Closed quang-huynh closed 7 months ago
Try the censored likelihood approach. https://cdnsciencepub.com/doi/full/10.1139/cjfas-2022-0159 sdmTMB::censored_poisson()
@jdunic can help :)
@seananderson the choice of approach doesn't seem to affect the trend too much, so perhaps we proceed with the parsimonious model?
I think the NB2 without hook competition is probably what we should go with. Joe’s CJFAS paper discusses why the instantaneous catch rate method is particularly bad for species that are unaffected by hook competition at low saturation levels with dogfish as an example. Plus, when we look at the CPUE vs hook saturation plots we see virtually no fall off of CPUE at higher levels of saturation, even when we split by year. As a result, the censored likelihood approaches end up not doing anything (what you saw before was purely Poisson).
We had some problems with convergence once we added the censored likelihood with the Poisson assuming there was some censoring at an arbitrary level, although it’s possible that’s my fault given how I implemented the likelihood.
I got a censored NB2 working yesterday, but that needs to be explored first before we can use it in an assessment.