McAllister et al : Using experiments and expert judgment to model catchability of Pacific rockfishes 
303 
Table 12 
Posterior means and coefficients of variation (CVs) of key stock assessment quantities for bocaccio ( Sebastes paucispinis ) obtained 
with a noninformative and informative prior for q gross (see Tables 6 and 9 for the inputs for the informative prior). See Stanley et 
al. (2009) for the stock assessment method applied. B (t) refers to the population biomass that provides the maximum sustain- 
able yield in tons. B 2008 (t) refers to the estimated population biomass in the year 2008 in tons. 
B msy ( t) 
CV 
B 2008 d) 
CV 
Replacement yield (t) 
CV 
Non-informative q gross prior 
24,146 
0.68 
4697 
2.27 
310 
1.24 
Informative q gross prior 
27,021 
0.66 
3022 
0.83 
236 
0.65 
it was felt that these other surveys covered most of 
the population’s range. Also, in contrast to the present 
study, in these other studies it was effectively assumed 
that inputs were obtained from only one expert and did 
not formally account for cross-expert uncertainty. 
All other studies so far have developed priors for q 
that have zero prior correlation (i.e., independence was 
assumed). In contrast, we developed a mixed-model 
structure for survey q that produces strong nonzero 
correlation in q aross values between different surveys — a 
necessary consequence because the information sources 
used to produce the q gross values for different surveys 
are not independent. The prior correlation between q gross 
values for different surveys was very high in some in- 
stances (up to 0.96) because different surveys were us- 
ing the same gear. This high correlation resulted in 
the same inputs for a given gear type feeding into the 
formulation of the q net factor across different surveys. 
It is important to include this correlation in the prior 
for q L , ross in a stock assessment because it accounts for 
the dependencies between the q gross values for differ- 
ent surveys. Use of the marginal prior variances and 
assuming independence, i.e., applying zero correlation, 
would overstate the amount of prior information avail- 
able about q gross . 
In contrast to the norm, of which experts tend to be 
overly certain, all captains in this study expressed con- 
cern about their estimates. They commented that there 
had been few opportunities in their careers to compare 
actual catches with acoustic signals for bocaccio. Three 
captains said that they could not provide an estimate for 
at least one question. All captains expressed that they 
would have been more comfortable estimating these val- 
ues for other schooling rockfish, particularly yellowtail 
rockfish and widow rockfish (S. entomelas) because of the 
greater opportunity to correlate acoustic observations 
with observed catches. Furthermore, they commented 
that for bocaccio, as well as other species, catchability 
would be influenced by factors such as location and 
bottom type, time of day, state of the tide, and whether 
the fish were present in large schools or were solitary. 
The Bayesian computations in our approach that vet 
the expert-specified inputs against survey-observed val- 
ues for the same quantities had the result of excluding 
the inputs from about half of the captains. This situa- 
tion is undesirable from the point of view of an attempt 
to include different viewpoints. However, it provides 
an empirical basis for screening the inputs provided 
by different experts. More conventional measures of 
experience (e.g., years of experience, total groundfish 
landings, and total bocaccio catch) showed either no 
correlation or a negative correlation with the amount 
of posterior weight placed on the captains. This finding 
indicates that practitioners should avoid applying ap- 
parently sensible criteria to formulate weights to inputs 
from different experts. Comparison of empirical data 
with the expert advice within the context of a model ap- 
pears to provide a reasonably objective way of screening 
such advice and should be considered instead. 
One of the most poorly understood parameters is 
the ratio of rockfish density in untrawlable to that in 
trawlable areas. In this analysis, a subjective prior was 
applied which ranged between 1 and 10, with a mode at 
3. This application had the effect of reducing the central 
tendency of the q gross by half for all surveys which would 
give larger estimates of population biomass. Doubling 
the width and mode of the input distribution for a fur- 
ther decreased the mean value for q gross , although by 
no more than about 37%. We have suggested a simple 
Bayesian approach to updating this prior, using esti- 
mates of a from experiments. Possible approaches to 
estimating a could include experiments designed to 
estimate relative density in trawlable and untrawlable 
locations with gillnets, hook-and-line sampling gear, or 
submersible vessels (Kreiger and Sigler, 1993). 
The q aross prior developed for British Columbia bocac- 
cio was applied in a recent stock assessment of this 
population. The availability of these priors was crucial 
because most of the survey index series were quite 
short and all had low precision. Although the prior CV 
was very high, i.e., no less than about 0.8, this mildly 
informative prior still helped to bound the range of 
plausible hypotheses about current stock size and re- 
placement yield. The posterior medians for q gross were 
also within the prior 95% Pis when a noninformative 
prior for it was applied, indicating that the uncertainty 
obtained in the priors was reasonable and that the 
priors were consistent with values indicated by the 
fit of the assessment model to the data. Thus, in this 
application, the method provided useful inputs for a 
stock assessment by bounding the range of values for 
estimated parameters and reducing uncertainty in key 
management quantities. The higher precision obtainable 
in stock assessment results when a noninformative prior 
