[R] Multilevel in R

[Douglas Bates]

"Fabrizio Consentino" <consentino at infinito.it> writes:

> Hello, 
> 
> I have difficulties to deal with multilevel model. My dataset is composed 
> of 10910 observations, 1237 plants nested within 17 stations. The data set is not 
> balanced. Response variable is binary and repeated.
> 
> I tried to fit this model
> 
> model<- glmmPQL( y ~ z1.lon*lun + z2.lat*lun + z1.lon*lar + z2.lat*lar + z1.lon*sca + z2.lat*sca +z1.lon*eta + z2.lat*eta,
>  random = ~ lun + lar + sca + eta | sta/piante, family=binomial, data=variabili)
> 
> where  y is presence (1) or absence (0) of a flowering
> 
> lun, lar, sca, eta are level 1 variables
> 
> z1.lon, z2.lat are level 2 variables.
> 
> but during third iteration it stop because there is a singular matrix in solve.
> 
> I stopped it after two iterations, however the results are not correct.
> 
> How can I fit this data? Are there other functions that I can use?
> 
> I would be thankfull for all the insights.

Start with a simpler model.  Try random = ~ 1 | sta/piante and see if
that converges.

You could also try function GLMM from the lme4 package.

-- 
Douglas Bates                            bates at stat.wisc.edu
Statistics Department                    608/262-2598
University of Wisconsin - Madison        http://www.stat.wisc.edu/~bates/