Tools and Methods for Addressing Multicollinearity in Energy Modeling

Introduction

Introduction After reviewing SEM measurement and verification (M&V) protocols from different programs, the ETSP team identified multicollinearity as a common statistical issue in industrial data sets.  Understanding that multicollinearity has the potential to affect the specification of regression-based energy models used to determine adjusted baselines, more consistent treatment of this issue may help improve confidence in SEM-based savings, and thereby address a potential market barrier.  Therefore, the team compiled this paper for the purposes of outlining the implications of multicollinearity in the context of SEM measurement and verification, and providing examples of how program implementers have successfully identified and treated the presence of multicollinearity among a set of predictor variables. 

Statistical Interpretation

Multicollinearity is present when two or more independent or predictor variables in a regression model are correlated among themselves.  When two independent variables tend to move together through time, including both variables  may not add appreciably to the explanatory power of the model, compared to just having one or the other in the model.  Essentially, the additional variable has little new information to provide.  While perfect multicollinearity involves two variables that are linear translations of each other (R²=1), partial-multicollinearity is more commonly observed in the context of regression-based energy models.

Download the full PDF above to continue reading.