245 – Working with Missing Data: Nonresponse, Imputation, and Suppressed Data
Capturing Additional Variability Introduced by Imputation Within the Agricultural Resource Management Survey
Darcy Miller
National Agricultural Statistics Service
Andrew Dau
National Agricultural Statistics Service
The Agricultural Resource Management Survey (ARMS) is conducted by the United States Department of Agriculture's (USDA) National Agricultural Statistics Service (NASS), and cosponsored by the USDA's Economic Research Service (ERS). The third phase of the survey, ARMS III, collects data to provide an annual snapshot of the financial health of the farm sector and farm household finances. Recently, the ARMS III imputation methodology was updated from using the mean of a stratum group to iterative sequential regression (ISR). With the previous imputation methodology, data that was imputed was treated as true values and thus underestimated the true variance of an estimate. With this new methodology, NASS can now evaluate the additional variability due to imputation. One method used to capture the additional variability is to impute values multiple times (producing multiple datasets) and use Rubin's formulas (1987) to calculate the variance. Research used Rubin's (1987) method to combine the datasets of ARMS III and analyzed variance estimates using pre-calibrated weights and calibrated weights to assess the increase in variability due to imputation.