Online Program

Identifying High-Priority Units for Analyst Imputation in a Monthly Economic Indicator Survey (308016)

Keywords: influence statistic, z test, imputation, advance survey

Currently, only a small set of units are imputed for the Advance Monthly Retail Trade and Food Services Survey (MARTS). These imputed values are computed by analysts using individual historic company ratios and subject matter knowledge, a process that is both time consuming and unique to the individual analyst. The units selected for imputation are subjectively selected from output listings but may be retained in the set of imputed units in later months even if they fall off the output listing. In general the number of units that received analyst imputation increased over time, adding an unmeasurable error component to the final estimates. We sought an automated, repeatable procedure for selecting units for analyst imputation each month that would both consistently identify exceptionally large units within an industry and identify units that are seasonally influential. This paper describes the methodology developed to identify influential units by industry for analyst imputation in MARTS.