« first day (2055 days earlier)      last day (3248 days later) » 

Glen_b
Glen_b
00:12
@mark999 I phrased that first one wrongly (it was in the wee hours for me when I typed it; I was not really awake); my idea was that the challenge to demonstrate the claim (with the corresponding rewards if it could be done) was intended for any person who made the claim, since it would be very useful if it were actually true.
Usually what is done in response to such a challenge is to simulate a collection of small sample cases for which the t is a reasonable approximation -- while encouraging (in that if you happen to use a t for cases very like those, you're usually not doing something silly), it doesn't demonstrate the claim that was made.
If one really wants to use a t-statistic in a small sample test, one could always use a permutation test (which I can type into the vanilla distribution of R from scratch in barely more time than it takes me to type the command to run a t-test anyway; if one sets up a function to do it, it can thereafter be done in exactly the same amount of effort as a t-test);
then one can abandon any worry about whether the t-approximation will be better than the asymptotic z-approximation and any worry about whether the asymptotic approximation is reasonable.
[Of course, if one is to do that, I'd then tend to wonder why not do a Wilcoxon test and get better power against the heavy-tailed cases where the asymptotic z-approximation doesn't tend to do so well at the null -- in any but tiny samples you have actual significance level closer to what you thought it was, and also better power at a given significance level, so it would seem odd not to use it.]
 
1 hour later…
mark999
mark999
01:38
Thanks for your additional comments @Glen_b.
 
8 hours later…
Glen_b
Glen_b
09:16
An activity sure to annoy me, but it seems to be pretty common in programs in the US in particular ... students get given problems and numerical answers. Students of course respond by randomly throwing any old combination of said numbers given in the question at the problem until they reproduce the answer. They either succeed (and stop) or fail (and stop), having gained not an iota of insight. What any of this has to do with actual scholarship or understanding is beyond my comprehension.
So many posts here are generated by students who have failed at the
"throw random combinations of the provided numbers at the problem". Actual reference to derivations of relevant results seems utterly absent -- we're expected to provide all the actual teaching and understanding apparently, from guessing what it is they're trying to do in the first place, to figuring out what they did and filling in all the gaps. I assume these students have notes and textbooks and classes ... so do many of these students simply fail to do the most basic of research steps and actually read them?
2
Indeed, while worked examples should certainly be given, often it seems as if providing solutions at all seems to be largely counterproductive (though I admit I often do it). As I often explain to students, the real world doesn't come with worked solutions ... your job is to convince me your answer is right, and any future boss you report to will expect the same.
I tell them that if you can't give convincing reasons that you have the answer right, you don't really understand it yet and haven't done enough work on it.
An employer wouldn't accept an unconvincing analysis, and I shouldn't either.
Skepticism should start at home. If you can't explain why you're right, you're probably wrong.
3
 
4 hours later…
Erosennin
Erosennin
13:05
Hello! I'd like to ask something regarding a question on difference between AIC and BIC, from here: stats.stackexchange.com/questions/577/…
The best rated answer states "BIC tries to find the TRUE model among the set of candidates"
this assertion is also challenged in the comment below the answer
I'm reading an article "Multimodel inference: Understanding AIC and BIC in Model selection" by Burnham and Anderson
stating that this assertion is false
I'm just wondering if somebody would like to comment on this, and if this should not be pointed out in the question im referring to above - in case somebody stumbles over this (as I did) and become confused
user777
user777
14:05
I'm actually amazed at how effective it is to search the page number of hte question you have.
Example: stats.stackexchange.com/questions/205044/…
 
4 hours later…
Erosennin
Erosennin
17:54
@user777 yes that is very fast indeed!
 
2 hours later…
Matthew Drury
Matthew Drury
20:03
@Glen_b "The first principle is that you must not fool yourself – and you are the easiest person to fool."
 
3 hours later…
Glen_b
Glen_b
23:22
@Erosennin I would suggest posting it as a question, linking to the things that differ (its possible to link to a comment but slightly tricky - if you can't see to do that, link instead to the post (Q or A) directly above it). Or if you want to make less of it, link each post to the other in comments to both pointing out the inconsistency -- but I really think it's worth a question.
@MatthewDrury indeed it is; I catch myself at it all the time. Erosennin's comment above might be another one.

« first day (2055 days earlier)      last day (3248 days later) »