0
$\begingroup$

Say I am running a phylogenetic generalised least squares (PGLS) regression with one predictor.

A model will return me a coefficient with a standard error - corrected for phylogenetic relatedness.

I also know there is uncertainty surrounding the phylogeny is use, so I run the same model multiple times using a sample of phylogenies from a MCMC generated posterior.

Is there a way to combine the standard errors from each model (which will presumably be slightly different from model to model) with the phylogenetic error from across all models?

$\endgroup$
  • $\begingroup$ have you considered doing your analysis in a Bayesian framework? This is easy to do using the BayesTraits program provided by Pagel and Meade here evolution.rdg.ac.uk/BayesTraitsV3.0.1/BayesTraitsV3.0.1.html. Also see this paper biomedcentral.com/content/pdf/1471-2148-12-102.pdf $\endgroup$ – Slow loris Dec 2 '17 at 15:25
  • $\begingroup$ BayesTraits doesn't offer the solution I want in this case unfortunately. $\endgroup$ – SamPassmore Dec 5 '17 at 10:48
  • $\begingroup$ Why not? It integrates across a Bayesian posterior distribution of trees, which it sounds like is exactly what you have. The resulting posterior distributions of parameter estimates account for phylogenetic uncertainty. $\endgroup$ – Slow loris Dec 6 '17 at 2:54
  • $\begingroup$ I am aware of BayesTraits capabilities. It does not implement the particular model that I use. I would like to know how to combine these errors in a frequentist framework. $\endgroup$ – SamPassmore Dec 6 '17 at 9:01
  • $\begingroup$ you said PGLS with one predictor, which BayesTraits most certainly does implement... $\endgroup$ – Slow loris Dec 6 '17 at 11:42
0
$\begingroup$

If you're opposed to doing this in a Bayesian framework, you could fit the model for each tree in your distribution and combine the results with model averaging. See this R tutorial accompanying the book chapter Multi-model inference in comparative analyses by Laszlo Garamszegi and Robert Mundry.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.