Huggins closed capture

[annaren]

I was wondering if somebody could tell me if I am approaching this the correct way:
I am analysing data on small mammal capture over 5 successive trap sessions and I would like to know if there is a habitat effect as I trap in two ifferent types of area. I have therefore used Huggins closed capture models using HABITAT as a covariate. I first ran Mo, Mb and Mt without the covariate and Mt had the most support. I then added in the covariate in the Mt model such that the design matirx looked like this:

Para t1 t2 t3 t4 habitat
p 1 0 0 0 habitat
p 0 1 0 0 habitat
p 0 0 1 0 habitat
p 0 0 0 1 habitat
p 0 0 0 0 habitat


I specified the User-specified option for habitat and it I then got real estimates for p 1-5 when habitat = 1, adn I then also when habitat = 0. These obviously gave different values which I expeced but how can I say if these are significantly differnt and thus if habitat has a significant effect.
Is this better than using a closed capture model with 2 groups for each habitat type?
Thanks

[ganghis]

Hi Anna,

With binary covariates (like your habitat variable for instance) there isn't any difference between using groups to represent habitat type vs. using habitat as an individual covariate. The full likelihood models are slightly more efficient than the Huggins-Alho procedure, so you might consider going with the 'group' approach using the 'full likelihood closed captures' data type (or related data type with less parameters).

Concerning model selection, I'd suggest doing a little more thinking about the models you are running - for instance, are the study plots close enough together to have common time effects? Your DM is really a t+g model but you may want a t*g model.

Are you really interested in whether capture probability differs between different habitats, or is this just a nuisance parameter on your way to estimating abundance? If you're interested in the former, you might look at the confidence interval for the beta parameter corresponding to the habitat effect and see if it includes zero - if interested in the latter, I'd run a somewhat 'balanced' model set (there's no real reason to be doing a stepwise selection procedure here), and do some model averaging.

Hope this helps.

Cheers, Paul Conn