FISHERY BULLETIN: VOL. 80. NO. 1 



tem to the upper limit of coho salmon spawning, 

 as estimated from Anadromous Fish Distribu- 

 tion Maps 5 and interviews with district fishery 

 biologists. Inasmuch as, intuitively, latitude 

 should be correlated with the temperature and 

 flow regimes of the stream systems, we deter- 

 mined the latitude at the mouth of each stream 

 system. Gradients were calculated from tide- 

 water to the upper limit of coho spawning as a 

 basis for estimating the difficulty of the spawn- 

 ing migration. Because estuary size and length is 

 an estimate of exposure to vibriosis (Harrel et al. 

 1976) and potential richness of feeding grounds 

 (Myers 1979), we measured the estuary lengths, 

 stream elevations, and distances on United 

 States Geographical Survey Quadrangle Maps. 

 Inasmuch as high flows could have an effect on 

 both the early life history and the smolting proc- 

 esses of juvenile coho salmon, we determined the 

 presence of a spring runoff from snowmelt by 

 interviewing district biologists. We obtained in- 

 formation on the distributions of other fish 

 species in Oregon stream systems from C. E. 

 Bond, 6 and on the distribution of Ceratamyxa 

 shasta from J. E. Sanders. 7 



We obtained temperature data from hatchery 

 records to help interpret the morphological data 

 for the hatchery stocks. The average tempera- 

 ture for the first month of incubation was used, 

 because previous studies have indicated that this 

 time is a period during ontogeny when morpho- 

 logical features may be most sensitive to the 

 effects of temperature (Taning 1952). 



Statistics 



We calculated averages for the morpho- 

 logical characters, enzyme gene frequencies, 

 and the proportion of females for each stock, and 

 used multivariate analysis of variance and Rao's 

 (1970) test for additional information to deter- 

 mine whether morphological characters differed 

 significantly among stocks. In Rao's test, the 

 statistical significance of each morphological 

 character is determined, given that the other 

 morphological characters are already in the 

 model. Because environmental data on spawn- 



5 Anadromous Fish Distribution Maps. Oregon State 

 Water Resources Board, Salem, Oreg. 



^arl E. Bond, Professor of Fisheries, Oregon State Univer- 

 sity, Corvallis. OR 97331, pers. commun. April 1979. 



7 James E. Sanders, Assistant Fish Pathologist, Oregon Dep. 

 Fish Wildl., Corvallis, OR 97331, pers. commun. February 

 1979. 



ing distance, estuary length, estuary size, basin 

 area, and gradient were skewed, we transformed 

 them to natural logarithms to stabilize the vari- 

 ance and improve normality. We standardized 

 the stock characters (z = 0, S 2 = 1) for the 

 cluster analyses, using the standard normal 

 standardization. This standardization expresses 

 the stock character as standard deviations from 

 the character means, thus giving equal weight 

 to each character. 



We calculated correlation coeffients (Snedecor 

 and Cochran 1967) between the stock characters 

 and the environmental data for all stocks, and 

 between the morphological characters and tem- 

 perature data for hatchery stocks only. The levels 

 of significance for the correlation coefficients 

 were also calculated as described by Snedecor 

 and Cochran (1967). 



We used two cluster analysis programs to dis- 

 play similarities among stocks. One, a nonhier- 

 archical divisive cluster analysis, minimized the 

 total sum of squares between observations and 

 the cluster means (Mclntire 1973). In the other, a 

 hierarchical agglomerative cluster analysis, 

 Euclidean distance was used as the dissimilar- 

 ity measure, and the clustering strategy was 

 group average (see Sneath and Sokal [1973] or 

 Clifford and Stephenson [1975] for terminology). 

 Standardized data were used in both programs. 



We used canonical variate analysis to investi- 

 gate the relation among the clusters from the 

 agglomerative cluster analysis (Clifford and 

 Stephenson 1975). Canonical variate analysis 

 produces canonical variables that project groups 

 of multivariate data onto axes separating the 

 groups as much as possible. We plotted the ca- 

 nonical variables against each other in two- 

 dimensional space to determine the relationships 

 among clusters and the discreteness of the 

 clusters. 



RESULTS AND DISCUSSION 



Morphological Characters 



Significant differences (a = 0.01) for each 

 morphological character (Tables 1-3) as indi- 

 cated by multivariate analysis of variance and 

 Rao's test of additional information existed 

 among the 35 samples which consisted of wild 

 and hatchery stocks from two brood years. When 

 morphological characters for each stock between 

 brood years were compared for each of the hatch- 

 eries that were sampled in both years of the study 



108 



