|
Abstract:
|
The National Agricultural Statistics Service (NASS), in conjunction with the Economic Research Service (ERS), conducts the three-phase Agricultural Resource Management Survey (ARMS) to study the well-being of farm households. Due to item nonresponse, some of the ARMS data are missing. Prior to 2015, a complete data set for use by NASS was formed by a mixture of machine imputation (conditional mean) and manual imputation. Since 2015, Iterative Sequential Regression (ISR), a multivariate imputation methodology, has been used for ARMS' third phase (ARMS 3). ISR is an in-house developed software program that requires a significant amount of support to maintain. Also, ISR has been developed for use on continuous and semi-continuous data, and NASS would also like to impute other data types including categorical and ordinal data. Hence, NASS is exploring alternative "off-the-shelf" imputation approaches, specifically, IVEware, a product of the University of Michigan, and SAS® PROC MI. Both imputation procedures are compared empirically for use in the ARMS 3 survey with attention not only given to data quality but also to ease of implementation and maintainability.
|
