Feature request: Consistent Student's versus Welch's t

Discuss the jamovi platform, possible improvements, etc.
Post Reply
User avatar
reason180
Posts: 274
Joined: Mon Jul 24, 2017 4:56 pm

Feature request: Consistent Student's versus Welch's t

Post by reason180 »

At appears that, currently, between-subject ANOVAs use Student's t for post hoc comparisons whereas repeated-measures ANOVAs use Welch's t for such comparisons. My feature request is to have a way to make the type of t test (Student's or Welch's) consistent across the different types of ANOVAs (between-subject or repeated-measures), perhaps by allowing the user to specify either Student's or Welch's post-hoc t for any given ANOVA.
User avatar
jonathon
Posts: 2613
Joined: Fri Jan 27, 2017 10:04 am

Re: Feature request: Consistent Student's versus Welch's t

Post by jonathon »

hi,

we actually use the emmeans package for all of this. it might be worth touching base with the emmeans author to check your theory, and see if there isn't some good rationale for this.

cheers

jonathon
User avatar
reason180
Posts: 274
Joined: Mon Jul 24, 2017 4:56 pm

Re: Feature request: Consistent Student's versus Welch's t

Post by reason180 »

OK. Given our long conversation on the emmeans Github, it looks like this is what's happening: The various authors of the various sub-packages employed within jamovi and within the emmeans package have made different decision--in the present case, a different decision about whether to use or not use the Welch-Satterthwaite method for calculating degrees of freedom (Welch-Sattertwaite is used in jamovi's ANOVA but not jamovi's repeated-measures ANOVA. It may be that neither way is wrong. They're just different.

So one way to take all of this is that we need to accept that there will be theses kinds of inconsistencies. Because R.

Still, it might be good to include notes in the jamovi documentation (or even in footnotes to jamovi analyses) explaining the inconsistencies, or at least explaining what was done in the analysis. [For example: "Note: d.f. is calculated using the Welch-Satterthwaite method."]
--
P.S. I'm told that the next version of EMMEANS will (at least for some of its models) be modified to use the Welch-Satterthwaite method to calculate d.f.
User avatar
jonathon
Posts: 2613
Joined: Fri Jan 27, 2017 10:04 am

Re: Feature request: Consistent Student's versus Welch's t

Post by jonathon »

thanks R180,

i'm feeling a bit spent after that discussion!

yeah, it's probably a good idea to update our documentation.

jonathon
Post Reply