(Academy of Science of South Africa (ASSAf), 2016-03) Weir-Smith, G
The longitudinal comparison of census data in spatial format is often problematic because of changes in administrative boundaries. Such shifting boundaries are referred to as the modifiable areal unit problem (MAUP). This article utilises unemployment data between 1991 and 2007 in South Africa to illustrate the challenge and proposes ways to overcome it. Various censuses in South Africa use different reporting geographies. Unemployment data for magisterial districts of census 1991 and 1996 were re-modelled to the 2005 municipal boundaries. This article showed that areal interpolation to a common administrative boundary could overcome these reporting obstacles. The results confirmed more accurate interpolations in rural areas with standard errors below 3300. Conversely, the largest errors were recorded in the metropolitan areas. Huge increases in unemployment between 1996 and 2001 statistics were also evident, especially in the metropolitan areas. Although such areas are more complex in nature, making it more difficult to accurately calculate census data, the increase in unemployment could also be the result of census taking methods. The article concludes that socio-economic data should be available at the smallest possible geographic area to ensure more accurate results in interpolation. It also recommends that new output areas be conceptualised to create a seamless database of census data from 1991 to 2011 in South Africa.