28 Michael E. Seidel 



mobilensis and 6 P. c. texana. All specimens were adults and ranged 

 from 27 to 57 mm skull length. Although the sex of some of the skulls 

 examined was not identified, it was evident that none of the samples 

 was skewed heavily toward males or females. Twenty-nine measure- 

 ments were made on each cranium and mandible (± 0.1mm): condyl- 

 obasal length (midline length of skull, from posterior aspect of occipital 

 condyle to anteriormost point of premaxillae); interquadratal width; 

 supraoccipital length; pterygoid width (least); interpterygoid process 

 width (greatest); orbital height and orbital width; narial height and 

 width; prefrontal length (midline); interorbital width (least); postorbital 

 length (width of postorbital arch, least distance from orbit to superior 

 temporal fossa); postorbital-quadratojugal breadth (least distance be- 

 tween superior temporal fossa and ventral ridge of quadratojugal); 

 jugal-quadratojugal length (least distance between orbit and tympanic 

 cavity); maxillary alveolar width (least); foramen magnum width and 

 height; basisphenoid-basioccipital length; interforamina stapedio- 

 temporale width (distance between the temporal-stapedial foramina); 

 anterior skull width (at anterior rim of tympanic cavity); posterior skull 

 width (at posterior rim of tympanic cavity); intersquamosal breadth 

 (distance between posterior aspects of squamosals); premaxillary height 

 (midline); temporal arch width (least distance between rim of tympanic 

 cavity and superior temporal fossa); otic capsule length (from posterior 

 rim of tympanic cavity); dentary-coronoid height; dentary breadth (mid- 

 line); dentary alveolar width (lateral); and lingual alveolar width (prox- 

 imal to median ridge of dentary). 



Due to intra- and interspecific size variation, regression analysis 

 was applied to all skull measurements to remove linearly-related effects 

 of size. Condylobasal length was used as the independent variable for 

 regression analysis of the other variables. The SAS General Linear 

 Models procedure produced residual values for each character; these 

 values were used as "size-free" variables. Using these 28 variables, the 

 SAS discriminant analysis procedure tested for homogeneity of within- 

 group covariance matrices. As the 28 characters measured showed no 

 evidence of heterogeneity of the within-group covariance matrices, 

 groups (taxa) were compared by step-wise discriminant analysis using 

 the computer program, BMD07M (Dixon 1974), which generates canon- 

 ical variates with maximum between-group variance relative to within- 

 group variance. The canonical variate means are plotted on the first two 

 axes, and analysis of variance describes significant differences between 

 groups (P < 0.05). Using canonical functions, the posterior probability 

 of each turtle belonging to its respective group is computed and classi- 

 fied accordingly. 



