The COVID-19 pandemic necessitated social distancing measures globally, including the United States. This study addresses the gap in understanding how people respond to such measures and how they resume normal behaviors after restrictions are eased. The research utilizes a large dataset of mobile device location data (100 million devices) from February 2, 2020 to May 30, 2020, covering the contiguous US, Alaska, and Hawaii. The primary objective is to provide insights into people's movement patterns to inform public health policy and improve epidemic modeling. Existing research often relies on modeling parameters estimated by MCMC, simulations based on synthetic populations, survey data, or dedicated surveys, lacking timely real-world observations. While some studies have analyzed mobility changes during the pandemic, they often focus on single indicators like travel distance, failing to capture the multifaceted nature of mobility changes and social distancing efforts. This paper proposes a comprehensive Social Distancing Index (SDI) to address these limitations.

The literature review highlights the importance of understanding people's actual behavior in response to interventions for accurately modeling transmission dynamics. Existing studies employ various methods such as Markov Chain Monte Carlo (MCMC) for parameter estimation, simulations using synthetic populations, survey data for contact pattern estimation, and dedicated surveys for behavioral responses. However, a lack of timely, real-world observational data is identified. Previous work evaluating mobility changes during the pandemic often focuses on a single metric, such as distance traveled, neglecting the multiple dimensions of human movement. The authors note the need for a more inclusive index that simplifies the information and facilitates communication among stakeholders, particularly during a crisis like the COVID-19 pandemic. The study examines existing indices, categorizing them as category-based (e.g., Pandemic Severity Index) or score-based (e.g., Bloomberg Global Health Index), and argues for the superiority of score-based indices in integrating multiple metrics for a more comprehensive evaluation.

The study constructs a Social Distancing Index (SDI) as a score-based index ranging from 0 to 100, where 0 indicates no social distancing and 100 represents perfect social distancing. The SDI integrates five key mobility metrics derived from the mobile device location data: percentage of residents staying home, daily work trips per person, daily non-work trips per person, distances traveled per person, and out-of-county trips. The benchmark values for these metrics are established using weekday data from the first two weeks of February 2020. The SDI calculation incorporates percentage reductions in these metrics from the benchmark, with an emphasis on reducing non-essential trips and factoring in the relative importance of different metrics. The formula is: SDI = [X₁ + 0.01 × (100 - X₁) × (0.25X₂ + 0.45X₃ + 0.3X₄)] × 0.8 + 0.2X₅, where X1 represents the percentage of residents staying home, and X2-X5 represent percentage reductions in other mobility metrics. Weight assignments (β1-β5) are determined based on observed ratios between resident and out-of-county trips, the relative importance of trip reduction vs. distance reduction, and the distinction between essential and non-essential trips. The study explores the sensitivity of SDI scores to changes in weight assignments, finding that while magnitudes change, the spatial and temporal trends remain consistent. The study also received IRB exemption due to the use of de-identified data. Data availability is noted, with mobility metrics and SDI computation codes published on Harvard Dataverse.

The SDI proved effective in capturing mobility behavior changes and reflecting the impact of government interventions. Initially, the national emergency declaration triggered increased social distancing, followed by a plateau and then a decline, possibly due to quarantine fatigue and economic pressures. The spatial analysis reveals significant variations in SDI scores across states. States with stay-at-home orders generally exhibited higher SDI scores than those without. The temporal analysis showcases five stages: pre-pandemic, behavior change, government orders, quarantine fatigue, and partial reopening. High SDI states (e.g., DC, Hawaii, New York, New Jersey, Maryland) generally had higher initial SDI scores and issued stay-at-home orders, while low SDI states (e.g., Wyoming, North Dakota, South Dakota, Arkansas, Montana) did not. Analysis of the top five and bottom five states regarding cumulative confirmed COVID-19 cases showed that while all experienced increased SDI after stay-at-home orders, the bottom five consistently had lower scores, indicating a role of local outbreak severity in people’s behavior. Early reopenings were associated with sharp declines in SDI and subsequent acceleration in confirmed cases in several states (e.g., New York, Massachusetts, Alaska). Spearman's rank correlation coefficients showed a stronger correlation between SDI and new infection rates than cumulative infection rates, suggesting attention to the outbreak’s dynamic evolution. County-level analysis revealed similar patterns, with New York counties showing strong social distancing and a slowdown in cases, followed by a relaxation after reopening. Counties with lower SDI scores and increasing confirmed cases raised concerns about potential outbreaks. Correlation analysis at the county level also showed stronger correlations in counties with higher SDI scores, suggesting a relationship between mobility and infection.

This study demonstrates the value of real-world observations, specifically mobile device location data, in understanding the impact of non-pharmaceutical interventions during the COVID-19 pandemic. The developed SDI provides a data-driven tool for policymakers to assess the effectiveness of policies and make data-informed decisions. It also empowers the public by increasing awareness of local risks. The findings highlight the significant impact of government mandates on social distancing behaviors and the potential for quarantine fatigue to lead to decreased adherence. Early reopenings also appear to have counteracted social distancing efforts and potentially increased infection rates. The results suggest a need for continuous monitoring of mobility patterns and adapting policies accordingly. The stronger correlation between SDI and new infection rates underscores the importance of dynamic monitoring and responsiveness to current outbreak dynamics.

This study presents a novel Social Distancing Index (SDI) based on real-time mobile device location data to quantify the changes in human mobility during the COVID-19 outbreak. The SDI provides valuable insights into the effectiveness of social distancing policies and the factors influencing people's behavior. Future work should focus on refining the SDI by incorporating regional differences, adding more mobility metrics, integrating trip purpose data, and incorporating the SDI into epidemiological models. Furthermore, investigating the relationship between specific mobility patterns and disease transmission would enhance the index's predictive power.

The study acknowledges several limitations. First, the basic mobility metrics might need adjustments to account for regional differences, particularly between rural and urban areas. Second, incorporating additional mobility metrics (e.g., trip purposes identified using POI data) and variables related to disease transmission could improve the index's comprehensiveness. Third, the weights assigned to different variables in the SDI could be further refined through expert surveys or longer-term observation of mobility patterns and COVID-19 evolution. The reliance on de-identified mobile device data means that granular, individual-level information on behavior is not accessible.