|
@talyarkoni | |||||
|
I wrote this paper arguing that the inferential statistics we report in psychology papers are so disconnected from the hypotheses they're meant to test that they may as well be made up. feedback is appreciated.
psyarxiv.com/jqw35
|
||||||
|
||||||
|
Michael Breakspear
@DrBreaky
|
22. stu |
|
Interesting and thought provoking paper! But, ironically, shouldn't it be called something like, "Under-estimation of posterior error through use of unrealistically restrictive statistical models in some psychology papers."
|
||
|
|
||
|
Tal Yarkoni
@talyarkoni
|
22. stu |
|
no. :)
|
||
|
|
||
|
Alex Shackman
@ajshackman
|
22. stu |
|
It's quite good. You're getting crankier, more Meehlish tho ;)
|
||
|
|
||
|
Tal Yarkoni
@talyarkoni
|
22. stu |
|
the Meehlian Catch-22 I've discovered is that if you say things with a smile, people think the problem must not be very serious, and if you say it while yelling, people dismiss you for being too angry
|
||
|
|
||
|
Pierre Bellec
@pierre_bellec
|
1. pro |
|
Quick thought. I feel you are taking stats too seriously. Fixed effect models are honest, at the end of the day you know only as much as your sample. Practical approaches to test model generalization (eg ncbi.nlm.nih.gov/pubmed/28601499) seem more useful to me than mixed effect models.
|
||
|
|
||
|
Tal Yarkoni
@talyarkoni
|
1. pro |
|
oh, sure. but I'm not arguing for any particular kind of model. I'm saying that researchers need to be clear and honest about the relationship between their models and their verbal claims. that's all.
|
||
|
|
||
|
Science in Crisis
@sciencecrisis
|
22. stu |
|
Congrats on your paper! I was wondering what your views are on generalization outside psychology. Is it a general problem or not (pun intended)?
|
||
|
|
||
|
Tal Yarkoni
@talyarkoni
|
22. stu |
|
based on my careful examination of psychology, I'm extremely confident that the problems I discuss generalize to at least the same degree to all other sciences
|
||
|
|
||
|
nick michalak
@nmmichalak
|
23. stu |
|
On p. 11, you talk about setting priors without terms in the model. How did you do that technically? (R's brms won't let me set priors that aren't in the model specification/data)
|
||
|
|
||
|
Tal Yarkoni
@talyarkoni
|
23. stu |
|
probably can't do it in brms, not sure. u could do it directly in Stan or some other probabilistic programming framework. my Python/PyMC3 code for those analyses is here: github.com/tyarkoni/gener…
|
||
|
|
||