People underestimate the probability of contracting the coronavirus from friends

The study investigates how perceptions of risk from different social ties influence intentions to follow physical distancing guidelines during COVID-19. Given that SARS-CoV-2 is predominantly transmitted through close contact, the authors posit that people may systematically underestimate the risk posed by socially close others (friends, family) compared to socially distant others (strangers, colleagues). The central hypothesis is that perceived probability of contracting COVID-19 from close versus distant others predicts intentions to physically distance from them. Specifically, H1: The perceived probability of contracting the coronavirus will mediate the extent to which people intend to physically distance from socially close (vs. distant) others. The paper tests these ideas across five studies, including large, representative samples from the US and Canada.

The theory section situates the research within work on physical distancing as an effective non-pharmaceutical intervention and on predictors of compliance, including political orientation and personal values. It emphasizes risk perception as a key determinant of public health behavior, comprising perceived threat, vulnerability, and consequences. Prior literature in the behavioral immune system suggests familiarity reduces perceived risk; people prefer contact with familiar over unfamiliar others, repeated exposure lowers perceived risk, unfamiliar targets and objects are judged riskier, and in-group members are perceived as less risky. Trust also reduces perceived risk judgments. Drawing on this literature, the authors propose that people will judge friends as less likely to spread the coronavirus than strangers and will be less inclined to physically distance from them. The review motivates a mediation pathway in which lower perceived infection probability from close others reduces distancing intentions.

Five studies (total N = 3395) were conducted. All studies obtained informed consent; several were preregistered. - Study 1A (US; N = 754) and Study 1B (Canada; N = 840) used online panels (Qualtrics and Lucid) to assess the perceived relative likelihood of contracting COVID-19 from a friend, family member, acquaintance, colleague, and stranger on a 0–100 scale (0 = far below average; 50 = average; 100 = far above average). Participants who had tested positive for COVID-19 were excluded. Covariates included work hours, perceived COVID-19 threat, self-reported staying at home/visiting others, knowing someone infected, and demographics. Linear mixed models compared target categories. - Study 2 (N = 296; MTurk) employed a scenario-based within-subjects design holding exposure constant: participants imagined interacting in a supermarket with a friend, family member, colleague, cashier, and stranger, then rated relative infection likelihood (0–100). They also estimated expected conversation time (0–60 min). A separate posttest (N = 57) validated perceived closeness (1–11), confirming friend/family as closer and cashier/stranger as less close. - Study 3 (N = 301; MTurk; preregistered) used a between-subjects scenario in a supermarket, controlling for mask use, distance (~1 m), interaction length (~15 min), and unknown vaccination/prior infection status. Participants indicated, assuming infection, whether a friend or a stranger likely infected them (binary choice). Additional items assessed components of risk (perceived likelihood of contraction, symptoms, severity, overall threat; 1–10). A posttest (N = 55) confirmed the closeness manipulation (friend closer than stranger). - Study 4 (N = 402; MTurk) used a between-subjects dinner invitation scenario. Participants generated a close friend’s name (with constraints to ensure unknown COVID-related behaviors) and imagined receiving an invitation from either that friend (friend condition) or a stranger (stranger condition) while controlling for mask and distance. Measures included perceived infection status of the other, worry the other has COVID-19, worry about contracting COVID-19 from the interaction, symptom expectations and severity if infected, overall risk (1–10), and intention to join the dinner (1–7). Alternative explanations were measured: fear of offending, missing the person, fear of missing out, general interest in the dinner, and perceived compliance with distancing/masking. Regression and mediation analyses (with and without controls) tested predictors of overall risk and intentions. - Study 5 (N = 802; MTurk; preregistered) used a park interaction scenario and a behavioral-intention proxy: an online social distancing task that recorded the pixel distance participants chose to stand from the target. Conditions included colleague vs friend, and friend conditions with interventions designed to reduce the friend-risk bias (e.g., reminders about distancing and risks posed by friends). Participants rated perceived risk using a multi-item scale (1–6), then completed the distancing task. Manipulation checks confirmed closeness across conditions. ANOVAs and mediation analyses tested effects of condition on risk and distancing intentions, and whether interventions increased intended physical distance by raising perceived risk. Across studies, analyses included linear mixed models, t-tests, ANOVAs, regressions, exact binomial tests, and bootstrapped mediation models.

- Across studies, participants judged close others (friends, often family) as less likely sources of COVID-19 infection than socially distant others (strangers, cashiers, colleagues), and this bias predicted reduced physical distancing intentions toward close others. - Study 1A (US) and 1B (Canada): Relative infection likelihood ratings placed friends lowest and strangers highest. Example means: US sample—friend 42.37, family 45.95, colleague 42.48, acquaintance 43.27, stranger 49.65; Canada sample—friend 37.27, family 41.77, colleague 38.58, acquaintance 39.06, stranger 49.62 (scale centered at 50 = average). Effects held controlling for covariates. - Study 2 (supermarket scenario): Holding exposure constant, close others were rated below average risk, and strangers/cashier above average. Example means: friend 4.91, family 39.66, colleague 50.06, stranger 55.33, cashier 59.45 (0–100 scale; friend and family lower than average; cashier highest). - Study 3 (between-subjects, supermarket): When forced to attribute infection, 77% chose the stranger and 23% the friend (exact binomial test p < 0.0001), indicating a strong bias to see strangers as the infectious source. - Study 4 (dinner invitation): Participants perceived the stranger as posing higher overall risk than the friend; they were more willing to join a friend for dinner than a stranger. Risk component analyses showed that the difference in perceived risk was explained primarily by higher perceived probability that the stranger (vs friend) had COVID-19; given infection, perceived transmission risk was similar. Regression showed overall risk was predicted by perceived infection likelihood and expected symptoms/severity. Mediation analyses indicated perceived threat mediated the effect of condition (friend vs stranger) on intention to join dinner, even controlling for alternative explanations (interest, missing the person, fear of missing out, fear of offending, and perceived guideline adherence). - Study 5 (park interaction and distancing task): Without intervention, participants perceived interacting with a friend as less risky than with a colleague and intended to keep less distance from the friend (ANOVAs significant). Interventions aimed at highlighting friends as potential sources of infection increased intended physical distance toward friends; mediation indicated the interventions operated by increasing perceived risk. Friend-focused messaging was effective but not more effective than a generic distancing reminder. - Overall: Across five studies (N = 3395), the robust pattern is that people underestimate the probability of contracting COVID-19 from close others compared to socially distant others, and this underestimation reduces physical distancing intentions. Informing people about this bias attenuates the effect and increases distancing intentions.

The findings support the hypothesis that perceived infection probability mediates distancing intentions across social ties: people see friends as less likely to be infected and thus less risky, leading to reduced intentions to physically distance. This bias persists even when exposure, interaction context, and precautionary behaviors (masking, distance, duration) are held constant. The work clarifies that the key driver is not necessarily perceived transmissibility given infection, but rather the perceived likelihood that close others are infected at all. This cognitive bias has practical implications: because people interact more frequently and for longer periods with friends, underestimating their risk can facilitate viral spread. Interventions that directly target this misperception can increase intended physical distancing toward friends. The results contribute to behavioral science on risk perception, in-group favoritism, and trust by showing how social closeness reduces perceived disease risk and affects compliance with public health recommendations. The research informs communication strategies: emphasizing that close others can pose similar risk as strangers can promote safer behavior during pandemics.

This paper demonstrates that individuals systematically underestimate the risk of contracting COVID-19 from close others, especially friends, relative to socially distant others, and that this misperception undermines physical distancing intentions. Across five studies, including large US and Canadian samples, the effect is robust to controls and across scenarios. Mechanistically, people judge friends as less likely to be infected, which explains lower perceived risk from them; this belief translates into a greater willingness to engage socially with friends (e.g., join a dinner) and to stand closer to them. Interventions that highlight the friend-risk bias or broadly remind people to distance can effectively increase intended physical distancing toward friends. Future research should test these effects on real-world behavior and transmission outcomes, validate behavioral distancing tasks, examine cross-cultural generalizability, and optimize targeted risk communications that correct underestimation of risk from close others without inducing undue anxiety or reactance.

- Sampling: Several studies relied on Amazon Mechanical Turk samples, which may skew younger and more educated than the general population, limiting generalizability. - Measures: Predominantly self-reported perceived risk and intentions rather than observed behavior; Study 5 used a behavioral-intention proxy task that was not externally validated. - Relative risk framing: Early studies asked risk relative to an average person, which could be misinterpreted (e.g., as multiple strangers) and inflate perceived risk from strangers; later studies addressed this but residual interpretive issues may remain. - Design constraints: Some studies used within-subjects assessments vulnerable to order effects or fatigue; subsequent between-subjects designs mitigated but do not eliminate all demand characteristics. - Context specificity: Scenarios (supermarket, dinner, park) may not capture all real-world contexts or evolving pandemic conditions (variants, vaccination rates, mandates).