在疫情防控常态化的今天，有必要对相关政策的执行开展研究。政策执行概念起源于政策过程理论，是政策从创设到终止之间必不可少的环节。自上世纪70年代以来，政策执行领域经过长期的发展，积累了丰富的研究成果，三大研究主题包括分析模型、研究视角、影响因素。根据研究视角的区分产生出“自上而下”“自下而上”两种具有代表性的流派，以及近年来兴起的、更加强调综合分析的政策执行网络研究。政策执行网络，为适应参与主体多元化、互动关系多向化、政策场景复杂化的现代趋势，更加深入理解政策执行问题提供了全新的研究路径。 采取政策执行网络视角，本文重点关注在中国的政治体制之下区级政府的政策执行与效果，以及这种效果如何受到来自纵向和横向双重网络效应的影响。基于2019年到2020年期间贵州省、市、区三级97个政府网站共330余万条的文本数据，本文应用“无监督式-半监督式”两阶段文本主题建模法刻画贵州省各级政府网站全量内容主题，反映与应对疫情相关的各类主题随时间的演变趋势。利用监督式机器学习法，构建衡量政府对疫情防控或复工复产关注程度的政策执行变量，验证政策执行差异的存在。通过融合包含贵州省各地车流量在内的多源异构数据，建立疫情期间贵州省83个区县面板数据模型，以政策执行为自变量、车流量为因变量，确证政策执行对政策效果具有显著影响。在此基础上，研究相邻区对本区影响的横向网络效应，并进一步引入市级对区级影响的纵向网络效应，得出双重网络效应对本区政策效果均具有显著影响和调节效应的结论。在更深入的分析中，发现纵向网络效应与横向网络效应同时存在时，以前者为主导。 本文是对政策执行网络理论的有益补充。除延续以往对横向网络效应的关注外，基于公共管理中的层级概念引入纵向网络效应，在国内的市-区两级政府主体之间开展了具体的分析。借助文本主题建模、机器学习等多源数据分析技术，不仅从实证角度展开探讨，更提供了来自以天为单位时间尺度上的证据。在应对疫情复杂背景下的分析，拓展了理论的适用范围，同时针对下一阶段有关工作提出了政策建议。本文采取的机器学习与计量模型相结合的多源数据分析方法，也可为政策信息学的更深层次应用带来启发。

Today, with the continuance of COVID-19 epidemic prevention and control, it is necessary to carry out research on the implementation of relevant policies. The concept of policy implementation originates from the policy process theory and is an indispensable step between policy creation and termination. Since the 1970s, the field of policy implementation has accumulated rich research results after a long period of development. The three major research themes include analytical models, research perspectives, and influencing factors. According to the distinction of research perspectives, there are two representative schools of "top-down" and "bottom-up", as well as policy implementation network research that has emerged in recent years and emphasizes comprehensive analysis. The policy implementation network provides a new research path for adapting to the modern trend of public administration, which includes diversification of participants, multi-directional interaction, and complex policy scenarios. Moreover, the network approach gives rise to a more in-depth understanding of policy implementation issues. Taking a policy implementation network perspective, this paper focuses on the policy implementation of China’s district governments, aiming to illustrate how policy effects are affected by policy implementation and its network effects. Based on more than 3.3 million text data from 97 government websites in Guizhou from 2019 to 2020, this paper combines unsupervised and semi-supervised topic modeling to describe the themes of government websites at the provincial, municipal and district levels. In addition, the paper reveals the evolution trend of various themes related to the response to the COVID-19 pandemic over time. Using the supervised machine learning method, the paper constructs policy implementation variables that measure the government's attention to epidemic prevention and control or resumption of work and production, verifying the existence of differences in policy implementation. Through the fusion of multi-source heterogeneous data including traffic flow and website contents, a panel data model of 83 districts in Guizhou Province during the epidemic is established. The model uses policy implementation as the independent variable and traffic flow as the dependent variable, to confirm that policy implementation has a significant impact on social functioning. On this basis, drawing on the concepts in social network analysis, the horizontal network effect from adjacent districts and the vertical network effect from the municipal level on the district level is studied. The results show that both of these two network effects have a moderating effect. In a more in-depth analysis, it is found that vertical network effect plays a more important role compared to the horizontal network effect. This paper is a useful complement to policy implementation network theory. In addition to continuing the previous focus on the horizontal network effect, the vertical network effect is introduced and a specific analysis is carried out between the city-district government entities in China. With the help of multi-source data analysis techniques such as topic modeling and machine learning, this paper not only carries out research from an empirical perspective, but also provides evidence from a time scale in days and weeks. The analysis under the complex background of responding to the COVID-19 pandemic expands the scope of application of the theory, and at the same time puts forward strategies for the next stage of work related to the epidemic. The multi-source data analysis method combining machine learning and econometric model adopted in this paper can also bring inspiration to following research, paving way for further application of policy informatics.