TEMPORAL NETWORK INFLUENCE MODEL WITH APPLICATION TO THE COVID-19 POPULATION FLOW NETWORK

• 成果类型: Article
• 署名作者: Zhang, Dongxue; Feng, Long; Wu, Yujia; Lan, Wei; Zhou, Jing
• 署名单位: Southwestern University of Finance & Economics - China; Southwestern University of Finance & Economics - China; Nankai University; Nankai University; Renmin University of China; Renmin University of China
• 刊物名称: ANNALS OF APPLIED STATISTICS
• ISSN/ISSBN: 1932-6157
• DOI: 10.1214/25-AOAS20252025
• 发表日期: 2025
• 页码: 1382-1402
• 关键词: panel-data models; human mobility; identification; selection

摘要

Ever since its outbreak, COVID-19 has been rapidly spreading around the world and has become a significant threat to public health. Past experience has shown that, because of the incubation period, the contemporaneous population flow does not affect the contemporaneous number of cases, but the time-lagged population flow can affect case numbers. Moreover, the population flow networks of different lags can exhibit varying influences on the transmission of COVID-19. However, most existing studies analyze only the static effects of a network, while ignoring its dynamic characteristics. To assess the dynamic effect of a population flow network, we propose a novel temporal network influence (TNIF) model. This model studies the influence of case numbers via three transmission mechanisms of COVID-19: cross-city transmission, within-city transmission and diffusion transmission. Consequently, three dynamic parameters are introduced to measure the three transmission influences, respectively. To estimate these influential parameters, we develop a quasi-maximum likelihood estimator and establish its theoretical properties. Additionally, to examine the changing pattern of the cross-city transmission influential parameter, we provide two max-type test statistics for testing the homogeneity and change point, respectively. We also demonstrate their asymptotic distributions. Furthermore, a Bayesian information criterion is developed to select the order of the within-city and diffusion transmission influential parameters. The effectiveness of the proposed TNIF model is evaluated through simulation studies. Finally, we apply the proposed TNIF model to analyze the COVID-19 data that collected from 283 Chinese cities from January 20, 2020 to April 23, 2020. We found that the population flow network from the previous period and two periods prior had a significant positive impact on the number of cases on the current period, while the network from earlier periods (three or more periods prior) had no significant effect. This finding further implies that the two-week lockdown policy implemented at that time was reasonable and could effectively slow down the spread of COVID-19 at a relatively low cost.