Tenure choice: Cross-sectional and longitudinal analyses

Abstract

Longitudinal data sets and modeling strategies have become more common in the social sciences. With their emergence, a tendency has arisen to criticize previous cross-sectional analyses. Some argue that cross-sectional analyses of certain geographical processes may have provided biased estimates of the model parameters. Others have gone further, suggesting that the results are misleading and must be viewed with suspicion. Our experience with both cross-sectional and longitudinal analysis of the processes of residential mobility and housing choice suggests otherwise. In our experience, it is not a matter of inferiority of cross-sectional analysis but of how (given the far fewer adequate panel data sets) each of these methods can enrich and complement each other. We will argue that longitudinal studies add to and enrich our understanding of the process of mobility and housing choice gained from earlier cross-sectional work. In particular, we want to emphasize that panel studies allow us to relate the event of changing residence to other events in the household life cycle, such as the advent of children, job change, increases in household income, and changes in marital status. Indeed, panel data allow us to demonstrate that the event of moving from a rental dwelling to ownership is embedded in a whole sequence of events. Those other events include forming a stable relationship (marriage), having children, and increasing the household income (Clark et al., 1994). In this article, we compare the results of analyses of residential mobility and tenure choice from cross-sectional and longitudinal data and models. In doing so, we place particular emphasis on our own research and experience. Our research has used existing data files. For instance, we have made use of the Netherlands Housing Demand Survey sample surveys (held at four-year intervals), the Annual Housing Surveys for the United States, and the Panel Study of Income Dynamics for the United States. Therefore, our research has always dealt with revealed preference that is, the actual housing choice of households and not with stated preference (see Timmermans, et. al., in this issue). Of course, existing data files do not allow