| Summary: | 碩士 === 輔仁大學 === 應用統計學研究所 === 99 === In meta-analysis a commonly used method which combines various significance tests on the basis of the observed p-values is the combined test procedure. The earliest proposed and most widely studied combination procedure is the Fisher’s combination procedure. Recently, a truncated procedure modifying Fisher’s method (truncated product method；TPM) was alternatively proposed. It was shown that the overall power of this truncated method will increase when a small subset of the test hypotheses is false. In this study, the truncated inverse normal method (TIN) and truncated inverse chi-square method (TIC) are investigated following the idea of TPM. Our simulation studies indicate that in some situations TIN and TIC have higher power performances than inverse normal method and inverse chi-square method. When the number of merged tests is limited, TIN outperforms the TPM. On the other hand, no matter what the test numbers by adjusting degrees of freedom of TIC the power of TIC can always get better power than TIN and TPM.
|
|---|
