Estimating causal effects of class size in secondary education: Evidence from TIMSS

The effectiveness of class size reduction on student learning has been of intense research interest and policy debate worldwide. Nevertheless, in observational and quasi-experimental studies the evidence has been largely inconclusive partially due to potential biases arising from non-random placement of students in differently-sized classrooms. This study used Third International Mathematics and Science Study (TIMSS) data from 2003, 2007 and 2011 to investigate the effects of class size on eighth graders’ cognitive outcomes over time. Five cognitive outcomes were used: mathematics, physics, biology, chemistry and earth science. The main independent variable was the teacher-reported eighth grade class size in each particular subject (i.e., mathematics, physics, biology, chemistry and earth science). Student covariates included gender, age, and socioeconomic status (SES). Teacher covariates included gender, education level, and teaching experience. School variables included eighth grade enrolment and the percentage of economically disadvantaged students in the school.

Instrumental variables (IV) were employed to analyze the TIMSS data. The instrument was the expected average class size in eighth grade in a school and was computed using the eighth grade enrollment in the school and the rule about maximum class size for eighth grade in a specific country (Angrist and Lavy, 1999). Once the instrument was constructed, two-stage least squares (TSLS) were utilized to analyze the data (Angrist, Imbens, & Rubin, 1996).

Overall, four European countries (i.e., Hungary, Lithuania, Romania and Slovenia) that met the following two criteria were selected. First, a country should have had in place a rule about the maximum number of students in a classroom in eighth grade and that rule should have been followed relatively well in order to ensure that the IV estimates would be valid. Second, to examine the consistency of the class size effects over time a country should have participated in all three TIMSS waves (2003, 2007, and 2011).

The results indicate that in Romania smaller classes had a statistically significant and positive impact on academic scores in mathematics, physics, chemistry and earth science in 2003. It is noteworthy that the small class effects obtained in Romania in 2003 were considerable in magnitude and typically much larger than the effects obtained from Project STAR (Tennessee class size experiment) (i.e., the effects were larger than one-fifth of a standard deviation). In all other years and countries the class size effects were not statistically significant.