interaction with individual covariate: strange results

[AndyT]

Hello,

I’m having a bit of trouble some strange results with a fairly straightforward recaptures only analysis:

I have data from two summers of marking, with monthly capture occasions in each summer, 5 occasions in summer 2004 and 4 in summer 2005. I am trying to look at differences in over-winter survival between different habitat types, as well as the effect of body condition (as an individual covariate) on survival.

I have structured my models such that survival is constant in the summers, and winter survival is different (i.e., a season effect, structurally similar to flood/no flood in the example in the MARK book).

The top model in my candidate set is:

phi(c+s+h+w+c*w+h*w)p(s+h+t)

where c=body condition, s=sex, h=habitat, w=season, t=time.

This model structure makes intuitive sense in terms of the interactions: You would expect body condition and habitat differences to affect survival differently in the summer and winter. However, the direction of the c*w interaction is puzzling me. When I run the model with a value of c at 1 SD above the mean, summer survival increases by a huge amount, but winter survival rates don’t really change.

I am especially puzzled by this because early on my analysis I ran a basic logistic regression on the effects of body condition on “return rates” from 2004 to 2005, and there was a strong positive relationship between body condition and over-winter “survival” (return rates).

If anybody can help me with some insight/directions re: the results I am getting from my MARK analysis, that would be greatly appreciated.

Thanks in advance,

Andy