[R-sig-ME] False convergence when running lmer with Zelig example for gamma.mixed

[Martin Schmettow]

Hi Hugo,

I've tried several times fitting a gamma model myself with lme4. Never
succeeded. 
Many others have reported similar problems fitting gamma models with lme4.
Seemingly, it never worked.
The easiest way of dealing with durations is log-transformation and then use
a Gaussian LMM.

CU, Martin.


> -----Original Message-----
> From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-
> models-bounces at r-project.org] On Behalf Of Hugo Mildenberger
> Sent: Tuesday, November 08, 2011 1:05 AM
> To: r-sig-mixed-models at r-project.org
> Subject: [R-sig-ME] False convergence when running lmer with Zelig example
> for gamma.mixed
> 
> Dear list members,
> 
> while testing Zelig example code related to mixed linear models, I got a
very
> strange result from "gamma.mixed". I therefore did run lmer directly to
see if
> the convergence problem was perhaps due to a parameter problem only,
> but it is not. Since the example should have been viable in the past, I'm
> wondering what is going wrong today? The only difference is the method
> specification, which was PQL, but PQL is not supported anymore. Using nAGQ
> > 1 makes no difference.
> 
>  # for coalition2 dataset only
>  require(Zelig)
>  data(coalition2)
> 
>  require(lme4)
> 
>  l.m <- lmer(duration ~ invest +
>                         fract +
>                         polar +
>                         numst2 +
>                         crisis + (1 | country),
>              data   = coalition2,
>              family = Gamma(link = log))
>  summary(l.m)
>  packageVersion('lme4')
> 
> Here is the output:
> 
> R version 2.13.1 (2011-07-08)
> [...]
> Warning message:
> In mer_finalize(ans) : false convergence (8)
> >  summary(l.m)
> Generalized linear mixed model fit by the Laplace approximation
> Formula: duration ~ invest + fract + polar + numst2 + crisis + (1 |
country)
>    Data: coalition2
>       AIC      BIC  logLik deviance
>  -3918661 -3918632 1959339 -3918677
> Random effects:
>  Groups   Name        Variance    Std.Dev.
>  country  (Intercept) 5.1269e+154 2.2643e+77
>  Residual             4.1745e+155 6.4610e+77
> Number of obs: 304, groups: country, 14
> 
> Fixed effects:
>               Estimate Std. Error t value
> (Intercept)  4.108e+00  8.941e+77       0
> invest      -3.882e-01  2.252e+77       0
> fract        4.980e-01  1.380e+75       0
> polar       -1.850e-02  1.017e+76       0
> numst2       3.696e-01  1.431e+77       0
> crisis       2.569e-02  4.590e+75       0
> 
> Correlation of Fixed Effects:
>        (Intr) invest fract  polar  numst2
> invest -0.020
> fract  -0.984 -0.043
> polar   0.227 -0.284 -0.262
> numst2 -0.439 -0.120  0.360 -0.039
> crisis -0.019 -0.283  0.000 -0.217  0.108
> >  packageVersion('lme4')
> [1] '0.999375.42'
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models