[Feature]: Add 2-parm catchability (Q) option with offset and slope
Rick-Methot-NOAA opened this issue · 2 comments
Describe the solution you would like.
When an environmental index is used to compare to a dev vector, including recruitment devs, there is a need for an offset parameter as well as the existing slope parameter.
This will be a step in implementation of #5
Input looks like:
#_Q_setup for fleets with cpue or survey data
#_1: fleet number
#_2: link type: 1=simple q; 2=mirror; 3=power (+1 parm); 4=mirror with rescale (+1p); 5=offset (+1p); 6=offset & power (+2p)
#_3: extra input for link, i.e. mirror fleet# or dev index number
#_4: 0/1 to select extra sd parameter
#_5: 0/1 for biasadj or not
#_6: 0/1 to float
#_ fleet link link_info extra_se biasadj float # fleetname
2 1 0 0 0 0 # Survey
3 5 0 0 0 0 # env
-9999 0 0 0 0 0
#
#_Q_parameters
#_ LO HI INIT PRIOR PR_SD PR_type PHASE env-var use_dev dev_mnyr dev_mxyr dev_PH Block Blk_Fxn # parm_name
-20 20 0 0 99 0 3 0 0 0 0 0 0 0 # LnQ_base_Survey(2)
0.001 20 1 0 99 0 2 0 0 0 0 0 0 0 # Q_base_env(3)
-1 1 0 0 99 0 2 0 0 0 0 0 0 0 # Q_offset_env(3)
Note that this new parameter is now labelled Q_offset. Previously the Q parameter associated with option 3 (mirror +) was called Q_offset. That is renamed to Q_scale because it is principally used to scale a CPUE to the relative area it represents.
Describe alternatives you have considered
force users to remember to zero-center their env data, but even zero-centering is an incomplete alternative to allowing for an offset parameter.
Statistical validity, if applicable
bias will be created if a non-zero centered env index is used
Describe if this is needed for a management application
bias will be created if a non-zero centered env index is used
Additional context
No response
Good to see this moving forward.
I wanted to note that I think the alternative of zero-centering the env data isn't necessarily a comparable solution anyway because the period of devs that are being informed by the index might not be zero-centered themselves. That is, you might have an index of recruitment deviations which you believe provides information on the relative strength of individual cohorts. If it's applied to a recent period where the other data sources suggest that recruitment was below average, a zero-centered index will pull them up if there's not an offset parameter available to make the scale of observed and expected line up. In some cases, we may want that pulling effect, in which case the offset could be fixed at 0.
test case here
test_env_offset.zip