评估留级影响：排除限制违反的警示案例

State laws that mandate in-grade retention for struggling readers are widespread in the U.S., and retention at the secondary school level is common in many countries. Researchers often use regression discontinuity (RD) methods to study such policies, leveraging strict performance cutoffs as an instrument to estimate the Local Average Treatment Effect (LATE) of retention on student achievement. In this paper, we document a likely threat to the internal validity of these studies. Examining two cohorts of Michigan students, we find that being flagged for retention increases reading performance by roughly 0.05 SD, a modest but meaningful impact. However, because being flagged increases the likelihood of actually being retained by only 3.4 percentage points, the implied effect of retention itself under standard assumptions would be an implausibly large 1.3 SD. Survey evidence suggests that flagged students receive more intensive reading support even if they are not retained, a violation of the exclusion restriction. Moreover, we estimate similar effects in districts that did not retain any students. These results raise concerns about potential bias in previously estimated retention effects and highlight the importance of carefully considering exclusion assumptions in analyses of multifaceted education interventions.