Different results between procD.lm and morphol.disparity

[Mera]

Dear all,

I have a somewhat basic question, but I don't seem to find a satisfactory answer by browsing literature.

The issue is that I'm getting different results between the results of the model in procD.lm and the pairwise of the groups (morphol.disparity). I only have two groups of individuals, so my rationale is that if my model says that there is an effect of my factor (significant factor), my two groups should have different shapes. However, when I do the morphol.disparity (or manual pairwise through "pairwise" and "test.type = "var"), the results are far from significant. Furthermore, if I analyze the data through PCA decomposition and MANOVA with the resulting PCs, I do find a significant pairwise comparison (and model). 

This has happened in both ways (e.g. morphol.disparity finding significant differences while model and PCA-MANOVA do not).

 

Can someone point me to why is happening this? which is the most "reliable" result or the place where I should look to find an explanation, would be most appreciated :). I paste below the relevant parts of my R script. 

Kind regards,

Melina  

Mod<-procD.lm(Coords~Size+Mes,
              iter = 999, data = gdf)
summary(Mod)

PW<-pairwise(Mod, groups = gdf$Mes)
summary(PW, test.type="var",confidence=0.95)

morphol.disparity(Mod, groups = gdf$Mes, data = gdf, iter = 999)

pca<-gm.prcomp(gpa$coords) #PCA analisis
PCs<-pca$x  #guardar los pca scores

Vars<-as.matrix(PCs[,1:39])

Comp1<-manova(Vars~gdf$Mes+gdf$Size)
summary(Comp1)

[Mike Collyer]

Dear Melina,

You are performing completely different tests.  ANOVA from procD.lm is testing (in this case) whether the coefficients for Size and Mes are significantly different from 0.  For Mes, this is a test for whether mean shape between groups differ.  The pairwise tests you perform are testing whether the variances (not the means) are the same between groups.  In fact, if you are testing differences between means, the variances should be approximately the same, so the non-significant result affirms that you are not confounding location and dispersion in your inference.  (For example, if one group had huge variance and another group small variance, but the the location of all shapes in the small variance group were found within the spread of the large variance group, a significant result for the test between means could not be trusted.)

If you change to test.type = “dist”, then you should get commensurate results with the ANOVA. morphol.disparity does not test means in any capacity.  Disparity is another word for variance (dispersion of shapes).  Sometimes it is of more interest to infer whether one group has more variation in shape than another.

Hope this helps.

Mike

--
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, visit https://groups.google.com/d/msgid/geomorph-r-package/8bc689a2-ca92-4c55-b0a6-63aa3b06931dn%40googlegroups.com.

[Mera]

Dear Mike,

Thanks a lot for your answer!! <3

Melina