|
Activity Number:
|
512
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307371 |
|
Title:
|
Performance of Pseudo-Rsquare Statistics in the Linear Mixed Model
|
|
Author(s):
|
Jean Orelien*+ and Lloyd Edwards
|
|
Companies:
|
SciMetrika LLC and The University of North Carolina at Chapel Hill
|
|
Address:
|
100 Capitola, Durham, NC, 27713,
|
|
Keywords:
|
linear mixed model ; Rsquare ; goodness-of-fit ; simulation ; longitudinal data ; repeated measures
|
|
Abstract:
|
In the linear mixed model (LMM), several pseudo-Rsquare statistics have been proposed. Vonesh et al. (1996) suggested the concordance correlation coefficient be used to measure the percent agreement between the observed and predicted values. Vonesh and Chinchilli (1997) offered a formula for computing the proportion of reduction in residual variation. Zheng (2000) introduced a statistic to measure the proportional reduction in penalized quasi-likelihood. Xu (2003) proposed three statistics to measure the proportion of explained variation. However, the performance of these different Rsquare statistics has not been sufficiently demonstrated. We will present results that show that they do not perform adequately. Explanations for the shortcomings are given. We propose a new Rsquare for the LMM that performed well in simulations and has intuitive interpretation.
|
