Page 402 - Special Topic Session (STS) - Volume 2
P. 402
STS507 Katherine Jenny T. et al.
event of no donors), or dollar value and additivity requirements for final
imputed data (no imputed values of less than $500 were allowed, but all
rounded values were required to add to the associated value of total receipts).
Imputation of detailed products was not addressed in the research study, nor
was calibration to industry totals.
3. Results
A new team was established for implementation. Leadership was provided
by project management experts with extensive familiarity with the subject
matter and with the planned EC processes. A large representation of (industry)
subject matter experts were included, as were production programmers. Four
methodologists from the research team were retained as consultants, with a
production methodologist recruited to develop the final specifications. [Note:
This methodologist directly supervised a research team member and was not
unfamiliar with the research project, even without working directly on the
team.] The team lead and one subject matter expert had participated on the
research team, but the remaining team members were recruited from other
EC projects and were not familiar with the earlier research or methods.
As with the research team, the project began slowly with an educational
component. Team members each had their own set of implementation issues
that needed to be addressed. The production programmers were concerned
about the computing resource demands of fully implementing hot deck
imputation; there were also staffing concerns as team members had to juggle
working on this project with other high priority projects. Fortunately, the
methodologists were able to provide concrete examples of efficient hot deck
systems, which partially alleviated these concerns.
Presentation of the hot deck methods was generally met with acceptance.
However, the subject matter experts balked at cell-collapsing procedures for
two reasons. First, they were not convinced that imputation cells with sufficient
observations should be combined with imputation cells with insufficient
observations. Instead, they insisted on using the original imputation cells in
the former case, reserving the collapsed imputation cells only for the latter
case. Second, they did not have resources to research alternative cell
collapsing procedures. Eventually, the implementation team compromised
under strong protest from several methodologists, agreeing to the
unconventional preferred cell-collapsing procedure but using a minimum cell
size of one establishment. Although the specifications called for
parameterized imputation cell definitions with three levels of collapsing, the
imputation cell definitions that were coded into the production program
varied little by industry.
The subject matter experts were emphatic on one point: they wanted to
maximize the use of unit-level reported data in the imputation procedures
391 | I S I W S C 2 0 1 9