ATTRITION IN LONGITUDINAL HOUSEHOLD SURVEY DATA: SOME TESTS FOR THREE DEVELOPING-COUNTRY SAMPLES

Longitudinal household data can have considerable advantages over much morewidely used cross-sectional data. The collection of longitudinal data, however, may bedifficult and expensive. One problem that has concerned many analysts is that sampleattrition may make the interpretation of estimates problematic. Such attrition may beparticularly severe in areas where there is considerable mobility because of migrationbetween rural and urban areas. Many analysts share the intuition that attrition is likely tobe selective on characteristics such as schooling and that high attrition is likely to biasestimates made from longitudinal data. This paper considers the extent of andimplications of attrition for three longitudinal household surveys from Bolivia, Kenya,and South Africa that report very high per-year attrition rates between survey rounds. Ourestimates indicate that (1) the means for a number of critical outcome and familybackground variables differ significantly between attritors and nonattritors; (2) a numberof family background variables are significant predictors of attrition; but (3) nevertheless,the coefficient estimates for “standard” family background variables in regressions andprobit equations for the majority of the outcome variables considered in all three data setsare not affected significantly by attrition. Therefore, attrition apparently is not a generalproblem for obtaining consistent estimates of the coefficients of interest for most of theseoutcomes. These results, which are very similar to results for developed economies,suggest that for these outcome variables—despite suggestions of systematic attrition from univariate comparisons between attritors and nonattritors, multivariate estimates ofbehavioral relations of interest may not be biased due to attrition.