Hi Max, I think you raise a great question and would be interested to hear others' takes on it. My thinking is it would be sound to calculate the vifs from the same model run through glm. This is because the vif is a transform of the R2 value from regressing each independent variable on all the other independent variables. As such, the dependent variable does not actually affect the vif calculation. PGLS techniques model the phylogenetic autocorrelation in the residuals of the dependent variable, probably the vif from a glm with no autocorrelation parameter is OK. Certainly I would expect if you have collinearity that is too high among your predictors for a normal glm then it is too high for pgls. I have seen this empirically once in my own work.
Here is a SAS conversation about that arrives at largely the same conclusion. The conversation is about random effects models, but that is relevant since many autocorrelation models can be reformulated as random effects models, and the two have many statistical similarities: https://communities.sas.com/message/170160 Now, I suppose there may be a different effect of collinearity on parameter estimates in the context of autocorrelation, so perhaps there is some need for a more sophisticated approach. It doesn't appear to me, however, that such an approach has been worked out mathematically. Best Luke Luke J. Matthews | Senior Scientific Director | Activate Networks, Inc. ---------------------------------------------------------------------- Message: 1 Date: Fri, 3 Jan 2014 13:39:20 -0500 From: Max Farrell <maxwellfarr...@gmail.com> To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Method for calculating VIFs with a 'pgls' object? Message-ID: <CAL5RxDOOMZ6VuaJfSLFSVnsVa6dg7-pB=KR=baukfku_zim...@mail.gmail.com> Content-Type: text/plain Hello, I am interested in calculating variance inflation factors for a pgls model. Is this possible? If not, would it be statistically sound to create the same model in a glm framework to calculate VIFs? I found this post in the r-sig-phylo archive, but it isn't clear how the inflation factors were calculated: http://www.mail-archive.com/r-sig-phylo@r-project.org/msg02094.html Max [[alternative HTML version deleted]] ------------------------------ _______________________________________________ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo End of R-sig-phylo Digest, Vol 72, Issue 2 _______________________________________________ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
