Unexpectedly high P values from pairwise tests on procD.lm

38 views
Skip to first unread message

Cruise Speck

unread,
Mar 15, 2023, 9:34:42 AM3/15/23
to geomorph R package
Hello,

I used procD.lm to create a model with my guppy body shape data, looking at the effects of population and treatment on shape, with log centroid size as a covariate.

fullModel <- procD.lm(guppyShape ~ log(CSize) * population * treatment, RRPP=TRUE, verbose = TRUE, iter = 9999)

The model summary shows the below output:

Screenshot 2023-03-15 140335.png

I then wanted to do a pairwise analysis to find out which population/treatment combinations differed significantly in shape. I used the below code:

gp <- interaction(population, treatment)
PW <- pairwise(fullModel, groups = gp, covariate = NULL)
summary(PW, test.type = "dist", confidence = 0.95, stat.table=TRUE)

Surprisingly, none of resulting P values were significant (most somewhere between 0.3 and 0.9). The same procD model as well as pairwise tests had already been done some years ago (I'm just reperforming some of these analyses), but they yielded very different results, with most of the pairwise P values below 0.05. However, they were done in an older version of geomorph and I don't know exactly what code was used to reach that output.

I wanted to know if there was anything I had done wrong here, or why the pairwise function might give me such high P values despite highly significant effects of population and treatment appearing in the model summary?

Any help is much appreciated and please let me know if any more info is required.

Thank you,

Cruise

Adams, Dean [EEOB]

unread,
Mar 17, 2023, 9:31:19 AM3/17/23
to geomorph-...@googlegroups.com

Cruise,

 

Like evaluating linear models, evaluating pairwise comparisons also requires a careful consideration of the full and the reduced model. For instance, your model has numerous interactions which do not seem to contribute much to the overall explanation of variance. Some statistical camps recommend removal of these, to fit a ‘common slope’ (or common group) model. That is one issue.

 

Another then is against what reduced model are you comparing things for your pairwise comparisons? The help file for pairwise shows a few options and provides some discussion, but one must think critically about this before embarking on the task.


Hope this helps.


Dean

 

Dr. Dean C. Adams (he/him)

Distinguished Professor of Evolutionary Biology

Department of Ecology, Evolution, and Organismal Biology

Iowa State University

https://faculty.sites.iastate.edu/dcadams/

phone: 515-294-3834

--
You received this message because you are subscribed to the Google Groups "geomorph R package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to geomorph-r-pack...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/geomorph-r-package/14997e3a-129f-4a8c-8bf9-c4e1de7f811bn%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages