On Thu, Mar 15, 2012 at 10:35 AM, S?bastien Bonthoux
<bonthoux.sebastien at gmail.com> wrote:
Dear D.Bates, I am using your package lme4 and the function lmer(). I link a metric of ecological community specialisation (gaussian distribution) with several land use variables and I add a random intercept because my plots are clustered. Can you explain me why I obtain a negative deviance (positive logLik) ? Is there any problem ?
Generally it is best to send questions like this to the R-SIG-Mixed-Models at R-project.org mailing list (see instructions at https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models). Many of those who read that list can reply and often much faster than I am able to. The short answer to your question is that a negative deviance for a model with a response measured on a continuous scale is not a problem. Probability mass functions for discrete random variables cannot exceed 1 but probability density functions for continuous random variables can. Thus the log-likelihood for a continuous response can be positive and the deviance negative.
Thank you for you reply. Best regards -- S?bastien Bonthoux Docteur en Ecologie - PhD in Ecology 02 54 78 05 74 Ecole Nationale Sup?rieure de la Nature et du Paysage 9 rue de la chocolaterie 41 000 Blois