An Assessment of Attrition in a Multi-Wave Panel of Households

Abstract

Attrition is an issue of great importance in empirical analysis using longitudinal data in which measurements are taken at two or more points in time on the same sample of units. Although attrition per se need not be a problem, any bias due to loss of sample size can have a profound effect on the usefulness of the empirical outputs of the study. For example, if in the current context of predicting automobile energy consumption the households that are lost at each recontact point are typically high-kilometre households then parameter estimates associated with a study in which vehicle use is endogenous could be significantly biased. If, however, there is no difference in the distribution of kilometres between the total sample and the continuing respondents, but there are some differences with respect to exogenous variables (e.g. number of workers, income, household size), it is not necessarily the case that attrition is a source of bias given the objectives of the study.