Survey research commonly encounters the problem of analyzing data that contains incomplete or missing information. Acock (2005) notes that this missing information can produce biased estimates, distort statistical power, and allow researchers to draw invalid conclusions. Fortunately, these risks can be effectively curbed in many situations with imputation. The imputation theory, developed by Rubin in 1987, has been proven to successfully estimate a parameter of interest, as well as accurately assess the variability of the estimate. The purpose of this paper is to test the use and applicability of the multiple imputation method PROC MI in SAS as a valid and robust missing data technique as compared to a multiple hot deck process. The Department of Defense Manpower Data Center (DMDC) applied these two imputation methods to the Federal Voting Assistance Program (FVAP) survey of Local Election Officials (LEOs) and analyzed their efficacy. The failure of the LEO survey data to meet the assumptions of PROC MI makes that method intractable. While the multiple hot deck method provides more plausible results, it struggles to incorporate complex logical relationships between questions.
[1]
D. Rubin,et al.
Multiple Imputation for Nonresponse in Surveys
,
1989
.
[2]
D. Rubin.
Multiple Imputation After 18+ Years
,
1996
.
[3]
A. Acock.
Working With Missing Values
,
2005
.
[4]
Nicholas J. Horton,et al.
Multiple Imputation in Practice
,
2001
.
[5]
Bruce Ellis.
A Consolidated Macro for Iterative Hot Deck Imputation
,
2007
.
[6]
Joseph L Schafer,et al.
Analysis of Incomplete Multivariate Data
,
1997
.
[7]
D. Rubin,et al.
Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper
,
1977
.