Risk attitudes and human mobility during the COVID-19 pandemic

The study investigates how regional risk attitudes shape behavioural responses to the COVID-19 pandemic, using mobility changes as outcomes. The core hypothesis is that risk-averse populations reduce mobility more, particularly around key information shocks such as the WHO’s pandemic declaration, and when opportunity costs of staying home are higher (weekends). The work is motivated by the idea that perceived risk and stable risk preferences influence precautionary behaviour more than calculated, age-adjusted health risks. The authors also examine whether the presence of more elderly (65+) in a region—who face higher health risks—moderates the relationship between risk attitudes and mobility reductions.

The paper situates its inquiry within literature on risk attitudes, risk perception, and behavioural responses to epidemics and disasters. Prior research shows that risk aversion leads to overweighting risks while risk seeking may underweight them, and that risk preferences can be relatively stable yet context dependent. Studies on epidemics (HIV/AIDS, epidemic modeling) highlight how perceptions and awareness shape spread and prevention, while disaster literature documents changes in risk-taking post-event. Cross-country work shows geographic variation in risk preferences. Framing effects, emotions, bounded rationality, and heuristics also shape risk judgments. In pandemics, precautionary behaviours such as social distancing occur in the face of uncertainty and are likely tied to underlying risk preferences and socio-cultural context.

Data combine: (1) Google COVID-19 Community Mobility Reports (six categories: Retail & Recreation, Grocery & Pharmacy, Parks, Transit Stations, Workplaces, Residential) for 132 countries (regional data for 51) from 15 Feb to 9 May 2020, expressed as percent change vs the median same weekday during 3 Jan–6 Feb 2020; (2) Risk preferences from the Global Preferences Survey (Gallup World Poll, 2012), measuring willingness to take risks via a combined self-assessment (11-point scale) and incentivized staircase lottery choices, aggregated to country and subnational regions (using regional values when available, otherwise country values); (3) COVID-19 cases/deaths (ECDC) and policy indicators (OxCGRT) including school, workplace, public transport closures, public events cancellations, stay-at-home, and restrictions on gatherings/internal movement (recoded dichotomously, with nationwide vs targeted robustness checks); (4) Controls: demographics (population density, urban share, % aged 65+), household size, unemployment rate, log GDP per capita, daily average temperature (GHCN-Daily, stations within 50 km of region centroid), and weekend definitions per country. The merged analytical sample includes 58 countries and 776 subnational regions in 33 countries with 58,284 to 67,073 region-day observations (depending on outcome availability). Empirical strategy: Random-effects linear models of daily mobility change on risk preference, including interactions with WHO pandemic declaration (11 March 2020), weekend indicator, and % aged 65+, controlling for daily cases per 1,000, days since first COVID-19 death (truncated at 0 before first death), demographics, economics, temperature, and policy dummies. Standard errors are clustered at the smallest geographic unit. Robustness: exclusion of regions with censored mobility values at varying thresholds; alternative codings; checks reported in SI Tables S7–S11. Data and code: OSF (https://osf.io/7bxqp/).

• Overall mobility fell across non-residential categories, especially early in the sample; park visits often increased. General risk-mobility relation over full period: risk-taking is positively associated with visits to Retail & Recreation (β=2.873, s.e.=1.180, 95% CI [0.561, 5.185], P=0.015) and Parks (β=7.667, s.e.=2.577, 95% CI [2.616, 12.718], P=0.003), with no significant relation for Grocery & Pharmacy (β=-0.481, P=0.650), Transit Stations (β=1.352, P=0.317), Workplaces (β=0.306, P=0.718), or Residential (β=-0.241, P=0.519).
• WHO pandemic declaration effects (level shifts): Additional reductions after 11 March 2020 of about 11.3 pp in Retail & Recreation (β=-11.328, s.e.=0.879, P<0.001), 7.3 pp in Parks (β=-7.303, s.e.=1.379, P<0.001), 12.0 pp in Transit (β=-11.998, s.e.=0.833, P<0.001), 8.1 pp in Workplaces (β=-8.103, s.e.=0.642, P<0.001); Residential increased by 3.6 pp (β=3.602, s.e.=0.285, P<0.001). Grocery/Pharmacy not significant in main model but significant in robustness checks.
• Moderation by risk after declaration: Interaction between risk and declaration is positive, indicating smaller pre-post differences in higher risk-tolerant areas (i.e., less responsiveness). Significant interactions for all except Residential: Retail & Recreation (β=6.715, s.e.=1.166, P<0.001), Grocery & Pharmacy (β=5.983, s.e.=1.013, P<0.001), Parks (β=11.910, s.e.=2.449, P<0.001), Transit (β=7.168, s.e.=1.422, P<0.001), Workplaces (β=4.020, s.e.=0.871, P<0.001).
• Weekends vs weekdays: Compared to weekdays, weekends show further average reductions vs baseline in Retail & Recreation (-4.27 pp), Grocery/Pharmacy (-3.92 pp), Parks (-4.39 pp), Transit (-0.72 pp); Workplaces reductions are larger on weekdays (+8.34 pp reduction relative to weekends), and Residential increases are larger on weekdays (weekend effect β=-3.346, s.e.=0.109, P<0.001). Risk mediates weekend effects: regions with lower risk tolerance reduce mobility more on weekends than weekdays relative to risk-tolerant regions (significant positive interactions for all non-residential and Workplaces; negative for Residential). Triple interactions show mediation strengthens post-declaration (e.g., Retail & Recreation β=5.036, s.e.=0.707, P<0.001; Grocery/Pharmacy β=4.273, s.e.=0.698, P<0.001; Parks β=5.989, s.e.=1.532, P<0.001; Transit β=4.697, s.e.=0.884, P<0.001; Workplaces β=4.008, s.e.=0.665, P<0.001; Residential β=-1.397, s.e.=0.290, P<0.001).
• Actual risk (share aged 65+): Higher elderly share associates with larger cutbacks in Grocery/Pharmacy (β=-0.597, s.e.=0.164, P<0.001), Transit (β=-0.364, s.e.=0.153, P=0.018), Workplaces (β=-0.447, s.e.=0.096, P<0.001), and a small decrease in Residential stay (β=-0.128, s.e.=0.049, P=0.009). Robustness suggests negative effects also for Retail & Recreation and Parks. Significant interactions: with more risk-loving populations and higher elderly share, Retail & Recreation reduced more (β=-0.388, s.e.=0.169, P=0.022); regions with fewer risk-takers and larger older populations increased staying home more (Residential interaction β=-0.183, s.e.=0.050, P<0.001).
• Controls behaved as expected: severity (cases per capita, declaration) and policy measures reduce out-of-home mobility and increase residential time; higher population density strengthens declines in Grocery/Pharmacy, Transit, Workplaces.

Findings show that risk attitudes meaningfully shape behavioural responses to COVID-19 beyond actual calculated health risks. Risk-averse regions curtailed mobility more, especially following salient information shocks (WHO pandemic declaration) and when weekend opportunity costs would usually raise out-of-home activity. Risk-tolerant regions were less responsive to environmental changes, maintaining higher mobility and showing smaller shifts pre- vs post-declaration. The differential responses imply that public health communication and policies must consider underlying regional risk cultures. The WHO declaration acted as a strong global signal, inducing additional reductions in mobility independent of government lockdowns, suggesting informational shocks can shift behaviour rapidly. The elderly share modestly intensified reductions in certain categories, and its interplay with risk preferences suggests communal risk considerations matter. Weekend-weekday convergence post-declaration reflects widespread economic and social reconfiguration (lockdowns, job losses), reducing typical weekly rhythms. Overall, aligning messages with risk perceptions and leveraging salient announcements may enhance precautionary behaviors and virus containment.

The paper demonstrates that regional risk attitudes are key predictors of mobility reductions during COVID-19, with risk-averse regions more rapidly and strongly adopting precautionary behaviors, particularly after the WHO’s pandemic declaration and when weekend opportunity costs are higher. The declaration itself induced a fundamental global shift in mobility independent of formal lockdowns. These insights underscore the importance of incorporating risk preference heterogeneity into epidemic response design and communication strategies. Future research should analyze individual-level data to unpack demographic and psychological moderators (e.g., age, gender, locus of control, overconfidence), examine how media and information dissemination shape behaviour across risk cultures, and study post-lockdown habit formation and persistence of behavioural changes.

Key limitations include: analysis at regional rather than individual level, precluding fine-grained examination of individual heterogeneity (age, gender, affective reactions, locus of control) and disentangling risk preferences from perceived vs actual risks; lack of detailed information on absolute baseline social mobility levels prior to the pandemic (only relative baseline used), which may mask heterogeneous losses and subsequent behavioural adjustments; potential unobserved psychological factors (e.g., overconfidence) and adaptation/habituation over time; and reliance on aggregated, anonymized mobility data with some censored values (addressed via robustness checks but still a constraint).