Age of Starting School, Academic Performance, and the Impact of Non-Compliance: An Experiment within an Experiment, Evidence from Australia*

This paper estimates the difference in academic performance of the oldest (treated) and youngest (control) students in a given grade. We employ Queensland Department of Education school administration panel data for the population of 2008–2016 state school students. Academic performance is measured using standardised test scores (National Assessments Program in Numeracy and Literacy) and teacher‐assessed measures of performance and effort for individuals in grades 3, 5 and 7. The empirical analysis employs a regression discontinuity design based on birth date relative to administrative rules on age of school enrolment. The class‐assigning mechanism operates via a known cut‐off date and results in the oldest child in the grade being almost a year older than the youngest. However, lack of enrolment compliance implies the regression discontinuity design is fuzzy rather than sharp, identifying the local average treatment effect rather than the average treatment effect. In particular, given the one‐sided nature of non‐compliance, we employ the partially fuzzy estimator of Battistin and Rettore (Journal of Econometrics, 142, 715–30, 2008). Exploiting a 2008 enrolment cut‐off date reform, which delayed school starting age by 6 months, we find a large increase in enrolment compliance. This exogenous change in the enrolment age enables one to gauge the importance of enrolment non‐compliance in estimating the treatment effect of being older versus younger in cohort. We find that pre‐reform the treatment effect is small and generally statistically insignificant and post‐reform it is sizeable and statistically significant.