3 ms·
As opposed to "experts" who always have rock solid statistical knowledge (/s): https://stats.stackexchange.com/questions/38403/what-causes-the-discontinuity-in-
by deepnotderp 5y ago
As opposed to "experts" who always have rock solid statistical knowledge (/s): https://stats.stackexchange.com/questions/38403/what-causes-the-discontinuity-in-the-distribution-of-published-p-values-at-p https://stats.stackexchange.com/questions/38403/what-causes-...
https://en.wikipedia.org/wiki/Misuse_of_p-values https://en.wikipedia.org/wiki/Misuse_of_p-values
I wish that people would stop disparaging attempts to contribute from people outside the specialized ingroup.
This stuff is hard, what counts as a hypothesis test for instance? Should datasets be locked and each view counted as a hypothesis? That's a very reasonable debate.
- rscho 5y agoContribute what? Mathematically incorrect statements don't seem very useful to me... After cursory examination, I don't understand your point regarding the stackexchange link. As for hypothesis testing, I also fail to see how that is supposedly poorly defined.
- deepnotderp 5y agoSlide 12 onwards illustrates my regarding hypothesis testing in the sciences: https://indico.cern.ch/event/995950/attachments/2171998/3676039/slides.pdf https://indico.cern.ch/event/995950/attachments/2171998/3676... Say you have a dataset with several variables and you’re looking for correlations between them. If 20 different groups separately download the dataset and test hypotheses, should their p-values be adjusted for multiple hypothesis testing?
- rscho 5y agoOk, now it's clear. So yes I agree that controlling false discovery rate is an unsolved problem that p-value doesn't address. Yes of course applied stats are a kind of 'soft science', at least regarding the interpretation of results. Statistical theory on the other hand, is very well defined. So that doesn't justify considering assumptions made from erroneous conceptual understanding. This is where we disagree, I guess.