Quantifying human mobility behaviour changes during the COVID-19 outbreak in the United States

The study addresses how people in the United States changed mobility behaviors in response to the COVID-19 outbreak, government social distancing policies (such as stay-at-home orders), and phased reopening, and how quickly behaviors relaxed when orders were lifted. The authors propose a Social Distancing Index (SDI) to capture multi-dimensional mobility changes using large-scale mobile device location data. Key questions include: How do people react to government actions and perform social distancing? What is the reopening readiness of each region? How can the risk of a second outbreak be assessed from mobility patterns? The objective is to generate a real-time, comprehensive, and communicable measure of social distancing to inform public health policy and improve epidemic modeling.

Prior research on non-pharmaceutical interventions often relies on modeling and simulation (e.g., MCMC-based parameter estimation, synthetic contact networks), survey-based contact patterns, or dedicated behavior surveys, with limited timely real-world observational evidence. Early real-time mobility studies during COVID-19 commonly focused on a single indicator such as distance traveled (e.g., social distancing scoreboards, effects of stay-at-home mandates, mobility by income). A single metric cannot fully characterize social distancing behaviors (e.g., trip frequency and spatial patterns). Indices in other fields include category-based (e.g., Pandemic Severity Index, Modified Mercalli Intensity Scale) and score-based (e.g., U.S. News State Rankings, Bloomberg Global Health Index); score-based indices better integrate multiple metrics. This motivates a score-based SDI aggregating multiple mobility dimensions to simplify communication and enhance policy relevance.

Data: Integrated mobile device location data representing person and vehicle movements from over 100 million devices across the contiguous U.S., Alaska, and Hawaii, from February 2 to May 30, 2020. Mobility metrics were generated via peer-reviewed, validated algorithms and integrated with COVID-19 case and population data. Benchmark period: weekdays during the first two weeks of February. Basic mobility metrics (state/county level): (1) Percentage of residents staying home (no trips >1.61 km from home). (2) Daily work trips per person (to/from imputed work location). (3) Daily non-work trips per person. (4) Distance traveled per person (km, all modes). (5) Out-of-county trips (count). Descriptive state-level ranges: staying home 13.0–58.0% (mean 26.1%), work trips 0.14–1.49 (mean 0.48), non-work trips 1.39–3.90 (mean 2.64), distance 15.6–113.4 km (mean 52.3), out-of-county trips 7–28,845 (mean 5,339). Social Distancing Index (SDI): A score from 0–100 indicating degree of social distancing relative to benchmark. Inputs X1–X5 represent changes relative to benchmark: X1 is absolute change in percentage staying home; X2–X5 are percentage reductions (0–100%, increases standardized to 0%). The SDI formula jointly considers resident and visitor behaviors: SDI = [X1 + 0.01 × (100 − X1) × (0.25 × X2 + 0.45 × X3 + 0.30 × X4)] × 0.8 + 0.2 × X5. Weighting rationale: Resident vs out-of-county trips ratio approximated at 4:1 (β5 = 0.2). Trip reductions weighted more than distance reductions; distance reduction weight β4 = 0.3. Non-work (often non-essential) trip reductions favored over work trips; within resident travelers, β2 = 0.25 (work), β3 = 0.45 (non-work), β4 = 0.30 (distance). Sensitivity analyses indicate magnitude changes with β4 but preserved spatial/temporal trends. Analyses: Computed SDI daily at state and county levels; visualized temporal patterns and policy dates (national emergency, stay-at-home orders, lifting, partial reopening). Computed Spearman rank correlations between SDI and infection rates (cumulative cases per thousand and new daily cases per thousand), using weekday SDI values to avoid weekend systematic differences.

- SDI validity and temporal patterns: SDI sensitively reflected behavior changes, with higher SDI on weekends (especially Sundays) and a notable increase on Memorial Day (May 25) compared to typical Mondays. Following the U.S. national emergency declaration (March 13), SDI increased nationwide beginning March 16 on weekdays and subsequent weekends. The range of SDI widened after March 16, indicating heterogeneous state responses. After the week of March 23, SDI plateaued; from April 6 onward some states showed relaxation, extending nationwide around April 13, suggesting quarantine fatigue and economic pressures. - Government orders and reopening: Stay-at-home mandates triggered strengthened social distancing; early lifting or partial reopening accelerated relaxation. Five stages emerged: pre-pandemic (before Mar 13), behavior change (Mar 14–22), government orders and holding steady (Mar 23–Apr 12), quarantine fatigue (Apr 13–Apr 26), and partial reopening/order lifting (Apr 27 onward). - State-level differences: On May 29, top SDI regions were District of Columbia, Hawaii, New York, New Jersey, and Maryland (all issued stay-at-home orders). Lowest SDI were Wyoming, North Dakota, South Dakota, Arkansas, and Montana (most without statewide stay-at-home mandates). Coasts tended to maintain higher SDI, potentially due to longer exposure and higher density. - Effects around reopening: All examined states increased SDI after orders; however, SDI remained lower in the lowest case states. Sharp SDI declines followed partial reopening/order lifting in New York, Massachusetts, and Alaska; Alaska’s case growth accelerated roughly two weeks post-reopening. Similar low SDI with rising cases in California, Montana, Oregon, and West Virginia raised concerns for a second outbreak. - Correlations with infection rates (state level): Spearman correlations between SDI and new infection rates were stronger than with cumulative rates, indicating responsiveness to evolving outbreaks. Examples (cumulative vs new): New York 0.546 vs 0.645; Hawaii 0.643 vs 0.711; New Jersey 0.571 vs 0.655; Montana 0.495 vs 0.574; Alaska 0.506 vs 0.597; Oregon 0.532 vs 0.600; Massachusetts 0.549 vs 0.652; West Virginia 0.522 vs 0.611. - County-level patterns: New York counties (e.g., New York, Nassau, Suffolk, Westchester) exhibited high SDI and flattened cumulative case curves; Middlesex MA, Wayne MI, and Hudson NJ showed similar slowing. Relaxation was observed after partial reopening and order expirations. Los Angeles CA and Philadelphia PA had lower SDI relative to peers with continued rapid case growth. - Correlations with infection rates (county level): Generally stronger correlations in counties with higher SDI. Examples (cumulative vs new): New York County NY 0.589 vs 0.696; Westchester NY 0.573 vs 0.686; Cook IL 0.549 vs 0.644; Middlesex MA 0.563 vs 0.675; Wayne MI 0.575 vs 0.656; Hudson NJ 0.581 vs 0.680. Counties with smaller SDI–new infection correlations tended to show increasing cumulative cases toward the study end.

The SDI effectively captures real-world social distancing behaviors and their temporal and spatial variation, addressing the need for a comprehensive, multi-metric indicator beyond single measures like distance traveled. Findings demonstrate that both policy interventions (stay-at-home mandates) and local outbreak severity influence mobility reductions. The stronger association of SDI with new infection rates indicates that populations respond to contemporaneous outbreak dynamics. Observed relaxation after policy easing underscores the challenge of sustaining distancing and the potential for resurgence, especially in regions with low SDI and increasing cases. Policymakers can use SDI to monitor compliance, time interventions, and assess reopening readiness, while communities can gauge local risk and adjust behaviors. Integrating SDI into epidemiological models (e.g., compartmental frameworks) can improve forecasting by providing data-driven inputs on contact-relevant mobility.

This study develops and validates a score-based Social Distancing Index (SDI) using large-scale mobile device data to quantify mobility behavior changes during COVID-19 across U.S. states and counties. SDI reveals clear responses to national and state policies, heterogeneous regional behaviors, and a tendency toward relaxation after reopening. Its stronger correlation with new infection rates highlights its relevance for real-time situational awareness. The work provides a practical tool for policymakers and researchers to monitor mobility, evaluate interventions, and anticipate outbreak risks. Future research should refine SDI by incorporating additional mobility semantics (e.g., trip purposes), accounting for regional differences in behavior, enhancing weight calibration (e.g., expert elicitation), extending observation periods, and integrating SDI with epidemiological models to improve predictive performance.

- Potential bias in defining “staying at home” due to urban–rural differences (e.g., rural residents may need longer trips for essentials and be misclassified as not staying home). - Limited set of mobility metrics; absence of trip purpose differentiation (essential vs non-essential) may limit interpretability. - Lack of direct transmission/contact measures; aggregate mobility cannot capture contact intensity or precise exposure patterns. - Weight assignments in SDI rely partly on assumptions; although sensitivity analyses suggest stable trends, absolute scores vary with weights. - Temporal scope limited to Feb 2–May 30, 2020; longer observation could inform more robust weighting and trend assessment. - Weekend vs weekday behavioral differences require careful handling (analyses used weekdays for correlations).