|
Activity Number:
|
500
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #306252 |
|
Title:
|
Bias-Reduced Multivariate Imputation: Use of the Locally Adjusted Predictive Mean Matching Method
|
|
Author(s):
|
Masato Okamoto*+
|
|
Companies:
|
Statistical Research and Training Institute
|
|
Address:
|
19-1 Wakamatsu-cho, Shinjuku-ku, Tokyo, 162-8668, Japan
|
|
Keywords:
|
predictive mean matching ; nearest neighbor imputation ; multivariate imputation ; predictive mean neighborhoods ; fractional imputation ; regression imputation
|
|
Abstract:
|
Donor imputation by the Predictive Mean Matching (PMM) method tends to yield biased estimations in multivariate cases. To improve the PMM imputation, I propose adjustment of imputed value by offsetting difference of (re-estimated) predicted means between the donor and the donee. The re-estimation of predicted means is performed in the enlarged neighborhood of the donee. This locally adjusted Predictive Mean Matching (laPMM) method is not a complete donor imputation method anymore, being in between donor and regression imputation in a sense. Empirical results based on a simulation study show a significant reduction of bias, which can be fully utilized by the fractional method for reducing MSE.
|
