[R-sig-ME] convergence problems and model averaging

[David Duffy]

On Tue, 17 Jan 2012, Diego Bilski wrote:

> Dear list,
>
> I have 1340 observations of animal survival (binomial), and 10 variables in
> my full model (with no interactions between variables). Using lmer() this
> full model didn't converged, then I removed 2 variables that were highly
> correlated (r=0.75 and 0.49) with two other, and this model presented no
> convergence problems.
>
> I standardized variables dividing by 2SD, in the "arm" package, and used
> the dredge() function of "MuMIn" to perform all combinations of variables,
> then selected those with deltaAIC <= 4. But from the 27 models with this
> deltaAIC, 19 presented false convergence warnings.

So, the 8 variable model works, but submodels of this are failing to 
converge?  If you have false convergence warnings, and you are interested 
in that model, you will need to crosscheck that solution using another 
package.  Since you just have a simple random intercept model, you can use
glmmML, glmmADMB, hglm...

Just 2c, 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