Evaluation of Integer Weighting for the 1997 Census of Agriculture

Abstract

The census of agriculture is an important source of statistics about the Nation’s agricultural production and provides consistent, comparable data at the county, state, and national levels. Each census record has weights which are used to produce totals for the entire population. The process of rounding weights to integer values has been in place for the last several censuses. When a record’s weight is rounded to an integer value, the totals represented by that record may or may not change dramatically. These changes may or may not become negligible when producing totals at the state or county level. This report compares totals calculated with the noninteger weights to the published totals (calculated with the integer weights) for the 1997 Census of Agriculture, evaluates how different these totals are, and examines how the differences relate to the standard error. The analysis examines a number of characteristics at both the state and county levels. The report also examines another weighting approach where the noninteger weights are applied to the record and the weighted data values are rounded at the record level. The difference between totals produced with these values and the noninteger weights is calculated and compared to the above differences. The reasons for rounding weights, to ease data review procedures and to ensure that publication totals add, are legitimate concerns. The author asserts that it is possible to address these two concerns and improve the totals produced when NASS revamps the census processing system for the 2002 census.