Timeline for How robust is the independent samples t-test when the distributions of the samples are non-normal?
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 13 at 19:10 | history | edited | Glen_b | CC BY-SA 4.0 |
added 37 characters in body
|
| Nov 24, 2022 at 22:49 | history | edited | Glen_b | CC BY-SA 4.0 |
deleted 2 characters in body
|
| Oct 9, 2012 at 20:32 | comment | added | Glen_b | @kjetilbhalvorsen The studies I've seen - including some simulation I've done myself (and I haven't looked at any for a good while; you may well have seen something I have not), the majority of the effect on power seemed to be mostly pushing the level up and down (which didn't affect the Wilcoxon). Given the generally good power properties of the Wilcoxon in these circumstances (particularly with heavy-tails), that's enough to have the Wilcoxon win on power - if you adjust the levels so they're similar, it surprised me how well the t- often did. | |
| Oct 9, 2012 at 16:08 | comment | added | kjetil b halvorsen♦ | I am not sure it is correct to say it is reasonably power-robust! It is reasonable level-robust, the significance level will be roughly correct, but for example wilcoxon tests can have much higher power for alternatives reasonably close to normality to be difficult to detect. This also depends on factors such as if there is equal number of observations in each group: robustness is much more fragile in the unequal-n case! | |
| Oct 9, 2012 at 2:44 | vote | accept | Archaeopteryx | ||
| Oct 9, 2012 at 2:44 | |||||
| Oct 9, 2012 at 0:55 | history | answered | Glen_b | CC BY-SA 3.0 |