pairwise phylogenetic test in Geomorph?

[chip...@museum.vic.gov.au]

Hi all,

Is it possible to run a post hoc pairwise phylogenetic comparison of group means in Geomorph similar to the advanced.procD.lm function? I get a significant result for shape variation among groups using both the standard (Procrustes) and phylogenetic ANOVA, so now I would like to know which groups differ from each other. If it can't be run in a single command, could I simply set up the procD.pgls myself for each pairwise comparison? If so, can the groups be specified in the command line somehow or I would need to prune the data set for each comparison? 

Thanks for any help,

Christy

[Michael Collyer]

Christy,

We do not currently have a function to do this, although we have one in development.  I know this is something you would like to do, but to date, nobody has examined the statistical properties of phylogenetic comparative methods (more generally and specifically, generalized least squares estimation) as applied to pairwise tests.  As soon as that happens in a formal way (the research is published), we will be eager to add such a function to our toolbox.

Because procD.pgls provides output that includes the phylogenetic correction (transformation) matrix, fitted values, residuals, and model design matrix, it is possible to create your own test for pairwise comparisons using the same elements we use in pairwise tests in advanced.procd.lm.  For example, one can randomize PGLS residuals and add them to PGLS fitted values in a permutation procedure, in order to generate distributions of pairwise differences.  This is logical, but again, we just do not know if such a procedure has appropriate statistical error rates.

Sorry this is not something we offer, but it is not because we have overlooked it - the theory just has not been developed yet.

Cheers!

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 post to this group, send email to geomorph-...@googlegroups.com.
Visit this group at https://groups.google.com/group/geomorph-r-package.
To view this discussion on the web, visit https://groups.google.com/d/msgid/geomorph-r-package/1728de68-73f6-40c4-9c43-8152264a2acc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Michael Collyer]

Hi Christy,

I think the important first question is how does PGLS change the conclusions you would make with OLS?  You can compare coefficients, or compare SS, MS, or R2 in your ANOVA tables, or maybe compare fitted values from both.  Here’s an idea, use two.b.pls to see how associated fitted values are from each.  If this correlation is high, or the ANOVA tables are similar (i.e., significant effects did not become non-significant after PGLS), then you can do the normal OLS pairwise comparisons with the ability to indicate that it’s okay to ignore phylogeny (if that is what your examination indicates).  If your effects were rendered non-significant due to phylogeny, there would be little sense in pairwise comparisons, anyway.  

To my way of thinking, phylogenetic “correction” is a first-order problem.  One does PGLS to ascertain if the same interpretations can be made after adjusting OLS coefficients (does ANOVA tell us the same thing?)  I suppose that there is a case where a significant ANOVA from OLS leads to e.g., 4 out of 6 pairwise comparisons that are significant and the still significant effect in PGLS might lead to only 2 out of 6 significant pairwise comparisons.  But in most cases, it is probably sufficient to simply ask if significant sources of variation remain significant after PGLS?  If so, go back to OLS and break things down further.

You mention between-group PCA.  This can be done easily with PGLS-fitted values, just like OLS-fitted values.  This is analogous to doing pairwise comparisons in a visual sense - just no P-values added to the distances between groups.  Caution!  Do not attempt to use PC scores from such an analysis for pairwise tests though.  This test would confound phylogeny and group differences in generating probability distributions.  We discussed this problem here: http://www.public.iastate.edu/~dcadams/PDFPubs/2015-Evol-AdamsCollyer-PIC-PGLS%20Shuffle.pdf

You also mention CVA… I would stay away from this in the PGLS sense.  One can get canonical axes after PGLS, project data onto them, and make plots, and that is fine.  But if one takes it a step further and tries to classify subjects, trouble could ensue, especially if using canned CVA/DFA programs.  The lda function in R is not so bad, because one can input prior probabilities.  I believe the CVA function in MorphoJ, for example, does not allow one to alter prior probabilities (it either assumes equal priors or bases them on group sizes - maybe somebody can correct me if I am wrong).  But it would be unfortunate to not have priors informed by phylogenetic relatedness.  We discussed this a bit here: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0146166

good luck!

Mike

On Sep 28, 2016, at 6:24 PM, Hipsley, Christy <chip...@museum.vic.gov.au> wrote:

Hi Mike,

 

Thanks for your response. It sounds like the test will have to wait. I’m only concerned that a reviewer will immediately demand it, since I get the impression that everyone expects phylogenetic comparative tests now even without knowing what they really produce. I can’t say how many times I’ve been asked to run a phylogenetic PCA, even though it’s been shown that it doesn’t do at all what one expects (create independent axes, relate back to the shape variables, etc). I could show the standard pairwise comparisons instead, plus I also have CVA and between-group PCA to identify where and what the main differences between groups are. Hopefully that will suffice!

 

Thanks again,

Christy

 

This e-mail is solely for the named addressee and may be confidential. You should only read, disclose, transmit, copy, distribute, act in reliance on or commercialise the contents if you are authorised to do so. If you are not the intended recipient of this e-mail, please notify postm...@museum.vic.gov.au by email immediately, or notify the sender and then destroy any copy of this message. Views expressed in this email are those of the individual sender, except where specifically stated to be those of an officer of Museum Victoria. Museum Victoria does not represent, warrant or guarantee that the integrity of this communication has been maintained nor that it is free from errors, virus or interference.

 

-- 
You received this message because you are subscribed to a topic in the Google Groups "geomorph R package" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/geomorph-r-package/wLLhEgjEPrE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to geomorph-r-pack...@googlegroups.com.

To post to this group, send email to geomorph-...@googlegroups.com.
Visit this group at https://groups.google.com/group/geomorph-r-package.

To view this discussion on the web, visit https://groups.google.com/d/msgid/geomorph-r-package/5782FAD1-8908-4474-B9C4-000CD1D6098B%40gmail.com.