Closed jysullivan closed 4 years ago
I think you mean Kristensen, not Kasper unless we are getting really informal with citations.
On Tue, May 21, 2019 at 8:26 AM Jane Sullivan notifications@github.com wrote:
Comments from BW to address:
- Clarify how we want to talk about recruitment and incorporating uncertainty into management in the introduction. Currently the text says,
Finally, strong recruitment from the 2014 and possibly 2013 and 2015 year classes were reported in the Federal assessment, prompting questions about how to treat the uncertainty in recruitment for State management (Hanselman et al. 2017, Sullivan and Williams 2018). A statistical catch-at-age model coded in Template Model Builder (TMB) will allow more flexibility in exploring recruitment using random effects (Kasper et al. 2016).
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/commfish/seak_sablefish/issues/38?email_source=notifications&email_token=ABSQNRWIHV2KULKLBREDEU3PWQPBRA5CNFSM4HOMWYG2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GVAJU7Q, or mute the thread https://github.com/notifications/unsubscribe-auth/ABSQNRW33GOXHHTDSTU4XZDPWQPBRANCNFSM4HOMWYGQ .
Fixed, thanks @Dana!
Comments from BW to address:
We have maturity data from the survey going back to 1997 (2007 was a typo). The query for the data and assumptions about which maturity codes I associate with mature and immature are found in
0_querynclean_data.r
.Here are the sample sizes by year:
[x] There was a question about whether or not proportions mature-at-age were used as inputs to the model. The answer is yes, they are. The vector of proportions mature-at-age is multiplied by the female numbers-at-age matrix, female weight-at-age, and annual survivorship to time of spawning (assuming February) in order to get the spawning biomass.
[x] There was a question on eqn 3 about why is this log transformed (more specifically how do I log transform 0?)
Within the
gam
function I used thefamly = quasibinomial
argument, which allows the response variable (the proportion female observed in the survey) to be a proportion instead of the normal 0/1 response used infamily = binomial
. The default transformation for the binomial and quasibinomial distributions islink = logit
. The rationale for how the logit link handles 0 (or 1) observations is best explained by this Cross Validated post (I especially like response 2). This explanation is for the binomial distribution, but the same applies for the quasibinomial, which is just a scaled version of the binomial.[x] There was a question about if I have plans to incorporate additional sources of mortality
I added some more text in this section and have added it to the long-term improvements section. To be honest I'm not sure exactly how this is normally done. Catch is fit assuming a single or time-varying selectivity curve. It's likely that adding sport and personal use harvest wouldn't affect this, but I'm not sure.