R help - Nov 2005 - AIC in lmer when using PQL

I am analysing binomial data using a generalised mixed effects model.  I
understand that if I use glmmPQL it is not appropriate to compare AIC
values to obtain a minimum adequate model.

 

I am assuming that this means it is also inappropriate to use AIC values
from lmer since, when analysing binomial data, lmer also uses PQL
methods.  However, I wasn't sure so please could somebody clarify this
for me.

 

I was also wondering how best to assess your minimum adequate model
without AIC values?  Do you simply have to rely on the p values
associated with the t-values/z-values?

 

Thanks very much.

Elizabeth Boakes

 

Elizabeth Boakes
PhD Student
Institute of Zoology
Regent's Park
London NW1 4RY
tel: 020 7449 6621 

 



_________________________________________________________________________
This e-mail has been sent in confidence to the named address...{{dropped}}

note that although PQL is the default method in lmer() for GLMMs, the 
recent version of the function allow also for Laplace or adaptive 
Gauss-Hermite approximations. In these cases it might be reasonable to 
compute AIC values depending on how good the approximation to the 
likelihood is; however, the use of AIC in mixed models can be tricky 
depending on the focus of your analysis, check e.g.,

Vaida, F. and Blanchard, S. (2005). Conditional Akaike information for 
mixed-effects models, Biometrika, 92, 351-370.

Regarding inference, I'd rely mainly on LRTs instead of Wald type 
p-values.


I hope it helps.

Best,
Dimitris

----
Dimitris Rizopoulos
Ph.D. Student
Biostatistical Centre
School of Public Health
Catholic University of Leuven

Address: Kapucijnenvoer 35, Leuven, Belgium
Tel: +32/(0)16/336899
Fax: +32/(0)16/337015
Web: http://www.med.kuleuven.be/biostat/
     http://www.student.kuleuven.be/~m0390867/dimitris.htm


----- Original Message ----- 
From: "Elizabeth Boakes" <Elizabeth.Boakes at ioz.ac.uk>
To: <r-help-request at stat.math.ethz.ch>; <r-help at
stat.math.ethz.ch>
Sent: Thursday, November 24, 2005 9:53 AM
Subject: [R] AIC in lmer when using PQL

>I am analysing binomial data using a generalised mixed effects model. 
>I
> understand that if I use glmmPQL it is not appropriate to compare 
> AIC
> values to obtain a minimum adequate model.
>
>
>
> I am assuming that this means it is also inappropriate to use AIC 
> values
> from lmer since, when analysing binomial data, lmer also uses PQL
> methods.  However, I wasn't sure so please could somebody clarify 
> this
> for me.
>
>
>
> I was also wondering how best to assess your minimum adequate model
> without AIC values?  Do you simply have to rely on the p values
> associated with the t-values/z-values?
>
>
>
> Thanks very much.
>
> Elizabeth Boakes
>
>
>
> Elizabeth Boakes
> PhD Student
> Institute of Zoology
> Regent's Park
> London NW1 4RY
> tel: 020 7449 6621
>
>
>
>
>
> _________________________________________________________________________
> This e-mail has been sent in confidence to the named\ > ad...{{dropped}}