On Testing Moderation Effects in Experiments Using Logistic Regression
17 Pages Posted: 13 Feb 2014 Last revised: 23 May 2023
Date Written: February 10, 2014
Abstract
Consumer researchers seeking to explain the probability of a binary outcome in an experiment often attend to the moderation of one treatment variable’s effect by the value of second. The most common approach for analyzing such data is logistic regression. However, due to the nonlinear transformation applied to the dependent variable in this method, the resulting interaction coefficients do not properly reflect moderation effects on the original probabilities. Significant moderation effects may result in non-significant interaction coefficients and vice versa. We illustrate the issue, discuss possible responses, outline the correct approach for testing moderation effects on probabilities, and demonstrate that addressing this problem makes a practical difference.
Keywords: logistic regression, nonlinear transformation, moderation, experiments
Suggested Citation: Suggested Citation