In jabbamodel/JABBA: Just Another Bayesian Biomass Assessment

JABBA Diagnostics

JABBA examines the root mean squared error (RMSE) to quantitatively evaluate the randomness of model residuals (Francis 2011; Breen et al 2003, Carvalho et al 2017). Analysts are encouraged to also conduct a visual examination of observed and predicted values to verify goodness-of-fit. Residuals exhibiting heteroscedasticity or serial correlation may suggest non-randomness. The outcomes of the test are written to a file diagnostics.csv in the user's output directory.

JABBA procures the root mean squared error (RMSE) for the series and returns the adjusted CV, or recommended weighting factor for that fleet. Accordingly, the RMSE is estimated as:

where Yt is the observed CPUE in year t on the log scale, Y(hat)t is the predicted CPUE in year t, and N is the number of CPUE observations.

Relevant Citations

Carvalho and Winker. 2015. Stock Assessment of South Atlantic Blue Shark (Prionace glauca) Through 2013 Collect. Vol. Sci. Pap. ICCAT SCRS/2015/153

Carvalho, F., Punt, A. E., Chang, Y. J., Maunder, M. N., & Piner, K. R. (2016). Can diagnostic tests help identify model misspecification in integrated stock assessments? Fisheries Research. http://doi.org/10.1016/j.fishres.2016.09.018

Chang, Yi-Jay; ISC. (2016). Stock Assessment Update for Blue Marlin ( Makaira nigricans ) in the Pacific Ocean through 2014. Sapporo, Hokkaido, Japan. Retrieved from http://isc.fra.go.jp/pdf/BILL/ISC16_BILL_2/WP1_Chang_final.pdf

Francis, R. I. C. C. 2011. Data weighting in statistical fisheries stock assessment models. Canadian Journal of Fisheries and Aquatic Sciences, 68: 1124-1138.

Implementation

Within the JABBA function, posterior model fits are extracted and used to calculate the RMSE, DIC, and SDNR on the log and standardized residuals, and the SDNR on the latter. These are saved to the Goodness of Fit csv file and plotted.

Resids = NULL
for (i in 1:series) {
  Resids = rbind(Resids, log(CPUE[, i]) - log(apply(posteriors$CPUE[, , i], 2, quantile, c(0.5))))
}
DIC =round(mod$BUGSoutput$DIC,1)
Nobs =length(as.numeric(Resids)[is.na(as.numeric(Resids))==FALSE])
DF = Nobs-npar
RMSE = round(100*sqrt(sum(Resids^2,na.rm =TRUE)/DF),1)


# Standardized Residuals
StResid = NULL
for(i in 1:series){
  StResid =rbind(StResid,log(CPUE[,i]/apply(posteriors$CPUE[,,i],2,quantile,c(0.5)))/
                   apply(posteriors$TOE[,,i],2,quantile,c(0.5))+0.5*apply(posteriors$TOE[,,i],2,quantile,c(0.5)))        
}
mean.res = apply(StResid,2,mean,na.rm =TRUE)
smooth.res = predict(loess(mean.res~Yr),data.frame(Yr=cpue.yrs))
lines(cpue.yrs,smooth.res,lwd=2)
DIC =round(mod$BUGSoutput$DIC,1)
SDNR = round(sqrt(sum(StResid^2,na.rm =TRUE)/(Nobs-1)),2)
Crit.value = (qchisq(.95, df=(Nobs-1))/(Nobs-1))^0.5


#Save Residuals 
Res.CPUE = data.frame(Resids)
row.names(Res.CPUE) = indices   
colnames(Res.CPUE) = paste(Yr)
write.csv(Res.CPUE,paste0(output.dir,"/ResCPUE_",assessment,"_",Scenario,".csv"))

#Save standardized Residuals 
StRes.CPUE = data.frame(StResid)
row.names(Res.CPUE) = indices   
colnames(Res.CPUE) = paste(Yr)
write.csv(Res.CPUE,paste0(output.dir,"/StResCPUE_",assessment,"_",Scenario,".csv"))

# Produce statistice describing the Goodness of the Fit
GOF = data.frame(Stastistic = c("N","p","DF","SDNR","RMSE","DIC"),Value = c(Nobs,npar,DF,SDNR,RMSE,DIC))
write.csv(GOF,paste0(output.dir,"/GOF_",assessment,"_",Scenario,".csv"))