Ten fold

12:22 AM

@Matt I wouldn't categorize things nearly so harshly. There can be perfectly legitimate reasons to treat something as linear even if it might not be (we do it every time we fit a linear model, for goodness sake). George Box's maxim (in Box and Draper as "Remember that all models are wrong; the practical question is how wrong do they have to be to not be useful") applies here as much as anywhere else. Such assumptions have their place!

They're not especially common, I think, but it can be a reasonable approach with ordered categories.

R has the nice feature of by default coding up the model matrix for ordered categories as orthogonal polynomials, so if you're really after a simpler model (or at least, one containing an additional assumption), like a linear one, you can do that, or if you want to fit a "full" model but see whether it's close to linear, you can do that.

Or if you want to investigate smooth deviations from linearity (eg imagine 6 ordered categories but where you think the means for the categories should progress nonlinearly but smoothly) you can do that (by, for example, only fitting the linear to cubic terms, say)

As an alternative of course, one could consider "nonparametric" smooths instead (and that may be better in many situations, though the orthogonal-polynomial route has some interesting/useful features)

One thing to keep in mind is that for many areas, data is hard to obtain, and power to pick up an effect (in particular, the impact of variance) may be more important that a bit of bias in a model. If you had a hundred times as many subjects, bias would rule and it would make no sense to consider anything but a full model. In very small samples, variance rules.

Antoni Parellada

1:17 AM

@Glen_b Very helpful. Thank you. I responded to your answer.

Silverfish

1:49 AM

@MatthewDrury "Suppose we have an eigenspace, but also a vector outside of the eigenspace that the matrix sends inside of the eigenspace. When that happens, the matrix cannot be diagonalizable." I might be missing something from the context, but diag(0,1) is obviously diagonalizable, being diagonal already. (1,1)^t is not in the eigenspace as it's mapped to (0,1)^t, but that is in the eigenspace (being mapped to itself, lambda=1).

Is this not a vector outside an eigenspace sent inside an eigenspace by a diagonalizable matrix? I think I've missed something.

On an unrelated note... I see it took just two minutes for me to go from 665 helpful flags to 667, hence almost entirely avoiding the dreaded sextuple Nelson. I envision a moderator hopping on one leg for the duration.

2

Antoni Parellada

3:05 AM

@Glen_b I would like to leave things unchanged, and not do anything at this point. Thank you.

Glen_b

@antoni Okay, that's fine

The abstract algebra thing looked interesting; I watched a bit of the one you posted here before becoming overly conscious that I should instead have been preparing some classes myself.

And I'm about to go read my partner's own notes for her classes (she just taught another one an hour or so ago), since it's assumed material for my students. She has some quite interesting examples.

General Abrial

4:33 AM

stats.stackexchange.com/questions/232419/… I give up

Glen_b

12:17 PM

No effort suggests "help vampire". Looks correctly closed, it's time to leave them to it

221

Q: The Help Vampire problem

What is Stack Overflow's long-term solution for the Help Vampire problem? Quote from article follows: Identifying Help Vampires can be tricky, because they look like any ordinary person (or Internet user, whichever is lesser). But by closely observing an individual's behavior using this hand...

Also

52

A: What is a help vampire?

Here are some further symptoms of a help vampire, viz a poster who treats the SO community as they would an online product support chat helpline, viz, a user who: Will continue to extend the original question with continued questions, even after the original question has been well answered. Pro...

(the one-box doesn't show it right for that second one but that's a link to an answer with additional symptoms

General Abrial

2:48 PM

@Glen_b I didn't know that was a term. I agree with the closure (I voted to close). I guess I'm just astonished at the persistence at asking the question in the face of much information about it.

amoeba

3:43 PM

It is funny that in the voting about When re-tagging old threads, how many should one do at once? the leading answer is up to 5 and the second leading is no limit at all. I invite more people to contribute there, because this is kind of important.

Glen_b

9:56 PM

@general I saw your vote -- I was letting you know I agreed with that judgement. It's closed, the OP needs to improve the question now. You've gone above and beyond already in comments, so you needn't worry.

@amoeba It is. I voted as soon as I saw it. Personally I'd have thought the split "should we have a limit or not" would be a first vote and then worry about how big it should be. Otherwise you could end up with a majority of people voting for a limit -- and even regarding no limit as the worst of all possible choices -- yet no limit might get up because enough limit options were available.

It's one of the big problems with first past the post voting -- if you offer similar options you dilute their vote.

None of this is intended as criticism of your actions. It's an observation on the nature of voting on issues.

Arrow's impossibility theorem tells us there's no ideal solution to this, of course.

Preference voting sometimes tends to leave you with the least hated rather than the most liked choice (which may actually be a good outcome in many cases, in that overall dissatisfaction with the choice is reduced)

I find it interesting that SE chooses to use a form of preference voting in its elections (STV, I think).

For elections like those it tends to give sensible results for a choice of moderator (in that polarizing candidates - ones both very popular and unpopular with different segments of the electorate - tend to find it harder to be elected over candidates with broader, if slightly weaker. appeal)

Matthew Drury

10:16 PM

@Glen_b I read what you said above, but I need to absorb it a bit.

I think I had misinterpreted a bit.

I was thinking the encoding enfoced an order as well as the "same magnitude" difference constraint.

but of course it doesnt enforce an order, as the estimated coefficient could be negative.

embarrasing

One real blind spot for me, which I am realizing more and more lately, is the use of stats for explanation instead (or along with) prediction.

I've been trying to read more lately.

I have some questions rolling around in my brain that I may eventually ask on the site.

There is so much I don't know.

@Silverfish of course, you are right.

I was too hasty with my reading of the answer to @AntoniParellada 's question and it's interpretation.

On the other hand, can this happen for anything but the zero eigenvalue?

Reasoning in an eigenbasis, it seems not?

Anyway, I'm going home for the day. It's been a tiring few days.

amoeba

@Glen_b That's a really good point about voting, perhaps I did not think about it long enough before posting my question. However, I should also say that we have only very limited resources in how we can set up a voting on Meta

I.e. I am no sure how one can conveniently set up a preference voting.

Anyway, currently 6 votes are for "five", 5 votes for "unlimited", and 1 vote for "ten", so overall it's 7 for limit and 5 for no limit, which is awfully close to 50/50.

May I ask what is your personal opinion?

Glen_b

10:41 PM

@Matt "There is so much I don't know." --- I find myself saying those words almost daily.

But we may be talking slightly at cross-purposes there. You shouldn't read my bit as anything but a comment on Kodiologist's response

@amoeba you can certainly ask.

... I'll consider whether I answer. I don't mind you knowing so much as I worry about influencing anyone else too much.

While first past the post can have the obvious effect of spoiler candidates (e.g. if the opposite side looks like winning, getting an even more extreme candidate from your position to run as well can hand you the election), but even with STV elections you can get a spoiler-like outcome, if the first choice can influence subsequent choices (influence the order you list your later preferences), you can get something like that happen.

E.g. in a national election, if Greens start to pick up a lot of votes a conservative party may induce people to set up some other "environmental" party that has more middle-of-the-road position on other social issues, who then encourage voters to put their later preferences toward the conservative side. That party is not there to be elected, but to drain preferences of people who might otherwise vote green away from the less conservative opponents.

It's a much less direct "spoiling" role but still effective enough that it does happen -- it's interesting that some kind of spoiling is possible either way.

amoeba

10:59 PM

got it

thanks

but what's wrong with "influencing anyone else"?

Glen_b

I often don't wish to do so.

As a mod, I'm conscious that sometimes people might give my opinion more weight than it's worth.

amoeba

I understand the sentiment but on the other hand we tend to select our most trusted / most devoted / etc people as mods, so it would be sad if they were then to drop out of policy discussions

Glen_b

The site should as far as possible be run by the users.

Rather than the mods.

Where I think it's important to contribute something I do still say so -- especially if I think there may be something for others to consider that they might not have.

In this case I think the pros and cons of the choice have already been made clear.

Often the best action a moderator can take is to do nothing (other than what an ordinary user can) and say as little as possible.

Silverfish

@MatthewDrury I reached the counterexample by reasoning about an eigenbasis (and immediately kicked myself for not seeing such a simple counterexample sooner, as one always does) but I didn't think carefully enough to see whether there would be another type of counterexample. It's an interesting Q actually.

Glen_b

Yes, it reduces the impact of a handful of users but those users have a lot of power and influence already.

@Silverfish what's the link on that?

Ah, never mind I think I can find it from Antoni's user page

Transcript for