The development of digital society makes machine learning more and more remarkable in social science research, but machine learning applications in public administration literature is still limited. This article firstly introduces the basic theories of machine learning, and then analyses the applications of machine learning in recently public administration leading journals, and finally summarizes challenges of using this method in public administration research. This article contends that machine learning can help both data collection and data analysis.In data collection, machine learning methods help to turn the non-numeric data into numeric data, as well as data clustering. In data analysis of relativity, machine learning is used to mitigate the dimensional disaster.In causal inference research, this method can analyse the heterogeneous causality, as well as average treatment effect with other causal inference methods such as IV, PSM, RDD, DID and SCM. However, machine learning is challenged in practicability, usability, explanatory and repeatability, which also draws risks from the perspective of public value. Even though, considering its great potential contribution,this article believes that using machine learning in public administration research will be a new trend.

社会数字化的发展使得机器学习方法愈加受社会科学研究重视。尽管机器学习方法对公共管理研究存在巨大价值,但该方法在现有公共管理文献中的运用仍相对有限。本文首先简要介绍了机器学习方法的基本原理和操作方法,然后整理分析了近年来国内外公共管理顶尖期刊上所发表文献中机器学习方法的具体应用,最后对该方法所面临的挑战和质疑进行了归纳和部分回应。本文认为,机器学习在公共管理研究中既可以作为一种数据收集与整理的研究方法,又可以作为一种数据分析的研究方法:首先,机器学习方法可以降低数据的收集成本,将非数值型数据转换为数值型数据,也可以用于非数值型数据的聚类;其次,在相关性分析的公共管理研究中,机器学习可以缓解 “维数灾难”,即变量数过多导致的多重共线性以及过拟合问题,从而分析变量之间的非线性驱动作用;在因果推断类型的公共管理研究中,该方法主要用来分析异质性因果,但也可以结合其他因果推断方法从而估计平均处理效应,解决工具变量法、倾向得分匹配法、断点回归法、双重差分法和合成控制法在公共管理研究应用中所遇到的实践难题。然而,机器学习方法在实用易用性、结果解释性、研究可复现性等方面同样遭受质疑, “决策不透明”的铁笼有时也会诱发违背公共精神的风险。尽管如此,考虑到该方法潜在的巨大贡献,本文预计在公共管理研究中结合机器学习方法将成为一种新的研究趋势。