Conditional probability methods (positive predictive power, negative p
redictive power, sensitivity, and specificity) were employed to differ
entiate children judged by teachers as belonging to Social Competence
(SC) or Low Social Competence (LSC) groups using the Social Skills Rat
ing System-Teacher (Gresham & Elliott, 1990). Social skills functioned
more effectively as exclusionary criteria (high negative predictive p
ower values) for ruling out membership in the LSC group than inclusion
ary criteria (lower positive predictive power values) for both males a
nd females. The majority of items best differentiating the groups invo
lved teacher-preferred social skills or behavior fitting a model behav
ior profile. Large differences were found in the number, but nor the t
ype, of social skills that predicted LSC group membership for females
and males. Results were discussed in the context of using conditional
probability methods at the local level to select children for social s
kills interventions and to select target behaviors for these intervent
ions.