Transient model output in MARK

[cristinvera]

Hi there,

I performed a GOF test on my models and it was clear that I needed to account for transience. This is my first time both working with this kind of data and using MARK so I followed chapter 7 in the MARK bible using a TSM model and adjuted my PIMs. I have run my model and I am currently checking the output. The only thing I am slightly confused about is that there are 6 survival parameter estimates in my output, but I am only looking at 3 survival parameters for 3 groups of animals. How do I distinguish between non-transients and transients in this output? Maybe I am interpretting this incorrectly but I am not really sure where to go for guidance on this!

I would be really grateful if someone could point me in the right direction or tell me if I am being utterly nonsensical!

[B.K. Sandercock]

A time-since-marking model without time effects would be expected to return six parameters for apparent survival if you have three groups. If you had five occasions in the encounter histories, the PIMs for apparent survival for your three groups should look like this:

Code: Select all

1 2 2 2
  1 2 2
    1 2
      1
3 4 4 4
  3 4 4
    3 4
      3
5 6 6 6
  5 6 6
    5 6
      5

You might expect a transience effect if you have a lot of encounter histories where an individual is marked but never seen again (e.g., 10000, 01000, 00100, etc). You can explore your data by looking at your parameter estimates, you would expect the transients to lead to a lower probability of apparent survival on the first diagonal of the PIMS so that phi1(1)<phi2+(2); phi1(3)<phi2+(4); phi1(5)<phi2+(6). You can get the same result with age effects on survival or site fidelity too. If you find the opposite pattern phi1>=phi2+ in any group, then you don't have much evidence for transience. You can combine the probabilities to estimate the probability that an unmarked individual is a transient: tau = 1-(phi1/phi2+). For more details see: Pradel, R. et al. (1997). Capture-recapture survival models taking account of transients. Biometrics 53: 60–72. Hope this helps, good luck.

Brett.

[cooch]

Brett is entirely correct (of course). Much of the material he nicely summarizes is in chapter 7 of the MARK book. See section 7.4. The calculation of proportions of transients and residents is presented in the -sidebar- on p. 44 of Chapter 7.

Note that a transient model (sensu Pradel) is simply a model structured for an extreme form of heterogeneity. You're pretending your population is a 'miixture' of two types of individuals - those that might stay in the population, conditional on surviving, and those that leave with probability 1.0. [In fact, you can approach this as a finite mixture estimation problem -- I should put that in the book at some point.]

The reason I mention this is that a TSM model is simply a structural approach to handling 'heterogeneity' - between young and old individuals, or residents, and transients. A TSM model might be well-supported by the data, but you as a biologist are responsbile for determing what that sort of model is actually telling you (i.e., it might not be transience - it could be some other form of unobserved heterogeneity.)