Hinton et al. : Use of gonad indices to estimate the reproductive status of Xiphias gladius 
83 
Table 3 
Comparison between the correct (from histological analy- 
ses [HA] of the ovaries) and estimated (from GI’s) classifi- 
cation of reproductive status of female swordfish. Individu- 
als were classified as reproductively active (A) or quies- 
cent (Q). Asterisks designate incorrect classifications based 
on GI’s, IC is the percentage of all individuals [ n deter- 
mined by H(i)] classified correctly, and AC is the percent- 
age of reproductively active individuals, within the n indi- 
viduals, that were classified correctly. 
H(i) 
n 
HA 
GI 
A 
GI 
Q 
IC 
AC 
1 
85 
A 
19 
2* 
95.3 
90.5 
Q 
2* 
62 
2 
48 
A 
10 
11* 
77.1 
47.6 
Q 
0* 
27 
3 
46 
A 
10 
11* 
76.1 
47.6 
Q 
0* 
25 
4 
52 
A 
17 
4* 
92.3 
81.0 
Q 
0* 
31 
5 
85 
A 
15 
6* 
92.9 
71.4 
Q 
0* 
64 
6 
85 
A 
21 
0* 
95.3 
100.0 
Q 
4* 
60 
cant term in logistic regressions that included EFL 
as a classification variable. 
Under H(l) and H(6), 95% of the 85 individuals 
were classified correctly (Table 3); further, under H(6) 
all individuals that were reproductively active were 
correctly classified, which was significantly (see fol- 
lowing discussion) more than the 91% of the repro- 
ductively-active individuals correctly classified un- 
der H( 1) and the 71% to 81% correctly classified un- 
der H(5) and H(4), respectively. Hypothesis H(6) 
placed about 4.7% of the individuals that were qui- 
escent in the active category, and H(l), about 2.5%, 
both of which represent relatively low rates of type- 
2 error. 
Although length-cubed is often the choice to stan- 
dardize GW, as in the “standard” expression for GI, 
length is also frequently used. Further, GW may be 
exponentially related to body size (de Vlaming et al., 
1982), in which case log transformation as in GI(2) 
is indicated. We examined these hypotheses in for- 
mulations of GI for female swordfish. The results of 
ANCOVA revealed significant (P<0.01) heterogene- 
ity among slopes and intercepts of the regressions of 
GW on EFL and on EFL 3 for reproductive classes (2, 
3, 4, [5, 6] and 8) of Taylor and Murphy (1992). At 
the same time, ANCOVA on the log-transformed data 
yielded only one significant (P<0.01) coefficient, that 
for the intercept of class (5, 6); however this coeffi- 
cient was only about 28% of the estimated intercept 
of the overall regression. Thus, the formulation of 
the gonad index that best conformed to the underly- 
ing assumptions (de Vlaming et al., 1982) was GK2). 
In addition, the maximum-likelihood test based on 
the values of LKLHD for the difference between 
methods showed that model GI(2) provided a signifi- 
cant ( j 2 ) 1), P=0.033) improvement over model GI(l). 
The estimate of OV obtained from maximum-likeli- 
hood analyses for GI( 2) was (1.366 < OV < 1.375). 
Note that although GK2) has a continuous distribu- 
tion, the interval estimate of OV is a function of the 
distribution of values of GI(2) in the sample data, 
and thus any hypothesized value in this range would 
yield LKLHD and tabled results identical to those 
shown for H(6). Further, because the solution for the 
function LKLHD is so knife-edged, the estimate of 
OV includes the 90% confidence interval, and the 95% 
confidence interval for the estimate of OV, (1.357 < 
OV < 1.375), differs only in the lower bound. 
Two points should now be clear. First, the classifi- 
cation methods that are based on GI(1) that have 
been used and published in studies requiring esti- 
mates of the reproductive status of female swordfish 
do not meet the underlying assumptions (de Vlaming 
et al., 1982) for use as an indicator of reproductive 
status, and they may also be viewed in some in- 
stances as overly restrictive, in that they may have 
excluded significant amounts of usable data from 
analyses that were already hampered by limited in- 
formation. This resulted, at least in part, from using 
values of GI that corresponded to a fully ripe and 
running condition of the gonad. Second, the classifi- 
cation methods, both previously published and de- 
scribed herein, were not impacted by minimum- 
length criteria which may be required to standard- 
ize comparisons of statistics that are based on gonad 
indices. 
Our results are conservative in the following re- 
spect. We have no knowledge of the frequency of 
spawning of female swordfish. Thus, because hydra- 
tion of eggs may occur over a very short period of 
time, by not including individuals in class 4 (“ma- 
ture ovaries” of Taylor and Murphy [1992]), some 
individuals that might be expected to spawn within 
a short period of time, and thus presumably within 
the general area of capture, may be excluded from 
consideration. The question of whether to include 
these individuals as reproductively active could be 
addressed by conducting a study of spawning fre- 
quency of female swordfish based on the condition of 
yolk development in the eggs, as has been done for 
yellowfin tuna, Thunnus albacares (Schaefer, 1996). 
Alternatively, it may be possible to determine 
whether or not class-4 individuals should be included 
