576 



Fishery Bulletin 97(3), 1999 



o 



D. 



O 



1.0 



0.5 



00 

 1.0 



0.5 



ao 



1 



c 5 - 



o 



0.0 

 10 



0,5  



ao 



10 

 5 



0.0 



B 



D 



H 



M" ' M ***'< ' 



5 10 15 20 25 30 35 40 45 



Size (mm) 



Figure 1 



Example output of simulated data on size frequency and maturity-at-size generated by one trial of the 

 MATSIMVL algorithm, for sample sizes of 500 (A and B ), 1000 (C and Dl. 3000 (E and Fi. .5000 (G and Hi. 

 and 10,000 (I and J) individuals. 



bias at low sample sizes and converged neatly to null 

 bias at large sample sizes (Fig. 2C). Likewise, the 

 three methods behaved exactly the same in length 

 of confidence interval, decaying exponentially as 

 sample size increased (Fig. 2C). Finally, both 

 resampling methods yielded asymmetrical (right- 

 tailed) confidence intervals, converging to the same 

 shape value (ca. 1.1) as sample size increased (by 

 definition, Fieller's method yields a symmetrical con- 

 fidence interval with shapes 1). 



Percentage success and bias of the Monte Carlo 

 method under our model for data generation were 



fairly insensitive to changes in natural mortality M 

 (Fig. 3) for sample sizes of 1000 and 5000 individu- 

 als. Percentage success remained close to the nomi- 

 nal 95'7( and bias was negligible. The length of the 

 confidence interval and asymmetry, however, in- 

 creased with increasing mortality, showing that es- 

 timation variance was directly proportional to natu- 

 ral mortality rate. 



Results of the Monte Carlo algorithm with real 

 data on female squat lobster are shown in Figure 4. 

 The Monte Carlo confidence interval for /r,,,,; was 

 fairly narrow (25.86 to 28.51 mm carapace length), 



