How can we evaluate the effectiveness of grammar schools in England? A regression discontinuity approach

Abstract

Previous research evaluating grammar school effectiveness has generally relied on snapshot or longitudinal regression models to deal with pre‐existing differences between grammar school pupils and those in non‐selective schools. These passive designs are based on correlations, and cannot demonstrate clear positive causal relationships between grammar school attendance and subsequent attainment. After accounting for the variables available for the analysis, pupils in different schools might still have distinct and unmeasured characteristics which threaten the validity of any conclusions drawn. Given that a randomised trial is not feasible under current conditions, this study addresses the limitations of previous research, using a regression discontinuity design (RDD) approach. This is the first use of RDD to attempt to make a robust causal inference about the effectiveness of grammar schools in one local authority in England. However, the authority, the Department for Education and the schools would not provide the individual data on pupils’ 11+ entry test scores, and the scores obtained could not be uniquely matched to Key Stage 4 outcomes. While the model presented suggests that there is an advantage to grammar school attendance, the incomplete data means that the study is more a feasibility trial of this strong design than any kind of definitive test intended to settle the debate on grammar school effectiveness. Conducting this design with national data on grammar school selection would create the most powerful evidence so far. To promote an effective and equitable education system for generations to come, those advocating the expansion of grammar schools should make the responsible decision to disclose all grammar school selection data for the purposes of research.