Online Program

Saturday, February 23
CS13 Theme 2: Data Modeling and Analysis #5 Sat, Feb 23, 9:00 AM - 10:30 AM
Napoleon A1&2

Confounders, Mediators, Moderators & Suppressors: Identifying and Testing for Different Types of Covariates (302455)

View Presentation View Presentation

Elaine Allen, Babson College 
Christopher A Seaman, Quahog Research Group 
*Julia E Seaman, UCSF 

Keywords: Covariates, Model validation, Pitfalls

Once a model is fit and relationship between dependent and independent variables is quantified, it is important to consider the role of additional variables in the model. We will consider four types of third variable effects—confounding, mediation, moderation and suppression where an additional variable may obscure, clarify, or change the nature of the relationship between independent variables and a dependent variable. Statistically the four types of covariates are similar but the changes in the relationship variables can lead to different models and conclusions.

By definition, confounders are variables related to two variables in the model that falsely obscures or accentuates the relationship between them. Mediators are variables that operate between the independent and dependent variable relationship and decompose this into two causal paths. With a moderator, there is a different relationship between independent and dependent variables when the covariate takes on different levels. Suppressor covariates increase the predictive validity of another variable (or set of variables) when it is included in the model. Examples of each and their interrelationships will be given.