Mensural Discrimination of Peromyscus 111 



1981) included: head and body length (body), tail length (tail), hind 

 foot length (foot), greatest skull length (SL), basonasal length (BNL), 

 rostral breadth (RB), nasal length (NL), interorbital constriction (OC), 

 zygomatic breadth (ZB), bony palate length (PL), maxillary toothrow 

 length (MTL), total toothrow length (TTL), palatal width (PW), pterygoid 

 breadth (PB), bullar depth (BD), and anterior palatal (incisive) foramen 

 length (PFL). We measured rostral length (RL) from the anteriormost 

 point of the nasals to the anterior edge of the zygomatic arch. Body 

 length was calculated as the difference between total and tail lengths. 

 We excluded ear length due to predominance of missing data. 



We performed statistical analyses with Systat 5.1a (Wilkenson 

 1989) and SPSS 4.01 (Norusis 1990). We tested normality and 

 homogeneity of variance by inspecting plotted residuals and by 

 Bartlett's test for homogeneity of group variances, respectively. 

 Differences among adult age classes and between sexes were tested 

 with analysis of variance, and type I error rates were corrected with 

 the Bonferroni adjustment (Rice 1989). We classified taxa using stepwise 

 discriminant analysis. Variables were included in the models based on 

 minimizing residual variance, prior probabilities were equal to sample 

 size, and varimax rotation was employed. Stepwise discriminant analysis 

 will find an optimal solution based on the data; however, depending on 

 where the analysis begins (i.e., which variables enter the model first), 

 it may find a local, rather than the global, optimum. To help avoid this 

 optimization problem, we removed variables that entered the model in 

 the first steps and repeated the analysis. In one case, that of discrimination 

 based on all external and skull measurements, we found that bullar 

 depth (BD) forced the model onto a local optimum. Therefore, we 

 eliminated this character from further consideration in that model. We 

 used stepwise discriminant analysis to produce two main predictive 

 functions from the smallest set of characters needed to separate all 

 four species — one for external and skull measurements and another for 

 skull measurements alone. In addition, we generated predictive functions 

 that used only one or two measurements to separate in pairwise comparisons 

 among species. 



We performed all analyses on raw data without transformation 

 (because transformation did not result in homogeneous variances) 

 and without removing size (Burnaby 1966, Rohlf and Bookstein 1987) 

 because this produced the simplest tool for the identification of 

 questionable specimens in the future. Although there was significant 

 heterogeneity of variances among species for some characters, standard 

 transformations (e.g., logarithm, etc.) did not homogenize it, and raw 

 data were more effective in discrimination than log-transformed data. 



