Panel surveys offer a valuable tool for researchers to explore the dynamics underlying individual and household behaviours. The Achilles heel of panel data is attrition. This paper examines the determinants and implications of attrition in the first two waves of South Africa’s National Income Dynamics Survey (NIDS). Multivariate tests in labour market and health specifications show that there is some moderate evidence of attrition bias in estimated coefficients based on the non-attriting sample. This bias can be seen in labour market specifications, in particular for men, and for Africans, and to a much lesser degree in health specifications, in particular for small samples of Whites. Researchers should take care when using the panel data set to generalise to the overall population.

The authors would like to thank the National Income Dynamics Survey for funding, Ingrid Woolard for her support and referral in the final stages, Natasha Curry for her invaluable editing and Salma Kagee, Rowan Clarke and Sam Day for helpful comments and input.