In the 2007 Census of Agriculture, adjustments for 
undercoverage and nonresponse were estimated 
independently. In 2007, as in earlier censuses, the 
NASS area frame was used to adjust for 
undercoverage. This process assumed that the area 
frame provided complete coverage and that all 
operations were correctly classified as farm/nonfarm. 
To determine the extent of undercoverage in 2007, 
the CML records were matched to the area-frame 
tracts designated as agricultural, non- agricultural 
with potential, or non-agricultural with potential 
unknown in June. The area- frame tracts that did not 
match a CML record were designated as being in the 
Not on the Mail List (NML) domain. In 2007, tracts 
that were determined to be non-agricultural without 
potential during the pre-screening phase of the June 
Agricultural Survey (JAS) were not considered in the 
NML domain construction. The NML domain tracts 
were sent a census form and, if a tract was associated 
with a farm, then that farm contributed to the 
correction for undercoverage. 
To adjust for nonresponse in 2007, each responding 
CML record was given a probability of being a farm 
using a classification tree. The inverse of this 
probability became the nonresponse weight for that 
record. For undercoverage, the adjustment provided 
State-level values. A State-level estimate was based 
on the weighted sum of the responders with an 
adjustment for the non-responders within that State 
plus the State-level undercoverage adjustment. 
Because State-level farm count estimates based on 
this two-step process sometimes had high standard 
errors and apparent biases, the national-level 
adjusted estimates were smoothed across States, 
producing initial State-level farm operation coverage 
targets. 
Research following the 2007 Census of Agriculture 
led to the realization that some area-frame operations 
were misclassified as farm/nonfarm, which was in 
conflict with the previous assumption that the JAS 
farm classification was the accurate classification. 
Further, because nonresponse could only occur if the 
operation was on the CML, undercoverage and 
nonresponse were dependent. Thus in 2012, NASS 
used capture-recapture methodology to adjust for 
undercoverage, nonresponse, and misclassification. 
To implement capture-recapture methods, two 
independent surveys were required. The 2012 
Census of Agriculture (based on the CML) and the 
A -10 APPENDIX A 
2012 JAS (based on the area frame) were those two 
surveys. Historically, NASS has been careful to 
maintain the independence of these two surveys. 
A second assumption was that the proportion of JAS 
farms with a given set of characteristics captured by 
the census was equal to the proportion of U.S. farms 
with those same characteristics captured by the 
census. 
For a farm to be identified as a farm, and thus 
captured by the census, it must be on the CML, 
respond to the census report form and, based on the 
census response, be classified as a farm; that is, the 
capture probability ttc is of interest: 
n c = 7i(CML, Responded, Farm on Census|Farm) 
Two types of classification error can occur. First, a 
farm can be misclassified as a nonfarm. This type of 
misclassification is accounted for in determining the 
probability of capture kq. The second type of 
classification error results when a response to the 
census is classified as a farm operation when it does 
not meet the definition of a farm. That is, some 
farms on the CML may be misclassified from their 
census report response and may be nonfarms. To 
account for the misclassification of nonfarms as 
farms, the probability of a farm on the census being 
classified correctly must be estimated; that is, 
n CCFC ■= 7i(Farm | Farm on Census) 
where CCFC represents Correct Census Farm 
Classification. To adjust for undercoverage, 
nonresponse, and misclassification, each CML 
record classified as a farm based on its response to 
the census report form was given a weight of the 
ratio of the estimated probability of correct 
classification of a farm on the census and the 
estimated probability of capture ( n CCFC I n c where 
the hat symbol ( A ) denotes an estimate). To estimate 
the number of farms with a given set of 
characteristics, the weights of CML records 
responding as farms on the census and having that 
set of characteristics were summed. This estimator is 
2012 Census of Agriculture 
USDA, National Agricultural Statistics Service 
