GOF with large losses on capture or an alternative?

[jschmutz]

I am estimating CJS survival for a population where individuals change states over time as dictated by observers (i.e., in one year a given bird is “seen”, whereas in another year it is “physically captured”, though there is actually 6 possible states). I don’t want to use a multi-state CJS model as the transitions are not ecological (i.e., I impose them by deciding whether or not we should physically recapture a bird vs just being content with seeing it without capture). I suppose I could use a multi-state CJS in the most parameterized form and then fix all the state transitions to the initial estimates and then force all estimation to be on the phi’s and p’s. But I worry that there would be bias in estimates, and lots of transitions can’t be estimated for lack of data (unless I use a simpler model for GOF by collapsing years). What does seem to work well is treat all recaptures (when transiting into a new state) as losses on capture, and then re-enter the bird as a “new bird” to carry on with its remaining history. Everything works well and the estimates seem as I expected. However, since almost every bird was re-encountered multiple times, about two thirds or more of the data are ‘losses on capture’, which don’t get included in a GOF. The only ones that would be included would be the ones that were never re-encoutered, which would lead to a bogus c-hat. So, neither path seems trustable. Any thoughts for GOF?