@Silverfish If they are both old threads, the first I'd do is cross link both. That way people have a chance to find the other answers even if they don't get related as duplicates. Then use your own judgement about how close they are.

Incidentally, it doesn't have to be the case that the newer one is marked as duplicates. If the answers for one are much better than the other, it should become the canonical one (or even if the question is much better in some way). If they're about the same, the older one should have priority.

@Glen_b Thanks. Yes, I already cross-linked them: I do that a lot as soon as I find threads that have a bearing on each other. I've gotten better with it as my familiarity with existing questions has improved

@Silverfish Cool. You're way ahead of me then. That sort of stuff makes the site much more useful (even though there's no credit for doing it).

@Glen_b (or anyone else) Also as a question of judgment, how close does stats.stackexchange.com/a/127037/22228 come to violating the spam policy?

I thought it was probably justabout on the right side okay, even though it contains a link plus clear solicitation for services

But I'm only leaning that way because the OP was clearly looking for such a service.

The main things are (1) it must answer the question. (2) the poster must reveal any affiliation. I haven't looked at (1), but (2) is minimally satisfied (since "we" implies he's part of whatever company or group did it).

At first glance it might fail as essentially a link-only answer, but probably not as spam.

But I'd have to take a closer look.

Was it something you flagged?

No didn't flag

Okay, thanks.

Since I thought it was probably alright.

Revealed affiliation, and directly relevant to the OP

Thing is, if I always let that sort of thing go, then I'll never find out if I was meant to be flagging them in the first place, if you know what I mean...

Okay, then probably nothing to do.

At least if I flagged and got rejected then I'd know.

Yes, you should ask about it in some fashion. You can always post on meta. Mods aren't the arbiters of site policy except by default.

Yes, that's true. Since you were about thought I would just ask directly. So long as I'm making a reasonable judgment on what I flag I'm happy. (Posting on meta would have had the advantage that other people could see how this kind of judgment call comes down.)

If you flagged it, I'd probably have marked the outcome as "no action required" and the flag as "helpful"

Yes, it's fine to ask someone if you're talking anyway. No problem with that

But you correctly identified why to ask on meta --- for everyone else with the same question.

We establish norms by discussing them and by watching what other people do (to the extent we can).

That kind of reviewing process is one of the things where we don't much observation practice!

Even if nothing is resolved in a meta question we at least see the range of opinions.

With reviews at least it is sometimes possible to back up through the queue after you vote and see what other people do. Can't do that with something you're just flagging, but you can see other posts that people have flagged and ask "would I have flagged this?"

I frequently check close and reopen votes to see whether my vote was broadly in line with other people's opinions; I also check down the list of my "suggested edit" approvals/rejections to see whether other people opted the same way, which gives a feel about what kind of edit requests are considered acceptable.

I can't see other people's flags (well, don't think I can, maybe that's a 10k thing?)

Yep. Of course, even if your vote was different, sometimes everyone else is wrong.

Oh, sorry of course you can't.

You can see moderator actions, but not whether that was a result of a flag.

There are plenty of things I was in a minority of one on and still reckon it was arguable either way.

Yeah, don't sweat those. That's how it's supposed to work.

I think "Late Answers" is something that is only reviewed by one reviewer

It's a noisy process. Occasionally the wrong decision gets made even when it takes 5 votes.

Fortunately almost all things can be undone in some way.

(Or if it is two on Late Answers, then unlike Suggested Edits I can't see what the other reviewer thought.)

Oh, I better get some of this late answer action while I can.

I am very curious how @gung reached 1000 on late answers! That's remarkable. I reviewed them whenever I saw them and don't think I'd got to 20 before the current windfall.

Perhaps different timezones are busier on the review queue.

I was only on 32 myself.

Well, a lot more than usual.

Possibly a side effect of the late answers... I know I flagged a few

There's a bunch from several people. Given the review queue is half what it was when you posted to meta, you'd expect a few flags to arise.

I flag as well, especially when the reason for the action I plan to take may not be obvious. The flag gives a reason for my subsequent action. It's a trick I learned from talking to mods of other SE sites.

We need to be as "open" as we can be about how we act. It's one way of making an audit trail for why you did stuff.

That's quite neat. I am very impressed by the moderation culture here.

It's basically just trying to follow the SE policy on moderation. I also happen to agree with it, which makes it easier to follow, but all credit to other people.

Aside from the technical issue of designing the moderation workflow, there is a lot of know-how involved in both creating a workable policy document and then getting folk in several dozen communities to abide by the culture.

Yes. There's also a degree of flexibility about how to interpret and implement the policies. To some extent the sites get to figure out how they work.

This is mostly a good thing.

Well, a lot of those late reviews I'm seeing needed review. That was probably a good change. Given the 20-a-day limit, the fact that 260-odd have already been cleared is amazing.

The Stackoverflow late answer queue is already empty

But they have 8200 in their close vote queue. It's an ongoing problem there.

Because they're users who didn't stick around, they often didn't know how to things like formatting. I did a lot of editing. But some of the content raised issues too, and obviously it went under the radar before.

Yeah there's a very large number of old answers that need attention; this was one set of criteria that found a chunk of them.

Thanks for raising it on meta, by the way, as well as thoughtful reviewing.

Thanks. I'm glad I got the screen capture when I did! Wouldn't look so impressive anymore!

It will be below 200 very soon. I expect you'll still get a chance at the tail of it once the 24 hours are up

Well, less than 24 hours... whenever it clocks over

It ticks over at 0:00 UTC so I've done 40.

(I think it's 0:00 anyhow.)

Ah. Well, there might still be some 22 hours from now.

So you might just get a chance at another 20.

The "active" page is a bit of a mess of nearly all old posts, but that can't be helped.

I have a question I am mid-way through preparing. I think I'll ask it when the current noise has died down a bit!

Might be a good idea.

I do wish some particular people better understood the distinction between "this is a bad/wrong answer" and "not an answer"

A genuine attempt at an answer that happens to be just flat wrong is not really something to flag, it's something to downvote.

And maybe comment on. ... In fact I see just such a suitable comment from Silverfish on one now. Someone else came along and flagged it as not an answer.

Of course it's an answer, it's just a bad one.

I have occasionally seen answers so wrong that they were utterly nonsense (as in, the answer itself makes no sense) and have been tempted to flag them as Not An Answer on the grounds that it can't possibly address the OP's issue. Something so beyond redemption it's at "not even wrong" levels.

Yes, those are "not an answer" I guess but "very low quality" may be a better flag

Occasionally there are one-liners that could be arguably converted to comments. But if I see a poor one I actually prefer it to stay an "answer", so it gets downvotes.

And of course there's some grey area near the border where there's plenty of room for difference of opinion about whether it's really an answer

I kind of wish review queues could be toggled (at the user's own behest) work like the mod queue.

The mod queue is different. You get a vertical list of flagged posts all at once, and can go up and down through the list. As they get handled they disappear. It means you can remove the obvious ones, but the hard ones you can think about and they're still there later. You can see what others have done. Maybe another mod added a comment or something after an hour and that might be sufficient, or it might help you figure out what you should do.

Being able to work through the list that way could make reviews "easier" to do I think.

I think it'd help on a lot of First Posts and some Close Votes where there is clarification going on in the comments.

It's not that it allows "conversation" but simply being able to see other people's actions makes it easier.

If you just happen to open a post on which there's a flag without going via the queue, stuff pops up from the bottom relating to the flag. That would be another feature I'd like to be able to turn on for review queues

I often find myself going back to posts that have disappeared from my review queue. And I'm not always engaged in the comments discussions so don't get pinged. With duplicate discussions, people often identify alternative targets, for instance, which you might not see the first time you look in on it.

Yep

Under 200 now. People are still busy on that queue.

Looks like there were a fair few flags handled overnight (my time)

Only to be expected I guess.

Gee I've had a lot of upvotes on old answers in the last day. Part of that may be the extra attention old questions are seeing. Outside that I can't explain it.

I had a couple of upvotes on older answers that I suspect come from questions getting linked from other questions.

I think I've had about 15 in the last 12 hours.

I'd usually see a few each day, but nothing like that.

Well that's the mod queue out of the way, only a couple of items acted on in some way but not finalized, which I'll leave for other mods to see. But I better get to other things I guess. Maybe a quick scan of the questions first in case there's something I can deal with quickly.

Have a good one. I've almost finished my Q but think I'll save it for a slower day on the activity list.

hello!

@Ghost hi.

@Glen_b how are you?

I am wondering if I should nuke my question on maths and repost it here math.stackexchange.com/questions/1454715/…

@Ghost Better not to repost. If you think it's better here, flag it for migration. It is on topic here, but there's some information that would help. I guess those clarifications can come once it gets here.

hmmm, I have just figured an answer - i.e. found a reference that explains it

@Erik -- Generally signatures and taglines should usually be removed as per SE policy, but you should edit everything else wrong while you're there. If the only issue with a post is one or two extraneous words (like "Hi--" or "thanks" or "--John") you're generally best to leave it; there's a competing principle of not making trivial edits.

@Ghost you're certain to reject, by the way

Any (reasonable) hypothesis test of equality of two histograms with n = 60 million? Yes, certain to reject

If you're doing some other kind of test (such as an equivalence test), maybe not, that depends on how the equivalence criterion is defined.

when we did the calculations - set one got a Bhattarcharyya coefficient/metric/etc of 0.99, the second 0.96 - but the third only 0.36

So?

You're looking at a measure of the effect size, not its significance.

Even the most trivial differences will become obvious at very large sample sizes ...

Oh, I guess I am assuming this is univariate.

If it's multivariate with large enough dimension, that will change.

All I am wanting to do is to show that there is a significant enough difference between each pair of histograms

each pair of histograms have the exact same amount of and width of bins

and each of the distributions are lognormal

When you say "significant" are you trying to use it in its nonstatistical sense?

Neither of the distributions are lognormal

You mean to say something else

These are derived from real data?

Yes these a real data - one histogram is a 'base' histogram - sort of a default so to speak

I am trying to work out if the calculated distance between this default and the 2nd measured histogram is statistically significant

The answer is yes, the distance is statistically significant. I'll put a dollar on it.

How many bins do you have?

256

And you want to use Bhattarcharyya distance as a statistic? Or is the statistic of choice for the test up for negotiation?

up for negotiation - I have just noticed that the Bhattarcharyya distance has been used in similar applications

these values are discrete by the way

Of course, if you're confident of lognormality you could ignore the binning and just test equality of log-mean and log-variance.

Wait, what? You just asserted they were lognormal.

How can you now say they're discrete?

Lognormal is continuous.

oh wait, thinking of something else

ignore the discrete comment

Of course the binning discretizes the distribution if you're treat the intervals as their centers

and no, am not all that confident of the lognormality

I think it's possible to make a permutation test out of Bhattarcharyya distance, but I'd be inclined to look at basing it off one of the more common two-sample goodness of fit tests.

when we did an initial look at the histogram pairs - using the B. distance - we got values of 0.01, 0.09 and 1.27 for the each pair

You say one is a "base". In what sense is it different from the other?

it is a baseline measurement

the conditions are changed slightly, for the 2nd histogram

But both are samples?

yup

Bhattarcharyya distance, if I understand it correctly, suffers from the same problem as chi-square -- they both ignore the ordering in the bins and so throw away vast amounts of information.

This will mean very low power to discriminate between two smooth densities.

Nevertheless at that sample size, you'll probably have plenty of power to burn.

it is indeed, a fairly chunky sample size

If I could be convinced there was any point to actually formally testing for a difference (since it's sure to reject, that would take some doing), I'd be inclined to look at something like a two-sample Anderson-Darling statistic but you'd have to deal with the heavy ties induced by binning.

I think I will have to continue the research - am looking up and reviewing papers of the same field of research - and see what they've used, how they've used it and what significance/justifications they come up with

Another alternative would be to take advantage of the approximate lognormality, take logs, construct some decent test (such as a likelihood ratio statistic) assuming normality and then if you were worried about it not being lognormal you could build a permutation test out of that to allow for some deviation from it.

hmmm that could work

I have lognormal means and standard deviations calculated

perhaps, as you said earlier, base it on a comparison of lognormal means and standard deviations/variances between the measured and baseline histogram data?

@Dawny33 Yes; asking and answering your own question is entirely allowed, and indeed often a very good idea. It's a way of telling the forum about a problem and a solution, which could match our goals exactly. Nothing rules out other answers, comments on the answer offered, etc., as appropriate. Similarly nothing rules out removing such questions or answers if they are inappropriate.

This particular case isn't on-topic, but that wasn't the point you raised.

@Glen_b I was talking more about the plea that I removed here: stats.stackexchange.com/posts/174696/revisions Based on what you've said it was probably a trivial edit since I didn't fix anything else. That not withstanding I should remove the "Please help" text as part of a larger edit correct?

@ Erik: Certainly remove stuff like that as part of a larger edit. Personally I don't see much harm in trivial edits while the review queue for suggested edits is as typically short as it is now.

And as @Silverfish implied, some posts are more worth perfecting than others.

@Erik Personally I usually wouldn't edit just to remove something like that (since to my reading of it, SE guidelines on edits say not to), but there's no harm if you're not bumping dozens of posts. If I can find anything else to fix (usually at least tags can do with some work), I probably would edit.

e.g. from the help "Tiny, trivial edits are discouraged." ... there's some grey area about how small is small. Scortchi is right to point out that the suggested edit queue is empty so it's not that big a deal in any case.

However, if you tend to operate in that grey area you'll probably hit some inconsistency in reviews of your suggested edits -- some people will reject as too small, and some will approve.