Multivariate Imputation Methods for Agricultural Resource Management Survey (ARMS) Data

• Activity Number: 489
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Sujit Kumar Ghosh*+ and Michael Robbins and Joshua D. Habiger and Darcy Miller
• Companies: North Carolina State University and National Institute of Statistical Sciences and National Agricultural Statistics Service/National Institute of Statistical Sciences and National Agricultural Statistics Service
• Keywords: agricultural data ; missing not at random ; multiple imputation ; semi-continuous data ; survey data

Abstract

ARMS is a detailed survey administered every year by the NASS and the ERS. The ARMS survey provides the only comprehensive examination of the economic well-being of farm businesses and households and thus is an essential component of agricultural policy making as well as the information used by agribusinesses. As is true with most large-scale surveys, a considerable amount of data is missing. The current procedure for imputing missing data values involves the use of conditional means calculated from complete surveys for similar farms, where similarity is measured by sales class, geographic location, and commodity specialization. This talk reports on research that investigates the application of novel multivariate multiple imputation methods. Using simulated data as well as applications to the ARMS data, performance of proposed imputation methods are evaluated relative to the current use.