What do surveys say about the historical trend of inequality and the applicability of two table-transformation methods? (2303.05895v1)

Abstract: We apply a pseudo panel analysis of survey data from the years 2010 and 2017 about Americans' self-reported marital preferences and perform some formal tests on the sign and magnitude of the change in educational homophily from the generation of the early Boomers to the late Boomers, as well as from the early GenerationX to the late GenerationX. In the analysis, we control for changes in preferences over the course of the survey respondents' lives. We use the test results to decide whether the popular iterative proportional fitting (IPF) algorithm, or its alternative, the NM-method is more suitable for analyzing revealed marital preferences. These two methods construct different tables representing counterfactual joint educational distributions of couples. Thereby, they disagree on the trend of revealed preferences identified from the prevalence of homogamy by counterfactual decompositions. By finding self-reported homophily to display a U-shaped pattern, our tests reject the hypothesis that the IPF is suitable for constructing counterfactuals in general, while we cannot reject the applicability of the NM. The significance of our survey-based method-selection is due to the fact that the choice between the IPF and the NM makes a difference not only to the identified historical trend of revealed homophily, but also to what future paths of social inequality are believed to be possible.