98 



Fishery Bulletin 104(1) 



(D 



20 

 1 5 - 

 1 

 5 

 00 



I -, 







0) 



> 



o 

 o 



0) 6 

 ^ 4 

 io 0.2 



Exclude DeMarIa index 



Selection favors older fish 



Year 



Year 



Year 



Figure 8 



Trends in relative spawning biomass (s/Sjqsj), apical fishing mortality rate (F). 

 and the probability of recovery is>s^Q,.^) for two sensitivity runs — one exclud- 

 ing the DeMaria index (left) and the other with the selection curve favoring 

 older fish (right). 



as it comes from the center of juvenile abundance; how- 

 ever most aciults were caught outside this area of abun- 

 dance. Thus, the relative contribution of juveniles and 

 adults to the overall catch is unclear and the directional 

 bias in the fitted logistic selection curve is uncertain. 



The only other age composition information that has 

 come to light comes from the study by Bullock et al 

 (1992). which was not designed to provide a random 

 sample of the catch and is probably biased towards 

 larger animals caught on offshore wrecks. In principle, 

 one could reflect this uncertainty more formally either 

 by developing a prior for the selectivity parameters or 

 else by weighting the results from the two selection 

 models. The scientists on the SEDAR stock assessment 

 review panel based their advice on the selection curve 

 derived from the ENP data," which is equivalent to 

 placing negligible weight on the curve derived from the 

 Bullock et al. (1992) data; however they recognized the 

 selection curve as an important source of uncertainty 

 that is difficult to address without adequate data. 



It is important to emphasize that the Bayesian ap- 

 proach adopted in the present study allows one to ex- 

 plicitly model the uncertainty about parameters such as 

 M, for which no data may exist, but a prior distribution 

 covering the plausible range of values may be specified. 

 There is, of course, the potential for introducing bias 

 when one or more of the priors are based on expert 

 opinion or otherwise subjective information. However, 

 the same sorts of bias can be introduced by conducting 

 sensitivity analyses where the unknown parameters are 

 fixed to various values selected by the analysts. Fur- 

 thermore, if unbiased data are in short supply, analyses 



