Timeline for Are there alternative statistics to a p-value in NHST?
Current License: CC BY-SA 4.0
15 events
when toggle format | what | by | license | comment | |
---|---|---|---|---|---|
Jul 14 at 11:22 | comment | added | Sam | @Dave More precisely, I seek to know whether there is such a statistic that is not equivalent to the p-value. | |
Jul 8 at 16:21 | comment | added | Dave | @Sam The test statistic? | |
Jul 8 at 12:08 | comment | added | Sam | @Dave The p-value is calculated assuming that the null hypothesis is true, but is used to reject it (or not). I am seeking for another such statistic, which is not the p-value | |
Jul 7 at 14:49 | comment | added | Sam | @GrahamBornholt The p-value is the probability of getting at least as extreme results as the experimental ones given the null hypothesis. I seek alternatives that would still be a probability value, would still be calculated given the null hypothesis, but would not use the criteria of "at least as extreme results" but some other criteria (and still make some sense). | |
Jul 6 at 0:44 | history | became hot network question | |||
Jul 5 at 21:32 | comment | added | Michael Lew | You will find many useful resources on this and related topics on this site. Perhaps start here: stats.stackexchange.com/questions/200500/… and here: stats.stackexchange.com/questions/488398/… | |
Jul 5 at 21:22 | comment | added | Michael Lew | P-values are not difficult to understand, but most attempts to introduce them to students are almost unbelievably bad, and they are usually contaminated with non-p-value stuff about error rate accounting of the all-or-none significant/not significant error rate stuff. | |
Jul 5 at 21:19 | comment | added | Michael Lew | ...cont. For the third bullet the reason to use the integral of 'more extreme' results is so that the p-value can be used as a ranking of unusualness of the observed results among all possible results according to the statistical model. You need to introduce the tests statistic to make that point, and that is often not done. | |
Jul 5 at 21:17 | comment | added | Michael Lew | ...cont. For the second bullet, the probability of exactly the experimenter's results is often (in theory) exactly zero. If you are interested in the probability of the results then you should be interested in the likelihood function, not a p-value. | |
Jul 5 at 21:16 | comment | added | Michael Lew | I think that all of your bulleted items are false. It is a statistical model that needs to be introduced for the first, not 'a random sample'. Think of resampling tests that yield p-values from an assumption of random treatment allocation instead of sampling from a population. | |
Jul 5 at 20:55 | answer | added | Greg Snow | timeline score: 4 | |
Jul 5 at 20:53 | comment | added | Graham Bornholt | "Have alternative statistics been suggested, the probability of which given the null hypothesis is the value to be considered for rejecting (or not) the null hypothesis?" Given what you acknowledge in your third dotpoint, the meaning of your question is unclear. Please clarify what you are requesting. | |
Jul 5 at 17:37 | answer | added | AdamO | timeline score: 4 | |
Jul 5 at 16:55 | comment | added | Dave | Given the null hypothesis, you want to know the probability of the null hypothesis? | |
Jul 5 at 16:43 | history | asked | Sam | CC BY-SA 4.0 |