Print

Parlin, Kathryn

Mathematics

2006

MS

Estimating the benefits and costs of energy efficiency programs is critical for ensuring well-designed and effective programs, and for maintaining public support. Impact evaluation is designed to quantify the savings associated with these efficiency efforts, an objective that is commonly achieved through the analysis of customers' bills. Linear regression of the efficiency interventions and weather effects on monthly use is a popular strategy to estimate savings, necessitating decisions regarding the structure and composition of the appropriate model(s). Within this context, model selection is an important, but often-ignored, aspect of impact evaluation. Without a clear and objective standard for identifying the "best" model, the researcher is left in the position of making a decision based on his or her frame of reference and lacks a strong foundation to support the choice of models. The information-theoretic approach provides the framework and theoretical justification for identifying the top-ranking model and counterbalancing the competing objectives of minimizing both the bias and variance in the model. Applying this model selection criterion, however, does not absolve the researcher of fully justifying of the range of candidate models. Defining the candidate models is the single most important step in this process, and consequently, the majority of this research is focused on narrowing the variety of options to those most suitable for impact evaluation.
This document includes an extensive discussion of the options for pooling cross-sectional time series in the context of estimating energy savings from conservation programs. The most viable options are then applied to a specific residential retrofit program in Vermont, and the information-theoretic approach to model selection is illustrated. This analysis supports the use of the fixed effects model for estimating program impacts, and also suggests that the model fit can be improved by addressing heteroskedasticity through weighted least square regression. The combination of a thorough review of the modeling options and the application of the information-theoretic approach to model selection gives us the ability to sort through the various choices and identify the best approach from the set of candidate models. This strategy creates a powerful tool for developing robust and defensible estimates of the impacts fiom energy efficiency programs.