[R-sig-ME] quasi-binomial family in lme4
john.maindonald at anu.edu.au
Wed Nov 10 09:50:50 CET 2010
I wonder if you have compared the results that you quote
with the result you get with observation level random effects
in a poisson model.
As I see it, use of observation level random effects should,
unless there is evidence that a multiplicative effect on the
scale of the response is a better fit, replace use of the quasi-
models in glm() as well as in generalised linear mixed models.
John Maindonald email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473 fax : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.
On 10/11/2010, at 6:39 PM, Gosselin Frederic wrote:
> Hi Florian,
> a different perpsective on the quasi-likelihood debate - that comes out sporadically on this list:
> (i) I globally agree with the previous repliers that a fully probabilistic solution looks better - at least aesthetically - than a quasi-likelihood;
> (ii) however, as I have already mentioned on the list (cf. below), earlier versions of lme4 give much more sensible results than the latest versions:
> This is why in the following papers:
> Elek Z., Dauffy-Richard & Gosselin F., 2010, Carabid species responses to hybrid poplar plantation in floodplains in France, Forest Ecology and Management, 260, 9, p. 1446-1455.
> Vuidot A., Paillet Y., Archaux F. & Gosselin F. (In Press) Influence of tree characteristics and forest management on tree microhabitats in France, Biological Conservation.
> we used version the R version 2.5.1 and the associated lme4 version (here with quasi-poisson, not quasi-bionomial).
> Hope this helps.
> Frédéric Gosselin
> Engineer & Researcher (PhD) in Forest Ecology
> Domaine des Barres
> F-45290 Nogent sur Vernisson
> [[alternative HTML version deleted]]
> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models