Re: [R] comparison of two logistic regression models

From: Prof Brian Ripley <ripley_at_stats.ox.ac.uk>
Date: Thu, 07 Jun 2007 18:10:38 +0100 (BST)

On Thu, 7 Jun 2007, David Barron wrote:

> You could use lmer in the lme4 package to fit the logistic regression > with random effect as it does report the AIC.

Indeed you could (lmer reports _an approximation_ to the AIC), but AIC comparison between these two models is not valid as whereas they are tested, the smaller model is on the boundary of the parameter space for the larger one, and that violates one of the assumptions in the derivation of AIC. From simulation studies I have heard seminar talks about, it makes a large practical difference as well.

Withou knowing why Anna-Maria wants to 'compare two models', I could not begin to offer advice. Generally one should test how well each does the task to hand, whatever that is.

> On 07/06/07, Anna-Maria Tyriseva <anna-maria.tyriseva_at_helsinki.fi> wrote:

>> Dear list members!
>>
>> Could you help me?
>> I would like to compare two models: a) logistic regression model, 3
>> factors as independents b) logistic regression model, 3 factors and one
>> random effect as independents (function glmmPQL). AIC are not available
>> with PQL and model comparison using ANOVA is not possible. What should I do?
>>
>> Thanks in advance.
>> Anna-Maria
-- 
Brian D. Ripley,                  ripley_at_stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

______________________________________________
R-help_at_stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Received on Thu 07 Jun 2007 - 17:16:05 GMT

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.2.0, at Thu 07 Jun 2007 - 17:31:47 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help. Please read the posting guide before posting to the list.