|
Activity Number:
|
75
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 6, 2006 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #306171 |
|
Title:
|
Multiple Imputation by Chained Equations: Predictive Mean Matching
|
|
Author(s):
|
Gerald Kolm*+ and Deborah Ehrenthal and Edward Ewen
|
|
Companies:
|
Emory University and Christiana Care Health System and Christiana Care Health System
|
|
Address:
|
7454 Woodruff Way, Stone Mountain, GA, 30087,
|
|
Keywords:
|
missing data ; multiple imputation ; predictive mean matching ; posterior distribution
|
|
Abstract:
|
Software for multiple imputation of missing data has become readily available. A number of packages have options for imputation methods and understanding the methods and their implications is crucial to obtaining reasonable imputed values. As an illustration, we compare differences in predictive mean matching with random draws from the predictive posterior distribution using multivariate imputation by chained equations (MICE) as implemented in the Stata routine, ice (Royston). The data base contains 3,594 records of current patients from an Adult Medicine Office, and includes demographic, clinical and laboratory variables. The amount of missing values range from 1% for blood pressure to over 35% for some lipid values. Results show that imputed values were not always consistent with observed values depending on imputation method and the distribution of the variables.
|
