Comparing the Performance of Alternative Imputation Methods for the Advance Monthly Retail Trade Survey

• Activity Number: 420 - Modern Modeling Approaches for Imputation Using Survey Data
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 4:00 PM to 5:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Katherine Jenny Thompson* and Stephen Kaputa and Nicole Czaplicki and Brian Dumbacher
• Companies: U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau
• Keywords: regression trees; RegARIMA forecasts; Bayesian hierarchical models; link relative estimator

Abstract

The Advance Monthly Retail Trade and Food Services Survey (MARTS) publishes early sales estimates of retail and food service companies approximately two weeks after the reference month. One month later, the preliminary estimate from the larger Monthly Retail Trade and Food Services Survey (MRTS) supersedes the MARTS sales estimate. The MARTS estimates forecast the preliminary – and more reliable -- MRTS estimates. Revisions are between the two corresponding estimates are expected and unavoidable due to differences in design and response rates, among other factors. However, large revisions are scrutinized when they reverse the direction of the previously published month-to-month change. Consequently, the U.S. Census Bureau is conducting a study investigating a suite of methodological enhancements to the current procedures designed to minimize the magnitude of such revisions. We share a simulation study designed to assess candidate sampling designs and evaluate alternative imputation methods for the MARTS. The study constructs sampling frames from historic MRTS response data, retaining irregular industry distributions and testing performance on real-data time-series.