bootstrap GOF: general model p and estimated c-hat

[simone77]

It was said to me (during a Workshop) that when you perform a bootstrap GOF, you can, more than calculating the c-hat by means of "observed deviance / mean bootstrap deviance", also have a look at how probably the model tested is adequately fitting the data.
Perhaps, in other words, it could be said that you can estimate the probability that the tested model is a good general model because it fits well the data.
You should be able to do it by looking at the proportion of simulated models with a deviance larger that the observed one. If you have 100 simulated models and 20 have a greater deviance (than the observed one) you could say that the model is adequate (as general model) with a p=0.2.

I am trying different analyses by organizing different dataset, splitting different groups, but it is being relatively frequent that I found a very low p (even p=0) by using the above criterion but, nevertheless, the estimated c-hat is very reasonable (c-hat< 1.5).

It seems to me that these results are contradictory among them, or perhaps something is escaping to my understanding.

Thanks for any response.