Spatiotemporal Heterogeneities in the Causal Effects of Mobility Intervention Policies during the COVID-19 Outbreak: A Spatially Interrupted Time-Series (SITS) Analysis

Abstract

Although there has been a growing interest in causal inference in geography studies, few studies have incorporated spatiotemporal heterogeneities with causalities. This study conceptualizes different patterns of spatiotemporal heterogeneity in the causal effects of policy interventions and develops a spatially interrupted time-series (SITS) quasi-experimental design to causally infer how the treatment effects of mobility control policies during the early stages of the COVID-19 outbreak vary across space and time, based on a five-month mobile phone big data set from Shenzhen, China. The modeling results reveal and distinguish significant temporal, spatial, and spatiotemporal heterogeneities in the policies’ causal effects. For example, we observed an abrupt decrease of 2.8 km in travel distance as a result of the first-level response to public health emergencies (i.e., FLR) and a decrease of 0.5 km as a result of the closed-off management of residential communities (i.e., COM), accounting for 44.6 percent and 7.2 percent of the baseline level before the pandemic, respectively. Such mobility reduction effects decayed at a rate of 0.033 km per day after the FLR and 0.076 km per day after the COM. For both policies, the abrupt effects were significantly larger in neighborhoods with a higher residential density and land-use mixture, lower average age, higher income, and higher marriage rate, whereas the gradual effect of the FLR decayed faster in similar compact neighborhoods. These findings demonstrate the importance of incorporating spatiotemporal variations with causality inference for fine-grained policy assessments, which can help policymakers determine when and where to implement which policies to mediate the mobility and the spread of the pandemic and plan for resilient neighborhoods in the postpandemic era.

尽管地理学研究愈发关注因果推断, 但很少将时空异质性和因果性相结合起来。本研究对政策干预的因果效应的不同时空异质性模式进行了概念化, 提出了空间中断时间序列准实验方案, 并基于中国深圳的5个月手机大数据, 因果式地推断了COVID-19爆发初期流动控制政策的治疗效果的空间和时间变化。模型结果揭示并区分了政策因果效应中的显著时间、空间和时空异质性。例如, 公共卫生突发事件一级响应导致出行距离急剧减少2.8公里, 居家封闭管理则导致出行距离减少0.5公里, 分别占COVID-19之前的44.6%和7.2%。这种流动性减弱效果, 在一级响应之后以每天0.033公里的速度衰减, 在居家封闭管理之后的衰减则是每天0.076公里。在居住密度较高、土地利用混合程度较高、平均年龄较低、收入较高和结婚率较高的社区, 这两项政策的剧烈影响更大, 而一级响应的渐进效应在小型社区的衰减更快。这些发现, 表明了在微观政策评估中结合时空变化与因果推断的重要性, 可以帮助决策者确定何时何地采取何种政策来调节流动性和流行病蔓延、规划流行病后的韧性社区。

Aunque se sabe del creciente interés por la inferencia causal en los estudios geográficos, pocos de estos han incorporado las heterogeneidades espaciotemporales con causalidades. El presente estudio conceptualiza diferentes patrones de heterogeneidad espaciotemporal en los efectos causales de las intervenciones de políticas, y desarrolla un diseño cuasiexperimental de series temporales espacialmente interrumpidas (SITS) para inferir causalmente cómo los efectos del tratamiento de las políticas para controlar la movilidad durante las etapas tempranas del brote de COVID-19 varían a través del espacio y el tiempo, con base en un conjunto de big data de telefonía móvil durante cinco meses en Shenzhen, China. Los resultados de la modelización ponen de manifiesto y distinguen heterogeneidades temporales, espaciales y espaciotemporales notables en los efectos de causales de las políticas. Por ejemplo, observamos un reducción abrupta de 2,8 km en la distancia del viaje como resultado de la respuesta de primer nivel a las emergencias de salud pública (esto es, FLR) y una disminución de 0,5 km como resultado del manejo cerrado de las comunidades residenciales (esto es, COM) que, respectivamente, representan el 44,6 y el 7,2 por ciento del nivel de referencia de antes de la pandemia. Tales efectos de disminución de la movilidad se redujeron a un ritmo de 0,033 km por día tras la FLR y a 0,076 km por día, después de la COM. Para ambas políticas, los efectos abruptos fueron significativamente más grandes en los vecindarios con densidades residenciales más altas y mezclas de uso del suelo, edad promedio más baja, ingresos mayores y tasas de nupcialidad más altas, en tanto que el efecto gradual de la FLR declinó más rápidamente en vecindarios similarmente compactos. Estos hallazgos demuestran la importancia de incorporar las variaciones espaciotemporales con inferencia de causalidad para evaluaciones políticas de grano fino que pueden ayudar a los dirigentes políticos a determinar cuándo y dónde implementar aquellas políticas con las que se medie la movilidad y la dispersión de la pandemia, y planear para conseguir vecindarios resilientes en la era pospandémica.

Key Words:

• COVID-19 pandemic
• human mobility
• interrupted time-series (ITS) design
• mobility intervention policies
• spatiotemporal causality inference

关键词：:

• COVID-19流行病
• 人类流动性
• 中断时间序列方案
• 流动性干预政策
• 时空因果推断。

Palabras clave::

• China
• diseño de series temporales interrumpidas (ITS)
• inferencia de causalidad espaciotemporal
• movilidad humana
• pandemia de COVID-19
• políticas de intervención en movilidad
• Shenzhen

Acknowledgments

We are grateful to the anonymous reviewers for their insightful comments on our work. We also highly appreciate the editor’s (Prof. Ling Bian) efforts devoted to this article.

Notes

1 We use the Hive SQL module on the DaaS platform provided by China Unicom to remove the error data. First, according to the probability distribution curve of the moving speed, we use 60 km/h as the threshold to exclude records with excessive moving speed. Second, we exclude movements with zero distance or speed. Finally, we check the origin and destination of each record, merging and adjusting those with disconnected origin–destination.

2 The interclass correlation coefficient of the null model equals 0.302, meaning that 30.2 percent of the variance is attributable to neighborhood-varying traits. The likelihood ratio test compared with ordinary linear regression is significant at the 0.1 percent level.

Additional information

Funding

The research reported in this article was funded by the National Natural Science Foundation of China (42171201), the Shenzhen Municipal Natural Science Foundation (Key Project) (GXWD20201231165807007-20200810223326001), and the Shenzhen Municipal Natural Science Foundation (JCYJ20190808173611341).

Notes on contributors

Wenjia Zhang

WENJIA ZHANG is an Assistant Professor in the School of Urban Planning & Design, Peking University Shenzhen Graduate School, Shenzhen, Guangdong 518055, China. E-mail: [email protected]. His research interests include urban behavioral geography, urban spatial structure, land use and transportation planning, as well as big data and machine learning approaches.

Kexin Ning

KEXIN NING is a Graduate Student in the School of Urban Planning & Design, Peking University Shenzhen Graduate School, Shenzhen, Guangdong 518055, China. E-mail: [email protected]. Her research interest focuses on the causality inference in urban and geography studies.