McAllister et al.: Using experiments and expert judgment to model catchability of Pacific rockfishes 
299 
Figure 9 
Probability density functions for trawl survey catchability (q gross ) for the 
(A) west coast Vancouver Island (WCVI) groundfish, (B) WCVI shrimp, 
and (C) U.S. triennial groundfish surveys, based on inputs provided for 
each captain with and without any Bayesian updating, and with and 
without the uncertainty factor applied. 
perch. Korotkov 1 used an underwater 
camera-mounted sled towed in front 
of the trawl to provide a ground-truth 
of actual fish density. He estimated 
a doorspread catchability of 0. 1-0.4 
for unspecified species of groundfish. 
Our estimates of q net for bocaccio for 
three different trawl net types ranging 
from about 0.1 to 0.3 are similar to 
his estimates. Scientists at the Alaska 
Fisheries Science Center (NMFS) have 
spent many years attempting to esti- 
mate catchability of the trawl used in 
their west coast groundfish surveys. 
Their most successful work was with 
flatfish for which they observed maxi- 
mum door-spread catchability for large 
arrowtooth flounder ( Atheresthes sto- 
mias) of 0.47 (Somerton et al., 2007). 
The catchability for a flatfish such as 
arrowtooth flounder could be expected 
to be higher than that for rockfishes 
because bottom trawl nets are gen- 
erally designed to capture flatfishes, 
which tend to stay very close to the 
bottom as opposed to many rockfishes, 
such as bocaccio, which tend to dis- 
tribute themselves higher in the water 
column. 
Previous efforts at indirectly form- 
ing a prior for q gross involved seeking 
expert judgment and, in some instanc- 
es, adding auxiliary data to the com- 
ponents of q gross and then integrat- 
ing these components within a Monte 
Carlo framework to formulate a pdf 
for q gross (McAllister and Ianelli, 1997; 
Boyer et al., 2001). Different approach- 
es were used to elicit information from 
experts. In some instances, the experts 
were interviewed separately to gain in- 
formation on key factors determining 
Qgross Punt et al. 1993, McAllister 
and Ianelli, 1997; Mosqueira, 2005). 
Another approach was to put several 
experts in the same room so that they could form a 
consensus on these factors (Boyer et al., 2001). An im- 
portant limitation to these approaches has been that 
the posterior distributions often tend to be very narrow 
as a result that too few experts were consulted or that 
divergent opinions were forced into a consensus. 
That experts often hold divergent views while each 
being certain about his or her knowledge has long been 
recognized as a problem when forming priors based on 
expert input. In the last decade, a number of analysts 
have suggested that it is important to retain this diver- 
1 Korotkov, V.K. 1984. Fish behaviour in a catching zone and 
influence of bottom trawl rig elements on selectivity. Int. 
Council. Explor. Sea, Council Meeting 1984. B:15. 
sity as an output of the analysis and that it is unwise 
to eliminate the diversity by averaging across experts 
(Burgman et al., 1993; Chrome et al., 1996; Uusitalo et 
al., 2005). Some researchers have advocated assigning 
weights to experts according to their level of expertise 
(Burgman et al., 1993); others have assigned equal 
weighting to expert input, providing that all of these 
experts initially qualify to provide expert judgment 
(Martin et al., 2005; Uusitalo et al., 2005). It may not 
be desirable to assign different weights when the num- 
ber of available experts is relatively few, because it is 
possible that a single expert may end up with all of the 
weight, thus defeating the purpose of the exercise. 
We recommend applying an uncertainty factor to the 
for q net variable obtained from each expert’s inputs to 
