316 



Fishery Bulletin 90(2). 1992 



Estimates of the first three size-classes were both 

 biased and imprecise. Poor estimates of the smallest 

 few size-classes were expected. These classes lacked 

 a catch history at the time of the last sample, so these 

 estimates of abundance (at the time of the last relative 

 abundance sample) were a function of the last-period 

 relative abundance sample only. 



Examples 



Two tests were used to discover what might be ex- 

 pected when assessing populations with no periodicity 

 in recruitment at all; recruitment dates were complete- 

 ly protracted imiformly through time (Fig. 1). Most con- 

 trol variables were the same in the two tests. Data were 

 assumed to be available in two-unit size intervals. A 

 120-unit asymptotic size fell in size-class 60, and a 

 30-unit recruitment size in class 15. The growth param- 

 eter k was left at 0.17. Continuous fishing was simu- 

 lated; the fishing mortality rate (F) for each period was 

 drawn from a 11(0.3,0.8) distribution. The expectations 

 of sampling efficiencies (q^) were arbitrarily chosen so 

 that their regression on size was sigmoid, reaching an 

 asymptote at size-class 30 (0.028, 0.031, 0.033, 0.038, 

 0.044, 0.053, 0.069, 0.101, 0.153, 0.190, 0.218, 0.234, 

 0.242, 0.247, 0.249, and 0.250). Catch estimates were 

 simulated to be imprecise (cv 0.4). A 10% growth mea- 

 surement error was simulated. Sampling intensities 



were the same in both tests; sample sizes for growth 

 parameter estimates and for relative abundance obser- 

 vations were such that a 95% CI was of width ±5%. 



Although the levels of population processes were the 

 same in both tests, the amoimt of process variability 

 was much higher in test 2. Normal growth variability 

 was simulated in test 1 and extreme variability in test 

 2. The rate of unobserved change in test 1 was con- 

 stant, but varied three-fold in test 2. The variance of 

 sampling efficiencies was set one order of magnitude 

 larger than that observed for commercial fishing gear 

 in test 1 and twice that in test 2. 



Error variances-of-abundance estimates were very 

 low in the case of normal process variability (Table 4). 

 Estimates of all but the smallest six size-classes were 

 biased by 10% or less, if at all, and were precise. Bias 

 (more than 10%) and imprecision of the smallest six 

 size-class estimates was expected because the smaller 

 fish were barely represented in the catch and appeared 

 in the relative abundance samples just once. Estimates 

 of the sampling efficiencies (the qs) tended to be im- 

 precise. Some were biased from 20% to 30% and a few 

 even more. The estimate of survival from unobserved 

 change was biased low (about 15%), yet precise. Esti- 

 mates did not tend to be correlated. The correlations 

 between the estimate of the unobserved change rate 

 (z) and other estimates (Table 5), particularly of the 

 C[s , were of interest because other studies found corre- 



