633 – Model-Based Statistics and Applications
Validation of 2015 Residential Energy Consumption Survey End-Use Estimates by Bayesian Calibration Models
Hiroaki Minato
U.S. Energy Information Administration
The 2015 Residential Energy Consumption Survey (RECS) national end-use estimates were produced in three steps: (1) the annual-level end-use energy consumption models, developed by energy-engineering experts (EEE), were applied to the 2015 RECS data and some weather data, outputting initial EEE end-use estimates; (2) the EEE end-use estimates were calibrated to the fuel-specific total consumption value for each sample respondent, followed by imputation for cases with missing total consumption values; and (3) the calibrated end-use estimates are multiplied by the survey weights to produce the 2015 national end-use estimates. In this paper, we keep (1) and (3) as fixed, but in (2) we implement two simpler but more direct Bayesian calibration methods than the official method in an attempt to increase the validity of the latter method and its published results. Uncertainty of the final official estimates are assessed against the posterior predictive distributions that are derived under one of the Bayesian models.