386 



Fishery Bulletin 103(2) 



simple linear fit to the estimates in the three papers 

 mentioned above. 



The estimates of M from model 1 appear more rea- 

 sonable for the aperiodic recruitment pattern. The cor- 

 relations between M and F R9 (the fishing mortality in 

 1989) are not as strong as in the simulation example. 

 This is encouraging and indicates that there may be 

 enough contrast in the effort pattern to separate fishing 

 mortality from natural mortality. 



The estimates of l r , and a, for model 3 are not sensi- 

 tive to M. However /; and M are quite strongly related. 

 In the case of constant recruitment r(t) and mortality 

 z(t)=Z, it is well known that k and Z are perfectly cor- 

 related, and only their ratio Zlk is able to be estimated. 

 The separation of M and k therefore relies on there be- 

 ing adequate contrast in recruitment and effort. 



For model 2 there is little difference between the two 

 recruitment models. The estimates of M show moderate 

 dependence on (k, I J, but without trend. These esti- 

 mates are generally somewhat higher than we expect 

 from prior studies. But for the natural mortality rate, 

 this is the first time we have obtained estimates of M, 

 which is larger than what we have assumed in previous 

 stock assessments, around 2.3 per year (Wang and Die, 

 1996). Estimates of F m are too variable to be relied up- 

 on. All models agree reasonably on the a, parameter. 



Our model assumes recruitment at a fixed length, 

 l , which has to be chosen. In Figure 3 the parameter 

 estimates for fixed (k, l^J are plotted against l Q for the 

 quasiperiodic recruitment model. Parameter estimates 

 are consistent for given l provided that all model as- 

 sumptions are satisfied. However, when l is too small 



