The literature on representative bureaucracy has demonstrated consistently that, under certain circumstances, minority clients fare better as the share of minority bureaucrats within an organization increases. However, despite decades of research, the underlying causal mechanism responsible for this observation remains unresolved. Scholars have suggested that it arises from bureaucrats acting in the interests of clients who share their characteristics; or that positive outcomes accrue to minority clients because similar bureaucrats are more likely to understand their values and motivations; or that diversity creates a culture in which even non-minority bureaucrats work to further the interests of historically underserved clients. Finally, some suggest that positive outcomes associated with representative bureaucracy may actually be a result of the actions of clientsthat are more responsive to bureaucrats that share their characteristics.

The failure to identify the causal mechanism underlying representative bureaucracy arises primarily from the choice of empirical strategy that has dominated this literature. The vast majority of studies examine only aggregate relationships—modeling the mean change in outcomes for some group of clients as a function of the percent of the organization’s employees from that same group. Using results from these analyses to “confirm” theoretical mechanisms fundamentally grounded in individual-levelbehavior obviously raises significant ecological inference concerns. It prevents the researcher from controlling for important predictors of the policy output. More importantly for our purposes, it also makes it impossible to empirically distinguish among the competing theories because, at the aggregate level, all can produce identical findings.

As an alternative, this study makes use of data that allows us to observe the behavior of individual clients and bureaucrats, as well as the aggregate characteristics of the organizations in which they interact. Specifically, we use nationally representative data from the Early Childhood Longitudinal Study, Kindergarten cohort (ECLS-K) to test for evidence of representation in the differential assignment of students to gifted programs by race. Prior studies using aggregate data have demonstrated a positive relationship between the fraction of black students receiving gifted services and the fraction of black teachers in a school. We instead make use of student-level data to predict the relative probability a given black (or Hispanic) student is referred to gifted services as a function of the race/ethnicity of the student’s classroom teacher, the racial/ethnic composition of the teachers at that school, and the interactions among the two. Importantly, the ECLS-K data allows us to account for students’ scores on cognitive assessments, parental education level, and other factors that are likely to be strong predictors of gifted assignment. Our results show that, even conditioning on test scores, black students are assigned to gifted services at higher rates when their classroom teacher is black but that the presence of black teachers in the school other than the classroom teacher has little effect. Own-race teachers matter most for assignment when students score moderately—but not exceptionally—well on cognitive assessments. These findings help illuminate the mechanisms underlying minority representation and confirm the importance of disaggregated data in identifying these pathways.