How do I perform multiple comparison tests (on the means) when you have unequal variances

[richnewman]

Let's say you are running a One-way ANOVA with 5 groups.  You test for equal variance and conclude that the assumption of equal variance does not hold.  You then evaluate Welch's test.  The p-value from Welch's test indicates that the 5 group means are not equal.  Is there a simple way to run a multiple comparison test (to compare the 5 means or a Dunnett's like test to compare 4 of the means to a control group) when you have unequal variance? 

Ideally, this would be an option found in the "red triangle" when using Fit Y by X (for the One-way ANOVA case).  Perhaps found in the Compare Means choices (see attached for a possible example).

Thanks

[cwillden]

Hi Rich,

Give this a try.  It at least covers the Turkey HSD situation.  I just updated the add-in a couple days ago.

https://community.jmp.com/t5/JMP-Add-Ins/Games-Howell-Test-Tukey-HSD-with-Welch-s-correction-for-Une...

[richnewman]

Perfect!  Thank you for your help.

[rc_hertzy]

A modified suggestion is, once the variances are found not to be equal, to allow modeling of the variance as function of the mean, then re-test homogeneity of the model-estimated variances. Then the means comparison is done as before but using the modeled variances.