[R-sig-ME] stability of glmmADMB

[Ben Bolker]

Pau Boon Kee <keepauboon at ...> writes:

> 
> Dear all
> 
> I have an enquiry on the stability of the glmm AD model builder package in R.
> 
> I wonder if anyone can shed some light on the possible reasons as to 
> why the results, estimated coefficients, 
> generated by two different operating systems, window 7 and OS X, 
> are different to a certain extent.
> 
> To a certain extent I meant, the directions
>  of correlation are still the same, 
> but the figures can be quite significantly different.
> 
> The following are my results:
> 
> windows 7:
> 
> intercept: -2.245552***
> link_DIS_mm: -0.021536
> landlock: 1.294218*** 
> 
> OS X:
> 
> intercept: -2.103106**
> link_DIS_MM: -0.029175
> landlock: 1.195673**
> 

  It's hard to say in general.  I agree that these
differences are a bit too big to be ascribed to 
numerical 'fuzz' alone, but they don't look big
enough to cause big differences in the interpretation;
what does "quite significantly different" mean?
How big are the estimated standard errors of these
estimates? How big are the differences in the 
maximum log-likelihoods between the two different
fits?