Trend in survival?

[Christian Ramp]

I am analyzing 18 years of mark recapture data of humpback whales and I have used the CJS model to estimate annual survival. Animals are marked as young & adult. In addition, there are a lot of transient, so the best fitting model has three ‘age classes’. The first one contains the first interval over all cohorts and is time variant, the second age class contains the second and third interval and is constant, while the third (adult) contains the rest of the data and is also constant. The probability of capture is time variant – no age classes.
Now I want to see if the survival rate of adult animals has changed over time. I have added a linear trend (as in chapter 7) to the third age class and this model is almost as well supported as the one with constant survival for the third age class (delta AIC = 0,07). However, while the constant survival rate is estimated to be 0.978 the trend shows a slow decline (starts at 0.99) and goes down to 0.95 for the last interval. This would be – if true – a drastic drop of survival, and hence would raise some conservation issues.

[I have also changed the model (via PIM) so that I have split the third age class in time spans (2,3 and 4) and all results show a decline in especially the last period. However, these models were less supported (delta AIC between 2 and 5) due to higher number of parameters. I have changed the trend dummy variable in the design matrix from 1,2,3,4,…,13 to 11111,2222,33333 and I got the same results (for phi) but due to less parameters it is only marginally less supported than the one with constant adult survival].

Finally my question: What do I do with a couple of models which have almost the same level of support but give very different results?
Or is there any mistake on my part (which is also possible)?

Any ideas/ suggestions/corrections would be really appreciated.
Thanks.
Christian