dcyphr | Effects of human mobility restrictions on the spread of COVID-19 in Shenzen, China: a modelling study using mobile phone data


One of the first steps utilized by governments trying to combat the spread of novel coronavirus is the restriction of public mobility. This tactic includes everything from limiting public transportation to banning public gatherings. While past studies have looked at the effect of mobility restrictions on certain subpopulations (school children or workers), none have observed information about a population as a whole. This study’s goal is to create several comprehensive coronavirus transmission models utilizing mobile phone data. These models will help “quantify the potential effects of various intra city mobility restrictions.” The results of this research can both improve government approaches to reducing COVID-19 spread and address social unrest related to shutdowns.


The models created in this study combined “the effects of intra city mobility with the force of infection.” In other words, statistics about virus transmission (R0, incubation period, infection duration, etc) were brought together with mobile data about population movement to predict epidemic curves. All of the mobile data was provided by the company Unicom (region of focus was Shenzhen, China) and was anonymous and aggregated (no access to personal information). 

The population analyzed in the study was broken down into several categories: vulnerable to SARS-CoV-2 infection, exposed individuals who are asymptomatic (assumed infectious), infectious and symptomatic individuals, and recovered individuals with no risk of further infection. Models were created for all ten regions in Shenzhen, China. Mobile phone data was collected during working days (Mon-Fri), with user location being accessed even if their phone was not in use (simply on). Location between 7am and 9am was assumed to represent a user’s home.

Although various sources publish conflicting values, this study assumed an incubation time of 5.2 days and a serial interval (time between an infector and an infectee presenting symptoms) of 7.5 days. An R0 of 2.68 with an infectious period of 12.7 days was used in a baseline model. Further models used an R0 from 2.0-4.0 and an infectious period between 8-15 days. The transmission modeling in this study involves a series of differential equations. The details and formulas used are available in the full publication. The levels of mobility restriction analyzed in the study range from 0%-20%-60%. Various types of restrictions were included in the models (widespread lock down, lock down in high risk areas, etc). Targeted restriction of infectious individuals was also analyzed (no simulation with 0% mobility was run as it is impossible to control all symptomatic patients). The effects of all these restrictions were discussed in the context of slowing epidemic growth (slower spike) and flattening of the epidemic curve (fewer peak cases). 


Based on the distribution of cases throughout Shenzhen, it was concluded that most cases were brought in from Wuhan. There was very little local transmission, largely due to the rapid implementation of health measures. The base model (R0 of 2.68) predicted an epidemic peak in March 2020 (with no mobility restrictions). Another model with the same conditions, but including a 20% mobility reduction demonstrated a peak appearing two weeks later, with a 33% reduction in peak incidence (lower spike). A 25% reduction in disease transmissibility, paired with a 20% reduction in mobility, demonstrated a peak delayed by 4 weeks, with a 42% peak incidence reduction. A 40% mobility reduction yielded a 66% reduction in peak incidence with a 6 week peak delay. A 25% reduction in disease transmissibility paired with a 40% mobility restriction resulted in a 76% peak incidence reduction, with a 7 week peak delay. A 60% mobility restriction resulted in a 91% peak incidence reduction, with a 14 week peak delay. Lastly, a 25% transmissibility reduction paired with a 60% mobility restriction produced a curve with no peak in the first half of 2020. Overall: increased mobility restriction paired with reduced disease transmissibility resulted in the most significant peak reduction (smaller, delayed peak). Transmissibility in all of these models is based on the infectious period and other measures to reduce disease spread.  

Mobility restrictions on particular subgroups were also assessed. A 20% mobility restriction of infectious individuals resulted in a two week peak delay, with 50% fewer peak cases. Inter-region mobility reduction of 80% led to a 4 week peak delay, with 50% fewer peak cases. Locking down the two regions with the most initial cases resulted in a four week peak delay. 


The study noted that the effectiveness of mobility restrictions significantly increased in models with increased “transmissibility reduction measures.” It also noted that the curves produced by the models largely depended on disease characteristics which still haven’t been established (many conflicting R0 values and incubation period values published). Mobile tracking was an effective way of representing population movement, as the region’s mobile phone use for individuals ages 15-65 is nearly 100%. 

The data published is valuable in weighing the benefits and limitations of mobile restrictions. Controlling the public’s ability to move around has many clear social and economic setbacks. 

One large limitation of this study is that 2019 movement data was used, which may not be consistent with movement patterns in 2020. Additionally, there is evident potential for error in using a single company’s tracking data to extrapolate and represent the whole city’s population. Lastly, as stated above, further research about the virus itself is essential to produce accurate models.