Who Makes Mistakes? Using Data Mining Techniques to Analyze Reporting Errors in Total Acres Operated

Classification trees were used to identify subgroups of respondents with higher error rates when reporting total acres operated on the 2002 Census of Agriculture. Separate trees were grown for operations exhibiting total acres summation errors, missing data, and nonequivalent sums of reported total acres. Terminal tree nodes demonstrating the greatest frequency of total acres operated errors identify characteristics of respondents and or operations that are more likely to make errors, suggest possible reasons for errors, identify content for future tests of alternative questionnaires and suggest ways to appropriately edit these items. Advantages of using classification trees over other analytic methods are also discussed.