66 Christopher W. Brown and Carl H. Ernst 



Initial stepwise discriminant analysis of the first 13 characters pro- 

 duced 5 discriminating characters. They were, in order of their increas- 

 ing ability to discriminate, characters BCB, VS, TCB, DSM, and PDS. 

 None were strongly correlated; the highest was 0.365 between VS and 

 DSM. The single discriminant function had a significant F-value (P = 

 0.01) of 16.103 and a canonical correlation of 0.722. These results, 

 including the value of the Wilks' lambda, are summarized in Table 3. 

 Thirteen (16%) of the 80 specimens used in this analysis were incorrectly 

 classified into the two groups: C. h. horridus, 6 (13.6%), and C. h. atri- 

 caudatus, 7 (19.4%). 



Discriminant analysis of all 19 characters produced 4 discriminat- 

 ing characters: MS, ADS, HL, and VS. Characters MS and HL had a 

 weak correlation of 0.633, the next highest correlation being 0.366 

 between HL and VS. The discriminant function had a significant F- 

 value (P = 0.01) of 48.577 and a canonical correlation or 0.866. These 

 results, including the value of the Wilks' lambda, are summarized in 

 Table 3. Five (7%) of the 70 specimens used in this analysis were incor- 

 rectly classified into the groups: C. h. horridus, 4 (11.1%), and C. h. 

 atricaudatus, 1 (2.9%). 



Group means and standard deviations of all 19 characters are pre- 

 sented for both nonintergrades and intergrades in Tables 4 and 5, 

 respectively. 



DISCUSSION 



One purpose of the factor analysis performed was to determine, by 

 inspection of the plotted factor scores, whether clusters of individuals 

 occur that correspond to subspecies. Analysis of the first 13 characters 

 reveals little or no clustering in any of the scatterplots. Factor 1 (ab- 

 scissa) versus factor 2 (ordinate) appears to have the best clustering of 

 the six graphs (Fig. 4A). Separation seems to occur along the vertical 

 axis. Crotalus h. atricaudatus tends to cluster in the first two quadrants, 

 while C. h. horridus tends to cluster in quadrants three and four, indi- 

 cating separation on the basis of dorsal scale rows (factor 2). However, 

 overlap is wide. More than 25% of the specimens of C. h. atricaudatus 

 lie below the first two quadrants. No other plot (Fig. 4B-F) demon- 

 strates any distinct clustering. On the basis of the first 13 characters, 

 therefore, no subspeciation can be recognized. 



The plots from the analysis of all 19 characters, however, show 

 contrary results. Factor 1 (abscissa) versus factor 2 (ordinate) demon- 

 strates clustering along the horizontal axis: C. h. atricaudatus tends to 



Fig. 5. Scatterplots of estimated factor scores for specimens from the factor 

 analysis of all 19 characters. Solid circles represent one or more specimens of 

 Crotalus h. horridus; open circles represent specimens of C. h. atricaudatus. The 

 X's represent specimens of both. Adult size measurements and postocular stripe 

 (characters SVL, TL, HL, and POS) correlated most closely with factor 1; tail 

 measurements (CS, R, and TL) correlated with factor 2; dorsal scale rows (ADS 

 and DSM) correlated with factor 3; and middorsal stripe (MS) correlated with 

 factor 4. 



