Publication bias and p-hacking in the effect of COVID-19 on learning
创建于 更新于
摘要
This paper re-examines the central estimate of COVID-19 learning loss with a comprehensive set of bias-correction methods to address publication bias, p-hacking, and mechanical dependence in standardized effect sizes [page::0]. Using Betthäuser et al.'s dataset, RoBMA and MAIVE yield an effect size around -0.12 SD, equivalent to about 30% of a school year, and robust to multiple bias corrections [page::0]. Despite some signals of selective reporting, the results remain economically meaningful and statistically robust, implying the learning deficit is not explained away by publication bias [page::0].
