Some Behavioral Issues
Panel surveys are most useful for the analysis of models of dynamic behavior, change over time, latent variable models, and measurement error. Panel data are usually treated as observations of a given economic unit for a fixed number of periods. This paper points out some of the ways that panel samples themselves are dynamic and may influence or be influenced by the behaviors under study. The discussion relies on the author's experience with the Panel Study of Income Dynamics (PSID) and other panel data and, in particular, on a study of the PSID's representativeness after 14 years. The author introduces the issues of sample dynamics that arise in family-oriented panel surveys; discusses the potential influence of entry and exit of sample members on behavioral outcomes; and considers the influence of recent behavior on the likelihood of leaving the panel, both for original sample members and new entrants. The findings suggest that the inclusion of nonsample individuals or of the earlier observations for those who leave the panel can dramatically affect the size and character of the sample used for the estimation of a particular behavioral model.