| Summary: | 碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 100 === P-value combination procedures were proved to be very powerful in integrating results from multiple individual studies, especially when detail information is not available except p-values and sample size. Fisher method and inverse normal method are the two most commonly used methods for combining p-values. Many studies indicated that when the sensitivities of the individual studies are different, power of combination method can be improved by choosing weights for p-values appropriately. In this study, a new combination method “weighted Lancaster method” was proposed by incorporating weights into the statistic of Lancaster method, and an alternative approximation for the distribution of the combined statistic without the condition that combined p-values should be independent is derived. Through simulation studies, we show that the approximation proposed for the distribution of the statistic of weighted Lancaster method works quite well and power of weighted Lancaster method is superior to weighted inverse normal method and Lancaster method under most situations studied whether combined p-values are independent or not.
|
|---|
