American Statistical Association
It is common in psychiatry and public health to assess interactions (or effect measure modifications). That is, it is often of interest to explore whether the effect of some exposure on some outcome differs in the presence or absence of another variable (the moderator). The most common way to operationalize the estimation of this moderation effect is through the inclusion of a cross-product term between the exposure and the potential moderator in a regression analysis. This cross-product term is commonly called an interaction term. Another method that is often used when the potential moderator is categorical is to perform stratified regression analysis at each level of the moderator and then compare the coefficients for the exposure variable. In the present talk we will emphasize that the answer to the question of whether there is or is not effect measure modification depends on what scale the interaction is considered. Specifically in the case of dichotomous outcomes this means whether we are examining risk differences (additive scale) or odds ratios (multiplicative scale). While it is common to test the statistical significance of the interaction term in a logistic regression, this is only a test for interaction on the logit (i.e. odds ratio - multiplicative) scale, and in general it is not consistent with the test for interaction on the probability (i.e. risk difference – additive) scale. We will demonstrate with worked examples how to test for both multiplicative and additive interactions.
Dr. Sharon Schwarz is Professor of Clinical Epidemiology in the Mailman School of Public Health at Columbia University. She is the training coordinator for the psychiatric epidemiology training program and teaches Epidemiology 2 and a doctoral level seminar in epidemiologic methods.
|Date:||Tuesday, February 28, 2012|
|Time:||3:30 - 4:30 P.M.|
New York State Psychiatric Institute
1051 Riverside Drive
6th Floor Multipurpose Room (6602)
New York, New York