Page 403 - Special Topic Session (STS) - Volume 2
P. 403
STS507 Katherine Jenny T. et al.
whenever possible. However, businesses are more likely to report broad
product categories than detailed categories. Restricting donor records to
establishments that provided usable values for broad and detailed products
was too restrictive for many industries and would likely lead to inefficient
estimates. An inspection of 2012 EC counts of reported broad and detailed
products within the most restrictive imputation cell definitions (industry by
state by unit type) confirmed this suspicion. The majority of imputation cells
contained at least five establishments that reported usable broad products;
this was not the case with the detailed products.
Moreover, the research team had not studied imputation methods for
detailed products. The lack of available reported data – and the differences in
types of detailed products between 6-digit industries – was a prohibitive
barrier. Methodologists on the team recommended using a ratio imputation
model for each detailed product, known in-house as “category average”
imputation. Although this method did not prove optimal with the referenced
broad product research, the model is supported by the literature and almost
always produced unbiased broad product estimates in the studied industries
(Garcia, Morris, and Diamond 2015).
Table 2 presents a categorization of donor and recipient establishments
based on the presence of usable broad and detailed products. Using 2012 EC
data, we estimated the percentage of donors that would fall in each of these
categories as follows: complete (86.6%), partial (7.4%), and minimal (6.0%).
Obviously these percentages may be different in 2017, given the changes to
the questionnaire.
Table 2: Establishment Classification for Imputation
Donors Broad products usable
Complete All broad and detailed products usable (contribute to category
average)
Partial All broad products usable and some detailed products usable
(contribute to category average)
Minimal All broad products usable; detailed products missing and required
(receive detailed products from category average)
Recipients Missing products
Full Need broad and detailed products (receive all products from hot
deck)
Partial Need some (designated) detailed products (receive detailed
products from category average)
Ineligible All products usable, but not “typical”; excluded from donor pool
To simplify the operational procedure, the implementation team decided
to create “complete” donor records (i.e. fill in the missing detailed products
for partial and minimal donors) prior to hot deck imputation. To accomplish
392 | I S I W S C 2 0 1 9