Human Mobility, Policy, and COVID-19: A Preliminary Analysis of South Carolina
Li Z., Huang X., Zhang J., Zeng C., Olatosi B., Li X., Weissman S., (2020), Human Mobility, Policy, and COVID-19: A Preliminary Study of South Carolina, preprint
Human movement is an important driver of geographic spread of infectious diseases (Kraemer et al., 2019). Prediction and control of the spread of infectious disease benefits greatly from our growing capacity to quantify human movement. (Hancock, 2014). COVID-19 has a high human-to-human transmission rate and can be transmitted during the incubation period. A recent study indicates that human movement predicts the spread and size of epidemics in China at the early stage and later the control measures mitigated the spread of the virus (Kraemer et al., 2020). Meanwhile, it is equally important for policy makers and emergency responders to gain a picture of how people follow the stay-at-home orders and how effective is the control measures curb the spatial propagation of virus.
Using geotagged Twitter data as the mobility data source, we present some preliminary findings and visualizations on population flows and human mobility changes during the pandemic at state level and county level in South Carolina. Potential associations between human mobility, state policies, and COVID-19 cases are also examined.
Overview of the national level mobility changes
Before zooming into South Carolina, let’s look at the state mobility changes across the nation. The average movement distances (in mile) for each state is derived from millions of geotagged Twitter data (collected using the official Twitter public APIs). The method for calculating the distance is detailed in our recent preprint paper available here (Huang, Li et al., 2020)(cross-day distance is used in the analysis).
Figure 1 shows the spatial pattern of average movement distance changes across the nation for the selected six days since March 2020. All maps use the same coloring group/legend so they are visually comparable. The first map shows the mobility pattern on March 1st before the pandemic declaration (March 11) and the National Emergency declaration (March 13). The mobility decreased on March 15, and April 26 saw a further decrease. With the lifting of restrictions across the nation, most states show increased mobility starting in May and continue to increase in June.
Figure 2 shows the daily average movement distance for the six selected states (South Carolina, North Carolina, Georgia, New York, and Washington) from Jan. 1 2020 to Jun. 30 2020. Note that the distance shown here are not normalized. It is good for showing the changes over time for each state, but comparison among states should be taken by caution as each state has its own normal daily travel distance (depending on state geographic size and location, people’s travel preference etc.). More analyses on the normalized mobility can be found here.
Figure 3 is a heat map visualization of the state level daily new COVID-19 cases from February to July 8, 2020. The COVID-19 cases data is downloaded from The New York Times. Each bar indicates a state and the vertical axis represents the dates. The six selected states in Figure 2 are highlighted with red rectangle in Figure 3. South Carolina has maintained a relatively low number of daily new cases until late May.
Figure 4 shows the daily average human movement distance and daily new COVID-19 cases from February to July in South Carolina. The data for movement distance is from Feb. 21 to Jun. 30 (blue dash line). The data for new cases is from Mar. 6 to Jul. 10 (orange dash line) . 7-day moving average of the two datasets are presented as solid lines. Some selected policies (measures) in the state are also indicated in the diagram. This figure reveals that policies (such as stay-at-home measures, and reopening facilities) have immediate impact on human mobility. Visually, the reduced mobility, together with other measures taken by the state, have contained initial phase of the outbreak in April and May as we saw a slight decrease in the number of daily new cases starting around April 10. The trend of the two lines also suggest the increase of human mobility from mid-April has a positive association with COVID-19 cases with a delayed effect. The increasing mobility after a series of restrictions being progressively lifted since April 20 followed by a second wave of infections again raises the question of whether the state opens too soon.
To further examine the association between daily average movement distance and daily new cases, a scatter plot is created using a 14-day lag between the two variables. The movement data is from Mar. 17 to Jun. 26 and new cases data is from Mar. 31 to Jul. 10. R-square of 0.59 indicates that the daily movement distance explains 59% of variations of the daily new cases in South Carolina during the study period. The Pearson’s r (linear correlation) between two variables is 0.756. This finding suggests and reinforces that integrating human mobility into predictive modeling is likely to provide a better picture about the future infected cases.
Myrtle Beach (located in Horry County of South Carolina) has become a coronavirus hot spot in June. Here, we further zoom in on Myrtle Beach to explore the potential role human mobility plays in the outbreak at the county level.
Figure 6 is a heat map visualization of the county daily new COVID-19 cases from March to July 10, 2020 in South Carolina. The COVID-19 cases data is downloaded from The New York Times. Each bar indicates a county and the vertical axis represents the dates. Horry County (Myrtle Beach) is highlighted with red rectangle in Figure 6. The heat map shows Horry County, along with several other counties including Greenville, Charleston, Lexington, and Richland, are experiencing high number of daily cases at the time of writing.
Figure 7 shows the daily movement distance (mile) in Myrtle Beach from Jan. 1 to Jun. 30, 2020. The method for calculating the distance is detailed in our recent preprint paper available here (Huang, Li et al., 2020) (cross-day distance is used in the analysis). Though with fluctuations, the effectiveness of the stay-at-home measure can be clearly observed: the average travel distance decreased dramatically around mid-March and stayed at a low level until late April when it started to bounce back and reached to previous level around mid-May.
Figure 8 shows daily average human movement distance and daily new COVID-19 cases from February to July in Myrtle Beach, South Carolina.
Some selected policies (measures) in South Carolina are indicated in the diagram. The data for movement distance is from Feb. 21 to Jun. 30 (blue dash line). The data for new cases is from Mar. 15 to Jul. 10 (orange dash line) . 7-day moving average of the two datasets are presented as solid lines. Again, the figure suggests the increase of mobility has a positive association with COVID-19 cases in Myrtle Beach with a time lag between the two variables.
Considering that Myrtle Beach is one of the top beach destinations in the U.S., two types of populations might contribute to the increase of human mobility observed since late April in Myrtle Beach: local residents and visitors from other parts of the state (and country). Figure 9 shows the number of daily Twitter visitors in Myrtle Beach from Jan. 1 to Jun. 30, 2020. The method for calculating the number visitors is detailed in a manuscript submitted to Applied Geography (Martin, Li, Ge, 2020, in review, link to the article will appear here once it is published). In general, the trend of daily visitors is similar to daily average movement distances.
For both daily movement distance and visitors, since late May the values have reached to the previous level of Jan. and Feb., and some days even
surpassed the values before the stay-at-home order (Figure 9). Does this mean Myrtle Beach has returned to the normal situation with regards to human mobility? The answer is no. It is understandable that coastal counties with tourist attractions exhibit clear seasonal pattern of visitations, and our twitter-derived population flows clearly reveal such patterns. Figure 10 shows daily number of Twitter visitors in Myrtle Beach in 2019 (blue line, from Jan. 1 to Dec. 31 ) and 2020 (from Jan. 1 to Jun. 30, orange line). As illustrated in this diagram, 2019 and 2020 show similar patterns in Jan. and Feb. (blue line and orange line generally overlap with each other before March). The 2020 pattern started to deviate from 2019 in early March due to the pandemic. In June 2019, the total Twitter visitors (samples) is 6,434, while in June 2020 the number is 3,755, about 58.4% of previous year.
Figure 11 shows the population flows (captured from Twitter data) to and from Myrtle Beach in May and June, 2020 (with travel distance greater than 50 miles). A further analysis reveals that 56.16 percent of the travels occurred within South Carolina, 23.67 percent from North Carolina, followed by Virginia (4.10), Ohio (2.22), Tennessee (1.83), Florida (1.53), West Virginia (1.53), Maryland (1.14), Pennsylvania (1.14), Kentucky (0.94), and others (5.73). Note that only travels within the continental U.S. are included in the analysis.
The spatial distribution of the population flows can be partially explained by the place connectedness among U.S. counties measured from historical population flows. Figure 12 shows the Place Connection Index (Li, Huang et al., 2020) between Myrtle Beach (Horry County) and other counties in the U.S. The index is derived from over 300 million geotagged tweets posted in 2019, and is controlled for population effect to reveal hidden spatial structures among places. Besides South Carolina, Myrtle Beach shows strong connections with some northern states including North Carolina, Ohio, West Virginia, Kentucky, Pennsylvania, and Maryland.
Understanding human mobility or population flows within a region or between different places (e.g., countries, states, counties, cities) could help us gain better insights into the current and future infectious risk at the population level during a pandemic such as COVID-19. Using South Carolina as a case study, this blog post reports our recent findings of human mobility changes at state and county level during the pandemic. Results suggest that declaration of the state of emergency and the stay-at-home order in mid-March have immediate impact on human mobility in South Carolina. The decrease of daily new cases in mid-April and the relatively low daily new cases in May indicate that the reduced mobility, together with other mitigation measures taken by the state, have effectively contained the initial phase of the outbreak in April and May. The preliminary findings also suggest the increase of mobility has a positive association with COVID-19 cases with a delayed effect at both state level and county level.
While Myrtle Beach has seen increases in human mobility (and visitors) since May, a comparison with 2019 mobility data reveals that Myrtle Beach only received about 58.4 percent of visitors in June 2020 compared to June 2019. The visitors are mainly from the counties (and states) that are highly connected to Myrtle Beach based on our 2019 Place Connection Index, reinforcing that human mobility patterns are to some extend predictable. Human movement patterns derived from historical big data can be used to assess the potential risks during such a pandemic. We are currently in the process of integrating the mobility information to predictive models to enhance model performance.
Finally, as geotagged Twitter data does not represent the whole population groups (e.g., biased toward young population, check Jiang, Li, Ye 2019 and Martin, Cutter, Li 2019 for more information on this topic), human mobility derived here must be interpreted with such limitations in mind. We are comparing and integrating human mobility derived from Twitter data with other data sources including Google mobility data, Apple mobility data, and mobile phone data to gain a more comprehensive picture of human movement during the pandemic.
Some personal thoughts
With the current alarming number of daily new cases in South Carolina, as a South Carolinian I think we cannot stress enough about the importance of following the safety measures recommended by SC Department of Health and Environmental Control, especially when we cannot reduce human mobility to the level as in March and April due to complex factors.
- Practicing social distancing
- Wearing a mask in public
- Avoiding group gatherings
- Regularly washing your hands
- Staying home if sick
Acknowledgement: The study is supported by NSF (2028791) and University of South Carolina COVID-19 Internal Funding Initiative (135400-20-54176). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of this post.
Project team: Zhenlong Li, Xiaoming Li, Dwayne Porter, Bankole Olatosi, Jiajia Zhang, Xiao Huang, Yuqin Jiang, Chengbo Zeng, Xinyue Ye
Kraemer, M. U. G., Golding, N., Bisanzio, D., Bhatt, S., Pigott, D. M., Ray, S. E., … & Pybus, O. G. (2019). Utilizing general human movement models to predict the spread of emerging infectious diseases in resource poor settings. Scientific reports, 9(1), 1-11
Kraemer, M. U., Yang, C. H., Gutierrez, B., Wu, C. H., Klein, B., Pigott, D. M., … & Brownstein, J. S. (2020). The effect of human mobility and control measures on the COVID-19 epidemic in China. Science.
Hancock, P. A., Rehman, Y., Hall, I. M., Edeghere, O., Danon, L., House, T. A., & Keeling, M. J. (2014). Strategies for controlling non-transmissible infection outbreaks using a large human movement data set. PLoS computational biology, 10(9).
Jiang, Y., Li, Z., & Ye, X. (2019). Understanding demographic and socioeconomic biases of geotagged Twitter users at the county level. Cartography and geographic information science, 46(3), 228-242.
Huang, X., Li, Z., Jiang, Y., Li, X. & Porter, D. (2020). Twitter, human mobility, and COVID-19. arXiv preprint arXiv:2007.01100
Martin Y., Li Z., Ge Y., Towards real-time population estimates: introducing Twitter daily estimates of residents and non-residents at the county level, Applied Geography (in review)
Yang C., Sha D., Liu S., Li Y., Lan H., Guan W., Hu T., Li Z., etc., Taking the pulse of COVID-19: A spatiotemporal perspective.,
arXiv preprint arXiv:2005.04224
Li Z., Huang X., Jiang Y., Ye X., et al. (2020), Measuring Place Connectedness using Social Media Big Data (to be submitted).