It's hard to know what any of this means without seeing your models. However, why would you use glmmADMB without random effects? Its documentation says it is for fitting mixed models.

I aware of that; but I would like to use the MCMC option that the library admb has.

I am still working on this issue.

Thanks,

Alex

Did you check the scale parameters? When you fit a glm with the family "binomial", a scale parameter=1 is assumed. I do not know how the admb works, but perhaps you could find a difference between both algorithms there. Try to run the same glm by using the quasibinomial family and compare the SE values obtained with the admb. It´s just an idea, l  

I will take a look. But the thing is that the summary table and the predicted values are almost identical. Only se.fit differ.

Thanks for the suggestion.

That´s OK. The predicted values should be identical since they are based on model coefficients, which should remain the same. Only the SE changes due to a different scale parameters.  

I think it is not related to scale parameters.

I have received the answer directly from Ben bolker:

"It looks like glmmADMB is getting the estimate of the correlation between the fixed effects wrong (I don't yet know whether this is a problem with the optimization in AD Model Builder or an actual bug in the translation of results from ADMB output to R)

Update: it turns out that you found a serious, long-standing bug in glmmADMB. I've just moved glmmADMB development to

https://github.com/bbolker/glmmadmb

so I'm hoping you can simply install the new version via

devtools::install_github("bbolker/glmmadmb")

and see if that solves the problem

cheers

Ben Bolker"