Identifying Bison Ration Groups by Multivariate Analysis of Blood Composition

Abstract

Discriminant analysis was used to classify animals into ration groups based on blood composition. Blood samples were drawn from groups of 3 bison (Bison bison) steers and 4 Hereford cattle steers receiving 4 rations each in summer and winter (N = 56). Four discriminant analyses, 1 for each combination of species and season, were used to classify blood samples into respective ration groups. Rates of correct classification, evaluated using a jackknife procedure, ranged from eleven of 12 winter bison samples to six of 16 summer cattle samples using blood urea nitrogen (BUN), alkaline phosphatase (ALP), and cholesterol (CHOL) as discriminating variables. Classification of summer cattle samples was improved to eleven of 16 by using BUN, total serum protein (TSP), 3-globulin (fl-GL), and hemoglobin (HGB) as variables. The overall correct classification rate was 79% (44 of 56 samples). J. WILDL. MANAGE. 51(4):893-900 Although it has been demonstrated frequently with wild and domestic animals that the levels of many blood components vary significantly with ration composition (Preston et al. 1965; Seal et al. 1972b; LeResche et al. 1974; Torell et al. 1974; deCalesta et al. 1975; Kirkpatrick et al. 1975; Manston et al. 1975; Parker and Blowey 1976; Seal et al. 1978b; Bahnak et al. 1979, 1981; Warren et al. 1981, 1982; Card et al. 1985), attempts to evaluate nutritional status from blood composition have had uncertain success (Adams et al. 1978, Franzmann and LeResche 1978, Lee et al. 1978, Rowlands 1980, Warren et al. 1981, Kopf et al. 1984). Most of these studies have involved primarily univariate analyses of blood data. A multivariate approach might be more successful, and blood data lend themselves to multivariate analyses because many components are routinely measured in a single blood sample. This paper is a statement of approach, involving the use of multivariate methods to analyze and interpret blood data. I used discriminant analysis as an example to illustrate the classification of samples into ration groups using blood data from bison and cattle. Discriminant analysis has been used widely to assign individuals to groups in taxonomy (Thorpe 1983, Gerasimov 1985), ecology (Stiteler 1979, Williams 1983), medicine (Lachenbruch and Clarke 1980, Brown 1984), and wildlife management (Cowardin and Johnson 1973, Hanson and Jones 1976, Crawford et al. 1981, Hale et al. 1982). I thank D. G. Peden for his support and for suggesting discriminant analysis to classify these blood samples, and gratefully acknowledge the statistical reviews of this manuscript by B. Collins and A. Sharma. This study was supported by Can. Wildl. Serv., Contract OSZ5-0285, and by the Alberta Environ. Cent. MATERIALS AND METHODS Sampling, handling, and feeding methods were described by Hawley and Peden (1982). Briefly, rations were formulated to contain high protein-high energy, high protein-low energy, low protein-high energy, or low protein-low energy. Groups of 3 yearling bison steers and 4 yearling Hereford steers, housed at the University of Saskatchewan, Saskatoon, were fed these rations ad libitum in late summer (24 Aug-4 Oct 1976) and late winter (26 Feb-9 Apr 1977). The Ist of a series of blood samples drawn from each animal in each ration group was selected for this study. Thus, there were 3 bison and 4 cattle samples for each of 4 rations in 2 seasons (N = 56). All samples were analyzed as described by Hawley and Peden (1982) to determine the levels of 20 blood components. These data were subjected to discriminant analysis in the present paper. Discriminant analysis is a multivariate technique for developing linear combinations of predictor variables in order to maximize differences among groups. In our data a single blood sample represented the analytical unit, or case, for which 20 dependent variables (the blood components) were measured. The different rations formed 4 groups, within which the cases were clustered. Discriminant analysis maximized the distinction among these groups. Classification functions were developed, 1 for each group, to determine in which group a case belonged. The analysis calculated the probability