model averaging when groups are handled as covariates

[vulpes]

Dear list,

I am using the Known Fate model to estimate survivorship for four groups of translocated populations of a small potoroid. The analysis has worked nicely and there are several competitive models from the candidate set, therefore I am model averaging. However, some of the ‘better’ models have included an interaction effect between an individual covariate and the groups (treatment groups). This is also fine, as I combined the groups as one group and used the treatment as an individual covariate (i.e. used three individual covariates to code for each of the four categories), as per the description in the Gentle Introduction to MARK (Chapter 11.7.2).

The problem is, when model averaging, the models which have used an interaction between the group and individual covariate have the same survivorship value for each group. This of course influences the model averaging outcome.

I am assuming this is what is alluded to (page 11-44 of the Gentle Introduction) as the ‘after the fact cost’ from using individual covariates when ‘handling attribute groups’.

Can anyone advise if I have missed something? Otherwise, I think the only solution (without going to RMark) is to manually model average as described in Burnham and Anderson (2002) and then also manually calculate the variance, std error, std dev and confidence intervals.

Is this the correct procedure?

In hope,

vulpes