Talk:Barnard's test

Untitled

This article never bothers to say what Barnard's test is. Michael Hardy (talk) 18:19, 23 April 2010 (UTC)

Clarification?

What does it mean in Purpose and scope: ″ because it considers more 'as or more extreme' tables by not conditioning on both margins.″? Should one delete the first 'more'? Xenonice (talk) 06:28, 29 December 2013 (UTC)

Assessment comment

The comment(s) below were originally left at Talk:Barnard's test/Comments, and are posted here for posterity. Following several discussions in past years, these subpages are now deprecated. The comments may be irrelevant or outdated; if so, please feel free to remove this section.

I'm very short on time right now and don't yet understand how to add external links in references... But I thought I should mention that I've found one of the papers mentioned in the references, which does not now have an external link: "Conditional versus Unconditional Exact Tests for Comparing Two Binomials" is at http://www.cytel.com/Papers/twobinomials.pdf. If someone who knows how wants to add it, great.

Last edited at 17:02, 4 July 2009 (UTC). Substituted at 01:47, 5 May 2016 (UTC)

"Criticisms" should actually contain some

The section on criticisms should contain criticisms of *Barnard's* test. It doesn't seem to contain any -- it seems it's just criticisms of Fisher's test (which belong elsewhere). Where's Barnard's own criticisms of unconditional tests in comparison with conditional tests for example? Indeed most of the article is tendentious rather than impartial/neutral. The claim that Barnard's test is preferred by "most researchers" is likely to be false. I've marked it as citation needed -- if none if forthcoming, that claim should be removed. Glenbarnett (talk) 09:23, 22 August 2017 (UTC)

Specific test or type of tests?

The article suggests at various points that Barnard's test is a type of tests, including all unconditional exact tests of contingency tables obtained by maximizing the p-value over the nuisance parameter. That seems to be motivated by Barnard's first publication (1945) where he shortly suggests this procedure. However, in Barnard's more detailed second paper (1947) he proposes a specific test with the C.S.M. procedure. In my opinion it is more adequate to call this test "Barnard's test" as it in fact is a test and not just a very general type of tests.

Please share your thoughts about this. If there is agreement, the article should be changed accordingly.

Muelsak (talk) 07:11, 7 May 2020 (UTC)

The issue is that Barnard (1945) suggested a general method for creating tests, which is now known as unconditional exact tests (applied to two independent binomials). Barnard (1947) later suggested a member of this class, the CSM test, but that is rarely used. I added the alternative name to the beginning with 3 references (one is mine, Fay and Hunsberger, 2021, which discusses the naming problem. I think that type of self referencing is acceptable under the Wikipedia guidelines, correct me if I am wrong), but did not explain about the CSM test, because that seemed like too much detail for the article.

Mpf3205 (talk) 03:05, 3 May 2021 (UTC)

Only 2x2 contingency tables

The article suggests that Barnard's test can be used for general contingency tables and is compared with Fisher's exact test in the specific case of 2x2 tables. However, both cited publications of Barnard only handle 2x2 tables and I don't think there is a generalisation of Barnard's C.S.M. test to larger contingency tables.

If noone has a reference for Barnard's test for larger tables, this should be changed.

Muelsak (talk) 07:17, 7 May 2020 (UTC)

I agree and have changed the article. Barnard's test in fact requires that the contingency table be a 2x2 one, and the one of the margins is fixed.

Mpf3205 (talk) 03:26, 11 April 2021 (UTC)

Too vague in some critical ways

The article presents many interesting and perhaps important facts about the test.

But (a) it doesn't present the test itself -- e.g., the expression for the PDF or a statement about WHAT the alpha level should be compared to; (b) it doesn't present even a single worked example, which could give readers insight and might even enable them to infer the PDF; (c) it doesn't present any case for using statistics (beyond the 4 counts in the contingency table, of course), yet the article ends with "It remains unclear which [sic] test statistic is preferred when implementing" the test, as though there are choices.

Is there really something superior (in any way) to using the 4 cell counts?

Can somebody please add something of content about the mechanics, use, or computations of the test itself? Jmacwiki (talk) 23:48, 29 August 2023 (UTC)

Article is argumentative, one sided, and tendentious without offering proper support for its positions

The article calls the very widespread view and long literature on objections to this form of analysis *specious* and makes the huge claim that Barnard was *forced* to recant (again, accompanied by another claim of that this, too was *specious*). The article doesn't appear to be taking anything remotely like a consensus or balanced view and there doesn't seem to be reputable sources for these statements offered (or at least no references in the places in the text where they occur). Having read Barnard's article where he did indeed recant in some detail on several occasions, the detail of his explanation of what was wrong and the style of his phrasing doesn't remotely seem to suggest any form of coercion or reluctance on Barnard's part. This phrasing and the insertion of seemingly unsupported allegations makes the article seem acutely tendentious rather than offering any kind balanced account of the long disagreement between two camps and Barnard's eventual conclusion - clearly stated - that he had been mistaken.

Unless there's very clear evidence in the statistics literature that the arguments Barnard himself offered against his own previous position are in fact *specious*, or that he was in fact coerced in some manner, these aspects must be removed. In any case the widespread presence of counterarguments to the unconditional camp would still need to be properly recognized; this sort of argumentative and one-sided account has no place in an encyclopedia. If people want to engage in these sorts of positions without offering clear evidence, they can do it on Twitter. This sort of polemic should not appear in an article on Wikipedia.

Glenbarnett (talk) 01:45, 13 March 2024 (UTC)

I see that I had previously complained about the article being tendentious approximately six and a half years ago (in 2017). If anything, this problem with the article seems to have grown worse in the meantime, rather than having been improved. Clearly the proponent or proponents of this one-sided view are not going to change to a more suitably encyclopedic tone.

I have no interest in engaging in an edit war with people who are prepared to treat Wikipedia this way. It's had every chance to become a reasonable account instead of a rant and has seemingly failed at every step (see the host of critiques over many years above). If the people engaged in this sort of behavior won't tone it back to a reasonable account of the disagreement, the article shouldn't be here at all. Unless the article is edited quite soon to approach some kind of reasonable and properly supported position and written in measured language, I suggest it should simply be removed altogether; better no article than one like this.

Glenbarnett (talk) — Preceding undated comment added 01:54, 13 March 2024 (UTC)

Further to my comment questioning the apparently unsourced (and pretty outrageous) claim that Fisher had somehow forced Barnard to recant in 1949: Barnard's own account in 1988 (26 years after Fisher's death) of his 1949 recanting (De Groot, 1988, full ref below) describes what led him to change his mind and quite clearly (p202, middle of second column, para. starting "It needs to be discussed...") indicates that Fisher did not know that Barnard had changed his mind when he wrote the paper recanting his position, and only learned of his change of heart later when Barnard sent him a copy of that paper. All other unsourced claims by the particular person who placed that libel of Fisher in the article should be treated as highly suspicious; they clearly have no qualms about misrepresentation (to put it generously), nor about trashing people's reputation in pursuit of their agenda. Please revert/remove every such claim in the article. A review of their other edits elsewhere might well be in order.

Morris H. DeGroot, "A Conversation with George A. Barnard", Statist. Sci. 3(2): 196-212 (May, 1988). DOI: 10.1214/ss/1177012905

https://projecteuclid.org/journals/statistical-science/volume-3/issue-2/A-Conversation-with-George-A-Barnard/10.1214/ss/1177012905.full

(pdf available at that link)

Glenbarnett (talk) 03:16, 16 March 2024 (UTC)

Criticism of Fisher's exact test sounds very opiniated and is not in line with the content of article on Fisher's exact test

It appears that this has already been pointed out several times, but this article makes relatively strong claims about the "falsehood" of Fisher's exact test, and fails to provide sufficient references to back up these claims. Moreover, the explanation of the shortcomings of Fisher's exact test and associated reasons to prefer Barnard's test are not in line with the discussion on the same topic in the article describing Fisher's exact test. Aufildelanuit (talk) 09:57, 29 September 2024 (UTC)