I am modeling survival for radio-collared ungulates in a known-fate context, a raggedy sampling design (few animals marked each year, 6 years of data). The study encountered a much-more-than-ideal amount of collar failure (mostly dropped collars), so there is quite a bit of right-censoring despite the relatively long-term study. Turns out that the relocation history (generated by students in an unnamed Asian country) also includes, for a few animals, a few re-sightings of ear-tagged animals made after collars no longer functioned. At first I considered using these resightings with appropriate interval censoring (resightings occur at longer intervals than I will use in the analysis), and p. 16-26 of Cooch & White (v.
seems to suggest this is appropriate. However, I am now inclined to the view that these resightings cannot be used at all -- at least within a known-fate approach - because animals that died subsequent to collar failure will be much less likely to be detected than those that continued living. There is no legal hunting here, so no dead recoveries. Short of ditching known-fate entirely and replacing with who-knows-what, I see no way to use these resightings without producing upward bias in estimated survival.
Any thoughts?
Rich Harris