Hello all,
Is AIC calculated incorrectly in lmer? It appears as though it uses
AIC = -2*logLik - 2*#parms, instead of -2*LogLik + 2*#parms? Below is
output from one of many models I have tried:
Generalized linear mixed model fit using PQL
Formula: cswa ~ pcov.ess1k + (1 | year)
Data: ptct50.5
Family: poisson(log link)
AIC BIC logLik deviance
224.8466 219.19 -114.4233 228.8466
Random effects:
Groups Name Variance Std.Dev.
year (Intercept) 0.0062643 0.079147
# of obs: 125, groups: year, 2
Estimated scale (compare to 1) 1.277183
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.1059628 0.1283976 -0.82527 0.4092
pcov.ess1k 0.0101182 0.0093962 1.07683 0.2816
A snip of my data:
cswa pcov.ess250 year
[1,] 4 7.14 2004
[2,] 4 19.26 2003
[3,] 1 3.66 2004
I'm using R 2.1.1 with Windows XP.
Thanks,
Richard
--
Richard Chandler
Department of Natural Resources Conservation
UMass Amherst
(413)545-1237
The calculation is being done in
print(data.frame(AIC = AIC(llik), BIC = BIC(llik),
logLik = c(llik),
deviance = -2*llik,
row.names = ""))
where llik is defined as
llik <- object at logLik
so the important question is whether the logLik slot has the correct
values for the number of parameters.
By the way, why are you calculating a random effect for year when you
only have two years of data? The estimate of the variance of that
random effect will have almost no precision.
On 10/7/05, Richard Chandler <rchandler at forwild.umass.edu>
wrote:> Hello all,
>
> Is AIC calculated incorrectly in lmer? It appears as though it uses
> AIC = -2*logLik - 2*#parms, instead of -2*LogLik + 2*#parms? Below is
> output from one of many models I have tried:
>
> Generalized linear mixed model fit using PQL
> Formula: cswa ~ pcov.ess1k + (1 | year)
> Data: ptct50.5
> Family: poisson(log link)
> AIC BIC logLik deviance
> 224.8466 219.19 -114.4233 228.8466
> Random effects:
> Groups Name Variance Std.Dev.
> year (Intercept) 0.0062643 0.079147
> # of obs: 125, groups: year, 2
>
> Estimated scale (compare to 1) 1.277183
>
> Fixed effects:
> Estimate Std. Error z value Pr(>|z|)
> (Intercept) -0.1059628 0.1283976 -0.82527 0.4092
> pcov.ess1k 0.0101182 0.0093962 1.07683 0.2816
>
>
> A snip of my data:
>
> cswa pcov.ess250 year
> [1,] 4 7.14 2004
> [2,] 4 19.26 2003
> [3,] 1 3.66 2004
>
> I'm using R 2.1.1 with Windows XP.
>
> Thanks,
> Richard
>
> --
> Richard Chandler
> Department of Natural Resources Conservation
> UMass Amherst
> (413)545-1237
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
R-project.org/posting-guide.html
>
Hi, my reply just concerns the usage of AIC in mixed models and not the lmer package. The "standard" AIC is actually unconditional. Vaida and Blanchard (2003, Proceeding 19 IWSM,101-105) discuss that a "conditional" version should be more appropriate in a mixed framework. I don't whether the paper has been pubblished elsewhere. regards, vito Richard Chandler wrote:> Hello all, > > Is AIC calculated incorrectly in lmer? It appears as though it uses > AIC = -2*logLik - 2*#parms, instead of -2*LogLik + 2*#parms? Below is > output from one of many models I have tried: > > Generalized linear mixed model fit using PQL > Formula: cswa ~ pcov.ess1k + (1 | year) > Data: ptct50.5 > Family: poisson(log link) > AIC BIC logLik deviance > 224.8466 219.19 -114.4233 228.8466 > Random effects: > Groups Name Variance Std.Dev. > year (Intercept) 0.0062643 0.079147 > # of obs: 125, groups: year, 2 > > Estimated scale (compare to 1) 1.277183 > > Fixed effects: > Estimate Std. Error z value Pr(>|z|) > (Intercept) -0.1059628 0.1283976 -0.82527 0.4092 > pcov.ess1k 0.0101182 0.0093962 1.07683 0.2816 > > > A snip of my data: > > cswa pcov.ess250 year > [1,] 4 7.14 2004 > [2,] 4 19.26 2003 > [3,] 1 3.66 2004 > > I'm using R 2.1.1 with Windows XP. > > Thanks, > Richard >-- ===================================Vito M.R. Muggeo Dip.to Sc Statist e Matem `Vianelli' Universit?? di Palermo viale delle Scienze, edificio 13 90121 Palermo - ITALY tel: 091 6626240 fax: 091 485726/485612