259 – Weighting
Statistical Calibration of Energy-Engineering End-Use Electricity Consumption Estimates: A Bayesian Multilevel Model Approach
Hiroaki Minato
U.S. Energy Information Administration
How do American homes consume energy? We tackle the question statistically with survey data and engineering estimates. Energy choices and home environments are measured for a national sample of homes by the Residential Energy Consumption Survey (RECS) conducted by the U. S. Energy Information Administration. From the survey response variables on building characteristics as well as energy end choices and uses, engineering models are formulated to estimate end-use energy consumptions. The sample homes' total energy consumptions are also captured in RECS. A statistical objective is to develop a robust probabilistic model to partition the total consumption into end-use consumptions, even when the engineering models may not be very accurate. In this paper, we try a Bayesian multilevel modeling approach for the partitioning, taking electricity as an example. By partially pooling sample homes by geography and end-use combination and by accounting for respondent selection through the weights, a multilevel model of the annual electricity consumption is built from engineering-based electricity end-use estimates. Our uncertainties about model parameters are incorporated as priors.