332 
Fishery Bulletin 115(3) 
tion models had 35 parameters that included log-scale 
mean recruitment (1), recruitment deviations (30), nat- 
ural mortality (1), survey catchability (1), and logistic 
parameters for survey selectivity (2). The models fit- 
ted age-composition data from the trawl survey with 
the multinomial distribution and biomass data from 
the trawl survey with the log-normal distribution. The 
sample sizes and CVs used to generate the age compo- 
sition and biomass data, respectively, from the trawl 
data were treated as known and used in the estimation 
models so that uncertainty was not misspecified. 
For each of the 100 replicates of recruitment that 
were generated, 100 replicates of survey age composi- 
tion and index were generated and fitted by the esti- 
mation models. For presentation we focus on the CV 
of the total biomass of the final year estimated by the 
SCAA model because this particular quantity allows 
consideration of the uncertainty in potential quantities 
of interest to management. 
Results 
Step 1 : distribution of sample sizes to determine age 
composition across multiple species 
Overall, the results of the distribution of sample size 
for age-composition among the species types investi- 
gated with the AFSC bottom trawl survey data were 
consistently similar across sampling goals (Fig. 1), 
sampling methods (Fig. 2), and whether or not aging 
error was included (Fig. 3, A-D). Proportionally speak- 
ing, the distribution of age samples was in general 
the smallest for roundfish, intermediate for flatfish, 
and largest for rockfish (Figs. 1-3). Upon combining 
sample sizes across species types, we found that the 
distribution of sample size for the collection of otoliths 
for age reading was around 10% for roundfish, 30-40% 
for flatfish, and 50-60% for rockfish (left panels, Figs. 
1-3). An interesting species that was a counter-exam- 
ple to the general results was Dover sole ( Microsto - 
mus pacificus ) in the GOA, which is the longest lived 
flatfish species investigated. In some sampling goals 
Dover sole resulted in a larger proportion of the total 
sample size than some rockfish species. Consistent pat- 
terns or large differences in the distribution of sample 
sizes in relation to location (e.g., among the GOA, AI, 
or BS) were not apparent across sampling goals, sam- 
pling methods, or cases of aging error for species that 
resided in more than one region investigated. 
A few minor differences resulted in the distributions 
of sample size for individual species across sampling 
goals (Fig. 1) and sampling methods (Fig. 2); however, 
the overall pattern of distribution by species type dom- 
inated the results of the distribution of sample size for 
collecting age samples. When aging error was included 
in the distribution of sample size, there were some dif- 
ferences in total sample size proportions among some 
of the roundfish and flatfish species but there were no 
differences when aging error was not included (for ex- 
ample, AI Atka mackerel (. Pleurogrammus monopteryg- 
ius ) or GOA Arrowtooth flounder), although, the over- 
all distribution by species type was again consistent 
(Fig. 3, A and C). When directly comparing across spe- 
cies, sampling goals, and sampling methods, we found 
that the sample sizes required when aging error was 
included were predominantly larger than when aging 
error was not included (Table 2). The slope parameter 
from a linear regression between estimated sample 
sizes that did and did not include aging error was sig- 
nificantly greater than 1 and the intercept parameter 
was significantly greater than 0, indicating that esti- 
mated sample sizes when aging error was included are 
larger than when aging error was not included (Fig. 
3E). On average, when aging error was included, the 
sample size needed to increase by around 10% for flat- 
fish and roundfish, and over 40% for rockfish to achieve 
the same level of uncertainty as when aging error was 
not included. 
Upon investigating the within- and between-length 
interval variance components across species there 
were patterns that emerged that could explain the re- 
sulting distribution of sample size across species (Fig. 
4). In general, the between-length interval variance 
was smallest for rockfish, intermediate for flatfish, 
and largest for roundfish. Alternatively, the within- 
length interval variance (under both proportional and 
fixed allocation) was, in general, smallest for round- 
fish, intermediate for flatfish, and largest for rockfish. 
Significant relationships resulted among all 4 life-his- 
tory statistics investigated and the proportion of total 
sample size across sampling goals (Fig. 5, including 
aging error, shown as an example). The weakest re- 
lationship was between the log of estimated sample 
size and the log of the growth rate at 50% of with 
coefficient of multiple determination (ix 2 ) values of 
0.61 (Fig. 5B). The strongest relationship was found 
between the log of median sample size and the log of 
the minimum lifetime growth rate with R 2 values of 
0.88 (Fig. 5D). 
Estimated sample sizes were comparable to AFSC 
bottom trawl survey sample sizes. Across the flatfish 
species investigated, the range between the average 
minimum and maximum sample sizes to achieve the 
sampling goals investigated was between 205 and 1374 
samples (Table 2), which contains the actual average 
sample size taken by the AFSC bottom trawl surveys 
and is approximately 500 samples per year (Table 1). 
For the rockfish species investigated, the range of av- 
erage minimum and maximum sample sizes to achieve 
our sampling goals was from 348 to 6494 samples (Ta- 
ble 2), which also contains the actual average yearly 
sample size of 630 samples taken by the AFSC bot- 
tom trawl surveys (Table 1). The range of the aver- 
age minimum and maximum sample sizes to achieve 
the sampling goals for the roundfish investigated was 
between 55 and 576 samples (Table 2), the average 
annual sample size taken by the AFSC bottom trawl 
surveys of around 1040 samples (Table 1) was larger 
than this range. The' estimated sample sizes were, in 
