returns that included multiple reports, respondent 
remarks, or that were marked out of business and 
report forms with no reported data were sent to an 
analyst for verification and appropriate action. 
Integrity checks were performed by the imaging 
system and data transfer functions. Standard quality 
control procedures were in place that required that 
randomly selected batches of data keyed from image 
be re-entered by a different operator to verify the 
work and evaluate key entry operators. All systems 
and programs were thoroughly tested before going 
on-line and were monitored throughout the 
processing period. 
Developing accurate processing methods is 
complicated by the complex structure of agriculture. 
Among the complexities are the many places to be 
included, the variety of arrangements under which 
farms are operated, the continuing changes in the 
relationship of operators to the farm operated, the 
expiration of leases and the initiation or renewal of 
leases, the problem of obtaining a complete list of 
agriculture operations, the difficulty of contacting 
and identifying some types of contractor/contractee 
relationships, the operator’s absence from the farm 
during the data collection period, and the operator’s 
opinion that part or all of the operation does not 
qualify and should not be included in the census. 
During data collection and processing of the census, 
all operations underwent a number of quality control 
checks to ensure results were as accurate as possible. 
Item Nonresponse 
All item nonresponse actions provide another 
opportunity to introduce measurement errors. 
Regardless of whether it was previously reported 
data, administrative data, the nearest neighbor 
algorithm, or manually imputed by an analyst, some 
risk exists that the imputed value does not equal the 
actual value. Previously reported and administrative 
data were used only when they related to the census 
reference period. A new nearest neighbor was 
randomly selected for each incident to eliminate the 
chance of a consistent bias. 
Record Matching Error 
The process of building and expanding the CML 
involves finding new list sources and checking for 
A -18 APPENDIX A 
names not on the list. An automated processing 
system compared each new name to the existing 
CML names and “linked” like records for the 
purpose of preventing duplication. New names with 
strong links to a CML name were discarded and 
those with no links were added as potential farms. 
Names with weak links, possible matches, were 
reviewed by staff to determine whether the new 
name should be added. Despite this thorough 
review, some new names may have been erroneously 
added or deleted. Additions could contribute to 
duplication (overcoverage) whereas deletions could 
contribute to undercoverage. As a result, some 
names received more than one report form, and some 
farm operators did not receive a report form. 
Respondents were instructed to complete one form 
and return all forms so the duplication could be 
removed. 
Another chance for error came when comparing June 
Agricultural Survey tract operator names to the 
CML. Area operators whose names were not found 
on the CML were part of the measure of list 
incompleteness, or NML. Mistakes in determining 
overlap status resulted in overcounts (including a 
tract whose operator was on the CML) or 
undercounts (excluding a tract whose operator was 
not on the CML). All tracts determined to not be on 
the list were triple checked to eliminate, or at least 
minimize, any error. NML tract operators were 
mailed a report form printed in a different color. In 
order to attempt to identify duplication, all 
respondents who received multiple report forms 
were instructed to complete the CML version and 
return all forms so duplication could be removed. 
Records in the 2012 JAS were matched to the 2012 
census using probabilistic record linkage. The 
records of operations with unresolved farm status 
were reviewed by the field offices. If farm status 
could not be resolved, the probability of an operation 
being a farm was imputed using a missing data 
model. The uncertainty associated with this estimate, 
with the exception of model uncertainty, was 
accounted for, but errors not found through this 
process were not. 
Model Uncertainty Error 
Five logistic models were developed in the process 
of adjusting the farm numbers for undercoverage, 
2012 Census of Agriculture 
USDA, National Agricultural Statistics Service 
