Fitting models with REML.

[pbastide]

This is a first attempt to add an REML option to the phylolm fit. I implemented this because I needed it in a project, and I figured I might share it with you. But maybe it does not fit within your development plan, and you'd prefer to stay with the curent "ML only" implementation? No problem if that is the case, just let me know.

Changes:

• definition of the internal logLik function changes when REML is TRUE.
• vcov is changed accordingly : even in the ML case, the REML estimate of sigma2 is taken (as was previously).
• tests of the REML and ML estimate against the gls function, for BM, lambda and OU models.

Questions:

• Is there anything to change elsewhere ? E.g. in the bootstrap code, or in other functions ?
• Would there be some more tests to add ? E.g. against other implementations, such as Rphylopars or mvMORPH ?
• Are there cases when this REML implementation would fail ? E.g. for non ultrametric trees ?

Thank you !

[rosbakh]

I no longer use this e-mail address and cannot guarantee that I will respond to your message in time.

Please use following address for future correspondence: @.***

---

Dr. Sergey Rosbakh

Ecology and Conservation Biology

Institute of Plant Sciences

Faculty of Biology and Preclinical Medicine

University of Regensburg

D-93040 Regensburg Germany

Phone: +49-(0)941-943-3105

Fax: +49-(0)941-943-3108 Email: @.***

[lamho86]

Thanks, Paul.

To answer your questions:

• Probably not. Bootstrap should still work.
• The test you added is sufficient.
• I can't think of any.