Modeling Marketing Data with Collinear Data

• Activity Number: 137
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and Marketing
• Author(s): Joseph Retzer*+
• Companies: MarketTools Inc.
• Keywords: collinearity ; lasso ; Bayesian ; random forest ; elastic net ; stepwise

Abstract

Data collinearity in multiple regression occurs when two or more regressors are highly correlated. The problem often manifests itself in inflated coefficient standard errors and general model instability. As the number of independent variables increases, other things being equal, the likelihood of collinearity increases. Marketing data models often contain related variables making the detection and accommodation of collinearity a modeling priority.

This presentation will explore various approaches to dealing with collinear data which fall under three general headings: (1) variable reduction, (2) model averaging and (3) use of prior information. A comparison of results, cautious recommendations and a discussion of common non-statistical considerations will be presented.