296 
Fishery Bulletin 108(3) 
(Fig. 10). s/blOAlthough the observed ratios for WCVI 
and QCS were about 4 and 9, the ratios obtained from 
the captains’ inputs ranged from about 1 to 32 (Table 
7). The CVs in the expertly determined ratios ranged 
from about 0.1 to 0.5, conveying a range of degrees of 
uncertainty between captains (Table 7). Where the cap- 
tains’ ratios deviated most from the observed ratios and 
had the smallest CVs, the posterior probability for the 
captain was effectively zero or very low. This situation 
occurred for six of the captains (Table 7). In contrast, 
ratios that deviated considerably from the observed 
ones but had high uncertainty, e.g., for a captain whose 
mean ratio was 0.7 and CV was 0.4, still retained some 
posterior weight whereas captains with mean ratios of 
1.3 and 1.5 but much smaller CVs had effectively zero 
posterior weight. The posteriors for the six captains 
that retained most of the weight ranged from about 0.06 
to 0.38. These posteriors showed, if anything, a negative 
correlation (about -0.10 to -0.30) with the three mea- 
sures of trawling experience (years of experience, and 
tons of groundfish and bocaccio landed) (Table 8). There- 
fore, although an expert’s amount of experience may be 
a reliable indicator of his or her technical proficiency 
(i.e., the estimates of total groundfish landings and to- 
tal bocaccio catch are strongly positively correlated with 
years of experience), experience does not indicate the re- 
liability of the information that he or she may provide. 
When Bayesian updating was applied without the un- 
certainty factor, most of the weight shifted from three 
modes for q net to predominantly two modes for each net 
type (Fig. 8). For example, the third mode over the high- 
est values in the shrimp q net posterior was eliminated. 
The central tendencies for the shrimp q net and q gross val- 
ues decreased by about 40%, whereas that for the DFO 
AWII nets increased by no more than about 5% with the 
Bayesian update. Under these same conditions, preci- 
sion in the q aross distribution for each survey increased 
only for the shrimp surveys (Table 6). Bimodality was 
no longer present in the q gross distributions (Fig. 9). In 
contrast to the instance with no uncertainty factor and 
no Bayesian updating, when Bayesian updating and the 
uncertainty factor were applied, the q net distributions 
by captain all overlapped for each of the 
three nets (Fig. 10). The mean value for 
q net for the shrimp trawl was lower and 
more uncertain than for the two ground- 
fish nets (Fig. 8). None of the q net and q gross 
composite distributions showed bimodality 
and results were not quite as precise as 
with the analogous case without the un- 
certainty factor (Figs. 8 and 9, Table 6). 
The CVs in the groundfish survey q gross val- 
ues ranged from about 0.77 to 0.83; and 
95% probability intervals (Pis) ranged be- 
tween 22- and 25-fold between the bounds 
(Table 9). The CVs for the shrimp survey 
q gross values were considerably higher at 
about 1.5 and 2.7 (95% Pis of about 84- 
and 3100-fold). This high uncertainty was 
largely due to large differences between the 
inputs provided by captains but also due to 
higher uncertainty in the fraction of stock 
in these surveys (Table 2). The U.S. trien- 
nial survey q gross had a high CV (1.7) and 
an 800-fold 95% probability interval (PI). 
This high uncertainty is also due mainly to 
the high uncertainty in the fraction of the 
stock in this survey (Table 2). 
One key factor is the ratio of fish density 
in untrawlable areas to that in trawlable 
areas, a. When this factor was set to 1 and 
the Bayesian update and uncertainty fac- 
tor were applied, the central tendencies of 
the posterior distributions for all of the 
surveys approximately doubled, indicat- 
ing that the effect of this parameter is to 
decrease q aross for all surveys and to lead 
to increased population biomass estimates 
(Table 6). Doubling the mode and maxi- 
mum value for a from 3 and 10 to 6 and 
20 caused the mean for q gross to decrease 
Figure 6 
Probability density functions for trawl survey net catchability (q net ) 
for each captain for the (A) Atlantic western (AWII), (B) shrimp, 
and (C) Nor’Eastern trawl nets, based on inputs provided for each 
captain without any Bayesian updating and without the uncertainty 
factor applied. 
