Message-ID: <Pine.LNX.4.64.1103190800180.31686@orpheus.qimr.edu.au>
Date: 2011-03-18T22:20:16Z
From: David Duffy
Subject: lmm WITHOUT random factor (lme4)
In-Reply-To: <AA818EAD2576BC488B4F623941DA74270398A0@inbomail.inbo.be>
On Fri, 18 Mar 2011, ONKELINX, Thierry wrote:
> What worries me is that the loglikelihood of a lm() model and the
> equivalent gls() model is different. Although both models should be
> mathematically identical. Assuming that the loglikelihood is calculated
> on the same way within a package, I therefore have more confidence in
> comparing two models from the same package, thus gls() versus lme().
> Furthermore, I get an error when doing an anova between a lm() and a
> lme() model.
> logLik(fm)
'log Lik.' -950.1465 (df=3)
> logLik(fm, REML=T)
'log Lik.' -946.8318 (df=3)
> anova(fm1, fm0, fm)
Model df AIC BIC logLik Test L.Ratio p-value
fm1 1 4 1794.465 1807.192 -893.2325
fm0 2 3 1899.664 1909.209 -946.8318 1 vs 2 107.1986 <.0001
fm 3 3 1899.664 1909.209 -946.8318
--
| David Duffy (MBBS PhD) ,-_|\
| email: davidD at qimr.edu.au ph: INT+61+7+3362-0217 fax: -0101 / *
| Epidemiology Unit, Queensland Institute of Medical Research \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia GPG 4D0B994A v