Methodology for Small Area Estimation with Special Reference to a One-Number Agricultural Census and Confidentiality: Results for Selected Major Crops and States

This report describes a flexible methodology for calculating spatial autoregressive model-based small geographic estimates, illustrating its utility with substitution computations for the suppression code (D) in order to allow more complete data tabulations to be released. Post-stratification figures reported here differ from those published in the 1997 Census of Agriculture, which was conducted by the National Agricultural Statistics Service (NASS). Estimated census totals differ somewhat, for several agricultural items of interest, from official estimates based upon the ongoing NASS survey program and administrative data. These discrepancies can be attributed to nonsampling errors, timing and process differences. One such source of nonsampling error is underenumeration, resulting from some small farms not being on the census mail list. Figures reported here have been adjusted for such errors. Once the best estimates of agricultural commodities have been established, another concern is preserving confidentiality while releasing as much of the data as possible. This issue arises when these data are post-stratified into counties, for small geographic area tabulations. The text describes applications of the estimation methodology to principal agricultural commodities for Michigan and Tennessee that have been adjusted in order to align the agricultural census results with reliable official totals. It also describes comparisons between the spatial autoregressive and other model-based estimates. One important finding is that currently anyone with a working knowledge of estimators like those discussed here already is capable of generating quite accurate synthetic values from public domain databases, and then substituting them for suppressed values.