threshold. Conversely, operations that report that 
they meet the threshold on the CHS might in fact not 
actually meet it and should be considered out of 
scope for the census. 
In order to measure the impact of misreporting scope 
status, NASS conducted a misclassification survey 
that consisted of a small sample of CHS respondents. 
A small set of screener questions was asked to 
determine the true scope for each of the operations 
selected for the misclassification survey. Using this 
methodology, misclassification adjustments were 
computed and used to adjust the nonresponse 
weights of the CHS respondents to account for 
reporting errors with respect to CHS scope status. 
Coverage Weighting Adjustments 
The target population for the 2014 CHS was all 
operations that had at least $10,000 of commercial 
horticultural production in 2014. Unfortunately, it is 
impossible to compose a list of operations that is 
complete. Due to this incompleteness of the mail list, 
data produced from it, even if perfectly corrected to 
account for nonresponse, will still have a tendency to 
be biased downwards because operations not on the 
list would not have any representation. This bias due 
to list incompleteness is called coverage bias, or 
more specifically, bias due to undercoverage of the 
sampling frame. 
To reduce the amount of this bias, an additional 
adjustment was calculated and applied to the 
nonresponse-adjusted weight for each responding 
operation. This was called the coverage adjustment. 
Coverage Adjustment Computation 
The majority of CHS respondents were also 
respondents on the 2012 Census of Agriculture. 
Operations that were respondents to both censuses 
were assigned the census of agriculture coverage 
adjustment computed for the operation in the 2012 
Census of Agriculture. The coverage adjustment for 
CHS respondents that did not match the census of 
agriculture were calculated using records with 
similar information that did match the census of 
agriculture. 
The coverage adjustment was then applied to the 
misclassification-adjusted nonresponse weight for 
2012 Census of Agriculture 
USDA, National Agricultural Statistics Service 
each CHS respondent record. This resulted in a fully- 
adjusted weight. The fully-adjusted weight attempts 
to correct for nonresponse and misclassification bias, 
as well as coverage bias. 
Summary Weights 
Most of the fully-adjusted weights for the 2014 
Census of Horticultural Specialties were not whole 
numbers (integers). Using these weights to create the 
estimates published in the tables would result in 
fractional values. These would be difficult to read 
and cause consistency problems between related 
tables. To avoid some of these problems, summary 
weights were created by randomly moving the fully- 
adjusted weights up or down to an integer in a way 
that preserved the overall sum of the fully adjusted 
weights. This process is called weight integerization. 
The resulting summary weights were used to 
produce the numbers published in the tables. 
MEASURES OF PRECISION AND 
ACCURACY OF THE ESTIMATES 
All numbers published in the tables are estimates of 
particular characteristics of the entire population of 
horticultural operations. The true values of these 
characteristics are unknown and unknowable. Even 
though an attempt was made to obtain a response 
from every operation selected for the survey and 
weight adjustments computed, the data produced by 
the census will not attain the true values. This is due 
to the fact that weight adjustments are imperfect and 
the assumptions on which those adjustments are 
made are imperfect as well. Hypothetically, if the 
entire census process was repeated over and over 
again, each replication of the census would almost 
certainly produce a different result for the same true 
population value every time. This is because each 
time the census is carried out, a different set of 
respondents would be obtained, response rates would 
fluctuate, and calculated weight adjustments would 
not be exactly the same. 
It is possible to obtain an idea of how much this 
variation would be on average by calculating the 
estimate’s variance. The estimated variance of an 
estimate gives a measure of the average squared 
random fluctuation that would be seen in an estimate 
if the census was carried out multiple times. Because 
the variance measures random fluctuation in squared 
Appendix A A - 5 
