Flexibility to handle 0 or 100% positive responses in a time slice #304
Replies: 3 comments
-
I suppose we'd have to add the complementary log-log link as an option. Nice to see the description relative to the poisson-link in the new vignette. |
Beta Was this translation helpful? Give feedback.
-
cloglog is already an option: Line 30 in 01ac913 although I'm not sure it's allowed as a valid link everywhere it should be or whether it works with all delta families yet. It should work with binomial() and was used in this paper https://cdnsciencepub.com/doi/full/10.1139/cjfas-2022-0279 Can you summarize what the approach was to 0 or 100% encounter rates? Mapping a parameter to a small or large value? |
Beta Was this translation helpful? Give feedback.
-
Thanks Sean. I'm looping in @James-Thorson-NOAA as he is more familiar with the details of how this was dealt with in VAST. Previously, as I understand it, we would fit a delta model with the cloglog, but it also required a lot of if() clauses to detect which parameters had to be mapped differently. I talked to Jim briefly about it today and it seems like we might be able to identify a cleaner way of doing it, perhaps leveraging the use of priors. |
Beta Was this translation helpful? Give feedback.
-
Maybe we already have this capability(?), but I was thinking about the modifications made in VAST that were necessary under that infrastructure to accommodate such data.
We do run into the case with all positives in some years for Bering Sea Pacific cod.
Beta Was this translation helpful? Give feedback.
All reactions