315 – Sampling Strategies for Rare and Hard-to-Reach Populations
Researching Forecast Models for Item Imputation in an EIA Survey
Joseph Conklin
Energy Information Administration
EIA's monthly natural gas survey collects volume and revenue data that support the estimates of state level volumes and prices published in EIA's Natural Gas Monthly. This paper and presentation focus on research over a nearly ten year period to improve imputations for volumes that are either not reported or of questionable quality. EIA considered (1) incorporating local heating degree data to improve predictions of distributor level volume, (2) developing multiple measures of success, (3) incorporating variables not included in earlier research, (4) identifying improvements likely to sustain themselves over the long term, and (5) testing novel ways of predicting the sales and transportation components of volume. While no procedure appeared superior after validation testing on 2011 data, a useful by-product of the research was linking the current sample selection code into the imputation code for the monthly natural gas survey.