Eveson et al.: Incorporating fishery observer data into an integrated catch-at-age and multiyear tagging model 



503 



may allow for more accurate correction factors to be 

 developed. 



Trade-off between number of tag releases and 

 observer coverage 



We first concentrated on the results for scenario 1, 

 and how changes in A'^ and d affected the accuracy of 

 the parameter estimates. For all parameters, biases 

 in the median of the estimates decreased rapidly as TV 

 increased, especially between 250 and 1000 releases 

 (Fig. 3). Biases also tended to decrease as d increased, 

 especially for Pj, A, and F^ to F3. In any case, only the 

 biases in natural mortality estimates at the lowest 



release numbers (7V<500) were large enough to be of 

 concern, and further investigation showed they were the 

 result of a large proportion of the estimates falling on 

 the lower bound of zero. 



As seen in the previous section, evaluating biases 

 could be complicated in some scenarios because of 

 natural mortality estimates hitting a lower bound of 

 zero, reporting-rate estimates hitting an upper bound of 

 one, and fishing morality estimates having right-skewed 

 distributions, especially at older ages. These problems 

 became more pronounced as A^ and d decreased, such 

 that with N=250 and 6=0.05 the median and mean 

 differed significantly for many parameters (e.g., M^ 

 had a median bias of -23% but a mean bias of -^23%; 



