Hochberg procedure under negative dependence

Jiangtao Gou, Ajit C. Tamhane

Research output: Contribution to journalArticlepeer-review

15 Scopus citations


The Hochberg (1988) procedure is commonly used in practice to test multiple hypotheses based on their p-values. It is a conservative step-up shortcut to the closed procedure (Marcus, Peritz and Gabriel (1976)) based on the Simes (1986) test. The Simes test is anti-conservative if the test statistics are negatively dependent in a certain sense. So practitioners are reluctant to use the Hochberg procedure under this condition and prefer to use the less powerful Holm (1979) procedure, which requires no dependence assumptions. But the Hochberg procedure is conservative by construction, so we may conjecture that it will remain so under certain types of negative dependence. In this paper we show that a slightly modified version of the Hochberg procedure controls the familywise type I error rate (FWER) if the p-values follow a multivariate uniform distribution which is a mixture of bivariate components each of which is negative quadrant dependent (NQD) (Lehmann (1966)) or positive dependent through stochastic ordering (PDS) (Block, Savits and Shaked (1985)). By negative dependence we will mean this distribution model, in particular, that its negatively dependent bivariate components are NQD. Simulations suggest that conservatism of the Hochberg procedure is likely to be true for more general negatively dependent distributions.

Original languageEnglish (US)
Pages (from-to)339-362
Number of pages24
JournalStatistica Sinica
Issue number1
StatePublished - Jan 2018


  • Familywise type I error rate
  • Multiple comparisons
  • Multivariate uniform distribution
  • Negative/positive dependence through stochastic ordering
  • Negative/positive quadrant dependence
  • Simes test

ASJC Scopus subject areas

  • Statistics and Probability
  • Statistics, Probability and Uncertainty


Dive into the research topics of 'Hochberg procedure under negative dependence'. Together they form a unique fingerprint.

Cite this