H.IORT ami SCHKKCK: I'HKNOTYI'IC DIFFERENCES AMONG ('OHO SALMON 



WOr 



80 



20 



43* 45* 



LATITUDE 'N 



4T 



Figure 5.— Transferrin gene frequencies for wild and hatch- 

 ery coho salmon stocks arranged by latitudes. 





STOCK 



Figure 6. — Dendrogram of the agglomerative cluster 

 analysis for all stocks of wild and hatchery coho salmon of two 

 brood years, 1976 and 1977. Euclidean distance was the dis- 

 similarity measure and group average was the clustering 

 strategy. Location codes are as in Figure 1. The other codes are 

 as follows: H6, hatchery stock of the 1976 brood year; H7, 

 hatchery stock of the 1977 brood year; and W7, wild stock from 

 the 1977 brood year. 



Oregon coast wild stocks (cluster 1), southern 

 Oregon coast wild stocks (cluster 2), stocks from 

 hatcheries that used wild stocks for the egg 

 source (cluster 3), stocks from large river sys- 

 tems (cluster 4), hatchery stocks and two wild 

 stocks from the northern Oregon coast (cluster 

 5), and three individual hatchery stocks from 

 California and Washington (clusters 6-8). 



Canonical variate analysis on the five larger 

 clusters produced three canonical variables that 



were significant (a = 0.05). When these three 

 variables were plotted against each other, only 

 clusters 1 and 5 (consisting of wild stocks and 

 hatchery stocks, both from the northern Oregon 

 coast) were not completely separate in three- 

 dimensional space. The other three clusters were 

 discrete, suggesting that intercluster differ- 

 ences were stronger than between clusters 1 and 

 5. Statistical testing for differences between the 

 clusters would not be valid because the necessary 

 assumption of randomness of data is violated. 



The results of the canonical variate analy- 

 sis must be interpreted with caution because the 

 variation within each cluster was reduced by our 

 using the averages of the morphological char- 

 acters. This reduction of variation facilitates dis- 

 crimination between clusters by canonical vari- 

 ate analysis, so that quantitative comparisons of 

 cluster discreteness cannot be made. Individual 

 phenotypes undoubtedly overlap between stocks 

 or between clusters; however, the multivariate 

 analysis of variance did indicate that significant 

 differences existed among the stocks for each of 

 the morphological characters. We characterized 

 the stocks by the average phenotypes in order to 

 estimate the phenotypes typical for each stream 

 system, and on that basis the results of the canon- 

 ical variate analysis suggested that there were 

 discrete differences between all clusters except 1 

 and 5. 



The results of the agglomerative and divisive 

 cluster analyses were similar. At the 13-cluster 

 level of the divisive analysis (Table 8), all but two 

 clusters were identical with clusters from the 

 agglomerative cluster analysis dendrogram. 

 The results of these analyses should be inter- 

 preted cautiously, because they are based on only 

 10 characteristics— a small number compared 

 with the total number of genetically related 

 characteristics possible. If other characteristics 

 had been used, the results might have differed. 

 Thus, we did not emphasize the exact order or 

 the levels of dissimilarity at which any two 

 clusters joined together; rather, we observed 

 only general trends in the clustering patterns. 



Three general trends are apparent in the 

 clustering patterns of the agglomerative cluster 

 analysis dendrogram. First, the stocks from the 

 larger stream systems (Columbia, Rogue, and 

 Klamath Rivers) were more similar to each other 

 than to the stocks from smaller streams. The only 

 exceptions to this trend were wild stock from the 

 Umpqua River and the Umpqua and Rogue 

 hatchery stocks. The Umpqua wild stock was 



115 



