[visionlist] Statistical Criticism is Easy; I Need to Remember That Real People are Involved

[JERMYN, IAN H.]

Hi Todd,

I hope it is OK for me to comment on this thread as a bit of an outsider.

> What I meant to say was that a p-value tells you how likely your data are given the null hypothesis, it doesn’t really say anything about the probability of the null hypothesis. So a SMALL p-value means that my data are unlikely given the null hypothesis, and a LARGE p-value means my data are likely given the null-hypothesis... but they could be even more compatible with some other hypothesis!

That is a nice summary. I would narrow it down even further: the p-value tells you the probability under the null hypothesis of getting the value of your chosen test statistic on the data, or any greater value (values which of course do not correspond to your data).

The data are always more compatible with some other hypothesis; the question is whether they are compatible with some other plausible hypothesis; but then perhaps we should weight these hypotheses according to their plausibility...starts to sound familiar...

Ian.




--------------

Ian H. Jermyn

E: i.h.jermyn at durham.ac.uk<mailto:i.h.jermyn at durham.ac.uk>

--------------

Department of Mathematical Sciences

Durham University

Science Laboratories

South Road

Durham DH1 3LE

United Kingdom


________________________________
From: visionlist <visionlist-bounces at visionscience.com> on behalf of Horowitz, Todd (NIH/NCI) [E] <todd.horowitz at nih.gov>
Sent: 16 November 2017 20:45
To: visionlist at visionscience.com
Subject: Re: [visionlist] Statistical Criticism is Easy; I Need to Remember That Real People are Involved


Oops!



What I meant to say was that a p-value tells you how likely your data are given the null hypothesis, it doesn’t really say anything about the probability of the null hypothesis. So a SMALL p-value means that my data are unlikely given the null hypothesis, and a LARGE p-value means my data are likely given the null-hypothesis... but they could be even more compatible with some other hypothesis!



thanks

Todd



From: Horowitz, Todd (NIH/NCI) [E] [mailto:todd.horowitz at nih.gov]
Sent: Thursday, November 16, 2017 2:00 PM
To: Pam Pallett <ppallett at gmail.com>; visionlist at visionscience.com
Subject: Re: [visionlist] Statistical Criticism is Easy; I Need to Remember That Real People are Involved



I thought that a a large p-value simply meant that my data were unlikely given the null-hypothesis, a statement which yields no evidence about either the null- or alternative hypotheses.



From: Pam Pallett [mailto:ppallett at gmail.com]
Sent: Thursday, November 16, 2017 10:29 AM
To: visionlist at visionscience.com<mailto:visionlist at visionscience.com>
Subject: [visionlist] Statistical Criticism is Easy; I Need to Remember That Real People are Involved



Hi All,



I came across a blog today by Frank Harrell, Professor of Biostatistics and Founding Chair at Vanderbilt.  His most recent post is the title of this email.  But as I'm reading through his blog, I'm hearing a lot that has been discussed and experienced by professors and postdocs subscribed to this list.  We are often very separated from our neighboring departments, and I actually found some comfort in the fact that these problems seem spread across the board (misery loves company). Even if we have been echoing these problems for over a decade with little effective change.



In his most recent post he says, "There are several ways to improve the system that I believe would foster clinical research and make peer review more objective and productive." I'm curious about what the people in the vision community think of these suggestions and whether they are realistic to implement in our field.  His list is at the bottom of the entry.  http://www.fharrell.com/2017/11/



For those experiencing TL;DR, here is the shortlist:

·         Have journals conduct reviews of background and methods without knowledge of results.

·         Abandon journals and use researcher-led online systems that invite open post-"publication" peer review and give researchers the opportunities to improve their "paper" in an ongoing fashion.

·         If not publishing the entire paper online, deposit the background and methods sections for open pre-journal submission review.

·         Abandon null hypothesis testing and p-values. Before that, always keep in mind that a large p-value means nothing more than "we don't yet have evidence against the null hypothesis", and emphasize confidence limits.

·         Embrace Bayesian methods that provide safer and more actionable evidence, including measures that quantify clinical significance. And if one is trying to amass evidence that the effects of two treatments are similar, compute the direct probability of similarity using a Bayesian model.

·         Improve statistical education of researchers, referees, and journal editors, and strengthen statistical review for journals.

·         Until everyone understands the most important statistical concepts, better educate researchers and peer reviewers on statistical problems to avoid<http://biostat.mc.vanderbilt.edu/ManuscriptChecklist>.



Best,

Pam Pallett
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://visionscience.com/pipermail/visionlist_visionscience.com/attachments/20171120/ee4275e1/attachment-0001.html>