By carefully examining our original data and models, Zoorob identified a potential outlier that should be scrutinized when evaluating our findings. We thank him for raising this important point and for engaging in collaborative, problem-solving research. Examining Zoorob’s analysis provided us the opportunity to return to our original study. We began by thinking about his first point: the raw data show that crime-related 911 calls did not appear to fall after the local press reported on Jude’s vicious beating by police officers. The Jude story was published in February, when both crime and crime reporting were at their lowest points of the year. (The enduring seasonal nature of crime has been empirically established at least since the eighteenth century [Anderson 2001; Field 1992; Hipp et al. 2004].) As temperatures warmed in subsequent weeks, crime increased but crime reporting did not rise at the same rate. That crime-related calls remained somewhat flat from January to March, despite a rise in recorded lawbreaking, suggests that the rate of 911 calls per crime decreased. This basic insight encouraged us to more directly control for the seasonal nature of crime.1 Accordingly, we estimated models that excluded the outlier Zoorob identified but controlled for average weekly temperature, as well as month and block-group fixed effects, as in our original specification.2 Doing so better enabled us to juxtapose rates of 911 calls during weeks with similar weather conditions and avoid comparing cooler days to hotter ones that see more crime. In Table 1, we report the results of these new models. Using AIC and BIC to guide model selection, we found that our revised models provide a better balance between fit and parsimony than our original model and those developed by Zoorob. Model 3 includes a quadratic post-Jude trend, as in the first two models. Model 4 includes a linear post-Jude trend.3 The third model of Table 1—which excludes the outlier week and controls for weekly temperatures—reports a negative linear term for the post-Jude period (p < .001) and a small, positive quadratic term (p < .01). Figure 1 plots the estimates from Model 3, showing that 911 calls declined initially following the Jude story before returning to expected levels a few months later. Likewise, Model 4 reports a significant linear decline in crime-related 911 calls following Jude’s story. These models (neither of which include the outlier week) indicate that crime-related 911 calls were much lower than expected in Milwaukee neighborhoods following news reports of Jude’s beating. The results are highly consistent with our original findings, reflected in the first column of the table. 895979 ASRXXX10.1177/0003122419895979American Sociological ReviewDesmond et al. research-article2020

中文翻译：

---

警察暴力对公民犯罪报告影响的证据

通过仔细检查我们的原始数据和模型，Zoorob 发现了一个潜在的异常值，在评估我们的发现时应该仔细检查。我们感谢他提出这一重要观点并参与合作解决问题的研究。检查 Zoorob 的分析为我们提供了返回原始研究的机会。我们首先考虑他的第一点：原始数据显示，在当地媒体报道裘德被警察恶毒殴打后，与犯罪有关的 911 电话似乎并没有下降。裘德的故事发表于 2 月，当时犯罪和犯罪报告都处于一年中的最低点。（至少自 18 世纪以来，犯罪的持久季节性性质已凭经验确立 [Anderson 2001; Field 1992; Hipp et al. 2004]。）随着随后几周气温升高，犯罪率有所增加，但犯罪率并未以同样的速度上升。尽管有记录的违法行为有所增加，但与犯罪相关的电话从 1 月到 3 月仍保持一定水平，这表明每次犯罪的 911 电话率有所下降。这一基本见解鼓励我们更直接地控制犯罪的季节性性质。 1 因此，我们估计了排除 Zoorob 识别出的异常值但控制每周平均温度以及月份和块组固定效应的模型，如我们原来的规范 2 做得更好使我们能够将具有相似天气条件的几周内的 911 呼叫率并列，并避免将凉爽的日子与犯罪率较高的炎热日子进行比较。在表 1 中，我们报告了这些新模型的结果。使用AIC和BIC指导模型选择，我们发现，与原始模型和 Zoorob 开发的模型相比，我们修改后的模型在拟合度和简约性之间提供了更好的平衡。与前两个模型一样，模型 3 包括二次后裘德趋势。模型 4 包括一个线性的后裘德趋势。3 表 1 的第三个模型——排除了异常周和每周温度的控制——报告了后裘德时期的负线性项 (p < .001) 和一个小的，正二次项 (p < .01)。图 1 绘制了模型 3 的估计值，显示 911 呼叫在 Jude 故事之后最初下降，然后在几个月后恢复到预期水平。同样，模型 4 报告了在裘德的故事之后与犯罪相关的 911 电话显着线性下降。这些模型（均不包括异常周）表明，在有关裘德被殴打的新闻报道后，密尔沃基社区与犯罪相关的 911 电话远低于预期。结果与我们的原始发现高度一致，反映在表的第一列中。895979 ASRXXX10.1177/0003122419895979美国社会学评论Desmond et al. 研究文章2020