the target populations and recognize sampled 

 populations which have defined boundaries in 

 space and time. From the characteristics of 

 the sampled populations we must infer the 

 nature and the bounds of the target populations. 

 Ginsburg (1938) recognized the sampling prob- 

 lem and emphasized the need for obtaining a 

 sample from as many locations as possible and 

 for avoiding excess representation from any 

 locality. He called his complete sample a com- 

 posite which was made up of constituent samples. 



After specifying the bounds of the 

 sampled population, the fish taxonomist must 

 further specify the method of sampling, because 

 fish vary greatly in their availability to all kinds 

 of gear, especially at different sizes and even 

 within a size group of a species. Inevitably the 

 nature of fishing gear makes the catch a non- 

 representative sample. Almost always a catch 

 consists of members of a school, or of an ag- 

 gregate, which more closely resemble one 

 another than they do the balance of the sampled 

 population . Frequently, too, the location of the 

 catch cannot be randomized in the specified area 

 or time because of economic factors. There- 

 fore the individuals within catches should be 

 expected to resemble one another more than in 

 a random sample. It follows that estimates of 

 variability from the usual samples will tend to 

 be too low . Estimates of the difference between 

 means of such samples in terms of either over- 

 lap or significance of the difference will tend to 

 be too high. 



Once we depart from a random sample 

 we forsake exactness and can only say vaguely 

 that the results must be treated with caution . 

 Therefore in any discussion of morphometric 

 data the sample characteristics should be speci- 

 fied in detail . 



EVALUATION OF INTERMINGLING FROM 



MORPHOLOGICAL STUDIES AND TAG 



RETURNS 



drawn from a single completely mixed popula- 

 tion. There has been a tendency to conclude 

 from this that the populations are distinct. 



These studies have not determined the 

 proportion of individuals within the groups that 

 are identical on the basis of the measurements 

 used. It must be recognized that this proportion 

 of individuals might have come from the same 

 parent stock. Of course, there is no proof that 

 they did, but then the data provide no proof that 

 they did not. It follows that the proportion with 

 identical characteristics is a maximum propor- 

 tion which might have come from the same 

 parent stock (intermingled) and that the true 

 proportion is equal or lower. 



Two recent authors have had an opportun- 

 ity to compare morphological studies with tagging 

 results. Both of them found statistically highly 

 significant differences in morphology between 

 populations which tagging studies showed to be 

 intermingling to a considerable degree. One of 

 these authors concluded that such results were 

 contradictory but we believe that such results 

 are not necessarily contradictory and we pro- 

 pose to examine them here in detail. 



In one comparison of morphological 

 studies and tag results Tester (1949) used the 

 vertebral counts of large numbers of herring 

 taken in five areas off the west coast of Van- 

 couver Island. He found statistically highly 

 significant differences in the number of vertebrae 

 among all areas, but the differences in mean 

 vertebral number were very small . We have 

 computed the overlap between the two most 

 widely separated areas which showed the great- 

 est differences in mean number of vertebrae 

 (pages 15-16, fig. 3), and even here the overlap 

 is almost complete, p = .465, cfi = 93%. Con- 

 sidering this, the findings from the tagging 

 studies, which indicated that about 45 percent of 

 the fish tagged in one area wandered to other 

 areas, are not surprising. 



In all of the recent morphometric studies 

 of tunas the conclusion has been reached that 

 statistically highly significant differences exist 

 between the samples being compared. The odds 

 against the null hypothesis have been such that 

 there is no doubt that the samples were not 



In the second example Roedel (1952) com- 

 pared samples of Pacific mackerel from five 

 areas between southern California and the Gulf 

 of California. He found statistically highly sig- 

 nificant differences in certain vertebral characters 

 and in head length and postulated that there were 



25 



