Giles: Multivariate Analysis of Pleistocene and Recent Coyotes 373 



Testing for the linearity of correlation is a real problem. It is often assumed that 

 morphological data of the sort used here are linearly correlated, since many statis- 

 tical techniques so require. However, it is also widely accepted that when a reason- 

 able size difference is involved, the distribution of the data may be more closely 

 approximated by the allometric formula, y = bx", which takes differential growth 

 rates into account. This dilemma can be resolved in two ways. The data may be put 

 in logarithmic form (e.g., Burma, 1948), making the allometric formula, log y- 

 log b - k log x, a form suitable for linear analysis. Olson and Miller (1951), fully 

 recognizing the importance of allometry, applied various tests that showed it to be 

 insignificant in their material. In an earlier study of my own, mammalian material 

 that was measured in a way similar to that I have used in this study and subjected 

 to other and possibly more sensitive statistical tests of significance than used by 

 Olson and Miller, indicated an allometric rather than a linear distribution of 

 measurements for a majority of dimensions (Giles, 1956a). In the present study 

 the probable allometric trend of the data has been ignored and the correlation 

 treated as linear. Testing for allometry would be a study in itself. The deviation, in 

 any case, would be relatively small and unlikely to affect the tests materially (e.g., 

 Mahalanobis, Majumdar, and Rao, 1949). 



In addition, random samples are necessary for this sort of statistical technique. 

 Every effort was made to exclude bias in choosing the specimens for measurement; 

 unfortunately, the analyst's possible bias may add to a bias created by the original 

 act of entrapment itself, but I have no reason to suspect that a relevant nonrandom 

 selection pressure operated to provide the samples utilized. Since the D 2 technique 

 is not suited for handling fragmentary data, measurements were restricted to 

 specimens providing the full series. 



MATERIAL 



The Recent coyote, a mammal of western and central North America, has been 

 placed in one species, Canis latrans Say, by current authorities (e.g., Nelson, 1932; 

 Grinnell, 1933; Young and Jackson, 1951). The coyote is readily distinguished 

 from all other canids in its range except the widespread gray wolf (C. lupus 

 Linnaeus) and the red wolf (C. niger Bartram) . The gray wolf greatly exceeds the 

 coyote in size, as well as differing in many details. The red wolf, which is restricted 

 to southeastern United States, is generally larger than the coyote, but some indi- 

 viduals are practically indistinguishable from coyotes where their ranges overlap 

 in eastern Texas (Young and Goldman, 1944). 



Jackson (Young and Jackson, 1951) recognizes 19 geographic subspecies of the 

 coyote in his recent revision based on 4,500 specimens obtained during predator 

 control campaigns. Four subspecies occur in California (fig. 1). Grinnell (1933) 

 did not feel justified in distinguishing Canis latrans clepticus Elliot as a separate 

 subspecies, but he recognized the other three subspecies. In California, C. I. clep- 

 ticus is found only in the southern part of San Diego County. Demarcation of the 

 four subspecies in all but insular groups is arbitrary; intergradation occurs at the 

 boundaries. 



California subspecies recognized by Jackson (Young and Jackson, 1951) and in 

 this paper are: Canis latrans ochropus Eschscholtz, the California Valley Coyote, 



