are p and phi confounded when using indiv. covariates?

[dd.snook]

Hello,

I'd like to fit a full-time CJS model where survival is a function of an individual covariate (e.g. weight) and was wondering if in this case the last phi and p will still be confounded.

I understand that in a model where both survival and recapture are a function of time only, {phi(time)p(time)}, the last phi and p are confounded. In the case where phi also varies with say weight, {phi(weight+time)p(time)}, would phi still be confounded with p in the last time interval? It seems to me that it wouldn't but wanted to double check this with those with more experience and expertise to see if I'm getting this right...

Sorry if this has been asked before but I couldn't find any similar posts with the keywords I used in the search.

Thanks a bunch!

Ed

[cooch]

Hello,

I'd like to fit a full-time CJS model where survival is a function of an individual covariate (e.g. weight) and was wondering if in this case the last phi and p will still be confounded.

Not entirely, although the specific structure of the confounding changes (to some degree, it depends on the temporal structure of the covariates, which makes it hard to generalize). Take the first example data set in chapter 11. Fit model phi(t)p(t) without the covariate. Then, try it again, building a model adding the covariate mass (m) into the DM. You'll notice it seemes to correct the confounding, but it doesn't really help - the CI's for the two 'separated' estimates of terminal phi and p are basically 0-1, suggesting the estimates aren't particularly meaningful.

My general suggestion is - leave terminal estimates of phi and t from a fully time-dependent CJS model alone. There may be particular interactions with individual covariates which might make them 'interpretable', but in general, probably not.

[dd.snook]

Hi Evan,

Thanks a lot, this was very helpful!

Ed

[simone77]

I have just bumped into this somehow old post (sorry to get late) and it made me wonder: how should this be handled in terms of parameters counting?
In my case, a CJS model with individual covariate, I have something very similar with a broad CI for the last Phi and P, I am not sure if I should count them as a single estimable parameter (product of confounded parameters) or as two. It can be important for the model selection (AIC).