Skip to main content

Timeline for The theory behind Tukey's HSD test.

Current License: CC BY-SA 4.0

12 events
when toggle format what by license comment
S Apr 3, 2023 at 9:18 vote accept ThighCrush
S Apr 3, 2023 at 9:18 vote accept ThighCrush
S Apr 3, 2023 at 9:18
Apr 3, 2023 at 9:18 vote accept ThighCrush
S Apr 3, 2023 at 9:18
Mar 29, 2023 at 6:32 answer added Sextus Empiricus timeline score: 6
Mar 29, 2023 at 6:02 answer added Thomas Lumley timeline score: 5
Mar 28, 2023 at 22:25 history edited kjetil b halvorsen
edited tags
Mar 28, 2023 at 12:35 history migrated from math.stackexchange.com (revisions)
Mar 26, 2023 at 16:00 comment added Henry That may depend on what you usually say when you reject the null hypothesis (personally I see that a step towards what to look at next). Given the the title of the test is HSD (honestly significantly different), I would report that those sample mean differences which exceed the critical value are significantly different in this test, and stop there.
Mar 26, 2023 at 14:47 comment added ThighCrush I understand using this to reject the null hypothesis that all means are equal. My question is: in practice this is used to conclude that specific pairs of means are not equal (ie H0 seems to become: one of these specific pairs identified is in fact equal, and we reject it.) Why? Or what is the exact reasoning?
Mar 26, 2023 at 12:38 comment added Henry In particular you might find ${\overline {y}}_{a }-{\overline {y}}_{b }$ is not statistically significant and ${\overline {y}}_{b }-{\overline {y}}_{c }$ is not statistically significant but ${\overline {y}}_{a }-{\overline {y}}_{c }$ is statistically significant. This non-transitive result is common: in a exam, you may often say the highest ranking are significantly better than the lowest ranking, without being able to say that the highest are better than those in the middle or that the lowest are worse than those in the middle.
Mar 26, 2023 at 12:34 comment added Henry It you take the null hypothesis (that the observations are all iid $N(\mu, \sigma^2)$, then rejecting the hypothesis because $\frac{{\overline {y}}_{\max }-{\overline {y}}_{\min }}{S\sqrt{2/n}}$ is too big is essentially saying that the particular difference ${\overline {y}}_{\max }-{\overline {y}}_{\min }$ is big enough (above the critical value) to be statistically significant. You might then say the same for any other differences in means above the critical value. Like all tests, this is suggestive rather than conclusive.
Mar 26, 2023 at 12:12 history asked ThighCrush CC BY-SA 4.0