[R-sig-ME] Advice on model selection procedure in glmer

[David Duffy]

On Thu, 5 Mar 2009, Achaz von Hardenberg wrote:

> Dear list members,
>
> I am analyzing data on Golden eagle productivity (PROD = number of
> chicks fledging for each eagle couple each year) to see how it is
> influenced by various meteorological factors, Nearest neighbor
> distance (NND) and reproductive success the year before (back1) and
> two years before (back2). for each nest/couple (Nterr) we have one
> PROD measure for each year of the study (about 20 years). PROD varies
> between 0 and 3.

>    AIC   BIC logLik deviance
>  246.7 271.8 -116.4    232.7
> Random effects:
>  Groups Name        Variance  Std.Dev.
>  anno   (Intercept) 0.0056248 0.074999
>  Nterr  (Intercept) 0.0324398 0.180111
> Number of obs: 266, groups: anno, 24; Nterr, 22
>
> Fixed effects:
>              Estimate Std. Error
> (Intercept) -1.016941   0.279004
> back1       -0.067498   0.146796
> back2        0.181103   0.136243
> NND          0.029392   0.053093
> Pcrescita    0.003070   0.003008
>
> removed 	df AIC	   BIC	  Loglik
> FULL		7  246.73  271.82 -116.37
> Back1		6  257.79  279.68 -122.89
> Back2		6  259.03  280.97 -123.52
> NND		6  281.81  304.34 -134.91
> Pcrescita	6  276.86  299.16 -132.43
>
> I would really like to know if you think this model selection
> procedure is correct.

I don't pretend to have any deep understanding, but it seems plausible to 
me.  I am curious about your including Back1 and Back2 as fixed effects, 
given they are partly measures of the "couple fitness" latent variable 
NTerr.  Do you actually need Nterr if these are included?  And since PROD 
has so few levels, you can check whether Poisson is a good representation, 
and how your predictions are (crossvalidation), maybe compare 
results from gamm (in the mgcv package). Are you interested in 
just predicting PROD for next year?

David Duffy.
-- 
| 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