RMark model averaging

[jlaufenb]

I'm trying to model average real parameter estimates using the FullHet data type for a closed pop analysis. I have 2 groups (M&F) and 10 occasions. Model specific estimates, model averaged estimates of p, and unconditional variances for all parameters are correct. However, the model averaged estimates for N-hat are way off. When I run the analysis in MARK or average by hand, I get MA N-hat's of 148 and 105 for F and M, respectively. RMark reports 67 and 47 even though the variances are calculated correctly. Any insights into why this is?

Thanks
Jared

[jlaake]

The actual real parameter for N in these models is f0 which is the number caught 0 times. To get N, f0 is added to M_t+1 which is the total caught. I've not programmed it to include M_t+1. You should find that if you add M_t+1 for each group to the value you got from RMark that it will equal the value you got in MARK model averaging.

--jeff

[jlaufenb]

I'm running a set of models on ~900 subsampled datasets to investigate subsampling intensity on bias, precision, model selection uncertainty, model averaging, etc. I have 2 groups, males & females. Is there a way to output M_t+1 for each group? I have outputed ess for data with a single group and divide by the number of occasions, but I don't think this would work for >1 group. I could use Huggins full heterogeneity models and code sex as an indiv. cov., but I would rather use group structure and the full likelihood estimators. Any suggestions?

Thanks
Jared

[jlaake]

Jared-

M_t+1 is simply the total number caught. If you are getting an N for each group then you should be able to sum the frequencies in each group to get the group-specific M_t+1. I presume you are doing the sampling in R and running RMark. If so you should be able to sum the frequencies for each of the sub-samples for each group. If you need more help with this contact me off-list.

--jeff

[gwhite]

Jared:
A back-door approach is to output the beta estimates, and exponentiate them to get f0 (assuming you are using the default log link for f0). If you also output the real parameter estimates, you can subtract f0 from the Nhat, and the resut is M_t+1. To check yourself, make sure that the value of M_t+1 is an integer and that your values match the M_t+1 values reported in the full output.

Gary