Impacts of social distancing policies on mobility and COVID-19 case growth in the US

The study investigates how specific state-level social distancing policies in the United States affected population mobility and, in turn, COVID-19 case growth. Against the backdrop of a patchwork of state and local directives early in the pandemic, the authors aim to quantify which policies most effectively reduce mobility, a proxy for opportunities for viral transmission. The research objectives are to: (1) quantify the effect on mobility of state emergency declarations, social distancing policies, and shelter-in-place orders; (2) identify which policies are most effective in reducing aggregate mobility; and (3) estimate how changes in mobility relate to subsequent COVID-19 case growth. Understanding these relationships informs policy decisions that balance infection control with economic and social impacts.

Prior epidemics (e.g., 2009 H1N1 influenza and Ebola) demonstrate that social distancing reduces transmission. Early COVID-19 mitigation in China, including stringent quarantines, effectively reduced spread. In the U.S., several studies have linked social distancing policies and reduced mobility to slower COVID-19 growth: e.g., increases in home dwell time and reduced mobility correlated with longer doubling times and reduced case growth 2–3 weeks later, using datasets such as SafeGraph and anonymized cell tower data. However, the relative effectiveness of specific state-level policies remained unclear. This study adds quantitative estimates of how distinct policy waves—emergency declarations, initial social distancing orders, and shelter-in-place—affected multiple mobility metrics and subsequent case growth across a broad U.S. population of opted-in Google users.

Design: Observational analysis combining policy timing data, anonymized and aggregated mobility data from opted-in Google users (Location History), and county-level COVID-19 case data, with causal inference via within-county regression discontinuity around policy enactment and mixed-effects modeling for case growth. Policy data: Official state documents (governors, health/education), aggregated via Kaiser Family Foundation State Data and Policy Actions Tracker, supplemented with state public health websites; cross-checked with American Enterprise Institute COVID-19 Action Tracker and New York Times Shelter in Place Tracker. Policies categorized: (1) state emergency declaration; (2) state-mandated school closures; (3) closure of non-essential businesses/services (including gyms/theaters, not necessarily all non-essential businesses); (4) limits on large gatherings (threshold varied; first order used if multiple); (5) bans on in-restaurant service (excluding capacity limits; allowing only pick-up/delivery; often included bars/clubs); (6) mandatory quarantine (stay-at-home/shelter-in-place for all residents; excluding high-risk-only orders). Orders effective after 12:00 p.m. were assigned to the following day. Mobility data: Same anonymized, differentially private dataset as Google COVID-19 Community Mobility Reports. County-level (incl. DC) daily data from Jan 3–Mar 29, 2020. Changes computed relative to day-of-week baselines from Jan 3–Feb 6, 2020. Primary metric: relative change in average hours spent away from places of residence (estimated as 24 minus population-averaged hours at residence). Secondary metrics: relative changes in visits to workplaces, retail/recreation, grocery/pharmacies, parks, and transit stations. Regression discontinuity (mobility effects): For each county and policy type, estimate change comparing the week after implementation to the 7-day period 9–2 days prior, with a 2-day washout before implementation to account for pre-implementation messaging. Effects of policies enacted on or before Mar 23 were evaluable (data through Mar 29). County-level standard errors derived from the variance of weekly averages in Feb 1–28, 2020; SEs of relative changes via delta method. State-level estimates are population-weighted aggregates of county estimates; national estimates are simple averages of state estimates. Causal interpretation applies within locations; cross-location comparisons are descriptive only due to differences in data coverage and demographics of opted-in users, among other factors. Case data and modeling: County-level COVID-19 cases from Johns Hopkins Coronavirus Resource Center. Case growth defined as week-to-week difference in log new cases (per Courtemanche et al.). Linear mixed-effects models tested association between weekly mobility changes and subsequent changes in case growth, adjusting for weeks since the county’s 10th infection and an indicator for the first week with 10 new cases. Multilevel structure nests counties within states. Models considered lags of 2, 3, and 4 weeks for mobility effects on case growth. Missing data handling: Less populous counties may have missing category-specific visit data due to privacy thresholds; impact deemed negligible because (1) state estimates are population-weighted, and (2) correlations between combined category metrics and lower-coverage alternatives exceed 0.95.

- Policy waves: Three waves identified—(1) emergency declarations in early March 2020; (2) mid-March initial social distancing (SD) orders (school closures, limits on gatherings, non-essential business closures, bans on in-restaurant dining); (3) late March shelter-in-place (SIP) orders. - Emergency declarations: Associated with a 9.9% decrease in time away from residences (95% CI: -10.1%, -9.7%); 11.4% fewer workplace visits (95% CI: -11.8%, -11.0%); 11.5% fewer retail/recreation visits (95% CI: -11.7%, -11.2%); 9.3% fewer transit station visits (95% CI: -9.6%, -9.0%); 3.5% fewer park visits (95% CI: -4.4%, -2.6%); and an 8.2% increase in grocery/pharmacy visits (95% CI: 7.9%, 8.5%). - First social distancing order(s): Additional 24.5% reduction in time away from residences (95% CI: -24.7%, -24.3%); 33.0% reduction in retail/recreation visits (95% CI: -33.3%, -32.8%); 27.9% reduction in workplace visits (95% CI: -28.3%, -27.5%) in the following week. Substantial heterogeneity across states and counties; states enacting multiple SD measures experienced greater mobility reductions. - Policy combinations: Among single initial orders, bar and restaurant limits were most effective, associated with a 25.8% reduction in time away from residences (95% CI: -26.3%, -25.3%). School closures and/or large gathering bans alone yielded smaller reductions. Non-essential business closures also contributed to larger reductions. - Shelter-in-place orders: Among states with SIP on or before Mar 23, time away from residences fell by an additional 29.0% in the subsequent week (95% CI: -29.4%, -28.5%). These effects are multiplicative with prior emergency and SD effects. - Mobility and case growth: A 5% decrease in time away from residences was associated with 9.2% fewer new COVID-19 cases 2 weeks later (95% CI: -11.0%, -7.3%). A 10% decrease was associated with 17.5% fewer cases 2 weeks later (95% CI: -20.9%, -14.1%). Associations strengthened at 3–4 week lags.

Findings demonstrate that state-level social distancing policies substantially decreased mobility across U.S. counties and that reduced mobility was followed by lower COVID-19 case growth in subsequent weeks. While emergency declarations produced modest mobility reductions, the enactment of specific social distancing measures—especially limits on bars and restaurants and closures of non-essential businesses—led to much larger declines, and statewide shelter-in-place orders produced additional sizable reductions. The temporal relationship between decreased mobility and reduced case growth (notably at 2–4 week lags) aligns with prior studies and biological expectations regarding incubation and reporting delays. These results underscore the effectiveness of policy-driven social distancing in reducing opportunities for transmission and highlight aggregate mobility metrics as useful leading indicators for anticipating changes in COVID-19 risk.

Using anonymized, aggregated, and differentially private mobility data from opted-in Google users, the study quantifies how successive waves of state-level social distancing policies—emergency declarations, initial SD orders, and shelter-in-place—reduced time away from home and visits to key venues, and how these mobility reductions predicted declines in COVID-19 case growth. Limits on bars and restaurants emerged as the most impactful single initial policy for reducing mobility. The work provides actionable, quantitative estimates to inform public health policy and illustrates the potential of aggregate mobility data as an early indicator of transmission risk. Future research could disentangle the independent effects of closely timed policy measures, assess heterogeneity at local and sub-county levels, evaluate representativeness and potential biases in mobility datasets, and examine longer-term dynamics and reopening policies’ impacts on mobility and transmission.

- Mobility data represent only smartphone users who opted into Google Location History and are processed with differential privacy; representativeness may vary by location. - Cross-location comparisons (across counties/states) are descriptive and may be confounded by differences beyond policy environments (e.g., demographics of opted-in users, data coverage, local context). - Focus on state-level policies; heterogeneous within-state effects exist, and many local jurisdictions implemented measures before state orders, potentially attenuating observed state-level impacts. - Close temporal proximity of multiple policies limits ability to estimate independent effects of individual measures. - Some less populous counties have missing category-specific visit data due to privacy thresholds, though impact on aggregate estimates is assessed as minimal.