Working Paper: CEPR ID: DP18159
Authors: Matej Opatrny; Tomas Havranek; Zuzana Irsova; Milan Scasny
Abstract: Class size reduction mandates are frequent and invariably justified by studies reporting positive effects on student achievement. Yet other studies report no effects, and the literature as a whole awaits correction for potential publication bias. Moreover, if identification drives results systematically, the relevance of individual studies will vary. We build a sample of 1,767 estimates collected from 62 studies and for each estimate codify 42 factors reflecting estimation context. We employ recently developed nonlinear techniques for publication bias correction and Bayesian model averaging techniques that address model uncertainty. The results suggest publication bias among studies featured in top five economics journals, but not elsewhere. The implied class size effect is zero for all identification approaches except Tennessee's Student/Teacher Achievement Ratio project. The effect remains zero for disadvantaged students and across subjects, school types, and countries.
Keywords: class size; Bayesian model averaging; publication bias
JEL Codes: C83; H52; I21
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| class size (C55) | student achievement (I24) |
| publication bias (C46) | perceived effectiveness of class size reductions (I21) |
| class size (C55) | student achievement (disadvantaged students) (I24) |
| class size (C55) | student achievement (different subjects) (I24) |
| class size (C55) | student achievement (different countries) (O57) |
| Tennessee STAR experiment (C93) | student achievement (I24) |