Page 399 - Special Topic Session (STS) - Volume 2
P. 399
STS507 Katherine Jenny T. et al.
data from a long, pre-specified list of potential products in a given industry –
some lists contain more than 50 potential products – and can write in
descriptions of other products that were not pre-specified. Product lists can
differ by industry within a sector. Furthermore, some product descriptions are
quite detailed, and some products are mutually exclusive. Consequently, some
establishments choose not to report any product data (complete product
nonresponse). Within an industry, reporting units tend to provide the same
small set of common products (Fink, Beck, and Willimack 2015).
The introduction of NAPCS marks a major departure from the prior
collections which explicitly linked product codes to industry, allowing for
different missing-data treatments for products by sector. Implementing a
NAPCS-based collection necessitated the development of a single imputation
approach for all EC products to allow production of cross-sector tabulations.
This paper describes the research process used to determine the product
imputation method and the process for implementing the research
recommendations into the 2017 EC production systems. Research and
implementation were accomplished by two different teams, with a small
fraction of membership overlap. Section 2 summarizes the research approach
and resultant recommendations. In Section 3, we discuss the implementation
of the recommended methods into a production system, specifically focusing
on some of the unaddressed or unforeseen – but important – details that were
excluded from the research study. We conclude in Section 4 with a few general
observations about implementing research-based results in a production
system.
2. Methodology
Thompson and Liu (2015) give an overview of the large scale research
project conducted to determine a single, unified imputation method for EC
broad products under NAPCS. The research was undertaken by a
commissioned team whose members included methodologists, subject matter
experts, and classification experts. The latter two groups developed the test
data used for all analyses and provided expertise on the 2012 EC procedures;
the 2017 collection procedures and NAPCS-based collection structure were
under development during the time of the research. The methodologists’
familiarity with the subject matter and expertise on the current procedures
ranged from completely novice (the majority) to extremely knowledgeable
about a selected subset of industry-specific procedures. Both team leads were
methodologists who were familiar with EC processing procedures and
methods in general but had little or no experience with the specific procedures
used in product processing.
The team divided the project into the three separate components listed in
Table 1, each lasting approximately two to three months. The project started
388 | I S I W S C 2 0 1 9