Munro: Correcting relative fishing power differences in trawl survey data 
545 
ing the variance of the mean CPUE (Fig. 4A). The 
noncorrection region of data thus extended far to the 
left (Fig. 30. When the nonstandard vessel was more 
efficient, the imposed fishing power difference had 
the opposite effect, increasing the variance of the 
mean calculated from uncorrected data (Fig. 4A). 
Correcting even moderate fishing power differences 
reduced the MSE (Fig 3C), but by reducing variance, 
not by correcting systematic error. This problem is 
purely a consequence of the fishing power mechan- 
ism in the simulations. An improved mechanism is 
needed to render the MSE-based decision rule a 
clearer tool. In this case, with highly skewed CPUE 
distributions, there may be a cut-off point beyond 
which a fishing power difference would not be im- 
posed, because the rare, large tows are more likely 
to be chance occurrences than indicators of fishing 
power. Note that the bias did behave as expected: 
the uncorrected data produced a minimum when the 
fishing powers were identical, and corrected data pro- 
duced a constant bias that was always lower than 
that of the uncorrected data (Fig. 4B). The pollock ex- 
ample illustrates how an MSE-based decision rule can 
be deceiving under an improperly specified model. 
The noncorrection region itself is an estimate, with 
error around the upper and lower bounds of the 
range. The observed fishing power difference or cor- 
rection factor will also be an estimate with its own 
error. Between these two sources of uncertainty, the 
upper and lower bounds of the noncorrection region 
are not as clear as they might appear in an applica- 
tion. Thus, a conservative approach would be to de- 
cide against using an FPC even when the observed 
value falls a little outside the noncorrection region. 
For this reason, and because improved precision gen- 
erally increases certainty, it would pay to narrow the 
noncorrection region as much as possible through 
estimators with less sensitivity to the rare, large 
observations that characterize much CPUE data and 
through careful modeling of the process leading to 
the fishing power difference. 
This decision strategy can also be used to re-evalu- 
ate old or changing fishing power problems. The con- 
cept of minimizing the MSE can accommodate 
changes in state-of-the-art estimation and allow the 
consequence of change to be examined. The decision 
rule remains valid regardless of improvements in 
simulation strategies, estimators, or understanding 
of processes leading to fishing power differences. The 
strategy can also be used to decide if experiments 
are warranted to calibrate the fishing power of re- 
search vessels. Such experiments tend to be very 
expensive yet limited in scope. Will it be possible to 
attain a sample size great enough to produce mean- 
ingfully narrow noncorrection regions? 
A decision not to apply an FPC by these methods 
does not in any way imply that the fishing power 
difference is trivial or unimportant. For instance, if 
one of two vessels in a survey had a catch rate 30% 
lower than the other, and both vessels had equal 
numbers of observations, then the estimated mean 
CPUE could be in error by as much as 15%. An error 
of this magnitude could have a profound influence 
on a heavily exploited, closely managed stock. Yet 
such a fishing power difference may go uncorrected 
because, to correct it, would involve a risk of error of 
even greater magnitude. A decision against correct- 
ing is only a conclusion about the feasibility of ap- 
plying an FPC, not the cost of having the systematic 
error in the data. Once a fishing power difference 
occurs in survey data, an irrevocable mistake has 
been made. The final estimate will have increased 
error, whether it is due to bias added by systematic 
error or due to variance added by estimating and 
correcting for the bias. This example illustrates the 
importance of maintaining standard survey tech- 
niques by illustrating the cost, in terms of precision, 
of correcting fishing power differences. 
The decision to apply an estimated FPC is diffi- 
cult. Statistical significance of a fishing power dif- 
ference does not necessarily permit inference regard- 
ing the consequence of applying a correction factor 
when estimating mean CPUE. These three examples 
demonstrate the usefulness of this decision procedure. 
Even under difficult circumstances, high variance and 
an estimator sensitive to rare, large observations, the 
noncorrection regions are easy to defend because they 
are based on a clear goal — that of minimizing the error 
of the estimate of mean CPUE. 
Acknowledgments 
Russ Kappenman demonstrated high endurance as 
patience, as well as providing several critical insights 
during the writing of this paper. Dave Somerton pro- 
vided support and encouragement for this work and, 
as an editor, helped to produce a much improved 
paper. Thanks to Jim lanelli, Bob McConnaughey, 
Susan Picquelle, and Pat Sullivan for very helpful 
critical reviews. Thanks to an anonymous reviewer 
whose careful reading and suggestions contributed 
greatly. 
Literature cited 
Aitchison, J., and J. A. C. Brown. 
1957. The lognormal distribution, with special reference to 
its uses in economics. Cambridge Univ. Press, Cambridge, 
176 p. 
