Brian Ripley sometimes on this list or elsewhere suggested to reparametrize as 1/k. I have used that with good results. But you should be aware that usually data contains very little information about k, so thhat if you do not have a lot more than 100 observations you coukld be out of luck. You should try to plot the likelihood as a function of k, possibly also the profile likelihood.
Kjetil On Thu, Dec 10, 2009 at 6:06 PM, Barbara Gonzalez <barbara.p.gonza...@gmail.com> wrote: > Thank you. > > I actually found fitdistr() in the package MASS, that "estimates" the > df, but it does a very bad job. I know that the main problem is that > the t distribution has a lot of local maxima, and of course, when > k->infty we have the Normal distribution, which has nice and easy to > obtain MLEs. > > I will try re-parametrizing k, but I doubt this will solve the problem > with the multiple local maxima. > > I would like to implement something like the EM algorithm to go around > this problem, but I don't know how to do that. > > Barbara > > On Thu, Dec 10, 2009 at 2:59 PM, Albyn Jones <jo...@reed.edu> wrote: >> k -> infinity gives the normal distribution. You probably don't care >> much about the difference between k=1000 and k=100000, so you might >> try reparametrizing df on [1,infinity) to a parameter on [0,1]... >> >> albyn >> >> On Thu, Dec 10, 2009 at 02:14:26PM -0600, Barbara Gonzalez wrote: >>> Given X1,...,Xn ~ t_k(mu,sigma) student t distribution with k degrees >>> of freedom, mean mu and standard deviation sigma, I want to obtain the >>> MLEs of the three parameters (mu, sigma and k). When I try traditional >>> optimization techniques I don't find the MLEs. Usually I just get >>> k->infty. Does anybody know of any algorithms/functions in R that can >>> help me obtain the MLEs? I am especially interested in the MLE for k, >>> the degrees of freedom. >>> >>> Thank you! >>> >>> Barbara >>> >>> ______________________________________________ >>> R-help@r-project.org mailing list >>> https://stat.ethz.ch/mailman/listinfo/r-help >>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html >>> and provide commented, minimal, self-contained, reproducible code. >>> >> > > ______________________________________________ > R-help@r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.