[R-sig-eco] correlated residuals

[Ben Bolker]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 11-03-31 03:05 AM, Alexandre Villers wrote:
> Good morning,
> 
> I'm trying to fit a very simple model: the proportion of occupied nest
> boxes according to food availability on a time serie (37 years). The
> problem is that the glm (binomial error family with logit  link)
> presents a marked overdispersion ( ~ 24). Checking the residuals showed
> me they are correlated with a lag of 3 years (preys have cyclic
> population dynamic) and accounting for this temporal correlation will
> certainly improve the model.
> I know how to cope with this with glmmPQL (with a corARMA correlation
> structure) but I "cheat" a bit given I don't really need random effect.
> 
> Is there a way to specify such correlation structure with gee or geepack
> ? Any link to a pdf or webpage explaining the "how to" would be
> appreciated...

  I don't know the answer to that question, but if the phase of the
cycle is consistent through the whole time period you could add a
sinusoidal component (with known phase and period) to the model ...

  Ben Bolker
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk2Ub+UACgkQc5UpGjwzenMi9wCfTurGG+rfLsuunzheRu6PBjoF
5F4AoIrwPKXt+z5BN1Js8ZitUlTrgKw4
=qImj
-----END PGP SIGNATURE-----