45 50 



LN LENGTH (CM) 



Figure I. — Blue marlin data from longline data are 

 plotted on a log-log scale to show the existence of two 

 growth stanzas. The straight lines were fitted by eye. 



was the only species exhibiting such a trend (Fig. I ) 

 and then only for the longiine-caught fish. Although 

 it was quite evident that two growth stanzas existed, 

 there were too few data to determine exactly where 

 the two stanzas met or overlapped. We arbitrarily 

 took the two data points at 135 cm FL (4.9 in natural 

 logarithms) as the overlap area, with the assumption 

 that the length-weight relationship for the older, 

 well-represented stanza should be accurately pre- 

 dicted even if it actually began at a smaller size while 

 that for the younger stanza is provisional. The 

 younger growth stanza was treated separately in the 

 subsequent analyses. 



Log-Linear Model 



The log-linear model (Equation 2) was fitted to the 

 data for all species (Table 2). The"F" testsforblack 

 marlin, sailfish, shortbill spearfish, and swordfish 

 were highly significant. Though the idea that a log- 

 linear relationship between weight and fork length 

 might not exist was rejected, this was a provisional 

 conclusion because the validity of the statistical tests 

 could not be checked. The proportion of the total 

 variation accounted for by the regression, R^, was 

 high for all species except for the shortbill spearfish, 

 where the usefulness of the relationship as a predic- 

 tor was not great. For striped marlin, although the 

 "/?-" value was high, the distribution of the error 

 term was not normal. The sample size was too small 

 to evaluate kurtosis, but since the more critical con- 

 dition of skewness was highly significant, tests of 

 significance could not be performed. For compara- 

 tive purposes, the log-linear model was fitted to the 

 pooled data for the blue marlin, and, as was the case 

 for striped marlin, the error term was not normally 

 distributed. For the blue marlin longline data, the 

 error term was not skewed, and there were too few 

 data to test for kurtosis. Tentatively accepting the 

 error term as being normally distributed, the "F" 

 test showed that the regression was highly signifi- 

 cant. For the trolling data, the error term was not 

 normally distributed; hence, tests of significance 

 could not be performed. Examination of the error 

 terms showed that there was one aberrant datum; 



Table 3. — Weight-length relationships for blue and striped marlins using the nonlinear model (Equation 3) 



pooled category indicates pooling of longline and trolling data. 



The data sets 



'** indicates significance at the 0.01 level. * indicates significance at the 0.05 level, and NS indicates not significant at the 0.05 level 



131 



