Objectives

Test the reliability of geotagged Twitter data for estimating block-level population metrics across place types. Evaluate whether the proportion of Twitter users on a block at a given time who are local residents, inter-metro commuters, or tourists is correlated with incidences of public violence and private conflict for four different time periods: weekday days, weekday nights, weekend days, and weekend nights.

Methods

DBSCAN* machine learning technique is used to estimate the home clusters of 54,249 Twitter users who sent at least one geotagged tweet in Boston. Public violence and private conflict are measured using geocoded 911 dispatches. ANOVA models are used to evaluate how the presence of our three groups of interests varies across three types of block-level land usage. Hierarchical linear regression models are used to evaluate whether the proportion of commuters and tourists at census tract- and block-levels are predictive of crime events across the four time periods of interest.

Results

We find evidence that Twitter data has limited reliability across residential blocks due to data sparseness. For non-residential blocks, we find that commuter and tourist presence at the block-level are positively associated with both public violence and private conflict, but that these effects are not stable across time periods. Commuters and tourists only effect violence during weekday days, and the effects of commuters and tourists on private conflict are only statistically significant during weekday days and weekend days.

Conclusions

Consistent with routine activities and crime pattern theories, the influx of outsiders in a given location impacts the likelihood of crime occurring there. While we find that data from Twitter users can be valuable for measuring block-level ambient populations, it appears this is not true for residential blocks. Future research may further consider how the characteristics of Twitter users may inform spatial patterns in crime.

中文翻译：

---

谁在何时何地“发推文”，以及它如何帮助理解地方的犯罪率？测量环境人口中游客和通勤者的存在

目标

测试经过地理标记的Twitter数据的可靠性，以估计跨场所类型的块级人口指标。评估给定时间在街区上的Twitter用户中本地居民，城际通勤者或游客的比例是否与以下四个时段的公共暴力和私人冲突发生率相关：工作日，工作日晚上，周末，以及周末之夜。

方法

DBSCAN *机器学习技术用于估计54249位Twitter用户的家庭集群，这些用户在波士顿发送了至少一条带有地理标记的推文。使用地理编码的911派遣量度公共暴力和私人冲突。ANOVA模型用于评估我们的三类利益的存在在三种类型的块级土地利用方式中如何变化。分层线性回归模型用于评估人口普查区域和街区级别的通勤者和游客比例是否可以预测四个感兴趣时段内的犯罪事件。

结果

我们发现有证据表明，由于数据稀疏，Twitter数据在整个住宅区的可靠性受到限制。对于非住宅街区，我们发现街区级别的通勤者和游客与公共暴力和私人冲突均呈正相关，但这些影响在一段时间内并不稳定。通勤者和游客仅在工作日内发生暴力，通勤者和游客对私人冲突的影响仅在工作日和周末期间具有统计意义。

结论

与常规活动和犯罪模式理论相一致，特定地点的外来人员涌入会影响那里发生犯罪的可能性。虽然我们发现来自Twitter用户的数据对于测量街区水平的环境人口可能很有价值，但对于住宅街区似乎并非如此。未来的研究可能会进一步考虑Twitter用户的特征如何为犯罪提供空间模式。