Using converging evidence in policy formation: The case of class size research

Alternative research designs can often provide evidence about important policy questions. It is often useful to know whether sources of evidence based on different research designs suggest similar conclusions, particularly since different actors in the policy arena may have different views about the validity of each research design. The case of the effects of class size reduction on academic achievement is illustrative. The evidence about the effects of class size reductions comes from three different research designs: small-scale randomised experiments, large-scale econometric studies, and the large scale longitudinal experiment in Tennessee, USA. The small scale experiments have high internal validity but may lack external validity. The econometric studies have external validity, but may lack internal validity. The single large-scale longitudinal experiment has high internal and external validity, but is a single, unreplicated study. This paper provides estimates of the effects of class size reduction derived from three different kinds of evidence, and argues that they are reasonably similar. Thus evidence from the three different research designs is mutually supporting. This converging evidence suggests that the effect of class size reduction from 24 to 15 produces modest, but lasting increases in academic achievement.