Economic Research and Education Policy: Project STAR and Class Size Reduction
Melbourne Institute Working Paper No. 37/16
The use of randomized controlled trials (RCTs) and related randomization strategies to eliminate selection biases in establishing causality is a key element of the “modern experimentalist paradigm” (MEP). Yet, its emphasis on precisely identifying causal factors often limits its capacity to provide an evidence base for policy. We illustrate this through a detailed look at Project STAR, an extensively analyzed, well-funded, large-scale, rigorous RCT commissioned by the Tennessee legislature to help it decide whether to mandate statewide class-size reductions (CSR) from kindergarten to the third grade. Project STAR randomly assigned students to classes of different size and compared test results across these classes, to obtain an unbiased answer to the research question, “Does reducing class size improve test scores?” However, this shed little light on whether reducing class size was a good use of increased education financing. Analyses of Project STAR ignored general equilibrium effects of CSR on both the demand for teachers and the value of test scores. Moreover, its emphasis on estimating average class-size effects in a particular setting diverted attention from their heterogeneity, and the need to understand how class size affects learning, and how its effect is moderated by circumstances. Rather than considering the full chain of evidence necessary for shaping class-size policy, Project STAR concentrated its effort on maximizing the accuracy of a single link in that chain; internal validity trumped policy relevance.