The use of p-values in combining the results of independent studies often involves studies that are potentially aberrant. This suggests the use of a statistic which trims some of the smallest and largest p-values. The authors provide two convenient approximations to the exact distribution of the trimmed Fisher statistic. The first utilizes Pearson's system of curve classification, while the second is a normal approximation.
Originally published in: Statistics & Probability Letters, v. 15, 1992, pp. 215-222.
This report is part of the RAND Corporation Reprint series. The Reprint was a product of the RAND Corporation from 1992 to 2011 that represented previously published journal articles, book chapters, and reports with the permission of the publisher. RAND reprints were formally reviewed in accordance with the publisher's editorial policy and compliant with RAND's rigorous quality assurance standards for quality and objectivity. For select current RAND journal articles, see External Publications.
The RAND Corporation is a nonprofit institution that helps improve policy and decisionmaking through research and analysis. RAND's publications do not necessarily reflect the opinions of its research clients and sponsors.