This paper discusses models for Prediction Analysis (PA) in longitudinal research. It describes PA as a non‐standard log‐linear model (von Eye et al., 1993). Models for predictions in longitudinal data are introduced including Equi‐Finality models and Equi‐Causality models of development. Models are described for two and more occasions of measurement. The relationship between formulating prediction hypotheses and model specification is discussed. Data examples illustrate model application and selection of log‐linear models for parameter estimation. The discussion focuses on types of variable relationships and their translation into testable hypotheses.
Notes
Address correspondence concerning this paper to Alexander von Eye, Michigan State University, Department of Psychology, 119 Snyder Hall, East Lansing, MI 48824–1117. The authors are indebted to Judith Glück, University of Vienna, Lars Bergman, University of Stockholm, and the reviewers for helpful comments on earlier versions of this paper. Parts of Alexander von Eye's work on this paper were supported by a grant from the Mary Louis Foundation.