|
Activity Number:
|
543
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #306768 |
|
Title:
|
The Median Test for Cluster-Correlated Data
|
|
Author(s):
|
Rick L. Williams*+ and Angela Pitts
|
|
Companies:
|
RTI International and RTI International
|
|
Address:
|
2816 Stone Gap Court, Raleigh, NC, 27162,
|
|
Keywords:
|
simulation ; Taylor series ; jackknife ; power
|
|
Abstract:
|
A simple method for conducting a large sample median test is demonstrated when the data are observed in clusters such that observations within clusters are correlated and observations between clusters are uncorrelated. In such situations, the standard assumptions underlying the median test are violated. The median test can be recast as a test of proportions which can then be analyzed using methods for proportions estimated from cluster correlated data. A simulation study demonstrates the methods ability to properly maintain the Type I error rate, measures its power for various alternatives and shows that ignoring the correlation of the data greatly misrepresents the Type I error rate. When comparing Taylor series linearization and jackknife variance estimators, it is found that tests based upon the jackknife estimator perform better than those based upon the Taylor series estimator.
|
