About the Book

Interaction Effects in Linear and Generalized Linear Models is a comprehensive and accessible text that provides a unified approach to interpreting interaction effects. The book develops the statistical basis for the general principles of interpretive tools and applies them to a variety of examples, introduces the ICALC Toolkit for Stata, and offers a series of start-to-finish application examples using ICALC to show analysts and students how to interpret interaction effects for a variety of different techniques of analysis, beginning with OLS regression. It emphasizes the challenges of interpretation for GLM’s with non-linear link functions—such as binomial and multinomial logistic regression, probit analysis, ordinal regression models, and count models— and offers solutions to overcome these challenges.
Purchase Book

Reviews & Comments

ICALC Description

ICALC stands for Interaction CALCulator, developed as a companion to Robert L. Kaufman’s Interaction Effects in Linear and Generalized Linear Models (Kaufman 2018, Sage). It is a set of tools designed to produce the calculations, tables, and graphics for each of the book’s principles of interpretation. The ICALC Toolkit for Stata is available at no charge (download here). It consists of a set-up tool plus four separate tools for understanding the effect of the focal variable on the outcome as moderated by the predictor(s) with which the focal variable interacts:
Specify the elements of your interaction model—the main effect variables, the two-way effect terms, and the three-way effect terms (if any). Optionally choose a display name and/or set display values for each of the interacting variables.
GFI Tool
GFI stands for Gather, Factor, and Inspect. Find the algebraic expression for the effect of the focal variable on the modeled outcome. Determine if the focal variable’s effect changes sign as it varies with the values of its moderators..
Create a table showing the focal variable’s effect and statistical significance as it varies across the values of the moderating predictor(s). Effects may be scaled in the original, as-estimated metric, or when relevant as factor changes and/or marginal changes.
Create visual counterparts to the significance region tables produced by SIGREG. Plot information about the varying magnitude and optionally the significance of the focal variable’s moderated effect.
Create tables and/or graphs of the predicted values of the outcome as it varies with the interacting predictors. Includes alternative displays for GLMs’ with non-linear link functions, such as logistic regression, ordinal regression models and count models.