[R-sig-ME] logLik (old-fashion way) for mixed-effects models

[Christian Salas]

Dear Prof. Bates:

I am aware of the logLik(lme.obj),  sorry if i was not clear before.
What I am aiming to find is a similar syntax (to the one that i use for 
lm) involving the residuals from a lme fitted object, but without using 
logLik(lme.obj), that allows me to compute the log-likelihood value of 
the fitted model. Probably this would require retrieving both the ML 
sigma for the errors and the sigma for the random effects of the lme.obj

thanks
c

-------------------------------------------------------------------------------
Christian Salas                     E-mail:christian.salas at yale.edu
PhD candidate                       http://environment.yale.edu/salas
School of Forestry and Environmental Studies
Yale University                     Tel: +1-(203)-432 5126
360 Prospect St                     Fax:+1-(203)-432 3809
New Haven, CT 06511-2189            Office: Room 35, Marsh Hall
USA

Yale Biometrics Lab                  http://environment.yale.edu/biometrics
-------------------------------------------------------------------------------





Douglas Bates wrote:
> On Thu, Apr 16, 2009 at 4:03 AM, Christian Salas
> <christian.salas at yale.edu> wrote:
>> If i already fit a lm() model, i can obtain the log-likelihood [i do not
>>  want to use AIC()] using the residuals from the model, and using the
>>  RMSE of the model as sigma for my normal pdf. This would be in R like
> 
> But the RMSE is not the maximum likelihood estimate of sigma.  It's
> the REML estimate but not the MLE.
> 
>>> sum(dnorm(-resi,mean=0,sd=sigma,log=T))  [1]
>>  if i fit a gls model i can do the same
>>
>>  for a lme() model, i know that we cannot just use the same loglik model
>>  [1], because they are different. I wonder if somedody already have some
>> syntax in R similar to [1] but for  mixed-effects models, i mean something
>> that compute the log-likelihood but without using lme() directly as
>> summary(lme.obj)$AIC
> 
> Assuming that you fit with method = "ML" then wouldn't it be simplest
> just to use
> 
> logLik(lme.obj)
> 
> It isn't clear from your question whether you want another approach
> involving residuals, etc. or if it is just the fact that you are not
> aware of the logLik generic.
> 
>>  thanks in advance!
>>
>> -------------------------------------------------------------------------------
>> Christian Salas                     E-mail:christian.salas at yale.edu
>> PhD candidate                       http://environment.yale.edu/salas
>> School of Forestry and Environmental Studies
>> Yale University                     Tel: +1-(203)-432 5126
>> 360 Prospect St                     Fax:+1-(203)-432 3809
>> New Haven, CT 06511-2189            Office: Room 35, Marsh Hall
>> USA
>>
>> Yale Biometrics Lab                  http://environment.yale.edu/biometrics
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>