Statistical issues in large complex household surveys conducted in development countries: the Lesotho National Household Health and Nutrition Survey

Paiker, Mark Joel
Journal Title
Journal ISSN
Volume Title
The statistical issues in conducting a National Household Survey such as the Lesotho National Household Health and Nutrition Survey (NH&NS) are considered from two aspects: 1) Issues related to data processing and the management of complex survey designs such as the NH&NS and 2) issues related to the weighting and analysis of such data. The problems with short-term consultants, the need for improved management skills and quality control are addressed in this thesis. To prevent common data errors found in a survey such as the NH&NS and to speed up the release of results, recommendations of improved data collection techniques are made. Since developing countries often lack the capacity to conduct large national surveys, the proposal of a well-conducted single round instead of a multi-round survey is investigated. Calculating weighting factors can be a complicated and time- consuming process for a Bureau of Statistics in a developing country, which might prevent the early release of the results. Therefore it is shown how appropriate weighting factors for a complex survey design are generated. It is also necessary to consider the effect of clustering and stratification, whether it is for presenting results as confidence intervals or for fitting models to the data. By applying the technique of portability to surveys such as the NH&NS, much time can be saved and the results released far earlier than when more convential analysis techniques are used. Chi-square analyses of two-way tables, log-linear modelling and logistic regression are the most common analytical techniques applied to the data from such surveys. The adjustments to these techniques to compensate for a complex survey design are demonstrated. These techniques are applied to the child nutrition, maternal care, disability and injury sections of the NH&NS, with an emphasis on the child nutrition data. Various programs have been written in SAS to perform these analyses. The results generated in this thesis using these techniques should be useful to the Lesotho Ministry of Health and serve as a reference for similar health and nutrition surveys in the future.