Imperial College London

Professor Nick Heard

Faculty of Natural SciencesDepartment of Mathematics

Chair in Statistics



+44 (0)20 7594 1490n.heard Website




543Huxley BuildingSouth Kensington Campus






BibTex format

author = {Heard, N and Rubin-Delanchy, P},
doi = {biomet/asx076},
journal = {Biometrika},
pages = {239--246},
title = {Choosing between methods of combining p-values},
url = {},
volume = {105},
year = {2018}

RIS format (EndNote, RefMan)

AB - Combining p-values from independent statistical tests is a popular approachto meta-analysis, particularly when the data underlying the tests are either nolonger available or are difficult to combine. A diverse range of p-valuecombination methods appear in the literature, each with different statisticalproperties. Yet all too often the final choice used in a meta-analysis canappear arbitrary, as if all effort has been expended building the models thatgave rise to the p-values. Birnbaum (1954) showed that any reasonable p-valuecombiner must be optimal against some alternative hypothesis. Starting fromthis perspective and recasting each method of combining p-values as alikelihood ratio test, we present theoretical results for some of the standardcombiners which provide guidance about how a powerful combiner might be chosenin practice.
AU - Heard,N
AU - Rubin-Delanchy,P
DO - biomet/asx076
EP - 246
PY - 2018///
SN - 0006-3444
SP - 239
TI - Choosing between methods of combining p-values
T2 - Biometrika
UR -
UR -
UR -
VL - 105
ER -