two sets of analyses relating to mobility change as defined by census
geography are presented for those children transitioning to high school in
2003. The first set of analyses compares children who move between schools
in the same area to those who move between schools in different areas. The
sample size for these analyses is 400, as the home address data is not used. The
second set of analyses compares those children who move away from a school
in their home area, and those children who move into a school in their home
area.
279
Children moving between schools in different areas
SAL
As was the case for children moving between primary schools, very few
children moving between a primary and high school remained in the same SAL
(n=27). Coloured children are more likely to remain in the same SAL than
black children (χ2(1)= 4.7763, Pr = 0.029). There is no evidence of a
relationship between whether children move between schools in the same or
different SALs and any of the other variables examined (gender, age at first
enrolment, grade repetition, maternal education, maternal marital status,
household SES in 1997 and 2003, change in household SES between 1997 and
2003, and poverty level of the SAL, SP and MP in which they child lives).
SP
125 (31.25%) of the children who moved from primary to high school moved
between two schools in the same SP, while the remaining 275 children moved
between two schools in different SPs. These proportions are again very similar
to those found amongst the children moving between two primary schools.
Coloured children were again more likely to move between two schools in the
same SP than black children (χ2(1)= 30.8927, Pr = 0.000). There is no evidence
for a relationship between moving between schools in different SPs and child
gender, age at first enrolment, grade repetition, maternal education, or maternal
marital status. In both 1997 (χ2(4)=12.3664, Pr=0.015) and 2003 (χ
2
(4)=16.7466,
Pr=0.002), children in the most extreme SES quintiles were most likely to be
moving between schools in the same SP, while children in the middle quintiles
were more likely to be moving between schools in different SPs. There was no
evidence for a relationship with change in household SES over time. Finally,
children living in poorer SAL (χ2(4)=8.0566, Pr=0.090), SP (χ
2
(4)=14.9720,
Pr=0.005) and MP (χ2(2)=9.9770, Pr=0.007) areas were more likely to be
moving between schools in two different SPs.
280
MP
Of the children moving from primary to high school, 284 (71.00%) move
between two schools within the same MP, while 116 (29.00%) move between
schools in two different MPs. Again, these figures are very similar to those
found for children moving between primary schools. Once again, coloured
children are significantly more likely to move between two schools in the same
MP than black children (χ2(1)= 9.8932, Pr = 0.002). Again, there is also no
evidence for a relationship with gender, age at first enrolment, grade repetition,
maternal education, or maternal marital status. In both 1997 (χ2(4)=11.8570,
Pr=0.018) and 2003 (χ2(4)=20.6648, Pr=0.000), the children from the poorest
households are more likely to move between schools within the same MP,
while children in the middle quintiles are more likely to move between schools
in different MPs. There is no evidence for a relationship with change in
household SES over time however. Children living in poorer SP
(χ2(4)=20.9806, Pr=0.000) and MP (χ
2
(2)=11.7971, Pr=0.003) areas are more
likely to move between schools in two different MPs than children in more
affluent areas, but there is no evidence for a relationship with SAL poverty.
Children moving into, and out of, schools in the same area as their
home
SAL
Altogether, 46 children transitioning to high school attend schools in the same
SAL as their home in either 1997 or 2003. Of these, 8 attend school in the
same SAL as their home at both points in time, while 28 do so only in 1997
and 10 only in 2003. Movement appears to generally be away from the home
SAL. Given the small numbers involved, no additional analysis is conducted at
this level.
SP
219 of the children transitioning to high school attend school in the same SP as
their home at some point. 98 attend a school in their home SP in both 1997 and
281
2003, 82 do so only in 1997, and 39 only in 2003. The analyses presented here
compare the group of 82 children attending only primary school in their home
SP, and the group of 39 children attending only high school in their home SP.
There is no significant difference between the two groups of children with
respect to race, gender, age at first enrolment, or grade repetition. Children
moving into the local SP for high school were more likely to have mothers
with intermediate levels of education, while those moving out of the local SP
were more likely to have mothers with particularly high or low levels of
education (χ2(4)=11.3383, Pr=0.014). There is no evidence that the two groups
of children differ with respect to any of the other variables considered
(maternal marital status, household SES in 1997 or 2003, change in household
SES over time, and area poverty in the home SAL, SP and MP).
MP
Almost all of the children transitioning to high school, 325, attend a school in
the same MP as their home at some point, and 236 are attending school in their
home MP at both points in time. Only 56 attend a school in their home MP in
1997 but not 2003, while even fewer, 31, do so only in 2003. The analyses
presented below compare the group of 56 children moving out of their home
MP for high school with the 31 children moving into their home MP. The two
groups of children do not differ with regards to gender, age at first enrolment,
grade repetition, or maternal education. Children moving to a high school in
their home MP are, however, substantially more likely to be coloured than
black (χ2(1)= 11.4483, Pr = 0.001), and are more likely to have married mothers
(χ2(1)=6.2554, Pr=0.012). Although the two groups of children do not differ
with regards to household SES in 2003, or change in SES over time, children
moving to a high school in their home MP were likely to come from either
particularly advantaged or disadvantaged households in 1997 (χ2(4)=10.9760,
Pr=0.024). Children moving to high schools in their home MP were also more
likely to live in particularly advantaged or disadvantaged SAL (χ2(4)=14.1911,
282
Pr=0.008), SP (χ2(4)=20.4014, Pr=0.000), and MP (χ
2
(2)= 14.8044, Pr=0.001)
areas.
8.5.3 Nearest school
The final set of tests compares those children in high school in 2003, who had
been attending their nearest school in 1997 and had moved to a school further
away by 2003 (n=41), and those children who attended their nearest school in
2003, but had not done so in 1997 (n=62). Movement towards the nearest
school appears to be somewhat more prevalent than movement away from the
nearest school, which may be explained by the smaller number of high schools
available. The two groups of children do not differ significantly with respect to
gender, race, age at first school enrolment, maternal education, maternal
marital status, or the poverty of the SAL, SP or MP areas in which they live.
They do, however, differ with respect to household SES in both 1997
(χ2(4)=7.8985, Pr=0.092) and 2003 (χ
2
(4)=11.5980, Pr=0.024), with children
from more affluent households more likely to be moving away from their
nearest school when transitioning to high school.
8.5.4 Conclusion: mobility change associated with transition to
high school
Somewhat surprisingly, many of the findings for children transitioning to high
school echo closely those for children changing between two primary schools.
This suggests that the two processes may not be as distinct as initially
hypothesized. Change in distance from home to school was related to race,
with black children typically moving further afield, as well as with household
SES, with children at extreme levels of both advantage and disadvantage
tending to move closer to home. There was also evidence that the children of
unmarried mothers experienced a greater increase in distance than those with
married mothers. Movement between different geographic areas was also
associated in the same ways with race and household SES, although in this
283
case an additional relationship with area poverty was also identified. Children
living in poorer areas were more likely to move between schools in different
geographic areas. Children moving away from their nearest school were more
likely to be black and to have mothers who had either particularly high or low
levels of education than their peers moving into schools in the same area as
their homes. Finally, children moving away from their nearest schools for high
school tended to be more affluent than those moving towards their nearest
school.
8.6 Conclusion
This chapter has provided an overview of how schooling mobility changes
over time, and how these changes relate to child and family variables. One of
the most striking findings of this chapter is the prevalence of school changes
during primary schooling. This high level of school changes – in a sample of
children with consistent residential addresses– has important mobility
implications. Most notably, it means that not only are many primary school
children travelling considerable distances, but also that these distances are not
constant with time, despite the widespread preconception that this is a
relatively stable phase of schooling. Children living in poor areas, whose
household SES increases with time, and whose mothers have intermediate
levels of education appear to be the most likely to move between different
primary schools.
A second important outcome of this chapter is that there are both similarities
and differences in the changes to mobility associated with the transition to high
school and those associated with a transition between two different primary
schools. The overall tendency appears to be for children transitioning to high
school to increase their travel distance somewhat, even as a higher proportion
of these children also begin to attend the school nearest to their home. By
contrast, although a similar proportion of children moving between primary
284
schools increase their travel distance, these increases are typically smaller, and
the average change in distance is negative. These primary school children also
become less likely to attend the school nearest to their home. Although the
patterns of change at each level of census geography is similar for children
moving between primary schools, and those moving to high schools, this
disguises the fact that children at the primary school level are on average
moving to areas closer to home, while children at the high school level are
moving to areas further away from home. Despite these differences, however,
the correlates of particular kinds of changes to mobility are extremely
consistent across the two groups of children.
Thirdly, the different correlates associated with the different types of changes
to mobility substantiate the notion that the three measures of mobility used in
this project capture slightly different aspects of the phenomenon. The evidence
continues to be consistent with the notion that there are two separate processes
of school choice, and by extension mobility behaviours, in play in
contemporary urban Soweto-Johannesburg. One involves substantial travel,
and typically involves those children with access to the greatest resources, and
those living in areas in which services are not available. The other involves
mobility at relatively local levels, and seems to engage children with more
intermediate levels of resources and maternal education.
285
Chapter 9: Modelling educational
mobility
9.1 Introduction
Having explored the bivariate relationships between educational mobility and a
range of variables at the child, household, community and school levels, this
chapter combines these variables to develop a multivariate model predicting
educational mobility. This is done through a series of regression analyses,
using child, household, community and schooling data as independent
variables, and the various measures of educational mobility as dependent
variables.
As most of the school attribute variables presented in Chapter 7 were closely
related to mobility, but were also highly correlated with each other, principal
component analysis (PCA) was conducted on all school attribute variables that
were consistently found to be significantly related to mobility (school quintile,
school fees, school enrolment, percent black learners, school sector, historical
DET status, and pass rate) to generate a school quality index. This process was
repeated for the school attended by each child in 1997 and 2003, as well as for
the nearest grade-appropriate school to the child‘s home in 1997 and 2003. In
all cases, the eigenvalues of the first two components of the PCA were both
greater than 1, and were therefore both retained.
Transformation of some additional variables was also conducted to allow for
their use in a regression context. Race was re-coded into a binary variable,
coded 1 if the child was black African and 0 otherwise. Gender was similarly
recoded to 1 if the child was a boy and 0 if the child was a girl. Poverty data
for the area in which a child lived was used in its raw form, as opposed to the
quintile form, as was household SES. The only non-binary categorical
independent variable used was maternal education, and this was converted to a
286
series of dummy variables, with a maternal education level of Grade 5 or
below used as the base category.
In the first section of this chapter, regression models are developed to predict
the straight-line distance between home and school for both 1997 and 2003. In
the second section, logistic regression models are used to predict the likelihood
that that children attend schools located in the same Sub-Place (SP) and Main
Place (MP) areas in which they live, again for both 1997 and 2003. In the third
section of the chapter, logistic regression models are again developed, this time
to predict the likelihood that a child will attend his or her nearest grade-
appropriate school. Finally, the implications of these various models for the
hypotheses around mobility presented in previous chapters are discussed.
9.2 Straight-line distance
As noted in Chapter 5, the distribution of straight-line distance from home to
school across the sample is highly non-normal. A range of transformations of
this variable were generated and tested for normality, but none were found to
be normal. The log transformation, however, was closest to a normal curve,
and as a result, is used as the dependent variable in this set of models. All of
the child, household and community level variables discussed in Chapter 5,
along with the PCA-generated variables relating to the quality of a child‘s
attended and nearest school, were considered for inclusion as independent
variables.
9.2.1 1997
Initially, a standard OLS regression was run, including all child, household and
community variables, as well as the derived scores for schools attended and
nearest to a child‘s home. This model, however, featured extensive
multicollinearity. The correlation matrix for all variables was examined, and
287
one major area of concern was identified. This was the strong relationship
between area poverty and the attributes of the school closest to the child‘s
home. As the attributes of the nearest school were also closely related to the
attributes of the school attended, a decision was made to retain the area poverty
level variables only. Due to the high correlation between poverty levels at the
various different levels of geography (SP and MP in particular), each was
tested for significance. , and the decision was made to retain only SP poverty
in the final model. Results for this model are described below, and presented in
Table 9.1.
Tests were conducted to explore concerns around cases exerting an undue
influence on the models. A number of outliers and cases with high leverage
were identified. As these cases appeared to be features of the data rather than
errors, they were not removed, but it should be noted that they may have
biased the results reported below. Although a Breusch-Pagan test found no
evidence of heteroskedasticity, both a visual examination of the variances and
a White test provided evidence of heteroskedasticity. The distribution of the
residuals also appeared normal on inspection, but a Shapiro-Wilk test rejected
this hypothesis. Due to these various concerns, the regression was repeated
using regression with robust standard errors, and also a robust regression
which weights cases differently to minimize the effects of cases with
particularly high influence. These results are also presented in Table 9.1, and it
is clear that they are not substantially different from the results of the standard
regression. For all models, tests indicated that omitted variable bias was
present, and this persisted even with the inclusion of a range of other variables.
However, a model specification link test failed to reject the assumption that the
model was correctly specified.
The results of these regressions, presented in Table 9.1 below, suggest that the
coefficients on race, maternal education, and the attributes of the school
attended are most significant in predicting the distance a child travels from
288
home to school in 1997. Across all models, black children are significantly
more likely to travel further than children of other race groups. The
relationship between distance and maternal education is somewhat more
complex, and slightly counterintuitive, as it suggests that children whose
mothers have very limited formal education (grades 5 or 6 only) tend to travel
the furthest, followed by those whose mothers have reached grade 11-12. As
shown by the coefficient on the first school attributes variable, children
attending more advantaged schools, are also likely to travel further than
children attending less advantaged schools. There is some weak evidence that
maternal marital status may also be related to mobility, with children of
married mothers tending to travel somewhat further. There is, however, no
evidence that household SES shapes distance, although it seems more likely
that the effect of household SES has already been captured in other relatively
highly correlated variables, such as school quality and maternal education, than
that it doesn‘t relate to mobility at all. There is also evidence that the poverty
level of the area in which a child lives plays a limited role in predicting travel
distance, with children in poorer areas tending to travel further. Overall, these
findings sustain the hypothesis that black children from homes where mothers
have at least some education, attending relatively privileged schools, but still
living in less advantaged areas, tend, overall, to travel somewhat further to
school than their peers.
Standard regression Standard regression
with robust errors
Robust regression
Black African race 2.053 (0.174) *** 2.053 (0.187) *** 2.438 (0.160) ***
Male gender 0.005 (0.092) 0.005 (0.094) 0.010 (0.084)
Later age at first
school enrolment
0.097 (0.094) 0.097 (0.094) 0.064 (0.086)
Maternal education
grade 5-7
0.793 (0.248) *** 0.793 (0.235) *** 0.527 (0.227) **
Maternal education
grade 8-10
0.395 (0.200) ** 0.395 (0.167) ** 0.242 (0.183)
Maternal education
grade 11-12
0.561 (0.210) *** 0.561 (0.184) *** 0.369 (0.192) *
Maternal education
post-school
0.404 (0.260) 0.404 (0.240) * 0.393 (0.238) *
289
Maternal marital
status
0.176 (0.104) * 0.176 (0.108) 0.184 (0.095) *
Household SES 1997 0.013 (0.032) 0.013 (0.034) 0.026 (0.029)
School attended
1997 attributes
component 1
9
0.644 (0.037) *** 0.644 (0.034) *** 0.696 (0.034) ***
School attended
1997 attributes
component 2
0.057 (.0481) 0.057 (0.042) 0.051 (0.044)
SP poverty (raw
score)
0.068 (0.038) * 0.068 (0.045) 0.080 (0.035) **
Constant -1.940 (0.247) *** -1.940 (0.223) *** -2.205 (0.226) ***
*significant at P<0.1
level
** significant at
P<0.05 level
***significant at
P<0.01 level
No. of obs = 742
F(12, 729) = 36.70
Prob > F = 0.0000
R-squared = 0.3766
Adj R-squared =
0.3663
Root MSE = 1.236
Number of obs = 742
F(12, 729) = 54.78
Prob > F = 0.0000
R-squared = 0.3766
Root MSE = 1.236
No. of obs = 742
F(12, 729) = 52.10
Prob > F = 0.000
Table 9.1: 1997 regression results. Figures in parentheses are standard errors.
9.2.2 2003
A similar process to that described above was used to develop the model for
2003. The major difference in model construction for 2003 is that a variable
indicating whether the child repeated a grade between 1996 and 2003 is
included. As was the case with the 1997 model, tests indicated concerns
around cases with particularly high influence, as well as the presence of some
heteroskedasticity. As a result, Table 9.2, below, reports the results of a
standard regression, along with the results of the regression re-run with robust
errors, and a robust regression to reduce the impact of particularly influential
cases. Once again, evidence for omitted variable bias could not be eliminated,
although a model specification link test failed to reject the assumption that the
model was correctly specified.
9 The strongest loadings for school component 1 in 1997 were historical DET status and
percentage black students, both loading negatively. Component 2 loadings were matric pass
rate and school fees, loading positively, and school enrollment, loading negatively.
290
The patterns identified in the 2003 regressions are very similar to those
identified for 1997, with race, school attributes, and maternal education
continuing to be the strongest predictors of distance from home to school. The
role of maternal education in shaping mobility has, however, changed
somewhat, becoming more linear. Maternal education at the grade 5 to 7 level
no longer contributes to mobility, and higher maternal education is typically
associated with higher mobility. Black children continue to travel substantially
further than members of other race groups, as do children attending more
advantaged schools. Evidence for a role of residential area poverty or maternal
marital status in determining mobility has largely disappeared in 2003. Overall,
however, the relative similarity of these results over time, despite the
substantial change in the age of the children, seems to suggest that the
determinants of mobility remain fairly consistent, throughout the primary
school years, and even into secondary schooling.
Standard regression Standard regression
with robust errors
Robust regression
Black African race 1.876 (0.216) *** 1.876 (0.269) *** 2.275 (0.209) ***
Male gender -0.153 (0.096) -0.153 (0.094) -0.212 (0.093) **
Late age at first school
enrolment
0.028 (0.100) 0.028 (0.100) -0.003 (0.097)
Grade repetition 0.018 (0.104) 0.018 (0.104) -0.026 (0.101)
Maternal education
grade 5-7
0.327 (0.255) 0.327 (0.249) 0.152 (0.247)
Maternal education
grade 8-10
0.398 (0.187) ** 0.398 (0.165) ** 0.324 (0.181) *
Maternal education
grade 11-12
0.488 (0.196) ** 0.488 (0.175) *** 0.378 (0.190) **
Maternal education
post-school
0.415 (0.248) * 0.415 (0.232) * 0.280 (0.240)
Maternal marital
status
0.086 (0.104) 0.086 (0.101) 0.064 (0.101)
Household SES 2003 0.112 (0.097) 0.112 (0.055) ** 0.077 (0.094)
School attended 2003
attributes component
110
0.696 (0.037) *** 0.696 (0.039) *** 0.746 (0.036) ***
10 For 2003, historical DET status and percent black students both loaded strongly negatively
on component 1, while school quintile and school fees loaded strongly positively. School
291
School attended 2003
attributes component
2
0.068 (0.053) 0.068 (0.053) 0.081 (0.051)
SP poverty (raw score) 0.071 (0.038) * 0.071 (0.047) 0.024 (0.037)
Constant -1.355 (0.283) *** -1.355 (0.282) *** -1.565 (0.274) ***
*significant at P<0.1
level
** significant at P<0.05
level
***significant at P<0.01
level
No. of obs = 543
F(13, 529) = 40.87
Prob > F = 0.0000
R-squared = 0.5011
Adj R-squared = 0.4888
Root MSE = 1.0728
Number of obs =
543
F(13, 529) = 43.06
Prob > F = 0.0000
R-squared = 0.5011
Root MSE = 1.0728
No. of obs = 543
F(13, 529) = 51.80
Prob > F = 0.0000
Table 9.2: 2003 regression results. Figures in parentheses are standard errors.
9.3 Census area geography
In this section, models are developed for whether children attend schools in the
same geographical area in which they live. Due to the very low numbers of
children attending school in their home Small Area Level (SAL), and the very
low numbers who do not attend school in their home Municipality (MN),
models are only pursued for mobility at the Sub-Place (SP) and Main-Place
(MP) levels. As the dependent variable used in these models is binary, a
logistic regression approach is used. Initially, the regression was run using the
same set of variables as presented in the distance to school regressions above.
However, tests indicated specification errors, which were reduced by replacing
the raw household SES scores with dummy variables for each quintile of
household SES. This improvement may be related to nonlinearities in the
relationship between household SES and the census area based measure of
mobility. In both 1997 and 2003, household SES quintile 1 (highest poverty)
was used as the base category. SP poverty was also treated as a categorical
variable, again to allow for nonlinearity in the relationship between SP poverty
and mobility behaviour. Additional tests for multicollinearity, goodness of fit,
and particularly influential cases were also conducted on all models before
they were finalised (Chen, Ender et al. 2011). Although at the end of this
enrolment loaded strongly negatively on component 2, while school fees again loaded strongly
positively.
292
process all of the models detailed below continued to provide some evidence
of specification error, there was no theoretically justifiable way to further
improve their specification.
9.3.1 Sub-Place level
Results for both 1997 and 2003 are presented in Table 9.3 below. Regressions
were run both with and without robust standard errors. However, as the
coefficients in both cases are identical, and significance levels did not change
substantially, only the results using robust standard errors are presented in
Table 9.3. Note that grade repetition was not included in the 1997 model. The
1997 model used the household SES poverty quintiles based on 1997 data,
while the 2003 model used the quintiles based on the 2003. Each model also
used the school attributes score based on the school attended at that point in
time.
The coefficients returned by the logistic regression for SP mobility in 1997
indicate that black children are again significantly more mobile, as black race
is negatively associated with the probability that a child attends a school in the
same SP as his or her home. By contrast, the association between maternal
education and mobility has disappeared. There is some evidence that children
in household SES quintiles 2 and 4 are more likely to attend schools outside of
their home SP. The evidence that the attributes of the school attended shape
mobility remains strong. The coefficients on both components of school
attributes indicate that children attending more advantaged schools are more
likely to attend a school located outside of their home SP. Finally, area poverty
is also significantly associated with mobility. Those children living in
wealthier areas are less likely than their peers living in poorer areas to be
attending schools in their home SPs.
The 2003 results are very similar to those for 1997. Black African race remains
a strongly significant predictor of mobility at the SP level, with black children
293
less likely to attend a school in their home SP. There is again no evidence that
maternal education is related to mobility at the SP level in 2003. Children
attending more advantaged schools, and those living in the least advantaged
areas, continue to be significantly more likely to be travelling to attend a
school outside of their home SP in 2003. Children with mid-range household
SES are also particularly likely to be attending school outside of their home
SP. There is also weak evidence that children with married mothers are more
likely to be travelling out of their home SP in order to attend school in 2003.
Overall, determinants of mobility at the SP level, in both 1997 and 2003,
appear to be fairly similar to determinants of distance from home to school,
with the exception that maternal education no longer appears to play a
significant role. This is critical, because of the close relationship between
maternal education and affluence, and suggests that household resource levels
may be less important in shaping mobility at the fairly small SP level of
geography, than in shaping distance from home to school.
1997: Logistic regression with
robust errors
2003: Logistic regression with
robust errors
Black African race -3.314 (0.480) *** -3.422 (0.725) ***
Male gender -0.222 (0.173) 0.005 (0.229)
Late age at first
school enrolment
-0.233 (0.178) 0.280 (0.238)
Grade Repetition 0.015 (0.245)
Maternal education
grade 5-7
-0.624 (0.434) -0.276 (0.573)
Maternal education
grade 8-10
-0.413 (0.358) -0.458 (0.436)
Maternal education
grade 11-12
-0.266 (0.380) -0.566 (0.471)
Maternal education
post-school
-0.258 (0.463) -0.231 (0.539)
Maternal marital
status
-0.164 (0.197) -0.433 (0.248) *
Household SES
quintile 2
-0.554 (0.244) ** -0.531 (0.323)
Household SES
quintile 3
-0.182 (0.250) -1.130 (0.330) ***
294
Household SES
quintile 4
-0.468 (0.274) * -0.269 (0.342)
Household SES
quintile 5 (most
advantaged)
-0.465 (0.376) 0.249 (0.391)
School attended
attributes
component 1
-0.947 (0.102) *** -1.240 (0.131) ***
School attended
attributes
component 2
-0.202 (0.099) ** -0.443 (0.147) ***
SP poverty quintile
2 (relatively low
area poverty)
-1.033 (0.349) *** -0.983 (0.627)
SP poverty quintile
3
-0.690 (0.422) -0.684 (0.645)
SP poverty quintile
4
-0.975 (0.427) ** -1.409 (0.669) **
SP poverty quintile
5 (highest poverty
areas)
-1.474 (0.464) *** -2.059 (0.698) ***
Constant 3.967 (0.535) *** 3.602 (0.760) ***
*significant at P<0.1
level
** significant at
P<0.05 level
***significant at
P<0.01 level
No. of obs = 742
Wald chi2(18) = 94.79
Prob > chi2 = 0.0000
Pseudo R-squared = 0.1940
Log likelihood = -410.34555
No. of obs = 543
Wald chi2(19) = 124.39
Prob > chi2 = 0.0000
Pseudo R-squared = 0.2849
Log likelihood = -252.24727
Table 9.3: 1997 & 2003 SP mobility regression results. Figures in parentheses are
standard errors.
9.3.2 Main Place level
Results for logistic regressions at the MP level for both 1997 and 2003 are
presented in Table 9.4 below. Again, only results with robust standard errors
are presented. Grade repetition was not included in the 1997 model. The 1997
model used the household SES poverty quintiles based on 1997 data, while the
2003 model used the quintiles based on the 2003. Each model also used the
school attributes score based on the school attended at that point in time.
295
In 1997, black children, and those attending comparatively advantaged schools
were less likely to be attending a school in their home MP. Although there is
weak evidence that maternal education at the grade 11 to 12 level is associated
with a higher probability of attending school outside of the MP in which the
home is located, there appears to be little role for maternal education in
determining MP level mobility. Similarly, evidence for a role of household
SES is also limited, with only children falling into quintile 2 being
significantly more likely to attend a school outside of their home MP. By
contrast, however, and as was the case at the SP level, there is strong evidence
that attending a more advantaged school, and living in an area with higher
poverty levels, were both associated with an increased probability of attending
school outside of the home MP.
The 2003 results for mobility at the MP level are fairly similar to those for
1997. Once again, black children are significantly more likely to be attending a
school outside of their home MP. There is some evidence that maternal
education at the post-school level increases the likelihood of a child attending a
school outside of their home MP. Children enrolled in more affluent schools
are also more likely to be travelling to a school outside of the MP in which
they live. Finally, children living in SP areas with higher levels of poverty are
more likely to be attending a school outside their home MP.
Overall, these results suggest that the determinants of MP mobility do not
change substantially over time, and indicate that black children, attending
fairly advantaged schools, but living in poor areas, are most likely to travelling
outside of the MP in which their home is located in order to go to school.
Interestingly, there was some evidence that particularly high levels of maternal
education were predictive of mobility at the MP level, which was not the case
for mobility at the SP level. This may be reflective of the typically longer
distances associated with MP mobility compared to SP mobility, which would
cause it to require greater levels of resource investment.
296
1997: Logistic regression with
robust errors
2003: Logistic regression with
robust errors
Black African race -3.820 (0.533) *** -6.816 (3.574) *
Male gender 0.037 (0.255) -0.019 (0.292)
Late age at first school
enrollment
-0.305 (0.244) -0.002 (0.341)
Grade repetition -0.091 (0.350)
Maternal education grade
5-7
-1.064 (0.802) -0.465 (1.043)
Maternal education grade
8-10
-1.057 (0.700) -0.441 (0.649)
Maternal education grade
11-12
-1.340 (0.707) * -0.529 (0.626)
Maternal education post-
school
-1.254 (0.773) -1.472 (0.735) **
Maternal marital status -0.056 (0.269) -0.137 (0.339)
Household SES quintile 2 -0.985 (0.389) ** -0.243 (0.504)
Household SES quintile 3 -0.484 (0.390) -0.798 (0.519)
Household SES quintile 4 -0.449 (0.403) -0.505 (0.518)
Household SES quintile 5
(most advantaged)
-0.598 (0.511) 0.029 (0.546)
School attended
attributes component 1
-1.324 (0.116) *** -1.931 (0.202) ***
School attended
attributes component 2
0.042 (0.110) 0.133 (0.180)
SP poverty quintile 2
(relatively low area
poverty)
-1.532 (0.611) ** -2.466 (0.812) ***
SP poverty quintile 3 -1.883 (0.694) *** -3.013 (0.820) ***
SP poverty quintile 4 -2.018 (0.708) *** -3.410 (0.809) ***
SP poverty quintile 5
(highest poverty areas)
-3.033 (0.723) *** -3.716 (0.836) ***
Constant 8.188 (0.904) *** 11.037 (3.790) ***
*significant at P<0.1 level
** significant at P<0.05
level
***significant at P<0.01
level
No. of obs = 742
Wald chi2(18) = 201.40
Prob > chi2 = 0.0000
Pseudo R-squared = 0.4315
Log likelihood = -232.40251
No. of obs = 543
Wald chi2(1) = 201.40
Prob > chi2 = 0.0000
Pseudo R-squared = 0.56102
Log likelihood = -148.17552
Table 9.4: 1997 and 2003 MP mobility regression results. Figures in parentheses are
standard errors.
297
9.3.3 Census area mobility discussion
At both the SP and MP level, and in both 1997 and 2003, black race is a strong
predictor for increased mobility, as is attending a comparatively advantaged
school, and living in a less affluent SP. Interestingly, maternal education,
which did predict distance from home to school, does not predict SP mobility
at all, and only the highest levels of maternal education are associated with MP
mobility. This may indicated that the role of household resources in
determining mobility is more limited when looking at travel between smaller
areas. This fits well with the argument that two patterns of mobility – one
involving substantial travel to historically advantaged schools, and demanding
more resources, and one involving more localized travelling and requiring
fewer resources – are evident in urban South Africa. SP mobility appears to fit
into the lower-resource pattern of mobility, while MP mobility is more closely
linked to the higher-resource pattern.
9.4 Nearest school analysis
The final set of models presented in this chapter use attendance at a child‘s
nearest grade-appropriate school as the dependent variable. Logistic
regressions were conducted using both the variable indicating whether the
child attended his or her nearest public school, and the variable indicating
whether the child attended his or her nearest public or independent school. As
the results in both cases were essentially identical, only the results for the
variable including both public and independent schools are presented here.
Constructing this model was challenging, as it was not possible to obtain a
particularly good fit for either the 1997 or 2003 models. This may be due to
omitted variables helping to determine whether a child attends his or her
nearest school, and may also be due to a fairly high level of randomness in this
particular outcome. However, given that both models pass all other goodness
of fit tests, and give no indication of specification errors, the low R-squared in
the context of a logistic regression is not necessarily of great concern.
298
Again, both a standard logistic regression, and a logistic regression with robust
errors were conducted for each time point, and provided very similar results.
Table 9.5, below, presents the results for the regressions with robust errors for
1997 and 2003. Grade repetition was only included in the 2003 model. The
1997 model was conducted using 1997 household SES and school attributes,
while the 2003 model used these variables for the 2003 time point.
The 1997 model indicates that black children are significantly less likely to
attend their nearest school. Children who start school late for their age are
more likely to attend their nearest schools than those who started early. This is
interesting, as it the effect of age at enrolment appears to be moving in the
opposite direction than that suggested in the bivariate analyses. The coefficient
may be driven by the fact that more affluent children are likely to start school
later, and are also more likely to attend their nearest school, even though their
average travel distance is greater than that of their less affluent peers. There is
no evidence for any relationship between likelihood of attending the nearest
school, and either household SES or maternal education. There is, however, a
strong significant relationship between the nature of the school the child
attends in 1997, and whether this is the school closest to his or her home.
Children attending more advantaged schools are less likely to be attending the
school nearest to their home. Finally, there is also evidence that children living
in areas with mid-range levels of poverty are least likely to attend their nearest
schools. Children living in the most disadvantaged areas are, however, still less
likely to attend their nearest school than their peers in more advantaged areas.
This is in line with the bivariate findings presented in earlier chapters.
As in 1997, black children in 2004 remain significantly less likely to attend
their nearest grade appropriate schools than children of other races. The
relationship between later age at first enrolment and attending the nearest
school has, however, disappeared by 2003. There is weak evidence that
299
children of mothers with limited primary school education (Grades 5-7) may
be more likely to attend their nearest school than children of mothers with no
formal education, but other than this, there is again no evidence for a
relationship between maternal education and whether a child attends his or her
nearest school. Similarly, there is also some evidence that children in
households with mid-range SES are more likely than the most disadvantaged
children to attend a school other than their nearest school, but other than this,
no evidence for a relationship between household SES and whether or not a
child attends his or her nearest school. A statistically significant negative
relationship remains between how advantaged the school a child attends is, and
the likelihood that the child is attending his or her nearest school. Finally, there
is also a strong negative relationship between area poverty and probability of
attending the nearest school, with children living in the poorest areas least
likely to do so.
The 1997 and 2003 results are once again largely consistent over time, and
highlight the role of race, school quality, and area poverty in determining
whether a child attends his or her nearest school. The very limited evidence for
any role of maternal education or household SES in determining whether a
child attends his or her nearest school also supports the argument that choosing
not to attend the nearest school is a form of school choice which requires fairly
little in terms of the investment of resources.
1997: Logistic regression with
robust errors
2003: Logistic regression with
robust errors
Black African race -2.251 (0.431) *** -1.477 (0.643) **
Male gender -0.138 (0.199) 0.145 (0.263)
Late age at first school
enrollment
0.504 (0.201) ** -0.150 (0.280)
Grade repetition -0.181 (0.285)
Maternal education
grade 5-7
-0.617 (0.509) 1.110 (0.641) *
Maternal education
grade 8-10
-0.371 (0.385) 0.151 (0.512)
Maternal education -0.388 (0.424) 0.242 (0.539)
300
grade 11-12
Maternal education
post-school
-0.325 (0.553) 0.636 (0.657)
Maternal marital status -0.242 (0.241) 0.100 (0.280)
1997 household SES
quintile 2
-0.281 (0.300) -0.215 (0.356)
1997 household SES
quintile 3
-0.326 (0.323) -0.858 (0.416) **
1997 household SES
quintile 4
0.194 (0.318) -0.172 (0.397)
1997 household SES
quintile 5 (most
advantaged)
0.062 (0.421) -0.311 (0.481)
School attended 1997
attributes component 1
-0.626 (0.099) *** -0.711 (0.143) ***
School attended 1997
attributes component 2
-0.233 (0.102) ** -0.658 (0.172) ***
SP poverty quintile 2
(relatively low area
poverty)
-0.693 (0.333) ** -0.566 (0.582)
SP poverty quintile 3 -1.550 (0.433) *** -1.263 (0.587) **
SP poverty quintile 4 -0.962 (0.403) ** -1.291 (0.599) **
SP poverty quintile 5
(highest poverty areas)
-0.914 (0.436) ** -1.612 (0.658) **
Constant 1.477 (0.541) *** 0.349 (0.733)
*significant at P<0.1
level
** significant at P<0.05
level
***significant at P<0.01
level
No. of obs = 738
Wald chi2(18) = 87.39
Prob > chi2 = 0.0000
Pseudo R-squared = 0.1241
Log likelihood = -327.56921
No. of obs = 541
Wald chi2(19) = 67.58
Prob > chi2 = 0.0000
Pseudo R-squared = 0.1631
Log likelihood = -207.7699
Table 9.5: 1997 and 2003 nearest school attendance logistic regression results.
Figures in parentheses are standard errors.
9.5 Conclusion
The data presented in previous chapters of this thesis have suggested that two
patterns of learner mobility are operating in the Johannesburg-Soweto area.
Firstly, there are children who are travelling particularly long distances to
attend school, typically at a fairly high economic and social cost. The second
pattern involves travel at a more local level, with children and their families
301
making active choices between those schools accessible from their homes. This
second pattern of mobility represents a substantially less resource-intensive
approach to engaging in school choice, and can be engaged in by a broader
group of children, and most notably those children whose mothers have only
fairly limited education. The results presented in this chapter are summarized
in Table 9.6, below, and provide additional support for presence of these two,
different, patterns of mobility.
Distance
from home
to school
School outside
of home SP
School outside of
home MP
Not attending
nearest school
Variables
associated
with
increased
mobility
Black race
Maternal
education
Higher
school
quality
Black race
1997 Household
SES quintiles 2
& 4 (1997 only)
2003 Household
SES middle
quintiles
Higher school
quality
Poorer SP area
Black race
Maternal
education Grade
11-12 (1997 only)
Maternal
education post-
school (2003 only)
1997 Household
SES quintile 2
(1997 only)
Higher school
quality
Poorer SP area
Black race
Younger age at
first school
enrolment (1997
only)
Very low
maternal
education (2003
only)
2003 Household
SES quintile 3
(2003 only)
Higher school
quality
Poorer SP area
Table 9.6: Summary of variables associated with increased mobility in the
regression models presented in this chapter
302
Evidence for the first, longer-distance pattern is provided in the regression
models for distance between home and school, and to a lesser degree, mobility
at the MP level of census geography. In addition to black race, and attending a
high quality school, distance from home to school is strongly related to higher
maternal education. The role of higher maternal education in increasing
mobility, however, largely disappears in all other models, with the exception of
mobility at the MP level. Maternal education, therefore, appears to be critical
to determining whether a child engages in a more costly form of mobility or
not. Although the absence of significant coefficients on household SES
variables is of some concern to the hypothesis that these forms of mobility are
pursued by more advantaged families, this is probably due to the strength of
the correlation between household SES and school quality, and the very
substantial coefficients attached to school quality in these regressions. Overall,
the models for distance from home to school, and for MP mobility suggest that
these forms of mobility are more demanding on families, and at the very least
require the additional social and human capital associated with a more highly
educated mother.
When looking at the results for SP mobility, and a child not attending his or
her nearest school, black race and the quality of the school attended continue to
be important predictors. Although, as noted, there is little or no evidence for a
role of maternal education in predicting these types of mobility, there is some
evidence that intermediate levels of household SES may play a role. These
combinations of significant variables support the argument that these are forms
of mobility that do not require particularly high levels of social or economic
capital, even though they are not completely cost free. The role of area poverty
in these models has also become more significant, suggesting that these forms
of mobility are likely to be more strongly driven by the nature of local
educational opportunities than by the resources available to a child and his or
her family.
303
Chapter 10: Conclusion
10.1 Introduction
This chapter begins with a brief overview and synthesis of the key findings
presented in this thesis, and then discusses the implications of these findings.
The original contributions – methodological, empirical and theoretical – that
this thesis makes to the existing body of scholarly literature are highlighted.
The contextual relevance of the study findings is discussed, with a brief
discussion of potential implications for school policy in South Africa. Finally,
limitations to the work presented here, along with suggestions for future work,
are presented.
10.2 Overview of key findings
Table 10.1 below outlines the key findings of this thesis, with respect to each
of the study‘s major aims. These findings are discussed at greater length in the
subsequent sections.
Objective Chapter Thesis Findings
To develop approaches to the
measurement of learner
mobility appropriate to the
South African context
3 & 5 -- Three different approaches to measuring
learner mobility were tested
-- Each approach provided different, but
complementary data
-- Using multiple approaches to measuring
learner mobility allowed for the identification of
two distinct patterns of mobility – one based
primarily on choice between local schools, and
one based on travel of substantial distances to
schools in more advantaged areas
To measure the extent of
learner mobility in post-
Apartheid Johannesburg-
Soweto, South Africa
5 -- Each approach to the measurement of learner
mobility provided evidence that learner mobility
is highly prevalent in Johannesburg-Soweto
-- Approximately a quarter of children travel
over 5km to school each way on a daily basis
-- Roughly 25% of children attend school outside
of the MP in which they live, while 60% attend
304
school outside of the SP in which they live
-- Less than 20% of children attend the grade-
appropriate school that is closest to their home.
-- This data suggests two distinct patterns of
mobility – one due primarily to choice between
fairly local schools, and involving relatively
limited travel, and one involving the choice of
historically advantaged schools substantially
further afield
To identify potential
determinants of learner
mobility at the child, family and
community level
6 &9 -- At the child level, bivariate analyses indicated
that mobility behaviour was related to race, age
at school enrolment, grade repetition, and
phase of schooling
-- At the family level, there was evidence for a
relationship between mobility and maternal
education and household SES
-- At the community level, there was evidence
for a non-linear relationship between mobility
and area poverty
-- Multivariate analysis suggested that race,
maternal education and area poverty where the
strongest and most consistent determinants of
mobility behaviour
To explore the relationship
between school attributes and
mobility behaviour
7 & 9 -- On average, children were found to be
attending a school more advantaged than the
school closest to their home
-- More advantaged children were found to be
attending more advantaged schools
-- Attending a comparatively advantaged school
was associated with greater engagement in all
forms of mobility
-- Children attending private schools, higher
quintile schools, schools with a lower proportion
of black learners, schools charging higher fees,
schools that did not historically fall under the
DET, and schools with higher pass rates were
more likely to be engaged in learner mobility
To identify whether and how
learner mobility changes as
children age
8 -- There was no clear evidence to suggest that
mobility behaviour changes substantially as
children age
-- Although there was some evidence that
children typically travelled slightly further to
high schools than primary schools, and were
slightly more likely to attend their nearest
school at the high school level, this appeared to
be due to differences in the sizes and
distribution of primary and high schools
To generate a preliminary 9 -- Models were developed using each of the
305
model of the determinants of
learner mobility
three definitions of learner mobility investigated
in this thesis
-These models suggested that black race and
attending a high quality school were strong
predictors of all forms of mobility
-- Mobility requiring substantial travel was also
predicted by maternal education, while mobility
at the more local level was most strongly
predicted by local area poverty
To develop an evidence-based
conceptual framework to
support ongoing research into
learner mobility and school
choice
3 & 9 &
10
-- The evidence presented in this thesis supports
the argument that child, family, community and
school level variables all play a role in shaping
school choice decision making
-- However, given that two patterns of learner
mobility are in operation in Johannesburg-
Soweto, and that they appear to be driven by
different variables, the evidence also suggests
that a conceptual framework which does not
differentiate between these forms of mobility
may be insufficient
Table 10.1: Overview of the key findings presented in this thesis
10.2.1 Developing approaches to measuring learner
mobility
The thesis made use of three different approaches to measuring learner
mobility: straight-line distance between home and school, whether children
attend school in the area in which they live, and whether children attend the
grade-appropriate school nearest to their homes. Although all of these
measures provided consistent evidence for high levels of learner mobility
amongst urban South African primary school children, they also captured
different aspects of this mobility. As such, the measures proved to be
complementary, providing evidence that learner mobility in Johannesburg-
Soweto should be understood to consist of two distinct components. These are
firstly the mobility involving fairly limited travel distances associated with
choice between a number of local schools, and secondly the mobility involving
much greater travel distances, and relating to the choice of schools much
306
further afield, typically in historically more advantaged areas than the child‘s
home.
10.2.2 Measuring the extent of learner mobility in
Johannesburg-Soweto
Although there has long been substantial reason to believe that South African
learners are motivated to travel relatively long distances to attend particular
schools (Cosser and du Toit 2002; Fiske and Ladd 2004; Maile 2004; Msila
2005; Woolman and Fleisch 2006; Msila 2009), this thesis presents the first
population-based evidence on what these distances actually are, and how
widespread engagement in mobility is. This evidence does suggest that primary
school learners in the Johannesburg-Soweto area are extremely mobile. At both
ages 7 and 13, over 25% of children were travelling to schools over 5km away
from their homes. Almost 60% of children were travelling outside of the
Census Sub-Place (SP) area in which they lived (roughly equivalent to a
suburb), to attend school in a different SP. At the Main Place (MP) level,
equivalent to a small town, roughly 25% of children were travelling to attend a
school in an MP other than the one in which they lived. Finally, fewer than
20% of children were found to be attending the grade-appropriate school
closest to their home.
These figures provide evidence that learner mobility, and school choice, is
widespread amongst primary school children and their families in urban South
Africa. Certainly, the numbers of children travelling on a daily basis to schools
over 5km from their homes is notable, and particularly at 7 years of age was
not anticipated by existing information. As corroborated by the overlap
between children travelling over 5km, and those mobile at the MP level, most
of these children are travelling to areas in very different parts of Johannesburg
from their homes, and will be attending school in a very different linguistic and
socio-economic context. The data presented also suggest that while some
307
families engage in school choice by sending children to schools further afield,
a substantial proportion of others engage in choice at a local level. A full 50%
of children lived less than 1.5km from their school, but less than half of them
were attending the school closest to their home. This suggests that even less
advantaged families are making active decisions in pursuit of the best possible
educational opportunities for their children, even in the context of an extremely
poorly performing public schooling system. Again, this provides support for
the argument that there are two distinct patterns of learner mobility in evidence
in Johannesburg-Soweto.
10.2.3 Potential child, household and community-level
determinants of learner mobility
This thesis tested a range of variables at the child, family and community
levels to explore their relationship to learner mobility. At the child level, race,
gender, age at first school enrolment and grade repetition, and school phase in
2003 were examined. For all definitions of mobility, race was strongly related
to mobility. Although small sample sizes meant that conclusions could not be
drawn about the mobility of white and Indian children, there was clear
evidence that black children were substantially more likely to engage in
mobility than their coloured peers. There was some indication that girls,
especially on reaching high school, tended to travel slightly further than boys,
but overall there was no compelling evidence for a relationship between gender
and mobility.
There was evidence that children who first enrolled in school at an older age
travelled further at age 7, though not at age 13. This may be due to less
wealthy parents sending their children to school at a younger age, to minimize
the need to provide childcare. It may also relate to wealthier parents enrolling
their children in primary school at a slightly older age, when they might be
expected to cope more easily with academic and social challenges associated
308
with attending a school outside of the local area. Grade repetition was also
strongly related to distance from home to school, at both ages 7 and 13, as well
as to mobility at the SP and MP levels. By contrast, there was no evidence for
a relationship between grade repetition and enrolment at the nearest school.
Finally, children attending high schools did travel further than those still
enrolled at the primary level, and were also significantly more likely to attend
their nearest school.
At the family level, the thesis examined maternal education, maternal marital
status, and household SES. Maternal education was strongly linked to all
measures of mobility, with the children of more educated mothers tending to
travel further, in both 1997 and 2003. However, there was some attenuation of
this relationship at the very highest level of maternal education, perhaps
because these families tended to live in more affluent areas, closer to high
quality schools. There was no clear evidence for a relationship between
maternal marital status and a child‘s engagement in learner mobility. Finally,
again in both 1997 and 2003, there was a strong relationship between
household SES and learner mobility, with children living in more advantaged
families being substantially more likely to engage in mobility, and tending to
travel greater distances.
Finally, at the community level, the poverty level of the area in which the child
lived was explored. Although this data was fairly complex, two general
patterns were discernable. Firstly, there was a clear, but non-linear,
relationship between distance from home to school, and area poverty. Those
children travelling furthest tended to be living in areas that were either
relatively affluent, or particularly poor. Secondly, however, there was also a
fairly linear, and positive, relationship between the likelihood of a child
attending his or her nearest school, and the affluence of the area in which the
child lived.
309
The relationships between learner mobility and the variables discussed here
substantiate the notion that two different types of school choice are in play in
contemporary urban South Africa. Firstly, there is clear evidence that certain
measures of mobility – particularly those relating to distance travelled – are
associated with indicators of affluence, such as socio-economic status,
maternal education, or living in a comparatively advantaged residential area.
The extent to which a child‘s family has the means to engage in mobility is
clearly one determinant of school choice involving mobility. Secondly,
however, there is also evidence of mobility, particularly at a relatively local
level, that is not strongly linked to affluence. Engagement in this more local
mobility appears to be more closely related to the poverty of the area a child
lives in, and by extension the quality of the local schools. This second type of
mobility is particularly evident in the data around whether or not a child
attends his or her nearest school, which suggests that even families with
relatively limited means appear to be making use of school choice to obtain the
best educational opportunities possible.
10.2.4 Potential school-level determinants of learner
mobility
The thesis also explored the relationship between a child‘s mobility, and a
range of attributes of the school he or she attended. The first important finding
here was that, on average, children attended schools that were more
advantaged than would be expected on the basis of the schools closest to their
homes.
The second important finding was that there was a very strong relationship
between child, family and community attributes, and the attributes of the
school attended, with more advantaged children tending to attend more
advantaged schools. Higher maternal education and household SES were
particularly strong predictors of attending higher quality schools. By contrast,
310
children living in areas with high poverty levels were significantly more likely
to attend poorer schools.
The third important finding was the clear evidence for a strong relationship
between mobility and the attributes of the school a child attended. Children
attending private schools, higher quintile schools, schools with a lower
proportion of black learners, schools charging higher fees, schools that did not
historically fall under the DET, and schools with higher pass rates, were more
likely to be engaged in learner mobility, largely regardless of how it was
measured.
This group of findings substantiates the notion that learner mobility and school
choice are typically used to enable a child to access higher quality education
than would ordinarily be the case. This appears to hold regardless of the type
of mobility explored. While more advantaged children remain substantially
more likely to attend more advantaged schools, it does appear to be the case
that even less advantaged children and families are able to make use of school
choice and learner mobility in such a way as to improve the educational
opportunities that they are able to access.
10.2.5 Changes in mobility as children age
While it was initially anticipated that mobility would increase substantially
between 1997 and 2003, as the children aged, became more independent, and
began to transition to high schools, this did not appear to be the case. Although
children moving from a primary school to a high school did tend to experience
a slight increase in distance, this appeared to be due to the smaller number of
high schools available. There was, however, little evidence that children‘s
travel behaviour changed substantially when the transitioned to high school.
Children who were enrolled in primary schools at both points in time did also
not experience any significant changes to their mobility.
311
Despite the relative consistency in children‘s mobility over time, school
change was widespread across the sample, with over half of the sample making
school changes other than those required by the transition to high school.
Children living in poor areas, whose household SES increases with time, and
whose mothers have intermediate levels of education appear to be the most
likely to move between different primary schools. Although changing between
two primary schools typically resulted in a small decrease in distance travelled
to school, substantial changes in mobility were not evident. This suggests that
school change during the primary school period may also be a strategy used in
pursuit of the best available educational opportunities by children and families
with resources to allow for substantial travel.
These findings also provide support for the hypothesis that two different
patterns of learner mobility are in play. Firstly, there is a group of children
travelling fairly long distances. These children typically experience more
stable primary schooling, and their mobility does not change substantially with
the transition to high school. Secondly, there is a larger group of children
attending relatively local schools. These children tend to experience more
school changes during the primary school years, although these changes
typically do not involve substantial changes in mobility. This data additionally
suggests that not only are at least two patterns of school choice in play, but that
the pattern in which a child is engaged appears to be fairly path-dependent.
The pattern of mobility in which a child engages at the beginning of their
schooling appears to remain relatively consistent even during the transition to
high school.
10.2.6 Predicting mobility
In Chapter 9 of the thesis, all of the variables discussed above were combined
to generate models for each of the forms of mobility behaviour discussed. The
results of these models are summarized in Table 10.2 below.
312
Variables Distance SP
mobility
MP mobility Nearest
school
Child level
Race 1997 Black race
associated
with greater
travel
Black race
associated
with
mobility
Black race
associated
with
mobility
Black race
associated
with mobility
2003 Black race
associated
with greater
travel
Black race
associated
with
mobility
Black race
associated
with
mobility
Black race
associated
with mobility
Gender 1997 -- -- -- --
2003 -- -- -- --
Age at first
enrolment
1997 -- -- -- Earlier
enrolment
associated
with mobility
2003 -- -- -- --
Repetition 2003 -- -- -- --
Household level
Maternal
education
1997 Maternal
education
between Gr5
& Gr12
associated
with greater
travel
--
Maternal
education
between
Gr11 & 12
associated
with
mobility
--
2003 Maternal
education
beyond Gr8
associated
with greater
travel
-- Post-school
maternal
education
associated
with
mobility
Maternal
education
between Gr5
& 7 associated
with less
mobility
Maternal
marital
status
1997 -- -- -- --
2003 -- Married
mother
associated
with
mobility
-- --
Household
SES
1997 -- SES quintile
2 & 4
associated
with
mobility
SES quintile
2 associated
with
mobility
--
2003 -- SES quintile
3 associated
with
-- SES quintile 3
associated
with mobility
313
mobility
Community level
Community
poverty (SP)
1997 Living in a
higher
poverty area
associated
with slightly
greater
travel
Living in a
high or low
poverty
area
associated
with
mobility
Living in a
higher
poverty area
associated
with
mobility
Living in a
higher
poverty area
associated
with mobility
2003 -- Living in a
higher
poverty
area
associated
with
mobility
Living in a
higher
poverty area
associated
with
mobility
Living in a
higher
poverty area
associated
with mobility
School level
School
resources
1997 Attending a
better
school
associated
with greater
travel
Attending
a better
school
associated
with
mobility
Attending a
better
school
associated
with
mobility
Attending a
better school
associated
with mobility
2003 Attending a
better
school
associated
with greater
travel
Attending
a better
school
associated
with
mobility
Attending a
better
school
associated
with
mobility
Attending a
better school
associated
with mobility
Table 10.2: Summarized results for the models of mobility developed in Chapter 9.
Results presented are for regressions with robust errors.
When all variables are controlled, being black, having a more educated mother,
attending a more advantaged school, and living in a comparatively
disadvantaged area all appeared to predict a greater distance between home and
school, in both 1997 and 2003. That is, more advantaged children living in
relatively disadvantaged areas, are likely to travel the greatest distances to get
to school. When mobility was measured in a way that picked up local level
school choice, all the variables listed above retained significance, with the
exception of maternal education. The loss of a significant coefficient for
maternal education suggests that household resource levels may be less critical
314
in shaping school choice and mobility at the local level. Again, this provides
support for the notion that two patterns of travel are in evidence, one requiring
substantially more in the way of resources than the other.
10.2.7 Developing an evidence based conceptual
framework for the study of learner mobility
Much of the evidence presented in this thesis has indicated that there are two
distinct patterns of learner mobility in operation in contemporary urban South
Africa. The first form appears to involve roughly 25% of children, and is fairly
resource intensive, involving often substantial travel, typically to well-
resourced schools in historically advantaged areas. The second form of learner
mobility is far less resource intensive, and relates primarily to children who are
attending local schools, but not the schools that are closest to their homes.
Typically these children are travelling less than 1.5km each way.
The conceptual framework proposed in Chapter 3 of this thesis is largely
appropriate, in that child, family and community variables all do appear to be
shaping decision making related to learner mobility, within the context of
history, geography and policy. In light of the strong evidence for two patterns
of mobility, however, it is appropriate to modify the outcome of the model to
indicate that, typically, one of two distinct paths is followed. A child and his or
her family are likely either to embark on the high resources mobility path, or
the low resource mobility path, and are likely to remain on this path throughout
the child‘s primary schooling, and during the child‘s transition to secondary
schooling. The revised framework is presented in Figure 10.1 below.
315
Figure 10.1: Conceptual framework revised on the basis of study findings
10.3 Key contributions
The findings summarized above provide a number of important original
contributions to the body of scholarly work on school choice and learner
mobility, both in South Africa and internationally. These contributions can be
categorized as methodological, empirical and theoretical, and are discussed in
these categories below.
10.3.1 Methodological contributions
The thesis has made three innovative methodological contributions to the study
of school choice and learner mobility. Firstly, as discussed above, it explored
three different approaches to the measurement of educational mobility:
straight-line distance from home to school; movement between different areas;
Historical, geographical and policy context
Decision
making process
- Desired
educational
outcomes
- Investment
required for
desired
outcomes
- Investment
constraints
Residential location
Child
Househol
d
& Family
Maternal
High resource
mobility –
selection of
historically
advantaged
schools
Low resource
mobility –
selection of local
schools
316
and whether or not a child attends the grade-appropriate school nearest to his
or her home. As far as I can determine, combining the use of these three
approaches has not previously been reported in either the South African or
international literature. By using these three different operationalizations of
mobility, it was possible to explore different dimensions of the phenomenon,
leading to the observation that there are at least two distinct patterns of
mobility in place in urban South Africa.
Secondly, this thesis is innovative with regards to the type of data used. As
discussed in Chapter 2, it is the first work of which I am aware to make use of
panel data to explore school choice and educational mobility in South Africa. It
is also the first work to make use of population level data for this purpose.
Finally, it is the first project on school choice in South Africa that I have been
able to identify that combines data drawn from a number of different sources to
simultaneously explore the relationship of household, community and school-
level variables to school choice and educational mobility. It has illustrated that
these types of data can be used for these purposes, and, as will be discussed
below, can provide novel theoretical and empirical contributions to current
knowledge. As such, the study contributes to filling the gap created by the lack
of empirical studies into the determinants of school choice, both in South
Africa and internationally, that was identified in Chapter 2.
10.3.2 Empirical contributions
As highlighted above, this thesis provides, for the first time, detailed
quantitative data on learner mobility in contemporary urban South Africa,
obtained at the population level. The study findings provide clear evidence of
how widespread learner mobility is, and furthermore that mobility is not
limited strictly to the most advantaged children, as was anticipated. Instead, it
suggests that two patterns of school choice and learner mobility are fairly
widespread: firstly, school choice requiring significant travel to historically
advantaged schools, and by extension the investment of substantial economic
317
and other resources; and secondly, school choice at a more local level, which is
less constrained by access to financial and social resources.
This has important implications for dominant narratives about the lives of
urban, working class children in South Africa. These children have typically
been portrayed as disadvantaged (which they are), and even where the
literature focuses on resilience, it is resilience in the context of hardship and
disadvantage (Barbarin and Richter 2001). This narrative tends to portray
children and families as passive in the face of difficult circumstances, and
largely devoid of choice with regards to services. The data presented here,
however, suggests that for a fairly substantial proportion of these children –
perhaps 30% - engagement in social mobility, particularly through education,
is evident. Although their home lives are located in a context of deprivation,
they attend school in more advantaged areas, socializing with more advantaged
children, and typically receiving a better education than they would in the
school nearest to their home. This thesis has also presented clear evidence that
the patterns of school choice and mobility identified are highly path-dependent.
Children who begin their primary schooling on a socially mobile path typically
continue to attend advantaged schools. By contrast, those who begin their
schooling close to home are also likely to remain at these more local schools at
least until the end of the primary phase.
Additional original empirical contributions include the finding that mobile
children typically attend a school more advantaged than the one nearest to their
home, and that certain groups of children (particularly black children, from
relatively well-off households and with more educated mothers and living in
relatively disadvantaged areas) are more likely to engage in educational
mobility than others. Finally, the finding that mobility behaviour is fairly
stable and consistent over time, even following the transition to high school, is
also novel, and largely unanticipated. This counters the widespread assumption
that mobility increases as children age and become more independent, and
318
particularly once they transition to high school, and provides further evidence
that mobility behaviour is fairly path dependent.
10.3.3 Theoretical contributions
Finally, at a more theoretical level, the thesis presents a preliminary theoretical
model detailing potential determinants of school choice and mobility. This
model raises some questions about the traditional market orientation of school
choice literature, both in South Africa and internationally. Certainly, in the
sample explored here, a large proportion of the children engaged in school
choice appeared to be doing so in ways that did not demand a very high level
of economic investment. That is, many children are travelling within a
constrained radius, making choices between a number of public schools with
much the same fee structures and associated costs.
While wealth appears to shape mobility, it is not as centrally important as
might have been expected. School choice work in South Africa has tended to
focus very heavily on those children able to access particularly advantaged
schools, which typically requires the investment of substantial financial and
social resources (Paterson and Kruss 1998; Sekete, Shilubane et al. 2001; Fiske
and Ladd 2004). The current thesis, however, suggests that a great number of
children are also engaged in another, less costly, form of school choice. As this
form of choice still has implications for the opportunities available to children,
is seems important that it receives closer attention in the future. At the
international level, work exploring school choice in developing countries has
also tended to focus very heavily on access to privileged schools, and paid less
attention to the ways in which children and families seek to maximize their
educational opportunities even in a context of very limited resources (Carnoy
and McEwan 2003; Tsang 2003; Elacqua, Schneider et al. 2006).). Again, this
thesis suggests that looking at school choices made by children facing resource
constraints may prove very valuable.
319
It also raises questions about the series of hypotheses, common in both
developing and developed country contexts, about schools and children who
are ‗left behind‘ in the context of choice (Bridge and Blackman 1978; Capell
1981; Henig 1994; Witte and Thorn 1996; Levin 1998; Goldhaber 1999; Hoxby
2003; Peterson, Howell et al. 2003). If engagement in school choice is as
widespread as this thesis suggests, that we have to widen our understanding of
choice and its parameters in order to grasp the degree of agency exercised by
parents and children, and how this might be leveraged to create higher demand
for quality education.
This feeds into the international debate, summarized in Chapter 2, about the
relationship between school choice, particularly in relatively unregulated
contexts, and educational segregation and inequality. This thesis suggests that
although engagement in choice may be extremely widespread, and relatively
unconstrained by economic and social resources, these resources still shape the
ways in which choice can be exercised. Different socio-economic groups
appear to access different forms of school choice, which in turn are likely to
result in differing educational outcomes for their children. The international
literature tends to suggest that more educated and advantaged parents are more
strongly involved in school choice than less advantaged parents (Carnoy and
McEwan 2003; Elacqua, Schneider et al. 2006). The findings presented here
suggest that this may not actually be the case. Instead, it appears that less
advantaged parents are highly engaged in choice, but simply lack the resources
to pursue choice in the ways that are likely to be most beneficial to their
children.
320
10.4 Contextual relevance
10.4.1 Relevance to South Africa
The findings of this thesis suggest a range of implications for individuals,
families, communities and society more broadly. These implications have
some relevance to policy, and suggest a range of avenues for additional
research and investigation. At the level of children and their families, the most
important implications relate to child well-being. Engaging in school choice
and mobility provides a child with access to educational opportunities they
might not otherwise receive. In the South African context, learner mobility is
also likely to be a path to social mobility for at least some children. However,
these educational and social benefits do come at some cost. Economic costs,
related to the cost of school fees and travel, may mean that a child‘s family has
fewer resources available to meet other needs. In the South African context,
the safety of a young child travelling substantial distances, typically alone, is a
potential risk, and travel time is certainly a cost. An additional risk relates to
the fact that for most children travelling substantial distances, their schooling
will take place in what is essentially a foreign social context, and often in a
language that they are not very familiar with. They will not have
neighbourhood friendships and it will be more difficult for their parents to
monitor their friendships and activities. In a society with as extensive societal
and educational disparities as South Africa, it seems like that over the long
term, the benefits of learner mobility for most children would tend to outweigh
the costs. In the shorter term, however, the risks and costs for individual
children are likely to remain quite high.
The implications of widespread learner mobility in urban South Africa are,
however, likely to be quite different at the community and societal levels.
Here, it seems likely that while the short-term costs and risks are reduced, the
longer term costs may be quite substantial. One group of particular concern is
those children who are not able to engage the more resource-intensive forms of
321
educational mobility, for whatever reason, and are obliged to attend schools
close to their homes. These children, along with the schools they attend, are
likely to be negatively impacted by the tendency of all the more advantaged
children in the area to attend schools further afield, as this will further reduce
the resources available at the local level. With the out-migration of more
advantaged children, and a relatively captive market of disadvantaged children,
poorly performing schools may lose the incentive to try to improve. Over the
longer term, high levels of mobility are also likely to have harmful
implications for community coherence, and may well contribute to growing
levels of inequality, both economic and educational, within historically
disadvantaged areas. The current distribution of educational opportunities,
which motivates the high levels of mobility identified in this thesis, are also
hugely inefficient, requiring large investments on the parts of families and
children, even while they potentially exacerbate already high societal
inequality.
This pattern, in which learner mobility appears likely to be beneficial for
individual children, particularly over the longer term, but also very costly at
the level of their community and society, poses real challenges for policy
makers. How to balance these various costs and benefits is a challenging
question, particularly in a society already so deeply challenged by inequality,
and a poorly performing educational system. This is further complicated by the
very strong incentives faced by individuals to continue engaging in mobility,
regardless of the policy environment. The findings presented in this thesis do
raise questions about the validity of certain core elements of South African
educational policy, such as the concept of our schools as ―community schools‖,
and the notion that a school‘s access to resources can be determined by looking
at its location, as opposed to the composition of its student body. The
complexity of the results presented, however, combined with the preliminary,
hypothesis-building nature of the study itself does, however, suggest that much
more needs to be known about how school choice operates in South Africa,
322
and how its costs and benefits play out at the individual and societal levels,
before making substantial changes to policy. For this reason, this thesis does
not provide a comprehensive discussion of policy relevance at this point.
10.4.2 International relevance
The findings of this thesis also have implications for both policy and future
research internationally. The thesis has highlighted just how widespread it is
possible for school choice to be in a developing country context with limited
regulation. South Africa has some distinctive features which limit the
generalizability of this study‘s results, such as the fact that in South Africa
school performance is very closely associated with geography, and the fact that
South Africa‘s public schooling system is much larger than those found in
most other countries at similar levels of development, and contains a subset of
very well performing schools. Nonetheless, South Africa‘s high level of
inequality, limited school choice regulation (or capacity to enforce regulation)
and a generally poorly performing public schooling system are all shared with
a number of other countries at similar levels of developments. The extent of
school choice in South Africa suggests that levels of choice in other similar
countries can also be expected to be high. Similar studies in other low and
middle income countries would therefore be extremely illuminating.
One question that would be particularly useful to ask is whether school choice
in other countries, particularly those where private schooling is more
widespread, is more strongly economically driven than choice in South Africa.
Additionally, understanding whether less advantaged children in other
countries also exercise local level school choice will be critical in determining
how best to finance public schools and enhance access to high quality
education. Understanding the extent of school choice, the forms which it takes,
and its determinants is essential to developing appropriate ways to capture this
engagement of children and families with educational systems in ways to
enhance the performance of both schools and their students.
323
If indeed school choice is widespread in other developing countries, this also
suggests a need to think about ways of protecting the most vulnerable groups
of children from potential choice-related harm. For example, if more
advantaged children are enrolling predominantly in private schools, this
deprives the public system of resources, and is likely to enhance societal
inequality. This raises questions as to whether voucher systems, such as those
found in Chile, or other innovative approaches to school funding should be
tested more widely in other developing countries.
10.5 Project limitations and future work
10.5.1 Sample composition
One of the major limitations of the thesis is the relatively constrained sample
that has been used for analysis. Although, as discussed, this was unavoidable
for practical reasons, an ideal next step is to broaden the study sample so that it
also includes children who do change their residential addresses during the
period under consideration. This more representative sample would ensure that
findings can be more broadly generalized. In addition, this sample would also
allow for the exploration of potential interactions between residential and
educational mobility.
10.5.2 Study end point and longitudinal analysis
A second limitation is that only a proportion of the study sample members
progressed to high school by 2003. As those children who had progressed to
high school by 2003 differed systematically from those who had not, this
introduced challenges around using the data to understand whether mobility
during primary and high school differed. Although preliminary analysis
suggested a great degree of path-dependency and consistency over time, it also
appeared to be the case that secondary school status had implications on
324
mobility behaviour, at the very least due to the different geographical
distributions of primary and secondary schools. Expanding the sample
longitudinally so that all study sample members reach secondary schooling
would facilitate analyses exploring the extent to which mobility changes at the
secondary school level, and why this is the case. It would also facilitate the use
of more complex longitudinal analyses, which would provide clearer and more
valuable data on the nature of changes in mobility behaviour over time.
The construction of a genuine longitudinal dataset, in which home and
schooling data were available for each year during a child‘s schooling would
also enhance the value of further analysis, by allowing for the use of more
advanced analytical techniques. This would also provide clearer evidence
around school change during primary and secondary schooling, and potential
motivating factors for this. Introducing data for a more contemporary cohort of
children, although likely to be extremely difficult and costly, would also allow
exploration of whether patterns of mobility have changed over time, since the
beginning of the post-Apartheid period.
10.5.3 Methodological approach
Finally, as with any methodological approach, the use of quantitative
secondary analysis in this project imposed a number of limitations. Perhaps
most importantly, it makes it extremely difficult to provide answers to
questions around the individual decision-making processes underlying the
decision to engage in learner mobility. However, it is valuable in highlighting
correlates of mobility, and by extension generating hypotheses about the types
of decisions which individuals may be making. These data-driven hypotheses
can then be tested by subsequent research using different methodological
approaches. Population based quantitative secondary analysis is also fairly
limited in the extent to which it can describe the implications and outcomes of
learner mobility. School and community level outcomes, in particular, can only
really be tested with data collected at those levels. To understand individual
325
level outcomes, a far more longitudinal approach, along with far more detailed
and specialised data, particularly regarding outcomes, than that used here
would be necessary.
10.5.4 Future work
There are a number of ways in which the work presented in this thesis could be
usefully extended. Firstly, as alluded to above, broadening the study sample
will produce more generalizable findings, and allow for an exploration of the
interactions between learner mobility and residential mobility. Secondly,
extending the dataset longitudinally will allow for the examination of learner
mobility in the high school period, as well as for the application of more
sophisticated tools for longitudinal analysis11. Thirdly, study methodology
could be developed further in a few directions, to strengthen study findings.
For example, it would be possible to generate a measure of ‗non-essential‘
travel to school, by looking at the difference between the distance to a child‘s
nearest school, and the school a child attends. This might provide a more
accurate measure of travel related to choice. Additional approaches to dealing
with the highly non-normal distribution of the travel data could also be
usefully explored, and in particular, more systematic approaches to dealing
with outliers. Developing measures of practical or substantive significance to
accompany the presentation of measures of statistical significance would also
contribute usefully to the interpretation of study findings.12
This study would also greatly benefit from the introduction of a qualitative
component. This could be used to test the hypotheses that this thesis has
generated. For example, do motivations for mobility (and constraints in
engaging in mobility), as experienced by children and their families, tie in with
11 Funding for this extension of the project has been obtained through an ESRC Pathfinders
grant, starting in June 2011.
12 I am indebted to my examiners for suggesting these methodological developments that
would contribute to strengthening future work.
326
the results presented in the previous chapters?13 Do children and their families
identify with the notion that there are two distinct patterns of school choice,
and by extension mobility, in play in contemporary urban South Africa?
Qualitative work would also allow for further unpacking of the decision
making process, providing insight into ways in which families select the
schools at which they pursue enrolment for their children. Finally, it would
provide a way to integrate the current work more strongly with the
international literature, by examining how and why school choice in the South
African context differs from experiences in other countries. A final
enhancement to work using the Bt20 data would be to develop more
sophisticated approaches to the measurement of variables used in the study. In
particular, alternative measures of school resource levels, community affluence
and coherence, and household SES could all usefully be explored.
An important set of questions which this thesis has not adequately explored are
those relating to the supply-side geography of schools in South Africa. A
clearer understanding of how schools with different attributes are distributed
across space would enrich our understanding of how and why children travel to
go to school. Incorporating supply side variables into the models of mobility
presented in this thesis would strengthen them considerably. For example, poor
supply of schools in local areas may be one of the reasons that mobility is
fairly high amongst the least advantaged sample members. An exploration of
the correlation between school density and population density in different areas
would also provide some indication as to whether supply side issues are behind
much of the mobility documented in this thesis. Additionally, mapping the
distribution of schools by the languages that they operate in will answer
questions about whether high levels of local level mobility might be explained
in part by language of schooling. These types of analyses can be accomplished,
13 Given the difficulty of accurately measuring school characteristics, answering this question
is particularly important, as it would address the concern that the high levels of mobility
documented in this study relate more to a highly stochastic schooling environment. Many
thanks to my examiners for highlighting this concern to me.
327
at least in part, with the data already compiled for this thesis, and would
provide some valuable extensions to study findings14.
Moving beyond the current study, similar work applied to populations in other
parts of South Africa would also be very useful in establishing the extent to
which the patterns identified here are prevalent throughout the country. A
similar study in a rural area, where school choice would be expected to be far
more limited due to the lower density of available schools would be
particularly valuable. Similar analyses applied to other low and middle income
countries would also help to shed light on just how widespread school choice
is in other contexts, and particularly the extent to which less advantaged
children are able to participate in it.
This study also suggests a number of additional, related questions that might
usefully be pursued. One question relates to the roles of schools in shaping
school choice outcomes. For example, to what extent are children not able to
enrol in the school they select, for example for reasons of overcrowding, lack
of social capital or knowhow, or even overt discrimination? A second question
relates to the implications of school choice for academic outcomes, both for
learners, and for entire schools. This issue lies at the core of much of the
international debate around school choice, and is a critical question which this
thesis has not been able to address. This issue is likely to be particularly
complex in the South African context, where the potential costs of mobility
(ranging from economic, to travel time, to learning in a language not used at
home) are extremely high, but variations in public school quality mean that
potential benefits are also substantial. Enhancing the dataset by the inclusion of
data on academic outcomes at the individual level would allow for an
exploration of the relationship between mobility, school choice, and academic
outcomes. A final set of additional questions relate to the implications of
14 I am indebted to my examiners for highlighting to me the importance that work of this type
would have, and suggesting ways in which it might usefully be approached in the near future..
328
learner mobility and school choice for social mobility. To what extent does the
ability to access education at a historically advantaged school determine the
opportunities available to a child as he or she moves through school, and then
into higher education or the workplace?
10.6 Conclusion
This final chapter has provided a brief overview and summary of the key
findings presented in this thesis. These highlight the original contributions that
this thesis makes to the scholarly literature on learner mobility and school
choice. Methodological contributions have included new approaches to the
identification and measurement of learner mobility, as well as the use of
population-based panel data, combined with data from other sources, to study
the phenomenon. At the empirical level, the study has contributed data on the
extent, correlates and determinates of learner mobility. Finally, at the
theoretical level, the study has contributed a conceptual framework to support
other work on the topic, and the insight that in contemporary urban South
Africa, there are at least two forms of learner mobility in play. The study also
feeds into broader international debates about the implications of school choice
to educational segregation and inequality. The chapter concludes by mapping
out a spectrum of further research possibilities.
329
Appendix A: Alternative data
sources considered for the thesis
A.1 Cape Area Panel Study (CAPS)
CAPS data collection began in 2002, and has been focused largely on a sample
of almost 5000 young adults who were then aged between 14 and 22. Most of
these participants have been interviewed four times since 2002, and additional
interviews have been conducted with their households, as well as some
additional households and older individuals. As indicated by the study‘s name,
the sample was drawn from the greater Cape Town area. Participants have
been asked a wide range of questions over the four waves of the study for
which data is available to date. These questions include school enrolment, both
current and historical, as well as reasons for enrolment at particular schools,
and for changes in school enrolment over time. The study‘s second wave also
includes a module in which perceptions of school quality for a range of
neighbourhood schools are explored for each child (Lam, Ardington et al.
2008; Lam, Ardington et al. 2008).
While the nature of the schooling data available in the study made it an
extremely strong candidate for use in this dissertation, the decision against
using it was finally made largely on the basis of its focus on the Cape Town
area, which is well known for having a far stronger educational system than
that found anywhere else in South Africa (Fiske and Ladd 2004)15.
Additionally, the Cape Town area differs substantially from the rest of the
country in terms of its population and its socio-economic conditions. For this
reason, it was felt that while the data from the study would certainly enable a
clear understanding of learner mobility in the Western Cape, this
understanding would be unlikely to travel well across the rest of South Africa.
15 Although the 2010 Matric examination results, in which Gauteng province outperformed the
Western Cape, may signal a shift in this pattern.
330
An additional concern related to the study‘s focus on data collection on youth
older than 14, as this would make it difficult to explore learner mobility during
primary schooling, and during the transition to secondary schooling.
A.2 Kwa-Zulu Natal Income Dynamics Survey (KIDS)
The KIDS study is a longitudinal dataset, with a focus primarily on poverty
and inequality in the KwaZulu Natal (KZN) region of South Africa. It
developed out of a cross-sectional study, the 1993 Project for Statistics on
Living Standard and Development (PSLSD), and the PSLSD data collected in
KZN in 1993, covering 1558 households, forms the first wave of KIDS.
Households interviewed for PSLSD were followed up, and re-interviewed in
1998, forming the second wave of KIDS data. A 3rd round of interviews was
conducted in 2004, with 865 households (including households previously
interviewed, and next-generation households which had split off from
households previously interviewed). During each interview wave, a detailed
household roster was completed, which in 1998 and 2004, along with socio-
economic information included current school enrolment of all school-aged
children. For each household, address information and GIS coordinates were
also collected (May, Carter et al. 1999; May, Aguero et al. 2007).
While the KIDS dataset did include all the information essential for this
dissertation, there were a number of ways in which the sample was not ideal.
Firstly, while the study itself is longitudinal in nature, this was focused at the
household level, rather than at the level of particular children or other
individuals. Therefore, while longitudinal data is available on some children,
this is not true of all children across the sample, and depends heavily on their
mobility and relationship to the household head. This would have made
identifying a sample appropriate to longitudinal analysis very challenging. The
relatively high levels of attrition experienced by KIDS (around 38% over 11
years), while unsurprising given the long intervals between the waves of data
collection and the longitudinal nature of the study, also raise concerns about
331
how representative the remaining sample is, particularly when considering that
the relevant data would only be available for a sub-sample of children, unlikely
to be randomly distributed amongst households.
Nonetheless, I had initially hoped to make some use of the KIDS data, to
provide some insight into the variations in learner mobility between urban and
rural areas, as it appeared to be the best potential source of data on the
enrolment of rural learners in South Africa. However, preliminary
communication with the KIDS research team revealed that a substantial
amount of preparatory work on the KIDS school enrolment data would be
required for it to become useable for this dissertation, which, in light of time
and resource constraints, resulted in a final decision not to use this data for the
current project.
A.3 National Income Dynamics Survey (NIDS)
NIDS was developed to answer very similar types of questions to those
addressed in KIDS, but at a national level. Areas of primary focus included
household wealth creation, demographic dynamics, social heritage, and access
to services and cash transfers. During its first wave of data collection NIDS
collected data on 7305 households, consisting of 28255 individuals, distributed
across all nine provinces of South Africa (Leibbrandt, Woolard et al. 2009).
While much of the essential data on residential addresses and schooling was
collected, the same limitations detailed in the discussion of KIDS, relating to
the focus on households rather than individual children or youth, continue to
hold. In addition, ethical protections on the NIDS data appear to prohibit the
release of residential address data to researchers altogether. However, the main
reason that it was not possible to seriously consider making use of NIDS data
for this dissertation was that formal data collection only began in 2008, and
data release only occurred from mid 2009. Waiting for this data would have
therefore substantially delayed this project. In addition, even the data available
332
in 2009 would only have been cross-sectional, preventing any longitudinal
analysis.
A.4 Agincourt Health and Demographic Surveillance System
(Agincourt)
As suggested by the name, the Agincourt data is concerned primarily with
health and demographics. Data is collected on the full population of around
82000 individuals living in the Agincourt sub-district of Bushbuckridge, South
Africa. A baseline census of the area was conducted in 1992 (Kahn, Tollman et
al. 2007). This data has subsequently been updated 12 times, most recently in
2008. Unfortunately, this dataset did not prove to be suitable for this project, as
the education data collected was extremely limited. School enrolment data was
collected at 4 points during the study, but was not in a form that made it
feasible to use for this study, particularly given the large sample size.
A.5 Africa Centre for Health and Population Studies
The Africa Centre is located in rural KwaZulu Natal, South Africa, and serves
as the base for a number of research project, including the longitudinal
collection of demographic and health data through the Africa Centre
Demographic Information System (ACDIS). ACDIS data collection began in
2000, and is ongoing, with data collection each year. ACIDS covers about 90
000 individuals in approximately 11 000 households, including those
household members who are not resident in the area. Unfortunately, the
educational data collected in this study is limited to attainment, and details
regarding the school enrolment of individuals are not available (Herbst, Newell
et al. 2010). For this reason it was not possible to pursue the use of this data set
for this dissertation.
333
Appendix B: Letter confirming
approval of ethics clearance for
thesis, received from the School
of Education, University of the
Witwatersrand
334
Wits School of Education
STUDENT
NUMBER: 386849
Protocol: 2009ECE01
7-May-09
Ms. Julia de Kadt
86 Glenroy Road
MANOR GARDENS
DURBAN
4001
Dear Ms. de Kadt
Application for Ethics Clearance: Doctor of Philosophy
I have pleasure of advising you that the Ethics Committee in Education
of the Faculty of Humanities, acting on behalf of the Senate has agreed
to approve your application for ethics clearance submitted for your
proposal entitled:
Learner Migration in South Africa: Dimensions and implications
for Educational Practice and Policy.
27 St Andrews Road, Parktown, Johannesburg, 2193 • Private Bag 3, Wits 2050, South Africa
Tel: +27 11 717-3007 • Fax: +27 11 717-3009 • E-mail: enquiries@educ.wits.ac.za • Website:
www.wits.ac.za
335
Recommendation:
Ethics clearance is granted
Yours sincerely
Matsie Mabeta
Wits School of Education
Cc Supervisor: Prof. B Fleisch (via email)
336
Appendix C: Relationships within
the study sample between
variables hypothesized to act as
determinants of learner mobility
C.1 Race and other variables
Given the small numbers of white and Indian children in the study sample, all
further discussion of race is limited to the black and coloured groups only.
There is no evidence for a relationship between race and gender. Similarly,
there is no evidence for a difference between age at first enrolment for black
and Coloured children, with slightly over half the members of each group
enrolling on time. Coloured children are slightly more likely to have reached
secondary school by 2003 than black children (χ2(1)= 3.4241, Pr=0.064), and
are somewhat less likely to have repeated any grades (χ2(1)= 4.0985, Pr=0.043).
Race is also related to the various household and family variables considered.
Maternal education is slightly higher for black children than coloured children
(Wilcoxon rank-sum test, Pr=0.0781). Black mothers were also less likely to
be married (χ2(1)= 44.4026, Pr=0.000) than coloured mothers. Wilcoxon rank-
sum tests indicate that ethnicity and household SES, for both 1997 and 2003,
are significantly related (Pr=0.0000), with coloured children having lower
household SES than black children. By contrast, however, black children live
in SAL, SP and MP areas with higher poverty levels than coloured children
(Wilcoxon rank-sum tests, Pr=0.0000).
337
C.2 Gender and other variables
Gender of the child is significantly related to age at first enrolment (χ2(1)=
5.1999, Pr=0.023) and grade repetition (χ2(1)=36.0374, Pr=0.000), with girls
more likely to start early or on time, and only about half as likely as boys to
repeat a grade. Given that girls tend to start their schooling earlier, and are less
likely to repeat grades, it comes as no surprise that they are also significantly
more likely to have reached high school by 2003 (χ2(1)=23.3055, Pr=0.000).
There is no evidence for any relationship between child gender and maternal
education, household SES in 1997 or 2003, or the poverty level of the area in
which the child lives. Although a weakly significant relationship between child
gender and maternal marital status is found (χ2(1)=3.5817, Pr=0.058), this
seems likely to be spurious.
C.3 Age at first enrolment and other variables
Predictably, there is a strong relationship between age at first enrolment and
phase of schooling in 2003 (χ2(1)=503.7355, Pr=0.000), with children who
started school late being unlikely to have reached high school at this point.
There is, however, no evidence of a relationship between age at first enrolment
and grade repetition. There is also no evidence that age at first enrolment is
related to maternal education, maternal marital status, or household SES in
1997. By contrast, there is a significant relationship between age at first
enrolment and household SES in 2003 (χ2(4)=15.8754 Pr = 0.003), with more
advantaged children tending to enrol later. Finally, there are significant
relationships between age at first enrolment and area poverty at the SAL
(χ2(4)=8.3949, Pr=0.078) and SP (χ
2
(4)=9.8775, Pr=0.043) levels, with children
living in wealthier areas being more likely to start school late. There was no
relationship between MP level poverty and age at first enrolment.
338
C.4 Phase of education in 2003 and other variables
Predictably, children who repeated a grade between 1997 and 2003 are
significantly less likely to have reached high school by 2003 (χ2(1)= 313.9621,
Pr=0.000). Children of more highly educated mothers are also more likely to
have reached high school by 2003 (χ2(1)=32.9057, Pr=0.000). There is no
evidence of a relationship between maternal marital status and the child‘s
phase of education by 2003. There is a weakly significant relationship between
phase of education in 2003 and household SES for 1997 (χ2(4)=8.9023,
Pr=0.064), but when 2003 household SES is used, this effect disappears. There
is no association between phase of education in 2003 and the poverty of the
area, SAL, SP or MP, in which a child lives.
C.5 Grade repetition and other variables
Children with less highly educated mothers are more likely to have repeated a
grade (χ2(4)=45.6920, Pr=0.000). Children whose mothers were unmarried at
their birth are also more likely to have repeated a grade (χ2(1)=3.7110, Pr =
0.054). Repetition is also significantly related to household SES, both in 1997
(χ2(4)=50.9931, Pr=0.000) and 2003 (χ
2
(4)=45.7106, Pr=0.000), with children in
more affluent households being less likely to have repeated a grade. Finally,
there is also a positive relationship between repetition and the poverty level of
the area in which a child lives, whether this is calculated at the SAL
(χ2(4)=33.9833, Pr=0.000), SP (χ
2
(4)=28.7967, Pr=0.000) or MP (χ
2
(2)=17.6835,
Pr=0.000) level. However, these relationships are not strictly linear throughout
all poverty quintiles, and are most marked at the extremes.
C.6 Maternal education and other variables
A positive, significant, but non-linear relationship exists between maternal
education and maternal marital status, both measured at the time of the child‘s
339
birth (χ2(4)= 16.3190, Pr = 0.003). The proportion of married mothers is highest
amongst mother with post-school education, followed by those with primary
schooling or less, while rates are lowest amongst mothers with partial or
complete secondary schooling. There is also a significant, positive relationship
between maternal education and household SES, both for 1997 (Kruskal-
Wallis test, Pr=0.0001) and for 2003 (Kruskal-Wallis test, Pr=0.0001).
Maternal education is also significantly related to area poverty, measured at the
SAL (χ2(16)=74.1141, Pr=0.000 ), SP (χ
2
(16)=57.4465, Pr=0.000 ), and MP
(χ2(8)=21.5601, Pr=0.006 ) levels, with more educated mothers tending to live
in more advantaged areas.
C.7 Maternal marital status and other variables
Maternal marital status is significantly related to household SES in both 1997
(χ2(4)=130.5849, Pr=0.000) and 2003 (χ2(4)=54.9065, Pr=0.000), with
mothers in more affluent households more likely to be married. Similarly,
mothers living in more advantaged areas are also significantly more likely to
be married, whether area poverty is measured at the SAL (χ2(4)=121.2915,
Pr=0.000), SP (χ2(4)=134.1958, Pr=0.000) or MP (χ
2
(2)=55.5876, Pr=0.000)
level.
C.8 Household SES and residential area poverty levels
Household SES in 1997 and 2003 are strongly related, with a correlation of
0.7655 (Pr=0.000). Household SES, measured in both 1997 and 2003, is also
strongly and inversely related to the poverty area in which the child lives (see
Table A3.1 below), with household SES tending to be higher in the areas with
the lowest poverty levels.
340
Relationship between
household SES and
residential area poverty
SAL poverty level SP poverty level MP poverty level
Household SES, 1997 χ2(16)=381.0466
Pr=0.000
χ2(16)=298.4818
Pr=0.000
χ2(8)=164.0330
Pr=0.000
Household SES, 2003 χ2(16)=229.4532
Pr=0.000
χ2(16)=230.2197
Pr=0.000
χ2(8)=68.0648
Pr=0.000
Table A.1: Significance of relationships between household SES and residential area
poverty levels
C.9 Conclusion
All relationships documented here appear to operate in the expected direction.
There is evidence that a number of the variables considered are strongly related
to each other, as expected. This implies that during the modelling component
of the thesis, attention must be paid to avoiding multicollinearity.
341
References
Ahmed, R. and Y. Sayed (2009). "Promoting access and enhancing education
opportunities? The case of ‗no-fees schools‘ in South Africa."
Compare: A Journal of Comparative and International Education 39(2):
203 - 218.
Alderman, H. and E. M. King (1998). "Gender differences in parental
investment in education." Structural Change and Economic Dynamics
9(4): 453-468.
Andrabi, T., J. Das, et al. (2009). What Did You Do All Day? Maternal
Education and Child Outcomes.
Astin, A. W. (1992). "Educational "Choice": Its Appeal May be Illusory."
Sociology of Education 65(4): 255-260.
Bagley, C. (1996). "Black and White Unite or Flight? The Racialised
Dimension of Schooling and Parental Choice." British Educational
Research Journal 22(5): 569-580.
Barbarin, O. A. and L. M. Richter (2001). Mandela's children: Growing up in
post-Apartheid South Africa. New York, Routledge.
Bifulco, R. and H. Ladd (2006). School choice, racial segregation and test-
score gaps: evidence from North Carolina's charter school programme.
. Annual Meeting of Allied Social Science Associations, Boston, MA.
Bosetti, L. (2004). "Determinants of school choice: understanding how parents
choose elementary schools in Alberta." Journal of Education Policy
19(4): 387 - 405.
Bradley, S. and J. Taylor (2002). "The effect of the quasi-market on the
efficiency-equity trade-off in the secondary school sector." Bulletin of
Economic Research 54(5): 295-314.
Bray, R., I. Gooskens, et al. (2010). Growing up in the new South Africa:
Childhood and adolescence in post-apartheid Cape Town. Cape Town,
South Africa, HSRC Press.
Bridge, R. and J. Blackman (1978). A study of alternatives in American
education: Vol IV: Family choice in schooling. Santa Monica, CA
RAND Corporation.
Bryman, A. (2004). Social research methods. Oxford, United Kingdom,
Oxford University Press.
Capell, F. (1981). A study of alternatives in American education: Vol VI:
Student outcomes at Alum Rock, 1974-1976. Santa Monica, CA
RAND Corporation.
Carnoy, M. (1999). Globalizaton and educational reform: what planners need
to know. Paris, France UNESCO.
Carnoy, M. and P. McEwan (2003). Does privatization improve education?
The case of Chile's national voucher plan. Choosing choice: School
choice in international perspective. D. Plank and G. Sykes. New York,
NY, Teachers College Press, Columbia University: 24-44.
342
Centre for Development and Enterprise (2010). Hidden assets: South Africa's
low-fee private schools. Johannesburg, South Africa.
Chen, X., P. Ender, et al. (2011). Stata web books: Logistic regression with
Stata, UCLA Academic Technology Services.
Chisholm, L. (2004). Changing Class: Education and Social Change in post-
Apartheid South Africa. Cape Town HSRC Press.
Chisholm, L. (2005). The state of South Africa's schools. State of the nation:
South Africa 2004-2005. J. Daniel, R. Southall and J. Lutchman. Cape
Town, HSRC Pres: 201-226.
Coleman, J. S. (1992). "Some Points on Choice in Education." Sociology of
Education 65(4): 260-262.
Cosser, M. and J. du Toit (2002). From school to higher education? Factors
affecting the choices of Grade 12 learners. Cape Town, HSRC
Publishers
Daun, H. (2003). Market forces and decentralization in Sweden: Impetus for
school development or threat to comprehensiveness and equity?
Choosing choice: School choice in international perspective. D. Plank
and G. Sykes. New York, NY, Teachers College Press, Columbia
University: 92-111.
De Jong, G., F. (2000). "Expectations, Gender, and Norms in Migration
Decision-Making." Population Studies 54(3): 307-319.
Denessen, E., G. Driessena, et al. (2005). "Segregation by choice? A study of
group-specific reasons for school choice." Journal of Education Policy
20 (3): 347-368.
Department of Education (2000). Brochure for the 2000 School Register of
Needs Report.
du Toit, J. (2003). Independent Schooling. Human Resources Development
Review 2003: Education, employment and skills in South Africa. A.
Kraak and H. Perold. Cape Town, HSRC Press: 380-395.
Elacqua, G. (2006). Enrollment practices in response to vouchers: evidence
from Chile.
Elacqua, G., M. Schneider, et al. (2006). "School choice in Chile: Is it class or
the classroom?" Journal of Policy Analysis and Management 25 (3):
577-601.
Fedderke, J. W., R. de Kadt, et al. (2000). "Uneducating South Africa: The
failure to address the 1910-1993 legacy." International Review of
Education 46(3): 257-281.
Filer, R. and D. Munich (2003). Public support for private schools in Post-
Communist Central Europe. Choosing choice: School choice in
international perspective. D. Plank and G. Sykes. New York, NY,
Teachers Colllege Press, Columbia University: 196-222.
Filmer, D. and L. Pritchett (2001). "Estimating wealth effects without
expenditure data—or tears: an application to educational enrollments in
states of India." Demography 38(1): 115-132.
Fiske, E. and H. Ladd (2000). When schools compete: a cautionary tale.
Washington DC, Brookings Institution Press
343
Fiske, E. and H. Ladd (2004). Elusive equity : education reform in post-
apartheid South Africa. Washington D.C. , Brookings Institute.
Fiske, E. B. and H. F. Ladd (2005). Racial equality in education: how far has
South Africa come? Terry Sanford Institute of Public Policy Working
Paper series, Duke University.
Fleisch, B. (2008). Primary Education in Crisis: Why South African
Schoolchildren Underachieve. Cape Town, South Africa, Juta.
Fleisch, B. and J. Schindler (2008). Gender repetition: School access,
transitions and equity in the 'Birth-to-Twenty' cohort panel study in
urban South Africa.
Fleisch, B. and J. Schindler (2009). Patterns and prevalence of school access,
transitions and equity in South Africa: Secondary analyses of BT20
large-scale data sources. CREATE Pathways to Access Research
Monographs. Brighton, United Kingdom, Consortium for Research on
Educational Access, Transitions and Equity.
Garcia, D. R. (2008). "Academic and Racial Segregation in Charter Schools:
Do Parents Sort Students Into Specialized Charter Schools?" Education
and Urban Society 40(5): 590-612.
Ginsburg, C., S. Norris, et al. (2009). "Patterns of residential mobility amongst
children in greater Johannesburg-Soweto, South Africa: observations
from the Birth to Twenty cohort." Urban Forum 20 397-413.
Ginsburg, C., L. M. Richter, et al. (2010). "An analysis of associations between
residential and school mobility and educational outcomes in South
African urban children: The Birth to Twenty cohort." International
Journal of Educational Development In Press.
Glazerman, S. M. (1998). School Quality and Social Stratification: The
Determinants and Consequences of Parental School Choice. Annual
Meeting of the American Educational Research Association. San
Diego, CA,: April 13-17, 1998.
Godwin, K., F. Kemerer, et al. (1998). "Liberal Equity in Education: A
Comparison of Choice Options." Social Science Quarterly (University
of Texas Press) 79(3): 502-522.
Godwin, K., S. Leland, et al. (2006). "Sinking Swann : public school choice
and the resegregation of Charlotte's public schools." Review of Policy
Research 23 (5): 983-997.
Goldhaber, D. (2000). "School Choice: Do We Know Enough?" Educational
Researcher 29(8): 21-22.
Goldhaber, D. and E. Eide (2002). "What do we know (and need to know)
about the impact of school choice reforms on disadvantaged students?"
Harvard Educational Review 72(2): 157-176.
Goldhaber, D. D. (1999). "School Choice: An Examination of the Empirical
Evidence on Achievement, Parental Decision Making, and Equity."
Educational Researcher 28(9): 16-25.
Goldring, E. B. and C. S. Hausman (1999). "Reasons for parental choice of
urban schools." Journal of Education Policy 14: 469-490.
344
Gorard, S. and J. Fitz (2000). "Investigating the determinants of segregation
between schools." Research Papers in Education 15(2): 115-132.
Gorard, S. and J. Fitz (2006). "What counts as evidence in the school choice
debate?" British Educational Research Journal 32(6): 797-816.
Gorard, S., J. Fitz, et al. (2001). "School Choice Impacts: What Do We
Know?" Educational Researcher 30(7): 18-23.
Greenberg, J. P. (2011). "The impact of maternal education on children's
enrollment in early childhood education and care." Children and Youth
Services Review In Press, Corrected Proof.
Greene, J., T. Loveless, et al. (2010). Expanding choice in elementary and
secondary education: A report on rethinking the federal role in
education, Brown Center on Education Policy at Brookings.
Hanson, K. and L. Litten (1982). Mapping the road to academia: A review of
research on women, men, and the college selection process. The
undergraduate woman: Issues in Educational Equity. P. Perun.
Lexington, Lexington Books.
Henig, J. (1994). Rethinking school choice: Limits of the market metaphor.
Princeton, NJ Princeton University Press.
Herbst, K., M. L. Newell, et al. (2010). "Study summary: ACDIS."
Hofmeyr, J. and S. Lee (2004). The new face of private schooling. Changing
Class. L. Chisholm. Cape Town, HSRC Press: 143-174.
Holmes, J. (2002). "Buying homes, buying schools: school choice and the
social construction of school quality." Harvard Educational Review 72
(2): 177-205.
Howe, L. D., J. R. Hargreaves, et al. (2008). "Issues in the construction of
wealth indices for the measurement of socio-economic position in low-
income countries." Emerging Themes in Epidemiology 5(3).
Hoxby, C. M. (1998). "What Do America's 'Traditional' Forms of School
Choice Teach Us About School Choice Reforms?" Economic Policy
Review, Vol.4, No.1, March 1998.
Hoxby, C. M. (2002). "School Choice and School Productivity (or Could
School Choice be a Tide that Lifts All Boats?)." SSRN eLibrary.
Hoxby, C. M. (2003). The economics of school choice. Chicago, The
University of Chicago Press.
Hoxby, C. M. (2003). Introduction. The economics of school choice. C. M.
Hoxby. Chicago, The University of Chicago Press: 1-22.
Hunter, M. (2010). Racial desegregation and schooling in South Africa:
contested geographies of class formation.
Johnson, D. (2007). "Building citizenship in fragmented societies: The
challenges of deracialising and integrating schools in post-Apartheid
South Africa." International Journal of Educational Development 27(3):
306-317.
Kahn, K., S. M. Tollman, et al. (2007). "Research into health, population and
social transitions in rural South Africa: Data and methods of the
Agincourt Health and Demographic Surveillance System."
Scandinavian Journal of Public Health 35(3 supp 69): 8-20.
345
Kanjee, A. (2007). Improving learner achievement in schools: applications of
national assessments in South Africa. State of the Nation: South Africa
2007. S. Buhlungu, J. Daniel, R. Southall and J. Lutchman. Cape
Town, HSRC Press: 470-499.
Kanjee, A. and A. Chudgar (2009). Accuracy of the poverty quintile system for
classifying South African schools. 2nd Monitoring and Evaluation
Colloquim, Gauteng Department of Education. Sandton, Johannesburg.
Kanjee, A. and A. Chudgar (2009). "School money: funding the flaws." HSRC
Review 7(4).
Karlsson, J. (2007). Mobility and equity: school transport, cost of schooling
and class formation in post-apartheid South Africa. Dilemmas of
implementing education reforms: Explorations from South Africa and
Sweden. C. Odora Hoppers, U. P. Lundgren, J. Pampallis, E. Motala
and E. Nihlfors. Uppsala, Sweden, STEP, Uppsala Universitet.
Karsten, S., C. Felix, et al. (2006). "Choosing Segregation or Integration?: The
Extent and Effects of Ethnic Segregation in Dutch Cities." Education
and Urban Society 38(2): 228-247.
Klasen, S. (2002). "Low Schooling for Girls, Slower Growth for All?
Cross‐Country Evidence on the Effect of Gender Inequality in
Education on Economic Development." The World Bank Economic
Review 16(3): 345-373.
Kolenikov, S. and G. Angeles (2008). Socioeconomic status measurement with
discrete proxy variables: is principal component analysis a reliable
answer?
Kristen, C. (2005). School choice and ethnic school segregation: Primary
school selection in Germany. Munster, Waxmann Verlag GmbH.
Ladd, H. (2003). Introduction. Choosing choice: School choice in international
perspective. D. Plank and G. Sykes. New York, NY, Teachers College
Press: Columbia University: 1-23.
Lam, D., C. Ardington, et al. (2008). The Cape Area Panel Study: a very short
introduction to the integrated waves 1-2-3-4 data, The University of
Cape Town.
Lam, D., C. Ardington, et al. (2008). The Cape Area Panel Study: Overview
and Technical Documentation, Waves 1-2-3-4 (2002-2006). University
of Cape Town.
Lam, D., C. Ardington, et al. (2008). Schooling as a lottery: racial differences
in school advancement in urban South Africa. Southern Africa Labour
and Development Research Unit Working Paper, SALDRU, University
of Cape Town.
Le, A. T. and P. W. Miller (2003). "Choice of School in Australia:
Determinants and Consequences." Australian Economic Review 36(1):
55-78.
Leibbrandt, M., I. Woolard, et al. (2009). Methodology: Report on NIDS wave
1, Technical paper no. 1. University of Cape Town, National Income
Dynamics Study.
346
Lemon, A. (2005). "Shifting geographies of social inclusion and exclusion:
Secondary education in Pietermaritzburg, South Africa." African
Affairs 104: 69-96.
Levin, H. M. (1991). "The economics of educational choice." Economics of
Education Review 10(2): 137-158.
Levin, H. M. (1998). "Educational Vouchers: Effectiveness, Choice, and
Costs." Journal of Policy Analysis and Management 17(3): 373-392.
Lombard, B. (2007). "Reasons why educator-parents based at township schools
transfer their own children from township schools to former Model C
schools." Education as Change 11(1): 43 - 57.
Lubienski, C., P. Weitzel, et al. (2009). "Is There a "Consensus" on School
Choice and Achievement?: Advocacy Research and the Emerging
Political Economy of Knowledge Production." Educational Policy
23(1): 161-193.
Magnuson, K. (2007). "Maternal education and children's academic
achievement during middle childhood." Dev Psychol 43(6): 1497-1512.
Maile, S. (2004). "School Choice in South Africa." Education and Urban
Society 37(1): 94-116.
Martin, P. (2010). Government-funded programmes and services for
vulnerable children in South Africa. Cape Town, South Africa.
May, J. D., J. Aguero, et al. (2007). "The KwaZulu-Natal Income Dynamics
Study (KIDS) Third Wave: Methods, First Findings and an Agenda for
Future Research." Development Southern Africa 24(5): 629-648.
May, J. D., M. Carter, et al. (1999). KwaZulu-Natal Income Dynamics Study
(KIDS) 1993-1998. Working Paper No. 21.
McMillan, J. and S. Schumacher (2005). Research in education: evidence
based enquiry, Allyn & Bacon.
Motala, S. (1995). "Surviving the System - a critical appraisal of some
conventional wisdoms in primary education in South Africa."
Comparative Education 31(2): 161 - 180.
Motala, S. (2009). "Privatising public schooling in post-apartheid South Africa
– equity considerations." Compare: A Journal of Comparative and
International Education 39(2): 185 - 202.
Motala, S., V. Dieltiens, et al. (2009). "Physical access to schooling in South
Africa: mapping dropout, repetition and age-grade progression in two
districts." Comparative Education 45(2): 251-263.
Msila, V. (2005). "The education exodus: the flight from township schools."
Africa Education Review 2(2): 173-188.
Msila, V. (2009). "School choice and intra-township migration: black parents
scrambling for quality education in South Africa." Journal of Education
46: 81-98.
Narodowski, M. and M. Nores (2002). "Socio-economic segregation with
(without) competitive education policies. A comparative analysis of
Argentina and Chile." Comparative Education 38 (4): 429-451.
347
Nelson Mandela Children's Fund (2005). Emerging Voices: A report on
Education in South African rural communities. Cape Town, HSRC
Press.
Nkomo, M., C. McKinney, et al. (2004). Reflections on school integration:
colloquium proceedings. Cape Town, HSRC Press
Noden, P. (2000). "Rediscovering the Impact of Marketisation: Dimensions of
Social Segregation in England's Secondary Schools, 1994-99." British
Journal of Sociology of Education 21(3): 371-390.
Norris, S. A., L. M. Richter, et al. (2007). "Field Report: Panel studies in
developing countries: case analysis of sample attrition over the past 16
years within the Birth to Twenty cohort in Johoannesburg, South
Africa." Journal of International Development 19 1143-1150.
Osborne, C. and S. McLanahan (2007). "Partnership Instability and Child
Well-Being." Journal of Marriage and Family 69: 1065-1083.
Pampallis, J. (2003). Education reform and school choice in South Africa.
Choosing choice: School choice in international perspective. D. Plank
and G. Sykes. New York, NY, Teachers College Press, Columbia
University: 143-163.
Pampallis, J. (2008). School fees. Issues in Education Policy. Johannesburg,
Centre for Education Policy Development.
Paterson, A. and G. Kruss (1998). "Educational migration and its effect on
access to schooling in South Africa." South African Journal of
Education 18 (3): 149-155.
Pendlebury, S. and N. Rudolph (2008). Children's access to education. South
African Child Gauge 2007/2008. P. Proudlock, M. Dutschke, L.
Jamieson, J. Monson and C. Smith. Cape Town, Children's Institute,
University of Cape Town: 74 -77.
Peterson, P., W. Howell, et al. (2003). School vouchers: results from
randomized experiments. The economics of school choice. C. M.
Hoxby. Chicago, The University of Chicago Press: 107-144.
Plank, D. and G. Sykes (2003). Choosing choice: School choice in
international perspective. New York, NY, Teachers College Press,
Columbia University.
Reddy, V. (2006). Mathematics and science achievement at South African
schools in TIMMS 2003. Cape Town, HSRC Press.
Reschovsky, A. (2006). "Financing Schools in the New South Africa."
Comparative Education Review 50(1): 21-45.
Richter, L., S. Norris, et al. (2004). "Transition from Birth to Ten to Birth to
Twenty: the South African cohort reaches 13 years of age." Paediatric
and Perinatal Epidemiology 18 290-301.
Richter, L., S. Norris, et al. (2007). "Cohort Profile: Mandela's children: The
1990 birth to twenty study in South Africa." International Journal of
Epidemiology: dym016.
Richter, L., S. Norris, et al. (2006). "In-migration and living conditions of
young adolescents in greater Johannesburg, South Africa " Social
Dynamics 32 (1): 195-216.
348
Richter, L., S. Panday, et al. (2009). "Factors influencing enrolment: A case
study from Birth to Twenty, the 1990 birth cohort in Soweto-
Johannesburg." Evaluation and Program Planning 32 (3 ): 197-203.
Richter, L., S. Panday, et al. (2009). "Adolescents in the city: material and
social living conditions in Johannesburg-Soweto, South Africa." Urban
Forum 20 319-334.
Rinne, R., J. Kivirauma, et al. (2002). "Shoots of revisionist education policy
or just slow readjustment? The Finnish case of educational
reconstruction." Journal of Education Policy 17(6): 643-658.
Rutstein, S. O. and K. Johnson (2004). The DHS wealth Index. DHS
Comparative Reports. Calverton, MD, Measure DHS+.
Saporito, S. (2003). "Private Choices, Public Consequences: Magnet School
Choice and Segregation by Race and Poverty." Social Problems 50(2):
181-203.
Schneider, M. and J. Buckley (2002). "What Do Parents Want From Schools?
Evidence From the Internet." Educational Evaluation and Policy
Analysis 24(2): 133-144.
Schneider, M., M. Marschall, et al. (1998). "School Choice and Culture Wars
in the Classroom: What Different Parents Seek from Education." Social
Science Quarterly (University of Texas Press) 79(3): 489-501.
Scott, D. W. (1992). Multivariate density estimation: theory, practice, and
visualization. New York, New York, John Wiley.
Sekete, P., M. Shilubane, et al. (2001). Deracialisation & Migration of learners
in South African schools: Challenges and Implications. Pretoria HSRC
Press.
Sinnott, R. W. (1984). "Virtues of the Haversine." Sky and Telescope 68(2):
159.
Smith, E. (2006). Using secondary data in educational and social research.
New York, NY McGraw Hill/Open University Press.
Soderstrom, M. and R. Uusitalo (2005). School choice and segregation:
evidence from an admission reform. Uppsala, Sweden, Institute for
Labour Market Policy Evaluation. 7
Soudien, C. (2003). Inclusion and exclusion in some South African schools:
Preliminary findings.
Soudien, C., N. Carrim, et al. (2004). School inclusion and exclusion in South
Africa: some theoretical and methodological considerations.
Reflections on school integration:. M. Nkomo, C. McKinney and L.
Chisholm. Cape Town, HSRC Press: 19-42.
South African Human Rights Commission (2004). The right to education,
South African Human Rights Commission.
South African Human Rights Commission (2006). Report of the public hearing
on the right to basic education, South African Human Rights
Commission
Spaull, N. (2011). A preliminary analysis of SACMEQ III South Africa.
Stellenbosch Economic Working Papers. Stellenbosch, University of
Stellenbosch.
349
Sujee, M. (2004). Deracialisation of Gauteng schools - a quantitative analysis.
Reflections on school integration: Colloquium proceedings. M. Nkomo,
C. McKinney and L. Chisholm. Cape Town, South Africa, HSRC
Publishers.
Teske, P. and M. Schneider (2001). "What Research Can Tell Policymakers
about School Choice." Journal of Policy Analysis and Management
20(4): 609-631.
Tsang, M. (2003). School choice in the People's Republic of China. Choosing
choice: School choice in international perspective. D. Plank and G.
Sykes. New York, NY, Teachers College Press, Columbia University:
164-195.
Unterhalter, E. (2005). Gender equality and education in South Africa:
Measurements, scores and strategies. Gender equity in South African
education 1994-2004: Perspectives from research, government and
unions. L. Chisholm and J. September. Cape Town, South Africa,
HSRC Press.
van der Berg, S., C. Burger, et al. (2011). Low quality education as a poverty
trap. Stellenbosch, Department of Economics, Stellenbosch University.
van der Berg, S., L. Wood, et al. (2002). "Differentiation in black education."
Development Southern Africa 19: 289-306.
Veriava, F. (2005). Free to learn: A discussion paper on the School Fee
Exemption policy. Cape Town, South Africa, Children's Institute,
University of Cape Town.
Viteritti, J. (2005). School choice: how an abstract idea became a political
reality, Brookings Institute.
Vyas, S. and L. Kumaranayake (2006). "Constructing socio-economic status
indices: how to use principal components analysis." Health Policy and
Planning 21(6): 459-468.
Waslander, S. and M. Thrupp (1995). "Choice, competition and segregation:
an empirical analysis of a New Zealand secondary school market,
1990-93." Journal of Education Policy 10(1): 1-26.
Weiher, G. and K. Tedin (2002). "Does choice lead to racially distinctive
schools? Charter schools and household preferences." Journal of Policy
Analysis and Management 21(1): 79-92.
West, A. and A. Hind (2007). "School Choice in London, England:
Characteristics of Students in Different Types of Secondary Schools."
Peabody Journal of Education 82(2): 498 - 529.
Witte, J. F. and C. A. Thorn (1996). "Who Chooses? Voucher and Interdistrict
Choice Programs in Milwaukee." American Journal of Education
104(3): 186-217.
Woolman, S. and B. Fleisch (2006). "South Africa's unintended experiment in
school choice: how the National Education Policy Act, the South Africa
Schools Act and the Employment of Educators Act create the enabling
conditions for quasi-markets in schools." Education and the Law 18:
31-75.
350
Yach, D., N. Cameron, et al. (1991). "Birth to Ten: Child health in South
Africa in the nineties: rationale and methods of a birth cohort study."
Paediatric and Perinatal Epidemiology 5: 211-233.
Yamauchi, F. (2004). Race, equity and public schools in post-Apartheid South
Africa: Is opportunity equal for all kids? IFPRI Food Consumption and
Nutrition Division Working Paper Series.
Zietz, J. and P. Joshi (2005). "Academic choice behavior of high school
students: economic rationale and empirical evidence." Economics of
Education Review 24(3): 297-308.