Talk:Bonferroni correction

Discussion of Holm-Bonferroni is inappropriate

If nothing else, this paragraph requires citations. Holm-Bonferroni is one of many sequential Bonferroni-type approaches to controlling the familywise error rate in a set of tests. Among these many approaches, it is not even particularly powerful---Hochberg's approach is, for example, guaranteed to be no less powerful and is in many cases more powerful. The paragraph should be changed to reference two classes of more pwoerful tests: (1) sequential Bonferroni-type tests that control the familywise error rate, and (2) Benjamini-Hochberg type tests that control the false discovery rate. Specific reference to an individual test such as the Holm's is out of place, as it appears to provide preference to one of many alternatives. —Preceding unsigned comment added by 71.48.130.128 (talk) 13:21, 27 October 2009 (UTC)

Thank you for your suggestion. When you believe an article needs improvement, please feel free to make those changes. Wikipedia is a wiki, so anyone can edit almost any article by simply following the edit this page link at the top. The Wikipedia community encourages you to be bold in updating pages. Don't worry too much about making honest mistakes—they're likely to be found and corrected quickly. If you're not sure how editing works, check out how to edit a page, or use the sandbox to try out your editing skills. New contributors are always welcome. You don't even need to log in (although there are many reasons why you might want to).--mcld (talk) 19:27, 27 October 2009 (UTC)

Šidák correction should be moved to its own article

Though related, the Bonferroni and Šidák tests rely on different assumptions and lead to generally different results. It is confusing to discuss both on a page labeled "Bonferroni correction." —Preceding unsigned comment added by 71.48.130.128 (talk) 13:24, 27 October 2009 (UTC)

This article is incorrect

It says "one way of maintaining the familywise error rate is to test each individual hypothesis at a statistical significance level of 1/n times what it would be if only one hypothesis were tested", i.e., alpha_corrected = alpha/n. The formula given is *another correction*, which is "applicable for two-sided hypotheses, multivariate normal statistics, and positive orthant dependent statistics, it is not, in general, correct (Shaffer 1995)". See http://mathworld.wolfram.com/BonferroniCorrection.html —Preceding unsigned comment added by Leonardo.hamilton (talk • contribs) 10:58, 20 February 2009 (UTC)

Extra references

I have added extra references to this article so that people can conveniently find out more about this method. I have also added a direct link to Thomas Perenger's critique of the Bonferroni method, as it is cited in the three other references I have added. Michael Glass 13:23, 9 June 2007 (UTC) I'm not sure about the point of adding this article by Perenger. It is a seriously flawed article in that you most definitely are not just rejected the global null hypothesis, but in fact also the individual hyphothesis. All of Perenger's arguments along of the lines of "And type II errors are no less false than type I errors" are rather pointless, if one pretends to perform multiple tests at a fixed type I error rate. One a whole the author just does not understand the Bonferroni adjustments. A much more reasonable criticism would be that Bonferroni-Holm is uniformly more powerful, but for example it's much harder to obtain valid Confidence intervals for the Bonferroni-Holm procedure (there's a recent paper on how to do it by Bretz et al., but these confidence intervals are surprisingly not allows contained within the Bonferroni ones). Baselbonsai 00:01, 16 August 2008 (CET).

Clarification needed

There is a need to clarify how the outcome of the multiple tests is judged .... "at least one test says reject", or "all tests say reject". Melcombe (talk) 17:06, 29 April 2008 (UTC)

Additive

There was some notation implying that you specify the tests to be of size alpha/n, when they should be of size 1-(1-alpha)^(1/n)... That is, one minus the chance of not rejecting each test. Why would these things be written as 1-(1-alpha/n)? Brusegadi (talk) 08:20, 14 February 2009 (UTC)

-Not developed by Bonferroni

Bonferonni correction was not developed by Bonferroni - it was developed by Dunn, based on a proof of an inequality. However, the proof of the inequality was not done by Bonferroni either, he extended it. I'll edit this when I have the time. Bonferroni correction is based on Boole's inequality - this is discussed in the article Stigler's law of eponymy by Steven Stigler. —Preceding unsigned comment added by Jeremymiles (talk • contribs) 05:04, 18 March 2010 (UTC)

Missing citation

The statement "Second, in certain situations where one wants to retain, not reject, the null hypothesis, then Bonferroni correction is non-conservative." could do with a citation Worik (talk) 00:47, 29 March 2010 (UTC)

I'd say that 'conservative', in its statistical sense, means 'rejecting the null hypothesis when it is true less often than indicated by the nominal p-value or specified Type I error rate', but i don't have a citation for that to hand either (other than conservatism (disambiguation), which is hardly a WP:reliable source). I may try to find one... Qwfp (talk) 19:45, 29 March 2010 (UTC)

"Lou Jost" test?

The only source for the "Lou Jost" test is Lou Jost's own website - his suggestion does not seem to have been subject to any peer-review. I have removed the reference and added tags; if someone cannot dig up a good source, I will remove that section. Angio (talk) 18:43, 15 November 2010 (UTC)