Session Title:  Small area estimation for official United States household statistics:  employment, inflation, and energy use

Description:  Household surveys have for decades been central to the U.S. federal statistical system.  They provide policy makers with crucial information on the nation’s labor force, consumption patterns, energy use, technology adoption, and other topics.  With voluntary household survey response rates declining and data collection costs increasing, statistical agencies look to administrative data sources and small area estimation techniques to produce estimates at detailed geographic levels.  We explore new techniques U.S. government statisticians are developing to bridge the gap between sparse survey data and the data users’ increasing interest in city and community-level information.

Speaker 1:  Tim Trudell, U.S. Census Bureau (Tim.Trudell@census.gov)
Title:  “Using linked survey and administrative data to estimate monthly state-level housing unit vacancy rates”

Abstract
The U.S. Current Population Survey (CPS), a major U.S. household survey, provides labor force and other economic statistics for the United States. Using administratively linked CPS microdata, we evaluate the potential for a small area model to produce monthly state-level housing unit vacancy estimates, consistent with published annual state and quarterly national estimates.

Speaker 2:  Vladislav Beresovsky, U.S. Bureau of Labor Statistics (Beresovsky.Vladislav@bls.gov)
Title: Challenges of estimating inflation in small areas in official statistics 

Abstract
The Consumer Price Index (CPI) survey is designed to measure inflation by collecting quotes in sampled Core-Based Statistical Areas (CBSA) of the U.S. The current design of the survey provides for reliable estimation of relative price changes with uncertainty measures in selected large CBSA and aggregated estimates in Census Divisions. We use Bayesian area level models to mass impute inflation measures for all CBSAs in the U.S. and produce aggregated estimates, for instance in states. We employ Bayesian model selection techniques and utilize spatial modeling to compensate for sparse representativeness of the available sample. We co-model point and variance estimates in sampled CBSAs to smooth out direct variance estimates. Model-based estimates of inflation changes are studied for different model assumptions.                 

Speaker 3: Matthew Sanders, U.S. Energy Information Administration (Matthew.Sanders@eia.gov)
Title: Estimating household energy consumption at the community level using the Residential Energy Consumption Survey data

Abstract
EIA’s Residential Energy Consumption Survey (RECS) provides comprehensive data on U.S. household energy use, including information on housing unit structure, heating fuels, and appliances.  The 2020 RECS was a single-stage stratified sample of nearly 18,500 households representing the 123.5 million housing units that are occupied as a primary residence in the United States. The relative standard error (RSE) for all fuel total at the state level was around 4%. While the 2020 sample was sufficient to provide direct survey estimates at the state level, in recent years there has been high demand for consumption and expenditure estimates at more granular or community level geographics. In this presentation, we will discuss applying small area estimation methods for producing community level energy consumption and expenditures estimates using RECS data and discuss estimating mean squared errors. 

Discussant:  Danny Pfefferman, University of Southampton, UK (d.pfeffermann@soton.ac.uk)

Chair and Organizer:  Janice Lent, U.S. Energy Information Administration (Janice.Lent@eia.gov)