[ADMB Users] latest glmmadmb and p values

[Ben Bolker]

On 10-10-25 02:16 PM, dave fournier wrote:
> It seems that the latest version of glmmadmb needs at least
> version 2.11 or R. Also for me (fedora 13)  the executable glmm
> did not have executable permission. In case one doesn't know you can
> search for it
> 
>     sudo find /  -name nbmm
> 
> and change the permission
> 
>    sudo chmod 755  fullpath/nbmm
> 
> Now people want p values  I see that if you run a model with
> 
>     fit=glmm.admb(....)
> 
> then
> 
>    fit$b holds the parameter estimates and fit$stdbeta  holds the
> estimated std dev of those
> parameters so that presumably p values can be calculated by applying to
> correct R function to
> 
>     fit$b /fit$stdbeta
> 
> Does someone know what the function is called?
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users

I think

   wald_Z_vals <- fit$b/fit$stdbeta
   2*pnorm(-abs(wald_Z_vals))

 should do it for a 2-tailed test (pnorm() gives the (lower) CDF of the
standard normal distribution)

  I would use this with some (much?) caution because it is (a) a Wald
test (i.e. less reliable for finite data than likelihood ratio or score
tests) and (b) a Z test (i.e., assumes 'large denominator degrees of
freedom' / large number of random-effect blocks / variances etc. known
without error) ...

 Ben Bolker