Causal Analysis of Data from Panel Studies and Other Kinds of Surveys

Abstract

This article presents new models and methods that are particularly appropriate for the causal analysis of data obtained in panel studies and in other kinds of surveys. For a given path diagram, we provide methods for estimating the magnitude of the various effects represented in the diagram and also methods for determining whether the path-diagram model (described by a given system of equations) is congruent with the data. We present an overall test of the entire system of equations in the model, as well as separate tests of each equation in the system, and we also provide methods for comparing the merits and demerits ofa variety of path diagrams pertaining to the data. To illustrate the various advantages of the techniques proposed, we shall reanalyze three sets of data: (a) The famous two-attribute turnover table (the 16-fold table) on voting intention analyzed earlier by Lazarsfeld (1948), Lipset et al. (1954), Campbell (1963), Lazarsfeld, Berelson, and Gaudet (1968), Boudon (1968), and Lazarsfeld (1970); (b) a two-attribute turnover table on student attitudes analyzed earlier by Coleman (1964); and (c) the four-way cross-classification pertaining to the "contact hypothesis" analyzed earlier by Wilner, Walkley, and Cook (1955), Davis(1971), and Goodman (1972b). New insights into these data will be obtained.