Randomization inference for before-and-after studies with multiple units: an application to a criminal procedure reform in Uruguay
提出一种适用于多单位前后比较研究的随机化推断框架，不依赖参数模型或长期趋势，用于评估短期因果效应。以乌拉圭2017年刑事司法改革为例，发现改革后第一周每日警情报告增加约25份（上升8%），效应具有统计显著性。
Abstract Learning about the immediate causal effects of large-scale policy interventions poses a significant challenge for quasi-experimental methods that rely on long-term trends or parametric modelling assumptions. As an alternative, we develop a randomization inference framework for before-and-after studies with multiple units, designed specifically for short-term causal inference and allowing for general assignment mechanisms. The method provides finite-sample-valid statistical inferences without relying on parametric time series models or extrapolation. We demonstrate its utility by analyzing a major criminal justice reform in Uruguay that switched from an inquisitorial to an adversarial system in November 2017. Our method relies on the key assumption of no local time trends near the policy adoption time, which is supported by several falsification tests in our empirical study. We find a statistically significant short-term causal effect: an increase of approximately 25 daily police reports (an 8% rise) in the first week of the new justice system. Our randomization inference framework provides a robust and flexible methodology for evaluating policy adoptions in before-and-after studies with multiple units.