|
Abstract:
|
Researchers are increasingly interested in "analyzing the analyses" conducted by others. Known as meta-analysis, these types of studies are conducted to bolster existing or proposed theories, or to highlight important findings that would otherwise be "lost in the crowd" of accumulated research. Several meta-analytic tests based on p-values have been proposed in the literature, most notably the Liptak, Fisher, Sidak, and Simes p-value combination tests. An important assumption underlying these tests is that p-values have an i.i.d. uniform(0,1) distribution. In this paper I examine through Monte Carlo simulation the effects of two common dependence structures on the nominal Type I error rates of these tests: compound symmetry and autoregressive order 1 (AR (1)). I then provide relevant discussion and suggestions for future research.
|
