FISHERY BULLETIN: VOL. 83, NO. 3 



axes. Confidence limits (95%) for the positions of dif- 

 ferent areas on these plots are presented as circles 

 with radius equal to the square root of 5.99/number 

 of fish in sample (Mardia et al. 1979). 



Analyses on the same combinations of parasites 

 were also done by calculating minimum spanning 

 trees (Gower and Digby 1981), and dendrograms 

 from nearest neighbor and centroid cluster analyses 

 (Clifford and Stephenson 1975), basing similarity 

 measures on logarithms of area means. Areas were 

 grouped in a similar way by all methods. Using 

 clustering algorithms which either ignored or allow- 

 ed for matches between areas where parasites were 

 not recorded did not significantly influence results. 

 For these reasons, and because only canonical 

 variate analysis provided some measure of reliabili- 

 ty for its conclusions (confidence rings), only the 

 results of the canonical analyses are presented 

 below. 



School integrity was examined by comparing the 

 variability in parasite numbers per fish between 

 schools, to that within schools, for the two areas 

 (Marquesas and east New Zealand) where the largest 

 numbers of schools were sampled. This showed 

 which parasites were strongly linked to schools, and 

 also allowed tentative estimation of the length of 

 time schools remained intact. In theory, for parasites 

 to show strong school associations two conditions 

 need to be met: the parasite must heavily infect some 

 schools and not others, and its life span in the fish 

 must be equal to or shorter than the life of the school. 

 Parasites which showed strong school-school associa- 

 tion were therefore likely to be shorter lived than 

 those not showing such associations, and other 

 evidence being equal, were considered less reliable as 

 population markers than related species. 



Two methods were used to compare within and be- 

 tween school variability in each of the two areas. 

 First, a series of univariate analyses of variance of 

 log (parasite numbers -f- 1.0) were done to calculate 

 the ratio of between school to within school 

 variances. The magnitude of these ratios, and the 

 corresponding probabilities that they do not differ 

 from 1.0, were interpreted as measures of school in- 

 tegrity. A limitation of this method was that the data 

 were only approximately normally distributed, par- 

 ticularly for rare parasites, and thus the derived pro- 

 babilities were also approximations. 



The second method, a median test, was based on 

 the binomial distribution. The number of parasites of 

 a particular species in each fish was transformed to a 

 zero if it was less than or equal to the median number 

 per fish for the area, and to a one otherwise. The 

 zeros and ones of each school were then considered 



as a binomial sample. If these samples showed 

 evidence of greater variation than expected by 

 chance (i.e., too many schools with nearly all zeros or 

 nearly all ones), then the schools differed with 

 respect to the distribution of the parasite. A statistic, 

 approximately distributed as a x^ random variable, 

 was calculated using GLIM (Baker and Nelder 1978) 

 to determine whether the binomial samples showed 

 evidence of differences. Its associated probability 

 was used as a measure of school integrity. The 

 method had the useful property of being independent 

 of the distribution of parasite numbers. For parasites 

 with a median per fish of <1, the test was based on 

 the presence or absence of the parasite, though ob- 

 viously the rarer the parasite the less sensitive the 

 test. 



It is possible that some schools were sampled 

 twice. If this did happen, the results of both methods 

 err on the conservative side. Only those species that 

 gave consistent results by both methods were used to 

 draw conclusions about school integrity. 



RESULTS 



Evaluation of Parasite Species 



Information was collected on 26 different types of 

 parasites (species or species complexes) from 15 

 areas. A summary of the raw data unadjusted for 

 fish length is given in Table 2. 



The parasite species were evaluated for their prob- 

 able longevity on or in skipjack tuna. For them to be 

 useful as markers they needed to be relatively long- 

 lived, preferably surviving for the life of the fish. 

 Nothing was known specifically about their longevity 

 in skipjack tuna, though data were available on 

 related forms (Table 3). In general, intestinal lumen 

 dwellers appear to be more easily lost than larval 

 forms encapsulated in the tissues. The 26 skipjack 

 tuna parasites were divided into four groups, those 

 considered "temporary", "semi-permanent", and 

 "permanent", and those not used at all. 



Four parasites were not used in any analyses. Two 

 of the nematodes, Ctena.searophis sp. and Spinitec- 

 tus sp. (Nos. 23 and 24 in Table 2), were found in the 

 gut of virtually every fish in which they were sought, 

 from every area. Their small size meant that the 

 number recovered was a function of the time spent 

 searching. They were only counted in every fifth fish, 

 as were the two larval cestodes from the large intes- 

 tine, Scolex polymorphus (large) and S. polymorphus 

 (small) (Nos. 25 and 26). Counting these was time 

 consuming, their apparent abundance may have 

 been inversely related to the state of preservation of 



346 



