dear all, a couple of month ago i have sent this message to [R-sig-phylo]:
>>> Le 02/11/2015 18:29, Antonella Soro a écrit : >>>> dear all, >>>> like kari allen in june 2011, i am trying to "to run Pagels GLS in R and I >>>> have consistently encountered >>>> an error message (see script and error message below). >>>> " >>>>> pagel.data<-corPagel(1,phy=gls.phylo1,fixed=FALSE) >>>>> pagel.gls8<-gls(Gst~log.Body_size+Diet+Behavior+Max_Geo+He,correlation=pagel.data,data=data,na.action=na.exclude, >>>>> method="ML") >>>> Error in corFactor.corStruct(object) : >>>> NA/NaN/Inf in foreign function call (arg 1) >>> i have now found out that the code above works for my data if i change the (initial) value of the parameter lambda from “1” to “0”. given this positive result, i am very much tempted to change that value and carry on with my analyses, but i wonder how entitled i am in making that change. what is the meaning of giving “1” as initial value of the parameter lambda, rather than any other value between “0” and “1"? and why would the outcome of the analysis depend so drastically (error vs not error) on that initial value? emmanuel paradies’s response to the same questions on the legitimacy of changing the initial value of the parameter lambda was: “Hi Antonella, I see no problem in changing the initial value of lambda as long as it gives an estimate. I do not know if there is a reason to expect a "regular" shape of the likelihood function for this kind of problem; so adjusting the initial values of the parameters to "help" the optimization algorithm is not a bad thing. You can send this question on r-sig-phylo; maybe others will have another opinion" any comment or a different opinion from the one of emmanuel is very much appreciated thank you in advance antonella _______________________________________________ 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/
