Causal Inference Methods and their Challenges: The Case of 311 Data
Causal Inference Methods and their Challenges: The Case of 311 Data
复制标题
DOI：
10.1145/3463677.3463717
复制
发表时间：
2021-06
期刊：
影响因子：
--
通讯作者：
F. Yusuf;Shaoming Cheng;S. Ganapati;G. Narasimhan
中科院分区：
文献类型：
--
作者：
F. Yusuf;Shaoming Cheng;S. Ganapati;G. Narasimhan
本文的主要目的是说明因果推理方法在行政数据中的应用以及这种应用的挑战。我们说明了应用贝叶斯网络的方法，311数据从迈阿密戴德县，佛罗里达（美国）。311个中心为居民提供非紧急服务。311数据很大且粒度很细。我们的目标是探讨这种特殊类型的服务请求中可能存在的公平问题和偏见。作为一个案例研究，人口特征（自变量）和请求量和完成时间（因变量）之间的关系进行检查，以确定的差异，如果有的话，从观察数据。实证分析表明，提供给任何特定人口、社会经济或地理群体的服务没有偏见。然而，由于数据缺失或不纯、不充分和潜在混杂因素，管理数据在推断因果关系方面确实存在各种挑战。讨论了应用因果分析技术分析311等行政数据的注意事项。
The main purpose of this paper is to illustrate the application of causal inference method to administrative data and the challenges of such application. We illustrate by applying Bayesian networks method to 311 data from Miami-Dade County, Florida (USA). The 311 centers provide non-emergency services to residents. The 311 data are large and granular. We aim to explore the equity issues and biases that might exist in this particular type of service requests. As a case study, the relationship between population characteristics (independent variables) and request volume and completion time (dependent variables) is examined to identify the disparities, if any, from the observational data. The empirical analysis shows that there are no biases in services provided to any specific demographic, socioeconomic, or geographical groups. However, the administrative data do have various challenges for inferring causality due to missing or impure data, inadequacy, and latent confounders. The precautions of applying causal techniques to analyzing administrative data like 311 are discussed.