|
Activity Number:
|
428
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #306513 |
|
Title:
|
The Relative Contribution Measures in Multilevel Modeling
|
|
Author(s):
|
Liyi Cen*+ and Zhen Chen and Daniel E. Polsky and Kevin G. Volpp
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
|
|
Address:
|
3504 Shetland Way, Westville, NJ, 08093,
|
|
Keywords:
|
variation ; fixed effect modeling ; multi-level modeling
|
|
Abstract:
|
The omega statistic was first introduced to compare the relative importance of two sets of explanatory variables in explaining the variation of the outcome. While there are many applications of omega statistic in fixed effect models, few exist in the literature in the context of multi-level modeling. When the data are nested, will the omega statistics be different between fixed effects models and multilevel models? How to calculate the omega statistics if the two sets of predictors are incomparable and correlated. In this study, we will address the above two questions by comparing the difference between standard logistic regression and multi-level modeling in predicting the variation in the response by patient level variables and hospital characteristics using simulations and Medicare file (MedPAR) in the context of examining mortality following hospital admission for a medical condition.
|
