Analyze pros/cons of tail compression

[k-doering-NOAA]

Imported from redmine, Issue #71957 Opened by @RickMethot on 2019-12-03 Status when imported: New

I am finding that tail compression is being used for sparse conditional age-at-length data. If all fish are in a single age bin, the compression of tails means that the expected values are also compressed to that one bin. This causes SS to not detect a gradient for the fit to that observation. Tail compression should be off when there is sparse data, and perhaps should always be off unless a user turns it on.

I will create a routine in readdata such that SS will count how many observations get compressed to 1 or 2 bins and issue a warning if this is more than 0.

[k-doering-NOAA]

comment from @k-doering-NOAA on 2019-12-03: A warning seems like a good idea. I just looked at the manual section on bin compression and think it is informative about how bin compression works, so doesn't seem any changes are needed there and a warning would be the right way to communicate this, in my opinion.

[k-doering-NOAA]

comment from @iantaylor-NOAA on 2019-12-04: I support the plan for the warning.

I also should confess that I'm unconvinced of the benefits of tail compression in general. Some analysis of the pros and cons would be useful and could either support my hypothesis that any benefits don't outweigh the drawbacks, or help me see the benefits. Some of the drawbacks are mundane, like getting in the way of aggregating observed and expected values across seasons or years, but in the face of uncertain benefits, I avoid using the option. Any such cost/benefit analysis would take some time and probably isn't a high priority, so doesn't solve the problem of users misusing the option in the short term, but worth considering nevertheless.