Post Hoc Test Values

[776655104]

For the post hoc tests in jamovi (e.g., in the ANOVA routine) does the no correction option give the Fishers protected or unprotected LSD?

Also, is there anyway to get the LSD value used, or the values used for any of the post hoc tests (e.g. Tukey's HSD values). Some disciplines like to display them when presenting graphs along with the means or include them in the legends. Current, you get P values so you know which means are significantly different from each other. but no actual post hoc statistic values.

[jonathon]

it doesn't look like we provide LSD.

we try not to clutter our analyses wire more obscure options, but if you wanted to add these additional statistics, i'd suggest making some modifications to the moretests module.

cheers

jonathon

[776655104]

Okay. I though the LSD was present (but I didn't know which version) since someone had mentioned that the last time I asked at the link below

viewtopic.php?f=2&t=1443&p=5072&hilit=lsd#p5072

Ravi also mentioned that he though so here

viewtopic.php?f=2&t=491&p=1903&hilit=lsd#p1904

What about actually giving the values used for the Post Hoc comparison tests already implemented though? For example, if I used the Tukey test it adds a column with the p values for that comparison between the means, but you do not know which critical value it is using for the comparisons for that particular analysis. This value has to be worked out anyway to decide which means are significantly different from each other, so I was just asking if there was anyway to get that value. This would usual be a value where if the difference between the two means is greater than the value then they are marked as significantly different from each other, if the difference between the two means is less than the value then the means are marked as not significantly different from each other. Would this be considered clutter?

[jonathon]

> Would this be considered clutter?

yeah, sorry

jonathon

[776655104]

Okay, I understand.

What does the no correction option for the post hoc tests do currently? Is it just doing individual t tests comparing each pair of means?

[jonathon]

i think that's how it works ... normally you generate a bunch of p-values, then you pass these p-values to a procedure that performs each of the corrections. so "no correction" is the p-values before they're modified by the tukey, the sheffe, the whatever, procedure.

we use emmeans to do all of this:

https://github.com/jamovi/jmv/blob/master/R/ancova.b.R#L450-L456

jonathon

[reason180]

Greetings.

I think jamovi's "no correction" is in fact Fisher's LSD. The p values for Fisher's LSD are in fact not corrected for familywise error ( https://www.graphpad.com/support/faqid/176/ (the "protection" is that you don't look at post-hoc results unless the relevant ANOVA main-effects and/or interactions are significant)). Therefore, I had assumed that "no correction" meant Fisher's LSD.

For the emmeans R package, I've not been able to find documentation about the meaning of "no correction." However, just now, using the Tooth Growth data set, I ran an ANOVA assessing tooth length as a function of dose (there was just one factor in the ANOVA). For the post hoc tests, I chose "no correction." I then ran the same ANOVA in SPSS but selected "LSD" (there's no "no correction" option in SPSS). The post hoc test results were identical for jamovi and SPSS. Consequently, I suspect that jamovi's "no correction" and LSD are exactly the same thing.

Perhaps the name of the option in jamovi should be changed from "No Correction" to "No Correction (LSD)"?

Regards,
--
Rich Anderson
Bowling Green State University

[jonathon]

oh yup, that's not a bad suggestion

[IlzeD]

Hi,

can you confirm what was said before "Consequently, I suspect that jamovi's "no correction" and LSD are exactly the same thing.

Perhaps the name of the option in jamovi should be changed from "No Correction" to "No Correction (LSD)"?

?
for me it is also relevant.

Ilze