Restricting interactions

[jlaake]

The following question was sent to me via email. Below is the question and answer.

Am using RMark to build models in the multi-strata robust design. I have 3 states, but only 2 transitions of interest.

My question: Is there a way to only model selected transitions as a function of an individual covariate? I can model all transition probabilities as a function of a covariate with the following:

Psi.strata.hatch=list(formula=~1+stratum:tostratum:hatch)

"hatch" is a binomial time-varying covariate indicating whether or not a nesting attempt was successful.

Is there a way to put the covariate effects only on the transitions of interest?

Sure. Use 0/1 variables in the Psi design data for stratum and tostratum. If you had strata A and B it could be something like A:toB:hatch. A:toB is only 1 when both A and toB are 1. This is a generic solution for restricting effects in any model. For those of you used to filling in DMs with the MARK interface, the A:toB is equivalent to only putting the hatch variable where A and toB are 1.

[mikehudson]

Dear Jeff,

Thanks - this is very useful. I used it successfully to restrict interactions in the transition design matrix, but I stumble when trying use it with recapture i.e. when attempting to build the following model:

constant pA, with a state:time interaction for pB and pC. I have tried:

p.A.dot.BxCxtime <- list(formula = ~A + B:time + C:time)

But this creates a model in which A is constant, but the estimates for B and C vary by time but these times varying estimates are shared by the two states i.e. there is not interaction. This approach did seem to have the desired approach when modelling transitions, but maybe I am mistaken!

I've tried several other permutations, but cannot seem to get what I want. e.g.:
p.A.dot.BxCxtime <- list(formula = ~A + B:C:time)

Any advice would be much appreciated.

Regards,
Mike

[mikehudson]

For some reason, this is now working. I must have been doing something wrong previously.

Sorry about that,

Regards,
Mike