[R] p.adjust not working correctly?

[David L Carlson]

Part of the confusion is that Bonferroni is often described as an adjustment to the significance level (sig-level) not the p-value. For example, to evaluate p-values when there are 8 tests, we compare the p-value to the sig-level/8 so the sig-level decreases. The p.adjust() function adjusts the p-value instead of the sig-level so we multiply the p-value by the number of tests (p-value * 8). As a result the p-values increase.

-------------------------------------
David L Carlson
Department of Anthropology
Texas A&M University
College Station, TX 77840-4352

-----Original Message-----
From: R-help [mailto:r-help-bounces at r-project.org] On Behalf Of S Ellison
Sent: Wednesday, August 3, 2016 6:05 AM
To: Alicia Ellis; r-help at r-project.org
Subject: Re: [R] p.adjust not working correctly?

> p.adjust for bonferroni p value correction does not appear to be working
> correctly:
You should re-check what a Bonferroni correction does, or at least reboot your intuition (mine needs rebooting all the time). All the p-values should _increase_ by a factor of n, with a ceiling of 1.0. Dividing by n would imply incorrectly that individual events have become less probable as the number increases.

The result you have obtained is what is supposed to happen.

S Ellison

> > p <- runif(50)
> >
> > p
>  [1] 0.08280254 0.08955706 0.19754389 0.52812033 0.68907772 0.21849696
> 0.02774784 0.23923562 0.03482480 0.76437481 0.87236155 0.76438604 [13]
> 0.37432032 0.89630318 0.01626565 0.08152060 0.55715478 0.47736921
> 0.77968275 0.17388127 0.37212900 0.18363170 0.51655538 0.14526733 [25]
> 0.60870820 0.13752392 0.92412799 0.73045115 0.89887453 0.33744577
> 0.84966571 0.97797283 0.20571554 0.29115022 0.75928867 0.12929511 [37]
> 0.64923057 0.68168196 0.19311014 0.83818106 0.85592243 0.56276287
> 0.80822911 0.53377044 0.44691466 0.59198417 0.36114259 0.35431768 [49]
> 0.10381694 0.57738429
> 
> > p.adjust(p, method = "bonferroni", n=50)
>  [1] 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> 1.0000000
> 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> [15] 0.8132824 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> 1.0000000
> 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> [29] 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> 1.0000000
> 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> [43] 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000 1.0000000
> 1.0000000
> 1.0000000
> 
> for the one corrected pvalue that is less than 1, it seems to have been divided
> by 0.02 and not 50.
> 
> I can do this one manually but would be nice if it worked:)
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.


*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:9}}