282 
Fishery Bulletin 119(4) 
UCD UU UU 
Table 4 
Mean, median, standard deviation, and coefficient of variance (CV) for estimates from the Patch 
model for the following parameters: dredge efficiency; clam density; the k parameter, which is the 
negative binomial dispersion parameter; effective area swept (EAS); and the number of tows of the 
dredge. Estimates are based on the 50 depletion experiments conducted in the field between 1997 
and 2011 for populations of ocean quahogs (Arctica islandica) and Atlantic surfclams (Spisula 
solidissima) off the mid-Atlantic coast of the United States. The standard deviation and CV val- 
ues for efficiency and density are the averages of the delta method uncertainties associated with 
parameter estimation. n=number of experiments. 
Statistic Efficiency 
Ocean quahog (n=19) 
Mean 0.586 
Median 0.629 
Standard deviation 0.113 
Coefficient of variance 0.357 
Atlantic surfclam (n=31) 
Mean 0.635 
Median 0.590 
Standard deviation 0.131 
Coefficient of variance 0.206 
negatively correlated with the number of tows and strongly 
positively correlated with the CV of the density estimate 
(CV)) (Figs. 4 and 5). In depletion experiments with Atlantic 
surfclams, as opposed to experiments with ocean quahogs, 
the CV, is significantly positively correlated with the CVp 
(Fig. 5). In the case of the Atlantic surfclam, no correlation 
exists between latitude and the efficiency estimates, but 
density estimates are negatively correlated with the lati- 
tude and efficiency estimates. 
Error estimates and Wilcoxon rank sum tests 
Field depletion experiments with parameter estimates 
that fall at or above the 80th percentile of their respective 
most comparable simulated experiments, for 1 or more of 
the 4 error estimates, are denoted by asterisks in Table 5. 
We used the 80th percentile, corresponding to a 90th per- 
centile one-sided threshold, to retain a high probability of 
including marginal experiments in the group flagged as 
suspect, recognizing that this threshold may entrap some 
experiments of higher quality. Effectively, the goal was to 
err on the side of removing a few “good” field depletion 
experiments rather than keep a few “bad” ones. 
Of the 50 depletion experiments, 24 experiments had 
estimates for 1 or more of the 4 error terms that fall at or 
above the 80th percentile. Experiments with estimates for 
error terms Err1 (Equation 9) and Err2 (Equation 10) at 
or above the 80th percentile are experiments that differed 
substantially from the chosen subset of simulations for 1 
or more of the 4 characteristics that describe the depletion 
experiments, the CV,, the CVx, the number of tows, and 
the EAS. These field experiments were not well described 
Parameter 
Density k 
(individuals/m?) 
EAS (m?) 
parameter 
7.724 13,688.81 
6.165 13,746.99 
3.045 13,471.86 
0.613 13,471.86 
12.097 13,570.90 
5.689 7325.56 
3.011 14,653.21 
0.351 0.104143 
by the most similar subset of simulations. The possibility 
that the range of values for EAS might influence the dif- 
ferential in the results for Errl and Err2 was tested by 
recomputing Err2 by using log.(HAS). The set of experi- 
ments flagged by Err2 did not change. 
The error terms Err3 (Equation 8) and Err4 (Equation 11) 
provide an alternate method from Err1 and Err2 to identify 
field experiments that potentially produced unreliable effi- 
ciency estimates. In this case, error in efficiency estimates 
from a subset of similar simulations, each with a known 
error in their efficiency estimate, were compared. Field 
experiments most similar to simulated experiments that 
yielded high values for Err3 and Err4 were flagged. 
When field experiments produced parameter estimates 
that fall at or above the 80th percentile for an error term, 
they were flagged by that error term (Table 6). When 
these field experiments were compared with the exper- 
iments that produced estimates that fall below the 80th 
percentile, some differences in Patch model estimates of 
gear efficiency and clam density came to light, highlight- 
ing that the 4 error measures are operationally different 
and can be used to evaluate experiment performance in 
different ways. For example, the experiments that were 
flagged by Err2, Err3, and Err4 produced lower average 
and median efficiency estimates than those from experi- 
ments flagged by Err1. 
The relationships between data from the field experi- 
ments flagged by one or more error terms and the rest of 
the data set were evaluated by using Wilcoxon rank sums 
tests (Table 7). Experiments flagged by Err1 did not differ 
significantly from the remaining experiments for any of 
the measured depletion parameters. In each case, the 
