366 



FISHERY BULLETIN OF THE FISH AND WILDLIFE SERVICE 



800 



1000 



1200 



1400 



1800 



TOTAL LENGTH 



Figure 9.- — Relations between distance from snout to insertion of ventral fin and total length. Solid circles represent 

 Hawaiian data; open circles represent Costa Rican data; solid triangles represent Godsil's west-coast data. Fine 

 solid line is linear regression line fitting west-coast data, while heavy solid line is linear regression line fitting Hawaiian 

 data. Fine broken line is Godsil's curvilinear regression for west-coast data, while heavy broken line is similar re- 

 gression fitted to Hawaiian data. 



and, if this be false, (2) that the regression coefB- 

 cients (slopes) of the regression lines fitting the 

 samples from the two regions are equal. As may 

 be seen from the variance ratios computed in table 

 3, both these hypotheses are to be rejected for each 

 dimension considered, the west-coast data in this 

 table including the measurements of both Schaefer 

 and Godsil. If we compare the Hawaiian data 

 with the data of Schaefer alone (table 4) we find 

 here also that for no character considered may the 

 data from the two regions be represented by a 

 single linear-regression equation. In two cases, 

 however, indicated by footnotes in the table, the 

 appropriate variance ratio indicates that there is 

 not sufficient reason from these particular data to 

 reject the hypothesis of equality of regression 

 coeflScients. In general, it is quite apparent that 

 for each character the regression lines are different 



for the two regions and that they differ in slope. 

 Comparison of the regression lines of the dimen- 

 sions of tuna from different regions is perfectly 

 straightforward so long as we are able to assume 

 that the sample regression lines are representative 

 of the tuna populations of the regions in each case. 

 As has been noted earlier, however, Godsil found 

 that repeated samples from the west coast yielded 

 regression lines (curvilinear) for which a null 

 hypothesis could not be supported. The same 

 thing is true if linear regressions are applied to his 

 data (table 5). -His various subgroups along the 

 west coast differ significantly among themselves, 

 and for each dimension they differ in respect of 

 the regression coefficients. As may be seen from 

 table 6, comparison of my Costa Rican data with 

 GodsQ's data from Costa Rica alone (his samples 

 4, 5, and 12) reveals that a single linear-regression 



