“Estimating Context-Independent Treatment Effects in Education Experiments” by Steven F. Lehrer
Steven F. Lehrer
In this study, we first document that the magnitude of the estimated treatment effect in Project STAR is substantially larger in schools where fewer students were assigned to small classes. The differences in student performance across schools cannot be explained by failure in randomization, other observed school level characteristics or differences in selective test taking. Further, we show that these achievement gains are driven by students in small classes from schools where fewer students were in small classes. The results are suggestive that there was a proportionate change in motivation or effort by teachers who teach small classes but not in regular classes. Second, we introduce an empirical strategy for experimental studies that aims at disentangling the pure educational effect from a specific treatment from that which is attributable to the interaction between the treatment and the social context in which the experiment takes place. Using minimal structural assumptions we disentangle the estimated treatment effect into components that are context specific and context independent. Our results indicate that between 50-70% of the estimated treatment effect in Project STAR is context specific.