Hello,
I'm fitting multistate models with 3 reproductive states (non breeder, failed breeder and successful breeder) to 13 years of data on a seabird (2004-2017, with 2014 missing).
Testing for Goodness-of-fit with U-Care, I don't find any problem of memory (WBWA non-significant) nor transience (3G.SR and 3G.Sm non-significants) in the data. Concerning trap-dependence however, I find immediate trap-happiness (M.ITEC significant but M.LTEC non significant). This is not so surprising because the birds are monitored from a tower which is also one of their only nesting places on this island.
If I'm correct, one of the ways to treat trap-dependence is to split the capture histories and then to use a model with a time-since-marking structure on p. So that's what I did. However, when I look at the residual plot of this model [ Phi(state*sex*t) p(state*sex*t / state*sex*t) Psi(state*sex*t) ] on MARK, I still find a strong structure (see picture). This makes me think that there is still an excessive lack of fit.
[img][IMG]https://nsa39.casimages.com/img/2018/05/08/180508112024513990.png[/img][/img]
Do you think it is really the case and would it thus be irrelevant to start model comparisons from this model ? If this is the case, do you have an idea of how I could solve the situation ?
An interesting feature of my data is that trap-dependence only starts to be significant after some point (year 2011), so maybe there is a better way to account for trap-dependence when the phenomenon is not constant over resighting occasions.
Many thanks in advance,
Teo