Using Log-Linear Models for Longitudinal Data to Test Alternative Explanations for Stage-Like Phenomena: An Example from Research on Adolescent Substance Use

Abstract

We show how log-linear models for multi-wave longitudinal data can be used to test hypotheses relating to stage-like relationships between variables. To illustrate our approach, we use an example from research on adolescent cigarette and marijuana experimentation. Previous research has documented that most adolescents experiment with substances in a stage-like sequence: first cigarettes and then marijuana. Several hypotheses have been suggested as potential explanations for this stage-like phenomenon. We show that traditional two-wave analyses give results that are difficult to interpret and that our multiple-wave analyses allow the testing of several theoretically interesting hypotheses. We also illustrate that log-linear models may be useful for testing hypotheses about stage-like phenomena in many other areas of psychological research. We suggest how several other techniques have the potential to be used as a multivariate analogue to the log-linear approach.