The established unit for reporting sociolinguistic data has been the sociologically determined group. Characteristically, only the mean rate of nonstandard usage for the group is reported for any linguistic feature. Such reporting obscures the possibility of linguistic heterogeneity within the group. Data from Los Angeles school children show that school classrooms are heterogeneous with respect to different forms of agreement usage. An alternative analysis is proposed in which children are grouped on the basis of linguistic criteria. Members of these groups share a common grammar. There is an implicational relationship among the grammars employed by such linguistically determined groups. The linguistically determined groups do not map exactly onto traditional sociologically determined groups. This relation may be expressed as the probability that a member of a sociologically determined group is a member of some linguistically determined group. This interpersonal variation may be distinguished from intrapersonal variation, or the probability that a rule of a particular grammar will apply. (Author)