Correlated predictors in regression models are a fact of life in applied social science research.
The extent to which they are correlated will influence the estimates and statistics associated with
the other variables they are modeled along with. These effects, for example, may include
enhanced regression coefficients for the other variables—a situation that may suggest the
presence of a suppressor variable. This paper examines the history, definitions, and design
implications and interpretations when variables are tested as suppressors versus when variables
are found that act as suppressors. Longitudinal course evaluation data from a single study
illustrate three different approaches to studying potential suppressors and the different results and
interpretations they lead to.
The CAUSE Research Group is supported in part by a member initiative grant from the American Statistical Association’s Section on Statistics and Data Science Education