Exploring the Performance of IVEware and Proc MI with Ordinal Categorical Data

• Activity Number: 300 - Innovations in and Applications of Imputation
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Government Statistics Section
• Author(s): Valbona Bejleri* and Andrew Dau and Darcy Miller
• Companies: USDA National Agricultural Statistics Service and National Agricultural Statistics Service and National Agricultural Statistics Service
• Keywords: ARMS III Survey; Assessing; Imputation; Missing; Ordinal data

Abstract

The USDA’s National Agricultural Statistics Service (NASS) uses surveys, the census of agriculture (conducted every five years) and data from other external sources to produce national and regional estimates. The presence of missing values due to item non-response is common among data collected through surveys. In an ongoing effort to mitigate issues with missing data, NASS has conducted several studies that have resulted on developing and/or adopting innovative approaches for re-weighting and imputation. In this paper, the performance of two multiple imputation methods (IVEware and Proc MI) is investigated and compared through a simulation study. Three missingness mechanisms are simulated using ordinal categorical responses from the Agricultural Resource Management Survey (ARMS III), a complex survey administered annually by NASS. Two simulated rates of item nonresponse at 30% and 50% are considered. The efficacies of the imputation methods are compared based on preserving the ordering of imputed values, the predictive accuracy of each method and the distribution accuracy. Also, the bias in several data characteristics that is attributable to imputation is investigated.