-4
$\begingroup$

My post is here: How can I test whether a single categorical variable significantly changes a linear regression model (0.05 level), without using the F test?

I have written a very clear, short, focused question that doesn't have any ambiguity, yet the post remains closed. I spent two days making minor changes, that I usually didn't agree caused ambiguity, to accommodate the feedback. All of the items have been addressed, there is no more feedback, yet the post remains closed.

There are two votes to re-open shown. I'm not sure if I need 3 or 5. I know I only need one vote from whuber, who closed it with one vote.

It seems the post is closed because some people (or someone) doesn't like my question, rather than my question being ambiguous.

The post was closed for ambiguity with a single vote from whuber, which I was not aware of until I recently reviewed the post's history. In the comments about the post being closed, whuber made a clear statement that "all you have needed to do to earn my vote is to state what hypothesis you wish to test." So I added the hypotheses. Then there was the complaint that "statistically significant" was in the hypotheses, which whuber agreed with. I can't even believe that. How does make my post hard to understand. The post remained closed, so it was closed for that??? So I removed "statistically significant," yet the post remains closed. I pointed out to whuber what he said, and he has not made any response since. He clearly hasn't followed up with his vote, as he can reopen it with one vote (as a moderator, just like he closed it with one vote).

Here is the hypotheses: (it originally read "no statistically significant" difference)

H0: There is no difference between the mean of Var1 (groups combined) and the means of Var1 in any of the three groups defined by the values of X.

HA: There is a difference between the mean of Var1 (groups combined) and at least one of the means of Var1 in the three groups defined by the values of X.

There's a reason for the wording, and if it's a problem (how?) I'll make it a standard ANOVA-type hypothesis, which says that the means of the groups are equal. I pointed that out in the comments, and this has never been replied to.

I need to know what is ambiguous, so I can get the post opened. The post is a short read. Or I just need vote(s) to reopen it.

$\endgroup$
7
  • 1
    $\begingroup$ Looking at the history it seems that it went into the reopen queue and received two votes to reopen but also two votes to remain closed which is why it remains closed. $\endgroup$
    – mdewey
    Aug 21, 2020 at 16:11
  • $\begingroup$ And what is the ambiguity, so I can resolve it? $\endgroup$
    – user280191
    Aug 21, 2020 at 17:00
  • $\begingroup$ I made some minor changes to the post, to address why someone doesn't understand that the null hypothesis says the overall mean of Var1 (groups combined) is equal to the mean of Var1 in each of the groups. I also added a statement that I'm not addressing whether the F-statistic from lm() addresses my hypothesis, as I'm trying to make the focus EXTREMELY narrow to remove the claim of ambiguity that is keeping my post closed. $\endgroup$
    – user280191
    Aug 21, 2020 at 21:33
  • 3
    $\begingroup$ It looks as if this is a dead issue, but while this thread remains open, one simple principle perhaps deserves emphasis. An attempt to clarify a question closed as unclear is not sufficient to reverse that closure. It has to appear clear to a moderator and/or those with enough reputation to vote to reopen. It is sad but not especially surprising that such asymmetry can seem obnoxious to those on the receiving end, as also do personal attacks or ridiculous exaggerations about the quality of the site. $\endgroup$
    – Nick Cox
    Aug 22, 2020 at 14:42
  • 1
    $\begingroup$ I’m voting to close this question because it has no enduring value. It refers to a question now deleted and so visible only to some and it concerns behaviour by people no longer on CV. $\endgroup$
    – Nick Cox
    Aug 24, 2020 at 8:26
  • $\begingroup$ @NickCox, it concerns the behavior of a single person (operating through many accounts) no longer on CV. $\endgroup$
    – gung - Reinstate Monica
    Aug 24, 2020 at 12:45
  • $\begingroup$ @gung Indeed. As a non-moderator that seemed implied to me, but I went for the slightly more benign interpretation. $\endgroup$
    – Nick Cox
    Aug 24, 2020 at 13:02

3 Answers 3

Reset to default
5
$\begingroup$

I don’t have unilateral close/reopen privileges, but if I did, I would take the stance that your behavior earned you a timeout.

Regarding the ambiguity, you still haven’t explained why you’re fine using Wald test p-values on individual parameters but oppose a global Wald test. Further, your examination of the individual parameters leaves me wondering if a global test really is what you seek, particularly given your interest in testing XA (what about it?). (EDIT) Finally, your first line asks if a categorical variable is significant overall. Does that mean that you want to test that some group has a nonzero mean? What if they all have the same nonzero mean?

I think you want an alternative to the usual ANOVA, of which Kruskal-Wallis, likelihood ratio, and permutation testing are three possibilities, but it’s not clear to me that those address the question that you want to ask.

$\endgroup$
1
  • $\begingroup$ No. Re: "Does that mean that you want to test that some group has a nonzero mean?" --- The null hypothesis is that the overall mean is equal to each of the individual group means. --- I really don't understand why that's so hard to understand. It's plain English. $\endgroup$
    – user280191
    Aug 21, 2020 at 21:13
4
$\begingroup$

It seems to me reading the comments to the original post that people got distracted by issues which were possibly not core to your question. Looking at the summary you give in your post here it seems to me that if I take the null and alternative hypotheses and the paragraph which immediately follows them it might be summarised as:

Is it meaningful to compare the overall mean with the mean of one of the sub-groups which comprise it and if it is, is that the same as testing for equality of the sub-group means?

If that is what you ask then perhaps it might be better to ask that as a new question.

There is also the issue of, if it is possible, how to program a computer to do it but as I am sure you know including that is liable to get people to vote to close it as off-topic so I would strongly advise resisting any temptation to include that.

Of course if I have mis-interpreted your question then that may help you to see how to re-word it to express what you do what to ask. Just for the record I had not seen the question before I saw this post on meta.

$\endgroup$
4
  • 2
    $\begingroup$ No, I am not asking if comparing the overall mean to the sub-group means is the same as comparing the sub-group means to each other. The null hypothesis says the overall mean is equal to each of the subgroup means. -- Can you suggest a revision of the null hypothesis so this is more clear? $\endgroup$
    – user280191
    Aug 21, 2020 at 17:04
  • $\begingroup$ I added a NOTE to my post, below the hypothesis, that I am not asking what you are saying above, and rather, I am testing whether the overall mean is equal to each of the sub-group means. Thanks for the suggested update (even though I don't understand how you read the null hypothesis the way you stated it). $\endgroup$
    – user280191
    Aug 21, 2020 at 17:23
  • 2
    $\begingroup$ @mdewey - How in the world do you read the null hypothesis that way? I'm talking about "Is it meaningful to compare the overall mean with the mean of one of the sub-groups which comprise it and if it is, is that the same as testing for equality of the sub-group means?" There is nothing in the question that is posted at stats.stackexchange that suggests that. $\endgroup$
    – user255758
    Aug 21, 2020 at 17:31
  • 1
    $\begingroup$ Thank you for noting "that people got distracted by issues which were possibly not core to your question." $\endgroup$
    – user280191
    Aug 21, 2020 at 17:48
-7
$\begingroup$

There is nothing ambiguous about your post. It doesn't need any revisions. You have maintained the same question from the original version, while addressing the claimed ambiguity that was brought up in the comments.

I don't have a 3,000 rep to vote to reopen your post.

$\endgroup$
6
  • 8
    $\begingroup$ This user's account was closed as a sockpuppet of the OP. $\endgroup$
    – whuber Mod
    Aug 21, 2020 at 21:15
  • 3
    $\begingroup$ @whuber, w/ all the accounts now deleted, it's interesting to see that all of the votes for this answer, all but 1 of the votes for the Q, & the downvotes for mdewey's A all came from the OP via their various socks. $\endgroup$
    – gung - Reinstate Monica
    Aug 23, 2020 at 11:51
  • 1
    $\begingroup$ FWIW, I don't think that it's good to delete this answer. The comments here are a reminder that sock-puppets are not permitted, and will be met with severe sanctions. $\endgroup$
    – Sycorax Mod
    Aug 26, 2020 at 16:13
  • 1
    $\begingroup$ @Sycorax The principle is good, but in practice how likely are people tempted to offend going to find this and be scared off by it? There are two kinds of antisocial behaviour on CV: new people behaving badly with outcome that they are removed quickly or leave the site in a huff on being reprimanded; and people convinced over a long period that they are right, or have a right to their views, and that moderation or peer pressure is undermining their freedom of expression. I don't think threads like this are a deterrent to either kind of behaviour. $\endgroup$
    – Nick Cox
    Sep 4, 2020 at 10:38
  • $\begingroup$ @NickCox The observation of antisocial behavior in a small number of users doesn't give us any information about the larger balance of users, who are likely a mix of pro-social people and people who are deterred by sanctions in incidents like this. $\endgroup$
    – Sycorax Mod
    Sep 4, 2020 at 16:19
  • $\begingroup$ I readily sign up to any idea that few people want to seem impolite or unappreciative of help. But it's the deterrent effect of such threads on people who might behave badly that I have to wonder about. I don't really expect testimonies to be easy to find, so we're just guessing. $\endgroup$
    – Nick Cox
    Sep 4, 2020 at 16:52