378 
Fishery Bulletin 117(4) 
In the summer of 2017, 3 experimental gears were 
tested, with each including a reduced-bar TED, 1 state 
fisheye, and a 3.81-cm codend (Table 1, Fig. 4). Following 
5 paired tows with less than desired reductions of finfish 
bycatch observed, the experimental gear in the trial was 
changed to include a reduced-bar TED, 1 state fisheye, 
and a 4.13-cm stretch mesh codend (Table 1, Fig. 4). The 
next tested gear, in the summer of 2017, was a reduced- 
bar TED with 2 state fisheyes and a 4.13-cm stretch 
mesh codend (Table 1, Fig. 4). In the autumn of 2017, 
2 experimental gears were tested. The first included a 
reduced-bar TED, 2 state fisheyes, and a 4.13-cm stretch 
mesh codend (Table 1, Fig. 4). Finally, a modification 
of that gear, consisting of a reduced-bar TED, 2 federal 
fisheyes, and a 4.13-cm stretch mesh codend, was tested 
(Table 1, Fig. 4). 
Analyses 
Any problematic tows (as described previously) were 
dropped. Catch per unit of effort (CPUE) was calculated 
as weight divided by tow time; observed weights were 
standardized to either a 2-h tow time (2015-2016) or a 1-h 
tow time (2017) to adjust for differences in tow times. The 
CPUE values were calculated with the assumption that 
there were no differences in spread ratio between trawls 
with various BRD and TED configurations. The 1-h tow 
time was used in 2017 because of the smaller vessel and 
net size. The average weight of catch was computed for 
each gear (control and experimental) and taxonomic group 
along with the differences in average weight of catch and 
percent change in CPUE. 
Paired t-tests were used to evaluate any differences in 
average CPUE values between the control and experi¬ 
mental nets by using the SURVEYSELECT procedure 
in the software SAS 9.4 15 (SAS Institute Inc., Cary, NC). 
The null hypothesis for the £-test was that there was no 
difference in CPUE between the control and experimen¬ 
tal nets (significance level of 0.05). The procedure was 
used to account for any uneven number of tows by gear 
type per vessel side made during trials and to randomly 
select an even number of tows to minimize side-bias. This 
procedure resulted in the random exclusion of tows for 
some analyses. 
One of the assumptions of the £-test is that the depen¬ 
dent variable (in this case, CPUE) is normally distrib¬ 
uted within each group. Because ecological data are often 
non-normal, a second test that does not require nor¬ 
mality of the data distribution was applied to the data. 
Specifically, a randomization procedure (Manly, 2007) 
was also used to compare CPUE between control and 
experimental nets for each gear and taxonomic group. 
The null hypothesis for the randomization test was that 
there was no difference in CPUE between the control 
and experimental nets. The test statistic evaluated was 
15 Mention of trade names or commercial companies is for identi¬ 
fication purposes only and does not imply endorsement by the 
National Marine Fisheries Service, NOAA. 
the difference in average CPUE between the control and 
experimental nets. Data were randomized and resam¬ 
pled 10,000 times for each gear and taxonomic group. 
The P-value from the randomization test for a 2-tailed 
test was the proportion of test statistics (including those 
computed from the original data) that were as large or 
larger in absolute value than the absolute value of the 
statistic computed from the original data. Code for this 
test was written and run in the R statistical program 
(vers. 3.4.3; R Core Team, 2017). 
Results 
A total of 412 paired tows were made by using 14 exper¬ 
imental gear configurations. Testing occurred on 39 trips 
across 104 fishing days by a total of 9 different industry 
fishing vessels. A trip was defined as the time period that 
began when a vessel departed port to conduct fishing oper¬ 
ations and ended with a return to port. Observers sampled 
and weighed over 142,000 kg of fish, shrimp, and other 
marine organisms. An even number of tows by side of ves¬ 
sel was not achieved for most gears (Table 2), and 81% 
of paired tows were made during the day (Table 2). The 
results of both the paired t-tests and the randomization 
tests for testing completed from 2015 through 2017 can be 
found in Tables 3-5. 
Paired f-tests 
None of the 5 experimental gears tested in 2015 lost a sig¬ 
nificant amount of shrimp (P>0.05; Table 3). Only with the 
Ricky BRD did we fail to achieve a significant reduction in 
finfish bycatch compared with that achieved by using the 
control gear (6.6% reduction with Ricky BRD, P=0.503). 
However, testing of this gear was discontinued after only 
15 tows. Only by using the configuration with double fed¬ 
eral fisheyes and a 4.76-cm codend was a greater than 40% 
reduction in bycatch (40.8%) achieved while maintaining 
shrimp catch (1.0% gain). 
In 2016, no significant shrimp loss was observed during 
tows of the 4 gears tested (Table 4). With the Virgil Potter 
gear, we achieved a significant gain in shrimp harvest over 
that during tests with the control net (9.9% gain, P=0.050). 
Significant reduction in finfish bycatch was observed with 
the use of all test BRDs in 2016, and the use of 3 of the 
4 gears resulted in bycatch reduction in excess of the 
40% target in a range of 43-57%. 
In 2017,1 of the 5 gears tested was not analyzed by using 
the t-test, because of low sample size. When 3 of the remain¬ 
ing 4 gears were tested in 2017, significant shrimp loss was 
observed (Table 5). Tests of the gear with a reduced-bar 
TED, double state fisheyes, and a 4.13-cm codend during 
summer, the gear with a reduced-bar TED, double state 
fisheyes, and a 4.13-cm codend during autumn, and the 
gear with a reduced-bar TED, double federal fisheyes, and a 
4.13-cm codend resulted in shrimp losses of 6.8%, 14.9%, and 
9.0%, respectively. By using 2 of these gears, the reduced- 
bar TED with a 4.13-cm codend and the reduced-bar TED 
