ABSTRACT
Any analysis using panel surveys must contend with the influences of attrition on sample representation over time. Attrition affects the results of longitudinal analysis by reducing the sample size, thereby diminishing the efficiency of estimates. More importantly, attrition may be of a selective nature and thereby increase the bias of certain estimates. Attrition may be biased with respect to the characteristics of nonresponse individuals or with respect to the recent behaviour of individuals. For example, household composition changes are the outcome of numerous demographic shifts and processes. Panel data potentially are rich sources of information about the causes and conditions which stimulate such changes. Consequently, knowledge of the sequence of transitions will lend insight into the demographic processes in evidence (Richards et al., 1987). An assessment of both characteristic and behavioural selectivity in attrition rates is an important preliminary stage for longitudinal studies of household dynamics.
