Log‐linear Analysis

Statisticians often characterize log-linear models as analogs of analysis of variance (ANOVA) procedures, and researchers familiar with ANOVA techniques may notice conceptual similarities between the linear and log-linear approaches. This chapter focuses on the general log-linear model, which treats all variables as outcomes, modeling the natural logs of cell frequencies. All forms of log-linear analysis belong to a special class of generalized linear models (GLMs), and GLM techniques are overviewed. The evolution of log-linear modeling is reviewed and a limited number of analyses in communication research are identified. Log-linear models can be applied to analyses containing more than three categorical variables, and the chapter uses a 2 x 3 x 3 x 3 model with a single covariate to demonstrate the processes involved. It concludes with the models for two and three nominal measures in addition to models for two and three ordinal variables.