Allocation and discrimination based on human odontometric data.

This study contrasts the confidence with which individuals may be grouped and then re-allocated on the basis of odontometric data. These data are derived from the mesiodistal and buccolingual diameters of 202 Lengua Indians of Paraguay (100 males, 102 females), 125 Caucasoid schoolchildren (59 male, 66 female), and 206 Negroes (106 male, 100 female). Multivariate intergroup discrimination is effected by means of canonical and stepwise discriminant analysis, whilst allocation is evaluated by means of posterior and typicality probabilities. Bias is reduced by means of a jackknifing procedure. High levels of discriminatory confidence (each Wilk's Lambda, P less than 0.01) are matched by high percentage correct classification (Caucasoid, 67.4-75.0%; Negro, 71.0-77.3%; Amerindian, 65.2-78.1%). However, these results are not matched by allocatory procedures: only 21.7% of caucasoids, 21.4% of Negroes, and 28.8% of Amerindians could be re-allocated with probabilities in excess of 80%. It is concluded that while multivariate discriminant techniques may be usefully employed in the separation of different populations, individuals may not be assigned with the same degree of confidence, even with an a priori knowledge of their group membership.