Scale issue in latest update

[ebabcock]

When I re-run the North Atlantic mako shark assessment from last year with the version of JABBA currently on github, I got estimates of K more than an order of magnitude higher than they were last year with the exact same code and data. When I reinstalled JABBA from a commit in June 2025, I got the same results I got last year with a reasonable K. I think some of the scaling corrections added in the last few months introduced an error. This is the code.

cpueval.csv
catchval.csv
seval.csv

First run with current JABBA

devtools::load_all()  #Current version 10 June 2026'
makoNSetup1<- build_jabba(catch=read.csv("catchval.csv"),
                                  cpue=read.csv("cpueval.csv"),
                                  se=read.csv("seval.csv"),
                                  assessment="SMA2025",
                                  scenario = "N1",
                                  model.type = "Pella",
                                  BmsyK = 0.597,
                                  shape.CV=0.1,
                                  add.catch.CV=TRUE,
                                  catch.cv=0.1,
                                  r.dist="lnorm",
                                  r.prior=c(0.085,0.2), 
                                  K.dist="lnorm",
                                  K.prior=c(100000,2), 
                                  psi.dist="lnorm",
                                  psi.prior=c(1,0.2),  
                                  sigma.est = TRUE,
                                  sets.var=1:8,  
                                  sigma.proc=sigma.proc.val,
                                  proc.dev.all=FALSE,
                                  sigmaobs_bound = 1,
                                  sigmaproc_bound =0.1,
                                  q_bounds=c(1E-10,1),
                                  P_bound = c(0.02,1.1),
                                  K_bounds=c(1000,5000000),
                                  harvest.label = c("Hmsy", "Fmsy")[1]
  )
makoNFit1<- fit_jabba( makoNSetup1,
                             quickmcmc = TRUE,
                             save.jabba = FALSE)
makoNFit1$estimates

mu lci uci
K 2425445.29109230 2224048.85093843 3103399.0609795
r 0.08447136 0.05803832 0.1238475
psi 0.93648575 0.74860548 1.0792224
sigma.proc 0.05000000 0.05000000 0.0500000
m 3.33500000 3.33500000 3.3350000
Hmsy 0.02500000 0.01700000 0.0370000
SBmsy 1447996.69100000 1327762.53000000 1852736.7280000
MSY 37298.13000000 24999.38300000 59826.1630000
bmsyk 0.59700000 0.59700000 0.5970000
P1950 0.93700000 0.74900000 1.0810000
P2023 0.96400000 0.79900000 1.1120000
B_Bmsy.cur 1.61400000 1.33800000 1.8630000
H_Hmsy.cur 0.02300000 0.01400000 0.0380000

Then used this code to load an older version

devtools::unload("JABBA")

# Find the SHA first via the GitHub API
commits <- jsonlite::fromJSON(
  "https://api.github.com/repos/jabbamodel/JABBA/commits?until=2025-06-15T00:00:00Z&per_page=1"
)
sha <- commits$sha[1]

# Install that exact commit
remotes::install_github("jabbamodel/JABBA", ref = sha)
library(JABBA)

#And rerun the same code and got

makoNFit2<- fit_jabba( makoNSetup1,
                       quickmcmc = TRUE,
                       save.jabba = FALSE)
makoNFit2$estimates

K 106770.4462201 88444.70517875 120101.0458655
r 0.1079827 0.07913354 0.1468269
psi 0.9511395 0.63832500 1.1361556
sigma.proc 0.0500000 0.05000000 0.0500000
m 3.3350000 3.33500000 3.3350000
Hmsy 0.0320000 0.02400000 0.0440000
SBmsy 63742.2140000 52801.70200000 71700.6140000
MSY 2043.9860000 1483.01000000 2874.2510000
bmsyk 0.5970000 0.59700000 0.5970000
P1950 0.9520000 0.63700000 1.1370000
P2023 0.6990000 0.52100000 0.9420000
B_Bmsy.cur 1.1710000 0.87300000 1.5770000
H_Hmsy.cur 0.5820000 0.32500000 1.0330000

This is what we got last year.

Thanks

[Henning-Winker]

Hi Beth,
I am sorry to hear - my apologies. I recall there was a temporary installation error when I pushed some extra plotting functions, but I thought it was resolved shortly after.

I just tested installing from github  devtools::install_github("JABBAmodel/JABBA") - Could you confirm?

There was initially an emerging issue with mixing of the mcmc chains in JAGS (see issue 28 - #28). It seems with some R update they became bimodal and needed much longer chains. I could resolve this by internally normalizing all biomass estimates and q. All trials showed that mixing/converge improved even more. As always, I also replicated the originally published SWO South ICCAT assessment and asked colleagues  to test on their assessments.

The issue seems to be with K_bounds, which I seem to have missed to accounting for during the rescaling,
If you switch them off #K_bounds=c(1000,5000000), (very large probably not needed) - all seems OK, with K = 100847.75.

My apologies for the K_bound oversight. I will fix this tomorrow morning and update here once pushed.
Until then I suggest not to use the option build_jabba(...,K_bounds=c(1000,5000000) )

Best
Henning

#devtools::install_github("JABBAmodel/JABBA")
library(JABBA)

sigma.proc.val = 0.07 #?

makoNSetup1<- build_jabba(catch=read.csv("catchval.csv"),
cpue=read.csv("cpueval.csv"),
se=read.csv("seval.csv"),
assessment="SMA2025",
scenario = "N1",
model.type = "Pella",
BmsyK = 0.597,
shape.CV=0.1,
add.catch.CV=TRUE,
catch.cv=0.1,
r.dist="lnorm",
r.prior=c(0.085,0.2),
K.dist="lnorm",
K.prior=c(100000,2),
psi.dist="lnorm",
psi.prior=c(1,0.2),
sigma.est = TRUE,
sets.var=1:8,
sigma.proc=sigma.proc.val,
proc.dev.all=FALSE,
sigmaobs_bound = 1,
sigmaproc_bound =0.1,
q_bounds=c(1E-10,1),
P_bound = c(0.02,1.1),
#K_bounds=c(1000,5000000),
harvest.label = c("Hmsy", "Fmsy")[1]
)
makoNFit1<- fit_jabba( makoNSetup1,
quickmcmc = TRUE,
save.jabba = FALSE)

makoNFit1$estimates

                   mu            lci            uci

K 100847.7548455 74552.23834561 142858.1960378
r 0.1029841 0.06985162 0.1572875
psi 0.8284019 0.66785482 1.0804945
sigma.proc 0.0700000 0.07000000 0.0700000
m 3.3350000 3.33500000 3.3350000
Hmsy 0.0310000 0.02100000 0.0470000
SBmsy 60206.3530000 44507.86600000 85286.6880000
MSY 1885.0970000 1364.77200000 2602.5840000
bmsyk 0.5970000 0.59700000 0.5970000
P1950 0.8280000 0.66500000 1.0780000
P2023 0.7010000 0.47600000 0.9720000
B_Bmsy.cur 1.1730000 0.79700000 1.6280000
H_Hmsy.cur 0.6350000 0.37100000 1.0860000

makoNFit1$pars
Median LCI UCI Geweke.p Heidel.p
K 100847.75484547336 74552.238345613325 142858.19603778844 0.208 0.975
r 0.10298412410 0.069851619166 0.15728747034 0.961 0.429
q.1 0.00001399067 0.000008690957 0.00002053930 0.924 0.727
q.2 0.00001534637 0.000009350228 0.00002255348 0.982 0.973
q.3 0.00001653813 0.000010338698 0.00002402939 0.712 0.853
q.4 0.00001426800 0.000008258006 0.00002269893 0.846 0.583
q.5 0.00001197615 0.000006313949 0.00002212640 0.364 0.474
q.6 0.00001521965 0.000008975087 0.00002310257 0.741 0.654
q.7 0.00001466758 0.000008713328 0.00002267759 0.847 0.857
q.8 0.00001574047 0.000007845541 0.00003603508 0.862 0.813
psi 0.82840185460 0.667854824042 1.08049448392 0.472 0.831

[ebabcock]

Thanks, Henning. I can now install with install_github. The results are correct without the K bounds.