EFD that ERS defined as manufacturers--about a third of the establishments inspected by FSIS1.
Excluded establishments included retailers, wholesalers, and other nonmanufacturers. About 60
percent of the population of plants selected by ERS responded to the survey. These included 131
ground beef, 73 hog carcass, and 72 broiler plants that underwent Salmonella spp. testing in 2000
and 73 cattle carcass plants. Theses plants amounted to about 44 percent of the cattle and hog
carcass and broiler plants and about 20 percent of the ground beef plants. The small number of
ground beef plants is due to the wide diversity of establishments that grind meat. For example,
many grocery stores and wholesalers grind meat as a side business. Additional plants were lost
when matching the ERS/EFD data with the LRD.
The ERS survey was not nationally representative, meaning that results cannot be
generalized. Two factors, however, suggest that the bias due to the use of a nonrepresentative
sample is small.2 First, the share of total output by respondents closely tracks the number of
plants that participated in the survey, and a regression analysis by the authors suggests that no
correlation exists between plant size and survey response. Second, the data were treated with a
post-stratification adjustment (Gelman and Carlin, 2000) in which the regression is adjusted with
a response weight equal to the reciprocal of the share of plants responding to the survey within
each of eight size strata for each industry.
Estimation and Model Selection
1 The EFD identifies the primary Standard Industrial Classification (SIC) of all establishments. An establishment
was assumed to be a manufacturer if it had a 2011, 2103, or 2015 SIC or slaughtered animals.
2 An anonymous reviewer asserts that a large degree of heterogeneity in the operations of establishments would
increase the bias.