Page 397 - Special Topic Session (STS) - Volume 2

STS507 Katherine Jenny T. et al.

Challenges in implementing a new imputation
method into production in the 2017 Economic
Census or what to do when the research
approach oversimplifies the problem
Katherine Jenny Thompson, William C. Davie Jr.
Economic Statistical Methods Division, U.S. Census Bureau.

Disclaimer: Any views expressed are those of the authors and not necessarily
those of the U.S. Census Bureau. The Census Bureau has reviewed this data
product for unauthorized disclosure of confidential information and has
approved the disclosure avoidance practices applied. (Approval ID: CBDRB-
FY19-ESMD-B00002)

Abstract
Beginning in 2017, the Economic Census (EC) will use NAPCS to produce
economy-wide tabulations of product sales. This change facilitated the
development of a statistically defensible and operationally feasible imputation
approach for these items, along with associated methodology for estimating
variances. Research teams were established to recommend methods for both.
To avoid confounding treatment effects with respondent size effects, both
studies restricted the analysis variables to a limited set of products within
industry. This greatly simplified the evaluations, but left potential
implementation challenges uninvestigated. This paper focuses on the
implementation of the recommended imputation methods in the 2017 EC,
discussing the modifications and enhancements needed to accommodate the
complete set of variables along with the implementation issues that motivated
each. We conclude by sharing our “lessons learned,” along with a laundry list
of topics for further investigation before the 2022 EC.

Keywords
Hot deck imputation; multiple imputation; production implementation

1. Introduction
Improvements to official statistics programs may require complicated
changes to existing methods or procedures to address new – emerging –
requirements or to accommodate new requests. Such changes are always
constrained. Budget constraints could force an overall reduction in sample
size; methodologists would need to revise the sampling design while
maintaining predetermined reliability levels on key statistics. Project sponsors
might commission the collection of additional data items, request preliminary
tabulations (and publication) of survey estimates, or desire subdomain

386 | I S I W S C 2 0 1 9