498 



Fishery Bulletin 105(4) 



be estimated with five release years). The unobserved 

 tag reporting rate was also constrained to be constant 

 over all recapture years (i.e., A, = A for ( = 1, ... , 5). It 

 seems reasonable that the tag reporting rate would be 

 constant, or at least similar, over the course of the ex- 

 periment. Exceptions would occur if there was a signifi- 

 cant change in the fishery or in tag-return promotional 

 activities during this time, or if the fishery involves 

 multiple fleets with different reporting rates so that the 

 overall reporting rate would change if the distribution 

 of catches among fleets changed. To account for such 

 situations, year-specific reporting rates were allowed for 

 in a later scenario (see next paragraph). The only other 

 constraints imposed were simple bound constraints to 

 keep all parameters positive and to keep the reporting 

 rate from exceeding one. Thus, the parameters esti- 

 mated were F, (/=!, . . . , 5), M, (t = l, ... ,4), Pj, and A. 

 Model performance will be affected by a large number 

 of factors, including the following: 1) the parameter val- 

 ues used for the mortality rates, cohort size, reporting 

 rates, catch aging error, and overdispersion factor; 2) 

 model parameterization (i.e., whether parameters are 

 assumed to vary with age, year, or both, or to have a 

 particular functional form); and 3) the design of the tag- 

 ging experiment (e.g., number of release and recapture 

 years; number of releases per year; level of observer 

 coverage). There are endless possibilities with regard 

 to these factors; therefore we have chosen a number of 

 scenarios that we feel are most illustrative for which 

 to present results (Table 1). All of these scenarios use 



scenario 1 as a base but include a variation on one of 

 the factors. 



Scenarios 2 through 7 investigate changes to the 

 parameter values (factor 1). In particular, scenario 2 in- 

 creases the natural mortality rate, scenario 3 increases 

 the fishing mortality rate, scenarios 4 and 5 decrease 

 and increase the reporting rate, respectively, and sce- 

 narios 6 and 7 decrease and increase the overdispersion 

 factor, respectively. Changes to the cohort size and vari- 

 ance of the catch aging error had less impact on the 

 results and are therefore not included here. 



Model parameterization (factor 2) can have a large 

 effect on how well parameters can be estimated. For 

 example, if natural mortality can be assumed to be 

 constant across ages (this is a fairly common assumption 

 in fishery models, at least over a limited range of age 

 classes), then the precision and accuracy of the natural 

 mortality-rate estimate should improve, which in turn 

 may lead to improvements in other parameter estimates. 

 Scenario 8 explores the benefits of having a constant 

 natural mortality rate. Another standard way of reduc- 

 ing the number of parameters in fishery models is to 

 model fishing mortality as a function of age by using an 

 appropriate selectivity curve. Scenario 9 considers the 

 situation where fishing mortality is constrained to be a 

 linear function of age. Note that we parameterized the 

 line in terms of F^ and F^ (i.e., F-=F^+ii-irHF^-F^)/4), 

 because this made it easy to constrain the fishing mor- 

 tality rates to be positive. Instead of imposing additional 

 constraints, scenario 10 relaxes the assumption of a con- 



