|
Activity Number:
|
271
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #301566 |
|
Title:
|
Random-coefficients Model with Missing Data
|
|
Author(s):
|
Yongyun Shin*+ and Stephen W. Raudenbush
|
|
Companies:
|
University of Michigan and University of Michigan
|
|
Address:
|
Dept. of Statistics, Ann Arbor, MI, 48109,
|
|
Keywords:
|
random coefficient ; maximum likelihood ; missing data ; hierarchical data ; EM algorithm ; Fisher scoring
|
|
Abstract:
|
A random-coefficients model is estimated with two-level hierarchical data when level-1 slopes vary randomly across level-2 units. We suggest the estimation method for a random-coefficients model when both individual and cluster level data are subject to missingness with a general missing pattern. Our method considers two-level data where the observations within each cluster are dependent. We maximize the observed data likelihood via a mixture of EM algorithm and Fisher scoring to obtain the maximum likelihood estimates (MLE) of the parameters using all available data. The key assumptions are that the data at the within-cluster or cluster level, or both, are missing at random; that parameter spaces for the complete data model and missing data mechanism are distinct; and that the data subject to missingness are multivariate normal conditional on all observed data. In order to translate back to the random-coefficients model of interest, we suggest two methods: (1) direct estimation of the parameters from the joint MLE by a subset regression on a disjoint subset of the complete data; and (2) multiple model-based imputation based on the joint MLE.
|
