Online Program

Interacting with Interactions: Understanding Interactions and Powering Studies to Detect Them (302744)

Keywords: interaction, model building, sample size

Practicing statisticians are frequently asked to look at the “interaction” of two variables. Interactions are not well taught in courses, and statisticians tend to let the software construct the interaction without input, creating problems with the interpretation. Yet, interactions are frequently “where the action is,” not only in clinical trials, where treatment-by-subgroup interactions are critical to investigate heterogeneous treatment effects, but in most other fields of statistical analysis. This presentation will cover the following objectives: (1) discuss and demonstrate what an interaction term is and what it is testing, using real data to show what statistical software is doing; (2) show the way the software (or statistician) determines the reference group through code specification is critical to the result and interpretation of the interaction term and discuss alternatives to passively accepting what the software selects; and (3) demonstrate how to determine the sample size (or power) for a study to detect interactions with two popular sample size software packages.