|
Activity Number:
|
141
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #301307 |
|
Title:
|
Using Shrinkage in an Exact Test for Clustered Binary Data with Small Numbers of Clusters
|
|
Author(s):
|
Patrick Gerard*+ and William R. Schucany
|
|
Companies:
|
Clemson University and Southern Methodist University
|
|
Address:
|
Dept of Applied Economics and Statistics, Clemson, SC, 29634,
|
|
Keywords:
|
sign test ; bootstrap ; permutation
|
|
Abstract:
|
The sign test is often employed with independent binary data. When binary data are clustered, the same hypothesis may be of interest, but the standard sign test performs poorly due to the dependence caused by clustering or blocking. Gerard and Schucany (2007) investigated a new test incorporating both permutation and an exact parametric bootstrap (EPB) to be used when the numbers of clusters are small. In this study we modify the EPB test using shrinkage of cluster level success probabilities. We compare the new test to the EPB as well as a classic permutation test for numbers of clusters ranging from 5 to 10 inclusive. We find that for most combinations of alternatives and correlations the new shrinkage test is the most powerful.
|
