[ADMB Users] Integrating R and ADMB to generalize TPL files

Ben Bolker bbolker at gmail.com
Wed Nov 7 10:39:09 PST 2012


On 12-11-07 01:13 PM, dave fournier wrote:
> I thought it might be interesting to model the increase in variability by
> using a different overdispersion parameter for each period. The results
> are.
> 
> 
> # Number of parameters = 8  Objective function value = 254.325 Maximum
> gradient component = 8.59709e-05
> # pz:
> 0.000100000000000
> # beta:
>  -10.8175827094 -2.23582249148 -3.42881880705 8.42578854173
> # tmpL:
>  0.436593520224
> # tmpL1:
>  0.000100000000000
> # log_alpha:
>  0.826338727816 0.0677913813376 0.657156858701
> 
> So this improves the fit better than using the extra parameters
> for the variances of the RE's. I'm not sure what is means though.
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users


  That had occurred to me as a possibility, but I hadn't gotten around
to trying it.

  What it means is that it is better to suppose that the additional
variability occurs at the *observation* level, varying both among groups
and among observations within groups, than at the group level ... it may
be nice to use a lognormal-Poisson for this rather than a negative
binomial (along the lines of Elston, D. A. and R. Moss and T. Boulinier
and C. Arrowsmith and X. Lambin. 2001. Analysis of aggregation, a worked
example: numbers of ticks on red grouse chicks. Parasitology 122(5):
563-569 doi:10.1017/S0031182001007740 ) because it allows a
decomposition of variance between the group and observation levels.

  Ben




More information about the Users mailing list