A systematic comparison of methods for combining p-values from independent tests

摘要: Six methods are studied for combining p-values from independent tests into a new test of the combined hypothesis. The methods—minimum (The Method Statistics, Williams and Norgate, London, 1931), chi-square (2)(Statistical Methods Research Workers, 4th Edition, Oliver Boyd, 1932), normal (Magyar Tudomanyos Akademia Matematikai Kutato Intezetenek Kozlemenyei 3 (1958) 1971), maximum (Wilkinson, Psycholog. Bull. 48 (1951) 156), uniform (J. Pyschol. 80 (1972) 351), logistic (in: Rustagi (Ed.), Symposium on Optimizing in Academic Press, New York, 1979, pp. 345–366)—are compared heuristically through simulation. Plots rejection regions two reveal much about tests’ relative strengths. simulations compare using different numbers tests, patterns evidence against null hypothesis, total strengths evidence, allowing broader recommendations than have been made past simulations. results indicate that most difficult kind problem is one which concentrated or very few being combined. For this case alone minimum function useful. does well problems where spread among more small fraction individual when weak. (2) best at least moderately strong relatively tests. combination provides compromise between these two. combinations generally poor power cannot be recommended use.