Comparison of Two Imputation Methods in the Survey of Doctorate Recipients

• Activity Number: 400
• Type: Topic Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Jeffrey H. Stratton*+ and John M. Finamore and Todd R. Williams
• Companies: U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau
• Keywords: hot-deck imputation ; Flexible Matching Imputation

Abstract

Nonresponse is a major concern in most sample surveys. Missing data are often imputed to provide users with a complete dataset. Many different methods of imputation have been proposed, and the methods used vary widely among different surveys. One of the most common techniques used to impute missing data is hot-deck imputation. Even having chosen this technique, there are many different ways to implement it. This paper examines the results of two different hot-deck imputation techniques applied to the 2001 Survey of Doctorate Recipients (SDR). The SDR is a longitudinal biennial survey sponsored by the National Science Foundation to measure the demographic characteristics of individuals who received a PhD in a science or engineering field from a U.S. institution. (Doctoral-level professional degrees such as those awarded in medicine, law, or education are not included.) The 2001 SDR included doctorate degrees earned between January 1, 1942, and June 30, 2000. We present results comparing the current hot-deck imputation method used in the SDR with a Flexible Matching Imputation (FMI) method proposed by Todd Williams of the U.S. Census Bureau. Both methods use model-based methodology to select important matching variables, but the FMI method is able to consider more variables. We use both methods to impute missing data in the 2001 SDR, and discuss their performance.