Talk:Nonparametric statistics

Interwiki links

Here's the situation in various editions of WP:

topic	en	it	nl
non-parametric method	en:non-parametric statistics	it:statistica non parametrica	—
non-parametric test	—	it:test non parametrico	nl:parametervrije toets → nl:verdelingsvrije toets
distribution-free method	—	—	nl:verdelingsvrije methode → nl:verdelingsvrije toets
distribution-free test	—	—	nl:verdelingsvrije toets

In light of these different divisions of the topic, it makes most sense to me to use the following interwiki links:

• from en:non-parametric statistics to it:statistica non parametrica and nl:verdelingsvrije toets
• from it:test non parametrico to nl:verdelingsvrije toets (and en:non-parametric statistics?)
• from nl:verdelingsvrije toets to en:non-parametric statistics and it:test non parametrico

--MarkSweep 22:24, 23 Dec 2004 (UTC)

Redirect

I would like this page to redirect to Category:Non-parametric statistics as I did with Category:Parametric statistics. There really isn't much in this article than a definition and some examples, and it doesn't provide anything useful other than links to articles that are actually non-parametric statistics. It doesn't matter that this redirect is unusual. It's more important that it's useful; that is, to direct the reader to the appropriate, specific article. — Chris53516 ^(Talk) 14:42, 15 February 2007 (UTC)

Hi Chris, thanks for your message. Rather than moving the text here to the Category page, shouldn't the conclusion be that this article needs to be improved? I don't think an article should ever redirect to a category, because they serve different functions. For instance, the definition given here for non-parametric statistics is very useful for someone who wants a quick idea of what the term means; and the article is there to provide that information. Category pages rarely have more than a few lines text, that's just not their function. I agree that the article is basically just a list at the moment, but it might improve if people start adding more information. For instance, on the history or methodology of non-parametric statistics; or when to choose which test. That stuff should go here, and not on the category page. Junes 12:46, 18 February 2007 (UTC)

My point is, what more would you say about the topic? The topic itself is really just a category of knowledge, and other than providing a list of topics that fit under it (which is what a category page would do), I don't see what else you would want to write. I say, if you can come up with something with which to expand this article, then it might be worthwhile. Otherwise, moving it to the category page is best. — Chris53516 ^(Talk) 14:49, 19 February 2007 (UTC)

I disagree, I think nonparametric statistics may be a vague field with unclear boundaries, but I think it is nonetheless a discipline, approach or perspective in and of its own right. The fact that there are a myriad of textbooks with titles like "Non-parametric statistics" (e.g. by Wasserman, Higgins, Gibbons & Chakraborti to name a few) seems testimony enough. If you want to get enough material to write a good page on this subject, discussing what identifies the nonparametric approach as a coherent approach, and what advantages it offers over parametric approaches, I suggest reading the prefaces and first couple chapters of each of these three books. I am actually working through these books right now and I may work on this page when I have the time...but I very strongly believe it should not be deleted or redirected. Cazort 22:56, 3 December 2007 (UTC)

Errors

There are many wrong statements on this page, however, I do not have time now to fix it. Therefore, I will mention only several crucial errors

1. Nonparametric tests are more powerfull than parametric when the assumptions of the parametric tests are not met. This is measured by, for example, Pitman asymptotic relative efficiency. If we compare rank sum test (or U test) with t test, the resuslts of ARE are as follows

normal distribution 0,955 (rank sum test is only 4.5% less efficient) logistic distribution 1,1 (rank sum test is 10% better) uniform distribution 1 Laplace distribution 2 exponential distribution 3 Cauchy ∞

The worst efficiency scenario for the rank sum test is 0,864; it means that U test will never have lower efficiency bound (relative to t test) than 86,4%. However, the upper bound for U test is infinity compare to the t test!

2. the robustness of the nonparametric tests is wrongly explained

.... —Preceding unsigned comment added by 212.200.159.132 (talk) 01:47, 21 November 2007 (UTC)

3. Rank methods are not always non-parametric. The Mann-Whitney-Wilcoxon U test is rather a "semi-nonparametric" because it is held under assumption that the two samples have the same scale parameter values. It is very important assumption. Two samples can strongly differ by scale and have the same medians. Then the Kolmogorov-Smirnov (which is a weaker test) will reject the null hypothesis and the Wilcoxon test - will not. — Preceding unsigned comment added by 83.5.8.69 (talk) 16:55, 14 January 2014 (UTC)

Spelling Error?

3. Why does this article (and other WP articles) hyphenate the prefix (`non-parametric` rather than `nonparametric`). Dictionaries that I've checked do not. Hobsonlane (talk)

Saturated model

I would like to include a link to a page that expands on the subject of non-parametric estimation with an example of the saturated model. Here you see the short article. Let me know if it would be ok.

"None"-parametric

"Note that the non-parametric model is not none-parametric."

What us this supposed to mean? Is it an "in" joke, or some kind of cutesy instructor mnemonic? I have no problem with including such features as long as they are explained from the ground up like everything else. 89.217.11.36 (talk) 09:36, 22 February 2015 (UTC)

Mann-Whitney nonparametric

Disclaimer: I'm not particularly expert. That said,

• I never heard presenting the Mann-Whitney U test as "semi-parametric",
• I'm not convinced by the above reasoning by 83.5.8.69 in point 3. The test will not reject the null of equality between two different but "similarly centered" distributions? Fine, but when it rejects, it gives me an interesting information that doesn't rely on any parametric assumption.
• the existence of a paper on a semi-parametric extension of such test... would suggest it was not orignially semi-parametric. --Toobaz (talk) 16:11, 12 June 2015 (UTC)

Link causes text to disappear

Hi,

In the part "Methods" the last link ("Wilcoxon signed-rank test"), does not bring you to the desired page, it just causes the text to disappear / re-appear. I tried fixing it myself, but I dit not manage.. I hope somebody can help!

Lizette — Preceding unsigned comment added by 81.220.54.32 (talk) 09:28, 30 July 2015 (UTC)

Opening of article

The opening of this article is fairly terrible. What is the "they" referred to in the 2nd sentence? Do non-para stats include certain types of inferential and descriptive stats? "Not based on parameterized families of probability distributions"...this can certainly be more easily explained to people who are coming to the article for a basic explanation. I won't rewrite this, but strongly encourage a simple re-writing of the opening sentences for clarity. Thereadletter (talk) 18:08, 8 December 2016 (UTC)