If you get such a large difference between ML and REML estimation in this sort of situation then probably either (1) something is wrong with the code (bad search algorithm?) or (2) you have something pathological in your tip data set and/or the tree (e.g., some really long singleton branches or really short sister branches). If you see a big outlier in the residuals, then this is a problem - I would not trust the results. How do the residuals look from a regular OLS analysis?
Cheers, Ted From: R-sig-phylo [r-sig-phylo-boun...@r-project.org] on behalf of Sergio Ferreira Cardoso [sff.card...@campus.fct.unl.pt] Sent: Saturday, April 18, 2015 10:59 AM To: Liam J. Revell Cc: r-sig-phylo@r-project.org Subject: Re: [R-sig-phylo] Non normal PGLS results Hello, Thank you very much for the answer. I assumed PGLS residuals had to be normal (I saw this thread: https://stat.ethz.ch/pipermail/r-sig-phylo/2012-May/002064.html). In fact, what I'm analysing are standardized residuals (standardized by setting the determinant of the covariance matrix equal to one). I thought something was wrong with the residuals because when estimating the Lamba using ML I get a completely different result from estimating it with REML (e.g., REML est: 0.945; M est: ~0). Do you have any idea of wat could be causing this? Once again, thank you for answering. Best regards, Sérgio. ᐧ 2015-04-18 18:22 GMT+01:00 Liam J. Revell <liam.rev...@umb.edu>: > Hi Sergio. > > We don't expect the residuals from PGLS to be independent draws from a > normal distribution, but multivariate normal with a correlation structure > given by the tree. > > Here I give some more explanation of this on my blog: > http://blog.phytools.org/2013/02/a-comment-on-distribution-of-residuals.html > . > > Let us know if this is helpful. > > All the best, Liam > > Liam J. Revell, Assistant Professor of Biology > University of Massachusetts Boston > web: http://faculty.umb.edu/liam.revell/ > email: liam.rev...@umb.edu > blog: http://blog.phytools.org > > > On 4/18/2015 1:15 PM, Sergio Ferreira Cardoso wrote: > >> Dear all, >> >> I'm performing PGLS's with an ultrametric phylogenetic tree (divergence >> time as branchlengths). I tied Pagel's Lambda transformation, OU >> transformation, regular GLS and OLS, to compare results. There is one >> problem: the residuals of my analyses are not normal. I tried to remove >> big >> outliers but it made things even worse, because without them there are >> even >> more outliers. >> >> So, what should I do? The results are consistent in both 4 tests. But can >> I >> trust the results of the PGLS? Is it particularly bad if PGLS residuals >> aren't normal? Does it critically afect my results? >> >> Thanks in advance. >> >> Best regards, >> Sérgio. >> >> -- Com os melhores cumprimentos, Sérgio Ferreira Cardoso. -------------------- Best regards, Sérgio Ferreira Cardoso MSc. Paleontology candidate Departamento de Ciências da Terra - FCT /Universidade Nova de Lisboa Geociências - Universidade de Évora Lisboa, Portugal [[alternative HTML version deleted]] _______________________________________________ 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/ _______________________________________________ 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/
