[R-sig-ME] increasing maximum iterations in glmmadmb function

[Ben Bolker]

[Please keep r-sig-mixed-models in the Cc: list ...]

Nothing stands out particularly strongly from this summary; the one
guess would be, since you have rather large parameter estimates
(abs(beta)>5) in a binomial-response model, that you are encountering
a problem of complete separation (i.e., that some combinations of
predictor variables have all- (or nearly-all)-negative or all-positive
responses, leading to infinite parameter estimates under MLE).
However, in my experience usually such cases give even larger
parameter estimates (e.g. abs(beta)>8).

Since this particular model is fairly standard (binomial GLMM,
standard logit link, no zero-inflation) it should be fairly easy to
run it with another platform/package (MASS::glmmPQL, lme4::glmer,
glmmML, MCMCglmm ...) and see if you get similar answers.


On Wed, Jan 27, 2016 at 3:11 PM, Reuben Smit <smit.reuben at gmail.com> wrote:
> Thanks for the help query, that is what I was looking for.
>
> The warning does not include anything but "Convergence failed: log
> likelihood -0.1037". You were right about increasing the iterations not
> helping, though.
>
> The following is the model summary, if you notice anything immediately,
> please let me know, otherwise I will continue with model diagnostics, etc.
> This model is a set of models for model based hypotheses. Thanks again for
> any help you can offer.
>
>
> Call:
> glmmadmb(formula = A_Neis_PA ~ HMU + Dist2bank + DistAC + Sinu +
>     (1 | Site_fact), data = mussel, family = "binomial")
>
>
> Coefficients:
>             Estimate Std. Error z value Pr(>|z|)
> (Intercept)   6.3850     2.0309    3.14  0.00167 **
> HMUB1         2.6482     0.7720    3.43  0.00060 ***
> HMUB2         2.9078     0.8030    3.62  0.00029 ***
> HMUC         -0.4973     1.2842   -0.39  0.69856
> HMUD          4.2994     1.0260    4.19  2.8e-05 ***
> Dist2bank    -0.1279     0.0337   -3.79  0.00015 ***
> DistAC        0.0165     0.0418    0.40  0.69220
> Sinu         -5.7189     1.6773   -3.41  0.00065 ***
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Number of observations: total=259, Site_fact=20
> Random effect variance(s):
> Group=Site_fact
>             Variance StdDev
> (Intercept)   0.2957 0.5438
>
>
> On Wed, Jan 27, 2016 at 10:53 AM, Ben Bolker <bbolker at gmail.com> wrote:
>>
>>
>>    Did you look at ?admbControl  ?
>>
>>    "failing to converge" is a pretty broad category; if for example the
>> warning message says something about a non-positive-definite Hessian at the
>> optimum, increasing the number of iterations may not help that much ...
>>
>>
>>
>> On 16-01-27 12:53 PM, Reuben Smit wrote:
>>>
>>> My binomial mixed model is failing to converge, so I would like to
>>> increase
>>> the number of iterations. However, upon research I cannot find
>>> solutions for glmmadmb function specifically. I must be missing
>>> something. Any guidance would be appreciated.
>>>
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>