The Multilevel Analysis of Trends with Repeated Cross-Sectional Data

The purpose of this paper is to develop a multilevel model for the analysis of trends within repeated cross-sectional samples. The model proposed by the authors uses a time series of macro-level variables to explain trends in the individual-level parameters estimated within each sample. This model falls within the general category of varying parameter models but in the present context additional methodological complications arise because of the potential for serial correlations among the errors in the macro-level equations and because the effects of the macro-level variables can be expected to differ across sectors of the population. The latter consideration leads the authors to propose a 3-level model permitting the effects of macro-level variables to be a function of a simple buffering variable that measures the sensitivity of different population groups to change. The authors develop a mixed estimator that allows for both types of methodological complications and they use it to describe and explain recent trends in the process of stratification. The results yield new insights into the structure of short-term trends in the mobility regime. (authors)