|
Activity Number:
|
285
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #303567 |
|
Title:
|
Testing the Equality of Conditional Correlations Across Numeric Variable(s) with Heteroscedastic Data
|
|
Author(s):
|
Xueya Cai*+ and Gregory E. Wilding and Alan Hutson and Yue Li
|
|
Companies:
|
Indiana University Purdue University Indianapolis and SUNY at Buffalo and SUNY at Buffalo and University of California, Irvine
|
|
Address:
|
Division of Biostatistics, Indianapolis, IN, 46202,
|
|
Keywords:
|
Correlation coefficient ; Equality test ; generalized linear model ; heteroscedasticity ; MOS SF-36
|
|
Abstract:
|
Practice for testing the equality of correlation coefficient across a numeric covariate is to group the data into k samples, and to compare the estimated correlation coefficient among the defined groups. Our group has previously presented a generalized linear model approach for directly estimating the correlation coefficient as a function of the numeric covariate(s). However, this approach was limited to the homoscedastic case. Here we extend the methodology to accommodate heteroscedastic data. The revised method not only maintains type I error control, but also shows improved power for statistical inference under heteroscedastic populations. A real data example is used for illustration.
|
