Ability Grouping in Middle School: Impacts on Performance and Behavior

Our intervention affected the way students were assigned to classrooms. Once the centralized allocation system to assign students to public middle schools in Mexico City had taken place, we received the final placement results, each student's admission test scores, and the number of seventh-grade groups/classrooms in each school. Based on these data, we defined three different models of group formation:
1. Tracking: Students were sorted according to their admission test score and grouped into classrooms based on their relative standing at the school level and class sizes. For example, suppose that 90 students are assigned to a school with three classrooms. In that case, the 30 lowest-achieving students are assigned to group A, the 30 medium-achieving students to group B, and the 30 highest-achieving students to group C. Whenever ties emerged, we randomized the allocation of students at the score cutoff to adjacent groups.
2. Bimodal: Students were first divided into terciles within each school based on their admission test scores. Then, two types of classrooms were formed: bimodal and homogeneous classrooms. Bimodal classrooms included students from the low- and high-achieving terciles. Students in the medium-achieving tercile were assigned to homogeneous classrooms. Using the same example above, with 90 students and three classrooms, we assigned 15 of the 30 lowest-performing students and 15 of the 30 highest-performing students to group A; the 15 remaining lowest-scoring and 15 remaining highest-scoring students were assigned to group B; all medium-performing students were assigned to group C.
3. Control: Students in control schools were randomly allocated to classrooms.

2015-06-01

2017-05-01

Academic performance, graduation on time

Behavioral responses from teachers (effort, class time management, self-percevied efficiency, curriculum coverage) and students (risky behavior, frienship sorting, effort, truancy)

We stratified the eligible universe of schools by school type (general or technical), quartiles of the average performance in a previous national exit exam and shift (morning or afternoon). Within each of these cells, we ranked schools by the total number of enrolled students, and we formed strata of size three. We then randomly assigned each school within each stratum to the control, tracking, or bimodal treatment arms. We conducted a single-blinded experiment in which school principals were aware they were participating in a study but were unaware of their treatment status at the beginning of the school year. Some may have inferred their status as the school year progressed. Teachers were assigned to classrooms by principals at the end of the preceding academic year, prior to the release of classroom composition. Therefore, teacher allocation was determined independently of group formation within the school. We implemented the three allocation models depending on the results of the randomization of schools into treatment arms. We produced a list of entering students for all schools in our sample with their corresponding group assignments. The Ministry of Education handled communication with schools' principals to ensure that the group assignment was implemented following the lists we had produced.

Randomization done in office by a computer

School

Yes

171 schools

38,006 students

57 schools per treatment arm