[ADMB Users] help with glmmADMB 0.6.4 - function maximizer failed

[Ben Bolker]

On 09/23/2011 10:39 AM, Rafael Mares wrote:
> Thank you very much for your replies.
> 
> Ben, is a lognormal Poisson in lme4 the "same" as adding an
> observation level random effect to a Poisson model? 

  Yes.

> If that's the
> case, I can run my model this way, with both random effects and all
> the data, no problem. The estimates seem reasonable to me.

  Because neg binomial "type 2" (i.e. parameterized as V=mu*(1+mu/k))
has the same mean-variance relationship as the lognormal-Poisson (see
Elston et al 2001), I would generally expect the LNP and NB2 results to
be *reasonably* similar.

> 
> The output from the glmmadmb model unfortunately doesn't mean much to
> me. But this is what it says right before the error:
> 
>  - final statistics:
> 9 variables; iteration 16; function evaluation 21
> Function value   8.0326e+03; maximum gradient component mag  -2.7007e-05
> Exit code = 1;  converg criter   1.0000e-04
> Var   Value    Gradient   |Var   Value    Gradient   |Var   Value    Gradient
>  1-26.04528  3.46314e-06 |  2-11.11848 -4.63363e-06 |  3  6.15104  1.90787e-06
>  4-18.30688  2.76734e-06 |  5-26.83680  4.50334e-06 |  6  5.36753 -1.32630e-05
>  7  5.04109 -2.67531e-06 |  8  6.48699  6.56383e-06 |  9 -0.09684 -2.70066e-05
> Hessian type 4
>  inner maxg = 0.0009837  Inner second time = 0.0009837  Inner f = 7024
>  f = 7024.448328266865 max g = 0.0009836751948745226
> Newton raphson 1  Error in glmmadmb
> 
> 
> The data seem ok to me, but I will try partitioning the data as you
> suggest to see if there are any particular data points causing
> problems. Thank you for your advice.
> 

  If I were you I would take Dave Fournier up on his offer, too.

  Ben