Multiple group association models with latent variables: an analysis of secular trends in abortion attitudes,...

This paper examines methods for integrating latent class models with association and conditional association models, including those in multiple groups. These methods are shown to be generalizable to a wide range of association models in which one or more variables is latent and shown to be especially well suited to modeling temporal changes (tends) in latent structures. Like the usual association model, association models with latent variables utilize the information inherent in the ordering of the categories. Consequently, these models require the estimation of fewer parameters and are preferred to loglinear models that may require the estimation of substantial numbers of parameters even for models with a moderate number of observed variables. Following a discussion of the formal model, an example is considered. This example focuses on the American public's approval (disapproval) of legalized abortion from 1972 to 1988.