Ten fold - 2017-04-24

« first day (2442 days earlier) ← previous day next day → last day (2566 days later) »

12:24 AM

@dbliss no doubt there are many people seeking one-on-one help, but there's not nearly enough people here to be offering one-on-one help as any kind of regular thing. If we answer a question here in chat, two days later or a week later we're answering essentially the same thing, and then another couple of weeks later, answering it again. Soon the effort put into answering the same question for the tenth or the twelfth or

2

the twentieth time gets too much and answers become overly compressed and even flippant. One of the reasons I now restrict most of my help-people-on-the-internet time to here is that it makes it easier to find (and to direct people to) older, good answers.

2

@Glen_b yeah, i think one issue is that CrossValidated seems to be run by a small number of people who answer and edit a huge number of questions.

It would be great if we could offer some kind of one-on-one service, but to be honest I'm here precisely to avoid continuing to do that.

@Glen_b i see. i can't really envision a future where there is never a need to hash out a stats issue in a one-on-one conversation. just like i can't envision myself doing research alone at my apartment without ever coming into my lab. there is a benefit to being able to talk things out with someone.

@Glen_b no matter how good the answers to generic questions are throughout the CV site.

Sure there is.

I don't doubt it for a moment.

But if there's far more people seeking that help than offering it, it's not practical for many people to do it.

right

you guys are overworked

i don't understand why that's the case

12:31 AM

By contrast with math.se - anyone who has done reasonably well on a fairly basic calculus subject and a fairly basic linear algebra subject (of which there are millions of people) can answer half the questions on math.se

You can understand enough to be useful by knowing a fairly limited number of things. For stats, often someone who has done a couple of first year uni subjects knows just enough to be dangerous.

yeah

Add to that that each area teaches its students statistics in slightly different ways.

i think there's a big divide too between statisticians and scientists who use stats.

Yep.

which could be bridged.

but it seems like statisticians are often not interested in the practical problems scientists face in specific domains

and scientists don't care so much about stats theory -- they just want the right test for their data set so they can move on

12:34 AM

I see statisticians working in universities actively discouraged by their administrations from working with people outside stats research

For example, where in some places they won't even count an interdisciplinary publication you get.

oh, i see. discouraged by their own departments.

No!

The depertment might well see it as good work.

At least at universities I've been at, there's a hierarchy of administration above the department that looks at things like tenure and promotion

gotcha

At a university-wide level.

i see

12:37 AM

Supposedly to keep it all fair.

There are far more administrators than there are actual people doing the teaching and research

Way, waay more.

what do you mean, like deans?

No.

University departments aren't their own fiefdoms

There's layers of management sitting above them trying to make them do stuff.

All those MBAs have to work somewhere.

Well, these days it's increasingly in universities.

ha. i find it hard to believe that administrators with a say in tenure decisions outnumber professors and post-docs, though.

but i have heard of "rules" related to tenure decisions that seem unfair.

Not all of the admin hierarchy have an actual say in those things; that's a relatively small proportion. But there's an army of them making policy documents, for example.

gotcha

12:42 AM

So you regularly get bizarre edicts that do nothing to promote actual scholarship.

i'm not exposed to much of that where i am.

Fortunately for me I avoid most of that; but I work with people who struggle with it daily.

gotcha

It means I earn less money though.

it's not as bad as like . . . what's happening in syria, though.

12:45 AM

Well, no. But let's not go toward godwinning the discussion.

Instead I spend more time working for free...

Let me check your link anyway

There is a certain amount of one-on-one that arises out of questions, as when a comment thread gets taken to chat

^ ha, hadn't heard of "godwinning." that's pretty funny.

From Godwin's law, but I expert you figured that out.

Your design is orthogonal?

yeah

You may find this post somewhat relevant:

8

Q: How to test an interaction effect with a non-parametric test (e.g. a permutation test)?

I have two categorical / nominal variables. Each of them can take only two distinct values (so, I have 4 combinations in total). Each combination of values comes with a set of numerical values. So, I have 4 sets of numbers. To make it more concrete, let us say that I have male / female and young...

hypothesis-testing p-value nonparametric permutation-test

i did see that. it seems to say you can't do it, more or less.

or wait, ha

maybe i didn't see this.

no, sorry, i did.

yeah, the TLDR seems to be, you can't do it.

12:54 AM

That's too strong a conclusion.

You can do it, sort of.

Ah, here's David Howell with a reasonable introduction

uvm.edu/~dhowell/StatPages/Permutation%20Anova/…

^ ah, many thanks

In particular, there can't be an exact test for interaction

that page seems to be exactly the kind of overview i was hoping for

But that doesn't mean you can't do anything.

i see. that makes sense.

12:56 AM

If you stick to the original variables (no transformation), interaction is still just as meaningful, but you have to look at approximate procedures.

You could also consider other resampling approaches

Like bootstrapping rather than permutation (as long as you have large enough sample sizes)

yeah, so what i'm doing now is the permutation test to compute significance, and then when i plot the data, i include bootstrapped CIs.

But that's getting outside what I have direct experience with

ok

Okay, that sounds sensible.

Edgington (referred to multiple times at that link) is a bit idiosyncratic.

... Oh have you seen Good's book?

no

1:01 AM

Permutation, Parametric, and Bootstrap Tests of Hypotheses, Phillip Good

sounds like something i should check out

It's one of the springer books, so yellow cover with brown on it

Not very thick -- would be under 300 pages I think, maybe 240ish?

he also has a book called Resampling Methods

t1.gstatic.com/…

ah, wait, that's a different book, no?

or did he just change the name for the third edition

anyway, i'll check these out.

1:08 AM

I think it grew out of his earlier book "permutation tests" rather than the resampling book (which I haven't seen)

ANyway, that book may have some useful information

If you can borrow a copy, I'd take a look

Actually, both it and his resampling methods books have third editions ...

ANyway, section 7.5.5 of that book is on interactions in multifactor anova

I don't recall anything from that section though

It's a long time since I looked at it and I wasn't looking for that stuff when I did

thanks!

uvm.edu/~dhowell/StatPages/R/RandomizationTestsWithR/…

Oh, okay that is the right link. I just wasn't expecting it to be Dave Howell's pages again

1:23 AM

i was starting to wonder whether you were dave howell.

That's a paper on multifactor ANOVA by Anderson and ter Braak

There's a section on interaction in it

cool, thanks

Well, the section on approximate tests, but that covers interaction

Sorry must go.

3 hours later…

4:19 AM

Any one can help me with this question?

0

Q: When to use mixed effect model?

Linear Mixed Effects Models are Extensions of Linear Regression models for data that are collected and summarized in groups. The key advantages is the coefficients can vary with respect to one or more group variables. However, I am struggling with when to use mixed effect model? I will elaborate...

regression mixed-model random-effects-model linear

I asked similar questions in CV differently but never get response. I think this is the best version I have...

Or anyone can give me some suggestions why this types of question attract less people to answer?

thanks!

8 hours later…

12:29 PM

They say you can't argue with results, but what kind of defeatist attitude is that? If you stick with it, you can argue with ANYTHING.

3

« first day (2442 days earlier) ← previous day next day → last day (2566 days later) »

Transcript for

Apr '1724

Ten fold

CrossValidated's general room for gossip, grumbles, and idle c...

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2024 Stack Exchange Inc; legal

mobile