Bootstrap GOF when model deviance is 0

[jhupp]

I’m comparing survival among marked Canada geese in 3 treatments and 2 cohorts over a three-year period. Our resighting probabilities were often (but not always) 1.0 for birds in a given treatment and year. Deviance of the fully parameterized model was 0 because of the absence of 1101 and 1001 encounter histories for some treatments. Hence deviance and c-hat were also 0 for simulations in the bootstrap GOF for the model.

My question is whether this indicates that the fully-parameterized model fits the data perfectly, and that adjustment of c-hat is not necessary. I can derive an estimate of c-hat if I conduct a bootstrap GOF on models that have slightly fewer parameters. However, the estimate for c-hat depends on how I constrain the model.

Any feedback is appreciated.

Jerry Hupp

Jerry_hupp@usgs.gov