196 
Fishery Bulletin 116(2) 
shaped selectivity means that there are older individu¬ 
als in the population that are not subject to fishing 
pressure. The eliminated data scenario assumes what 
might be considered a more precautionary assumption 
for selectivity in the absence of composition data. 
Sensitivity to adding survey data 
Additional simulations were conducted to evaluate the 
impact of having only fishery information versus indi¬ 
ces of abundance and length- and age-composition data 
available from both a fishery-independent survey and 
a fishery. The operating model generated a highly un¬ 
certain survey (coefficient of variation: 0.40) that was 
conducted on a biennial basis with low sample sizes 
(n- 10 per year) for length- and age-composition data 
starting in year 40, 10 years before the first assess¬ 
ment in year 50. The survey selectivity was assumed 
to be fixed at an asymptotic shape, selecting fish at 
smaller sizes in relation to the fishery selectivity. All 
other specifications for the fishery within the operating 
model and the assumptions applied by the estimation 
method were the same as those detailed previously. 
Performance measures 
The outcomes of the simulations for each case and 
data scenario were summarized by using 5 metrics 
that were selected to evaluate the effect of data on 
estimation of indicators of stock status (e.g., relative 
spawning biomass) and management quantities (e.g., 
rebuilding catch). 
1 The relative errors (REs) for estimated parameters, 
calculated as 
RE = (8) 
whereE = the estimated quantity of interest; and 
T = the true value from the operating model. 
The REs for spawning biomass and relative spawn¬ 
ing biomass were calculated for each simulation for 
the ending year estimate each time the simulated 
stock was assessed. 
2 The percent root mean square error (RMSE), a mea¬ 
sure of precision and bias, was calculated to assess 
the overall level of error given the amount of data 
available: 
RMSE = jizu'-^jjr^, (9) 
where n is the number of simulations (re=100) 
3 The average (over simulations) of the total catch 
while the simulated stock was recovering to the tar¬ 
get biomass. 
4 The annual average variability of the catches (AAV), 
defined as 
A4V = 100 It ^ t ~ Ct+1 L (10) 
where C t = the catch during year t. 
5 The percentage of simulations with stocks that re¬ 
built to the target biomass and percentage of simu¬ 
lations with stocks that remained overfished at the 
end of the management period. 
Results 
Assessment performance with time-invariant parameters 
The full and reduced data scenarios performed simi¬ 
larly while simulated stocks were rebuilding and after 
stocks had rebuilt, and the trends of the relative er¬ 
ror for spawning biomass and relative spawning bio¬ 
mass were generally consistent between the full and 
reduced data scenarios (Fig. 3, A, B, D, and E ). The 
median estimates of spawning biomass and relative 
spawning biomass were less than the true values dur¬ 
ing rebuilding for both scenarios (Fig. 3, A, B, D, and 
E). As expected, the full data scenario had less arnong- 
simulation variability in the differences in spawning 
biomass and relative spawning biomass between the 
operating model and estimation method during the 
rebuilding period than the variability in the reduced 
and eliminated data scenarios (Fig. 3, A-F). However, 
the among-simulation variability of errors in biomass 
metrics was similar between the full and reduced data 
scenarios by the end of the management period, when 
a majority of the simulated stocks were estimated to be 
rebuilt and data collections had returned to historical, 
higher sample sizes for the reduced data scenario. 
The eliminated data scenario in which no data were 
available during the rebuilding period resulted in me¬ 
dian (across simulations) estimates of spawning bio¬ 
mass and relative spawning biomass errors that were 
similar to the true values but were highly imprecise 
at the start of the management period (years 50-74) 
(Fig. 3, C and F). The eliminated data scenario, in the 
absence of new data during rebuilding, projected the 
simulated stocks on the basis of the historical data and 
new catches until the simulated stock was rebuilt, at 
which time data collection resumed and allowed the 
estimation method to estimate population status. The 
median estimates of spawning biomass and relative 
spawning biomass for the eliminated data scenario 
were less than the true values, and had high among- 
simulation variability in error as simulated stocks be¬ 
gan to be projected to be rebuilt and data collection re¬ 
sumed. In contrast to the full and reduced data scenar¬ 
ios, the estimates of spawning biomass and the relative 
spawning biomass for the eliminated data scenario had 
little improvement in the among-simulation variability 
in error estimates by the end of the management pe¬ 
riod (Fig. 3, C and F). 
Even when data collection continued at reduced 
levels in the reduced data scenario, the estimates of 
steepness varied in relation to the steepness estimates 
from the full data scenario. The full data scenario re¬ 
sulted in generally median unbiased estimates during 
the rebuilding period and small positive median bias by 
