[R-sig-ME] Robust Standard Errors

[Ben Bolker]

Zahwa Al Ayyash (Student <zsa11 at ...> writes:

> 
> Dear list,
> 
> Can we extract the robust standard errors from 
> Poisson/Negative Binomial/Zero Inflated/Hurdle Models
> in glmmADMB?
> 
> If yes, how is this doable?
> If no, are there possible consequences of using the standard errors that
are returned currently?
> 

  Not easily.

  You could try to implement one of the standard approaches for 
estimating heteroscedasticity-consistent standard errors:

http://en.wikipedia.org/wiki/Heteroscedasticity-consistent_standard_errors

I don't know how much trouble you'll get into if you try to estimate
this naively (e.g. you need the inverse of X X^T) ...

  The possible consequences of using the non-robust standard errors
are, as usual, that if the model is misspecified you'll get the wrong
answer.  You have to decide for yourself how much this worries you.
An equal and possibly bigger problem is whether using Wald estimates
of the standard errors for inference is OK in this case.  You might
want to look into profile confidence intervals (hmm, ADMB has
these but they don't seem to be implemented in glmmADBM) or post-hoc
MCMC (which *is* implemented).