|
Activity Number:
|
342
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 9:00 AM to 10:50 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #300223 |
|
Title:
|
Permutation Tests for Linear Models in Meta-Analysis: Robustness and Power Under Non-Normality and Variance Heterogeneity
|
|
Author(s):
|
Kristine Y. Hogarty*+ and Jeffrey D. Kromrey
|
|
Companies:
|
University of South Florida and University of South Florida
|
|
Address:
|
7132 Wrenwood Cir., Tampa, FL, 33617-8462,
|
|
Keywords:
|
meta-analysis ; robustness ; permutation tests
|
|
Abstract:
|
With the growing popularity of meta-analytic techniques to analyze and synthesize results across sets of empirical studies has come concern about the sensitivity of traditional tests in meta-analysis to violations of assumptions. This is particularly distressing because the tenability of such assumptions in primary studies is often impossible to evaluate. Permutation tests for linear models may provide a robust alternative. Monte Carlo methods were used to investigate Type I error control and statistical power of four permutation tests and a traditional WLS approach. Factors investigated included characteristics of both the populations from which samples were drawn (distribution shape, variance heterogeneity, effect of moderating variables, and correlation among moderating variables) and the corpus of studies in each meta-analysis (sample size and number of studies). The WLS procedure was the most powerful when the assumptions were met, but Type I error inflation was evident under violations. The Freedman and Lane and the Manly procedures provided the best Type I error control and power when assumptions were violated, while the Ter Braak procedure was overly conservative.
|
