A General Model for the Analysis of Surveys

Abstract

This article shows how the combined use of direct estimation methods and indirect testing procedures, which was advocated by Goodman (1970-1971a), can beapplied in survey analysis. this approach to survey analysis replaces a "causal system" diagram (of the kind used, e.g., by Davis 1971) by a more quantitativelyexplicit model of the system, which then leads to a more direct and more unifiedassessment of the relationships among the variables in the system. The methods presented in the present article can also help the survey analyst to determine whether his survey data support or negate a given hypothesized causal system; and in some cases these methods can be used to determine alternative causal systems that provide better descriptions of the phenomena under investigation. To illustrate the application of these methods, we shall reanalyze data that were analyzed recently by Davis (1971) and earlier by Wilner, Walkley, and Cook (1955) in their well-known study of the "contact hypothesis." We also include in the present article some new results on how the relationship between two given dichotomous variables is affected by the introduction of additional variables.