Abstract
Why do people perceive risk differently during the same public health crisis? Using longitudinal, nationally representative panel data from the Understanding America Study spanning two years of the COVID-19 pandemic in the United States (March 2020--February 2022), we show risk perception can be a result of both social experience and social construction. Specifically, we examine how risk exposure within close social networks and political party affiliation jointly shape perceived risks of infection, hospitalization, and death. Results indicate that each additional social network member’s infection, hospitalization, or death predicts higher perceived risk of the corresponding outcome (all p ≤ .01). Republicans report lower perceived risks than Democrats on average, but the association between social network infections and perceived infection risk is steeper among Republicans (p < .05), which suggests that risk exposure and political identity also interact in creating risk perception. Party-stratified models corroborate that within-person increases in network exposures are followed by higher perceived risks. Phase-specific analyses reveal shifting salience of network experiences and partisan identity across the pandemic. Risk exposure in social networks and political identity (partisanship) appear to operate jointly in shaping health risk perceptions, highlighting the need for tailored risk communication strategies to support effective preparedness for future pandemics.
Keywords: Perceived Risk, Social Networks, Political Identity, Pandemic Preparedness
INTRODUCTION
COVID-19 caused more than 1.2 million deaths in the United States (U.S.) alone (CDC, 2026). Despite policies to encourage protective health behaviors, adoption was uneven, and the pandemic exacted a devastating toll (CDC, 2024, 2026). Understanding Americans’ risk perceptions is critical given their role in shaping health behavior during public health crises. In this paper, we examine two central inputs to risk perception under uncertainty: social network exposures (information) and political affiliation (identity). Using repeated measures from the Understanding America Study spanning March 2020–February 2022, we test whether increases in network exposures are followed by changes in perceived risk of infection, hospitalization, and death, and whether these associations vary by political party.
Theories of Health Behavior and Risk Perception
Health behavior frameworks such as the Health Belief Model (HBM) provide a lens for interpreting perceived susceptibility and severity as meaningful components of health threat appraisal (Janz & Becker, 1984; Rosenstock, 1974; Sugg Skinner et al., 2015). In this study, susceptibility is operationalized as perceived risk of infection and severity as perceived risk of hospitalization or death (Botzen et al., 2022; Dryhurst et al., 2020; Gachabayov et al., 2022; Stone, 2014). Both dimensions consistently predict engagement in protective health behaviors (Carpenter, 2010; Sulat et al., 2018). We focus on two factors expected to influence these risk perceptions during the COVID-19 pandemic: informational exposure through close social networks and identity-based influences associated with political affiliation.
Two well-established mechanisms shape risk perceptions during public health crises: availability-by-recall, whereby proximate experiences increase the cognitive salience of risk, and identity-based motivated reasoning, whereby political affiliation structures how information is interpreted under uncertainty. We draw on these complementary mechanisms to examine how informational exposure through close social networks and partisan identity jointly relate to perceived COVID-19 risks over time.
Theories of risk perception and empirical evidence indicate that informational cues and social identities shape threat appraisal (Gachabayov et al., 2022; Kiviniemi et al., 2022; Stroebe et al., 2021). Social network exposures, such as infection, hospitalization, or death among friends and family, may elevate perceived risk by increasing the cognitive salience of threat through the availability heuristic (Pachur et al., 2012; Tversky & Kahneman, 1974). Although affective processes such as empathy or emotional contagion may also matter, our measures of infection, hospitalization, and death most directly capture informational exposure. Political affiliation may also reflect embeddedness in like-minded informational environments characterized by shared media consumption and interpretive frames.
Risk perceptions also depend on familiarity with disease and trust in information sources, both of which influence the ease with which threats come to mind (Botzen et al., 2022; Dryhurst et al., 2020; Gachabayov et al., 2022; Stone, 2014). Classic experimental work demonstrates that knowing someone who experienced a health outcome strongly predicts perceived risk frequency, consistent with availability-by-recall (Lichtenstein et al., 1978). Applied to COVID-19, higher prevalence of infection, hospitalization, or death within one’s close social network should therefore predict higher perceived risk of corresponding outcomes.
Empirical evidence supports this expectation. Studies across multiple countries have found that direct experience with COVID-19 exposure through close ties are positively associated with perceived susceptibility and severity (Botzen et al., 2022; Dryhurst et al., 2020; Gachabayov et al., 2022).
Political Affiliation and Risk Perception
Political affiliation is especially consequential for risk perception in highly politicized contexts, as it structures trust in information sources and motivates selective acceptance of congruent cues while discounting incongruent signals. The politicization of COVID-19 in the U.S. has contributed to known, preventable morbidity and mortality (Bolsen & Palm, 2022; Wallace et al., 2022). Politicization, trust in, and exposure to different information sources has been linked to COVID-19 risk perception and engagement in risk mitigation behaviors (Bolsen & Palm, 2022; Entradas, 2022; Gadarian et al., 2021; Reynolds et al., 2022; Siegrist et al., 2021; Wong & Jensen, 2020). Political affiliation more reliably predicted COVID-19 attitudes and behaviors than did county-level deaths (Gadarian et al., 2021). Republicans and conservative news consumers were less likely than their counterparts to social distance, wear masks, or support vaccination (Barrios & Hochberg, 2020; Simonov et al., 2020), and the more strongly a person identified as Republican, the lower their perceived risk of COVID-19 and the less likely they were to adopt risk-mitigating behaviors (Kiviniemi et al., 2022). Vaccination messaging from in-group party leadership was trusted by unvaccinated Republicans more than identical messaging from out-group leadership (Pink et al., 2021). These differences in perceptions and behaviors translated into real population health consequences. The politicization of COVID-19 ultimately manifested in mortality disparities between Republicans and Democrats, with an excess COVID-19 death rate estimated to be 76% higher among Republicans than among Democrats, including the period after widespread vaccine availability during which this disparity widened from 22% to 153% (Wallace et al., 2022).
Collectively, this prior research suggests that political affiliation is associated with COVID-19 attitudes, behaviors, and risk perceptions with considerable disparities in health outcomes (Botzen et al., 2022; Dryhurst et al., 2020; Gachabayov et al., 2022). However, less is known about how partisan identity operates alongside personally experienced information from close social networks, or whether the relative influence of these informational and identity-based processes varies over time. Existing studies have also relied predominantly on cross-sectional designs, which limits our understanding of how social network exposures and risk perceptions evolve during a prolonged public health crisis like the COVID-19 pandemic.
Longitudinal approaches are needed to provide a fuller picture of how these effects unfold over time. Known social network infections could reasonably be expected to increase over time as national infection rates increase. Perceived severity, or perceived risk of hospitalization or death due to the virus, might also increase alongside hospitalization and mortality rates. Alternatively, perceived hospitalization or death risks could taper with time as the virus stabilizes and becomes less lethal, as was observed with the omicron variant compared to delta (Lorenzo-Redondo et al., 2022; Sigal et al., 2022). Perceived death risk might also decrease over time with improvement in hospital resourcing and symptom management and witnessing higher recovery rates within one’s social network. Beyond these epidemiological and experiential influences, political affiliation may have affected how individuals interpreted changing pandemic conditions over time.
An unanswered question is how politicization shaped perceived risk of COVID-19 over an extended period of time. A recent longitudinal study found that political conservatism was inversely related to risk-mitigating health behavior over five weeks during the spring 2020, promptly following national lockdowns in March; this relationship was mediated by perceived risk of infection and perceived severity. Effects were stronger among U.S. respondents than others, which may underscore the powerful role of U.S. politicization on Americans’ risk perception and health behaviors (Stroebe et al., 2021). Understanding these dynamics over a longer period of the pandemic may be helpful to informing effective outbreak response strategies in the future.
Present Study and Contributions
Our contribution is to use a large, nationally representative panel with repeated observations over two years to test whether changes in social network exposures predict changes in perceived risk of infection, hospitalization, and death, and whether these associations vary by political affiliation. We evaluate main and interactive effects of informational (network) and identity (partisan) predictors on risk perception. Further, because the pandemic unfolded in distinct socio-political phases, phase-specific analyses examine whether these associations varied across early lockdowns, the politicized pre-vaccine period, and the vaccine rollout under new federal leadership. Conceptually, we contribute to the social network and health literature by shifting attention from network structure to networked experience, positioning close ties as a key informational context through which health risks are interpreted under uncertainty. Our first hypothesis was:
H1: Consistent with theories of risk perception, greater known social network COVID-19 experiences (infections, hospitalizations, and deaths) will be positively associated with risk perceptions of each corresponding outcome.
We also analyze the role of political affiliation on perceived risk of infection, hospitalization, and death over time. Political identity shapes how individuals process information under conditions of uncertainty. Motivated reasoning theory suggests that individuals evaluate risks in line with their partisan commitments, especially when issues are politicized (Kunda, 1990; Taber & Lodge, 2006). During COVID-19, Republicans exhibited lower trust in scientific and governmental institutions and greater reliance on partisan media, resulting in systematically lower risk perceptions than Democrats (Bolsen & Palm, 2022; Wallace et al., 2022). Therefore, our hypothesis to examine associations between political affiliation and risk perceptions was:
H2: Consistent with research on partisan-motivated reasoning and trust in institutions during COVID-19, Republicans will report, on average, lower perceived risk of infection, hospitalization, and death over time compared to Democrats.
In today’s fragmented media environment, partisan homophily and selective exposure foster motivated reasoning, making people more receptive to congenial cues and more skeptical of out-group messages (Iyengar & Massey, 2019). Experimental evidence also shows strong in-group cueing among Republicans: unvaccinated Republicans increased intentions to vaccinate after Republican-elite endorsements and reacted negatively to Democratic endorsements (Pink et al., 2021). While broad, population-wide attitude shifts from partisan media are often limited and concentrated among the most engaged (Prior, 2013). The overall pattern suggests that, where institutional trust is lower and in-group cues dominate, Republicans will, on average, down-weight institutional signals of danger and report lower perceived COVID-19 risks than Democrats. In testing for interactions between political affiliation and social network exposures on risk perceptions over time, we hypothesize:
H3: Political affiliation will modify the effect of social network exposures on perceived risk of infection, hospitalization, and death. We expect network exposures to be more influential among Republicans, reflecting selective trust in personal information and partisan polarization in institutional trust.
Additional phase-specific analyses examine whether the relative salience of informational and identity-based predictors varies across major phases of the pandemic. Pandemic phases were selected to coincide with major shifts in the epidemiologic, policy, and informational context of COVID-19 in the U.S. Phase 1 (March through May 2020) corresponds to the initial outbreak period, characterized by widespread stay-at-home orders, school and business closures, and substantial uncertainty regarding disease transmission and severity. Phase 2 (June through December 2020) captures a reopening and pre-vaccine period marked by increasing political polarization surrounding COVID-19 mitigation measures and divergent public health messaging. Phase 3 (January 2021 onward) began with a transition in federal leadership (President Biden’s inauguration), new CDC leadership (Dr. Rochelle Walensky appointed director), and increasing vaccine availability) (CDC, 2024). Therefore, we hypothesize:
H4: Associations between social network exposures, political affiliation, and perceived risk will vary across phases of the pandemic as the epidemiologic and socio-political context evolved.
Data and Methods
Study Sample
Data were collected as part of the Understanding America Study (UAS), a nationally representative panel study of adults living in the U.S., launched in 2014 (Kapteyn et al., 2020). Probability sampling was used to recruit from the sampling frame of U.S. postal addresses.
Invitations were mailed to recruit participants. UAS surveys were self-administered via computer, tablet, or smartphone. Tablets and Internet access were provided to those who did not already have access. During March 2020, the COVID-19 Longitudinal Survey was initiated as part of UAS. COVID-related survey data were first collected between March 10th through the 31st (Wave 1) and repeated in subsequent biweekly waves beginning April 1st (Wave 2) and recurring through February 2022, the last wave included in present analyses.
Measures
Primary outcomes include perceived risk of COVID-19 infection, hospitalization, and death. These were assessed at every wave using a visual linear scale from 0-100%. Respondents were encouraged to give their best guess. Primary predictors include known COVID-19 infections, hospitalizations, and deaths among one’s social network of friends and close family. These three predictors were assessed following an item about social network size. Appendix A provides additional measurement details.
Our known social network infection, hospitalization, and death counts are best interpreted as informational exposures that increase the cognitive availability of COVID-19 risk within egocentric networks. Although affective pathways (e.g., empathy, emotional contagion) may co-occur, our measures most directly capture availability-by-recall rather than affect. We operationalized exposures as raw counts of infected, hospitalized, or deceased close ties rather than as proportions to reflect both survey design and theoretical alignment. The UAS denominator item asked respondents to count only living close ties, while a numerator item included deaths, so ratios would bias rates. Recall error affects both numerators and denominators, but ratios compound these errors and unstable denominators can distort associations (Brewer, 2000; Marin, 2004; Marsden, 1990). Counts thus serve as indices of salient, recalled experiences that align with the theorized availability-by-recall mechanism (Lichtenstein et al., 1978; Pachur et al., 2012).
Another key predictor is political party affiliation. We recoded survey responses so that the variable had four levels, with 1 = Democrats, 2 = Republicans, and 3 = Independent, and 4 = Other political affiliation. The variable was measured at only one time point.
Additional variables and covariates were collected at each wave and included in the regression analyses. We adjusted for demographics plausibly related to both perceived and actual risk and to network composition: age (baseline vulnerability), gender (documented risk perception differences), race/ethnicity (structural exposure and informational channels), education and household income (health knowledge/resources), employment type (occupational exposure), immigrant and U.S. citizenship status (information access and trust), disability status (health vulnerability), and marital status (household network exposure).
Analysis
Statistical analyses were performed in Stata, version 17 (StataCorp, 2021). Descriptive analyses were conducted to check recoding, variable frequencies, and distributions. Primary analyses used random-effects panel models with robust standard errors clustered by respondent. All models included state and month fixed effects and individual random effects. The rationale for using random-effects models is that one of the key predictors, political party affiliation, was measured only once and therefore could not be estimated in individual fixed-effects models. Alpha was set to .05 for all significance testing.
Models 1-3a tested H1 and H2 by estimating associations between known social network exposures (COVID-19 infections, hospitalizations, and deaths), political affiliation, and perceived risk of infection, hospitalization, and death. These models adjusted for age, gender, race/ethnicity, education level, household income, employment type, immigrant status, U.S. citizenship, disability status, and marital status. Models 1-3b tested H3 by adding interaction terms between social network exposures and political affiliation. Phase-specific models tested H4 by estimating the same associations within the three pandemic phases described above.
The dataset is an unbalanced panel with respondents contributing varying numbers of observations due to dropout and replenishment. Missingness was handled with listwise deletion. Hospitalizations and deaths were not fielded in waves 1-2, which represents the primary source of missingness in those models. After fielding, per-wave item nonresponse was generally ≤10%, and missingness was comparable across partisan groups. Model-wise, 84.7% of rows were available for infection-risk models versus approximately 20% for hospitalization- and death-risk models.
Three approaches were used to check for robustness. First, we estimated lagged outcome models, which account for prior risk perceptions to help isolate effects of social network exposures and changes in risk perceptions over time. Because lagged outcomes absorb substantial within-person variation, these models generally provide conservative estimates of covariate effects (Achen, 2000; Angrist & Pischke, 2009). The second approach was to estimate individual fixed-effects models that absorb all stable respondent characteristics. Because fixed-effects models eliminate time-invariant predictors, we could not estimate the main effect of political affiliation. Instead, we stratified the fixed-effects models by party (Democrat, Republican, Independent, and Other) and tested whether within-person changes in network exposures predicted within-person changes in risk perceptions, adjusting for time-varying covariates and month indicators. Third, because the primary random-effects models (xtreg) do not accommodate the UAS post-stratification weights, we estimated weighted generalized estimating equations (xtgee) models using respondent-averaged post-stratification weights to assess whether the primary findings were robust to population weighting. We also evaluated the linearity assumption for social network exposure variables by re-estimating the random-effects models using binary indicators of any exposure and quadratic terms for social network infections, hospitalizations, and deaths.
RESULTS
Descriptive Statistics
The analytic sample consisted of 8,983 respondents with 171,571 observations for Model 1a, 8,856 respondents with 159,794 observations for Model 2a, and 8,854 respondents with 159,765 observations for Model 3a. Participants were predominantly White (78%), female (59%), and averaged 51.47 years of age (Table 1).
Table 1.
Sample Characteristics
| Variable | n | % |
|---|---|---|
| Age | ||
| 18-34 years | 33,853 | 17.49 |
| 35-54 years | 72,545 | 37.48 |
| 55-64 years | 39,949 | 20.64 |
| 65+ years | 47,224 | 24.40 |
| Gender | ||
| Female | 113,758 | 58.71 |
| Male | 79,999 | 41.29 |
| Race | ||
| White | 150,758 | 78.39 |
| Asian | 10,530 | 5.48 |
| Black | 14,875 | 7.73 |
| Hispanic (non-white) | 6,518 | 3.39 |
| American Indian or Alaska Native | 1,638 | 0.85 |
| Mixed | 8,003 | 4.16 |
| Education | ||
| High school or below | 40,814 | 21.07 |
| Some college | 70,656 | 36.48 |
| Bachelor degree | 48,090 | 24.83 |
| Graduate degree | 34,145 | 17.63 |
| Household Income | ||
| Below $25,000 | 36,817 | 19.05 |
| $25,000 to $59,999 | 57,097 | 29.54 |
| $60,000 to $99,999 | 48,401 | 25.04 |
| Above $150,000 | 50,999 | 26.38 |
| Employment | ||
| Not working/Unknown | 82,904 | 42.79 |
| Government | 21,572 | 11.13 |
| Private | 60,282 | 31.11 |
| Non-profit | 15,333 | 7.91 |
| Self-employed | 13,667 | 7.05 |
| Political Affiliation | ||
| Democrat | 101,969 | 40.26 |
| Republican | 81,898 | 32.34 |
| Independent | 59,637 | 23.55 |
| Other | 9,759 | 3.85 |
| Immigrant | ||
| Non-immigrant | 103,940 | 53.64 |
| First, second, or third generation | 84,582 | 43.65 |
| Unknown | 5,236 | 2.70 |
| Citizenship | ||
| U.S. citizen | 187,692 | 96.90 |
| Non-U.S. citizen | 6,010 | 3.10 |
| Disabled | ||
| Yes | 16,129 | 8.33 |
| No | 177,520 | 91.67 |
| Marital Status | ||
| Married (live together) | 106,109 | 54.79 |
| Married (live separately) | 2,285 | 1.18 |
| Separated | 3,263 | 1.68 |
| Divorced | 28,506 | 14.72 |
| Widowed | 9,780 | 5.05 |
| Never married | 43,718 | 22.57 |
Frequencies and percentages are reported across all observations of the panel study where participant n = 9,423.
On key variables pertaining to the research question, 40.26% of respondents were Democrats, 32.34% Republicans, 23.55% Independents, and the remainder indicated Other political affiliation. The mean number of known social network members who had been infected was 3.03, hospitalized 0.48, and died 0.21. The mean perceived risk of COVID-19 infection was 20.88%, hospitalization 20.76%, and death 16.33%.
Descriptive analyses indicated modest but significant partisan differences in reported network infections. Democrats reported approximately 2.9 infected close others on average, compared to 3.35 for Republicans, with Independents in between (Kruskal-Wallis p < .001). Partisan differences in hospitalizations and deaths were small and not statistically significant. Figures that visualize these partisan differences over time are in Appendix C (Figures C1-C3). These descriptive patterns may provide relevant context for interpreting partisan moderation of network effects.
Main Effects
Significant main effects from the multiple linear regressions to predict risk perceptions of infection, hospitalization, and death support hypothesis H1 (Table 2, Models 1-3a). Model 1a revealed that with adjustment for state and month fixed effects and individual random effects on covariates, there was a significant main effect of the number of known social network members who were infected on perceived risk of infection on average over time; i.e., for every additional close friend or family member infected over time, perceived risk of infection increased, on average, by 0.032 units (SE 0.007, p < .001). Political affiliation was also a significant predictor of perceived risk of infection, such that compared to Democrats, Republicans and Independents had significantly lower perceived risk of COVID-19 infection by 1.239 units (SE 0.389, p < .01) and 0.912 units (SE 0.356, p < .05) respectively. Other significant predictors of perceived risk of infection over time included age, gender, race, education, household income, employment type, immigrant status, and disability. Model 1a explained a statistically significant proportion of the variance in infection risk perception over time, R2 = 0.053, W(102) = 4,689.75, p < .001. Rho was equal to 0.510, which means that approximately half of the variance in perceived risk of infection was due to differences across panels.
Table 2.
Predicting Perceived Risk of COVID-19 Infection, Hospitalization, and Death
| Model 1a: Perceived Risk of Infection |
Model 2a: Perceived Risk of Hospitalization |
Model 3a: Perceived Risk of Death |
||||
|---|---|---|---|---|---|---|
| Variables | Est. | SE | Est. | SE | Est. | SE |
| Social network infections | 0.032 | 0.007*** | ||||
| Social network hospitalizations | 0.105 | 0.032** | ||||
| Social network deaths | 0.177 | 0.054** | ||||
| Political affiliation (Democrat) | ||||||
| Republican | −1.239 | 0.389** | −1.380 | 0.476** | −0.846 | 0.428* |
| Independent | −0.912 | 0.356* | −0.893 | 0.396* | −0.532 | 0.350 |
| Other political affiliation | −1.210 | 0.668 | −0.277 | 0.624 | −0.131 | 0.567 |
| Age (18-34 years) | ||||||
| 35-54 years | −1.287 | 0.520* | 2.015 | 0.540*** | 1.501 | 0.469** |
| 55-64 years | −2.696 | 0.628*** | 5.810 | 0.720*** | 4.740 | 0.690*** |
| 65+ years | −3.903 | 0.645*** | 7.412 | 0.815*** | 5.330 | 0.825*** |
| Gender (male) | −2.188 | 0.357*** | −1.329 | 0.464** | −1.973 | 0.440*** |
| Race (white) | ||||||
| Asian | −0.958 | 0.832 | 0.142 | 0.986 | 0.609 | 0.907 |
| Black | −3.160 | 0.722*** | −0.485 | 0.922 | 0.171 | 0.866 |
| Hispanic (non-white) | 0.418 | 1.056 | 0.519 | 1.552 | 1.673 | 1.469 |
| American Indian or | ||||||
| Alaska Native | −4.732 | 1.861* | 0.582 | 2.951 | 2.401 | 2.853 |
| Mixed | −0.950 | 0.826 | 0.229 | 1.427 | 0.534 | 1.182 |
| Education (high school or below) | ||||||
| Some college | −0.769 | 0.554 | −2.782 | 0.840** | −3.223 | 0.689*** |
| Bachelor degree | −2.290 | 0.598*** | −6.765 | 0.791*** | −7.719 | 0.715*** |
| Graduate degree | −2.380 | 0.642*** | −7.376 | 0.843** | −8.453 | 0.786*** |
| Household income (below $25,000) | ||||||
| $25,000 to $59,000 | −0.502 | 0.423 | −0.542 | 0.458 | −0.828 | 0.406* |
| $60,000 to $99,000 | −1.449 | 0.491** | −1.844 | 0.544** | −1.447 | 0.484** |
| Above $150,000 | −1.708 | 0.546** | −1.553 | 0.615* | −1.282 | 0.530* |
| Employment type (not working/unknown) | ||||||
| Government | 1.645 | 0.614** | −0.293 | 0.622 | −0.300 | 0.555 |
| Private | 0.660 | 0.360 | −0.777 | 0.376* | −0.727 | 0.341* |
| Non-profit | 1.626 | 0.681* | −0.125 | 0.846 | −0.437 | 0.864 |
| Self-employed | −1.204 | 0.547* | −0.692 | 0.577 | −0.678 | 0.536 |
| Immigrant origin (no) | ||||||
| 1-3rd generation | 0.970 | 0.405* | 1.757 | 0.554** | 1.944 | 0.561** |
| Unknown status | 1.723 | 0.979 | 1.586 | 1.208 | 1.897 | 1.296** |
| U.S. citizen | 1.071 | 1.020 | 0.501 | 1.224 | 0.384 | 1.319 |
| Disabled | 1.945 | 0.771* | 7.266 | 1.224*** | 6.222 | 0.952*** |
| Marital status (married, live together) | ||||||
| Married, live separately | 0.097 | 1.434 | 2.097 | 1.283 | 0.489 | 1.121 |
| Separated | −0.494 | 1.258 | 2.647 | 1.582 | 0.542 | 1.444 |
| Divorced | 0.373 | 0.566 | 0.919 | 0.746 | 0.312 | 0.664 |
| Widowed | 0.678 | 0.811 | −0.163 | 1.283 | 1.158 | 1.692 |
| Never married | 0.732 | 0.501 | 0.745 | 0.606 | 0.137 | 0.559 |
| Constant | 25.044 | 2.066*** | 17.371 | 2.915*** | 14.479 | 2.458*** |
| n (Observations) | 171,571 | 159,794 | 159,765 | |||
| n (Respondents) | 8,983 | 8,856 | 8,854 | |||
Source: Understanding America Study (UAS) COVID-19 National Sample Longitudinal File, 2020-2021. Note: Model controls for state and month fixed effects and individual random effects. The final month was omitted from Models 2a and 3a due to collinearity.
p < .05;
p < .01;
p < .001. SE = robust standard error.
Model 2a results are also presented in Table 2, which show predicted risk perception of hospitalization over time. While adjusting for state and month fixed effects and individual random effects on covariates, there was a significant main effect of the number of known social network members who were hospitalized due to COVID-19 on perceived risk of hospitalization on average over time; i.e., for every additional close friend or family member who was hospitalized with COVID-19, perceived risk of hospitalization increased, on average, by 0.105 units (SE 0.032, p < .01) over time. Political affiliation was a significant predictor of perceived risk of COVID-related hospitalization, where compared to Democrats, Republicans and Independents had significantly lower perceived risk of hospitalization by 1.380 units (SE 0.476, p < .01) and 0.893 units (SE 0.396, p < .05) respectively. Other significant predictors of perceived risk of hospitalization over time were age, gender, education, household income, employment type, immigrant status, and disability. Model 2a explained a statistically significant proportion of the variance in hospitalization risk perception over time, R2 = 0.121, W(101) = 4,495.31, p < .001. Rho was equal to 0.690, which means that nearly 70% of the variance in perceived risk of hospitalization was due to differences across panels.
Table 2 also shows Model 3a results predicting perceived risk of death due to COVID-19 over time. With adjustment for state and month fixed effects and individual random effects on shown covariates, there was a significant main effect of the number of known social network members who had died due to COVID-19 on perceived risk of death due to the virus on average over time; i.e., for every additional close friend or family member who died, perceived risk of death increased, on average, by 0.177 units (SE 0.054, p < .01) over time. Political affiliation was a significant predictor of perceived risk of COVID-related death, where compared to Democrats, Republicans reported significantly lower perceived risk of death due to COVID-19 by 0.846 units (SE 0.428, p < .05). Other significant predictors of perceived risk of death over time included age, gender, education, household income, employment type, immigrant status, and disability. Model 3a explained a statistically significant proportion of the variance in death risk perception over time, R2 = 0.115, W(101) = 2,687.18, p < .001. Since rho was 0.742, approximately three quarters of the variance in perceived risk of COVID-related death was due to differences across panels.
Results from robustness analyses support findings from the main analyses. Social network exposures retained positive predictive associations in all three lagged outcome models (Table D1). Fixed-effects models stratified by political affiliation produced similar findings (Table D2). Across partisan groups, within-person increases in network exposures were generally associated with higher perceived risk. Weighted GEE sensitivity analyses produced comparable results for both main-effect and interaction models (Tables D3-D4), with the primary associations between social network exposures and corresponding risk perceptions remaining positive and statistically significant across all outcomes.
Additional sensitivity analyses examining alternative exposure specifications also yielded substantively similar conclusions. Binary indicators of any known infection and hospitalization remained positively associated with corresponding risk perceptions. Quadratic specifications suggested model nonlinearities, with evidence of diminishing marginal effects for infections and hospitalizations and a modest accelerating effect for deaths at higher levels of exposure. However, across specifications, greater social network COVID-19 exposure was generally associated with higher corresponding risk perceptions, and the overall interpretation of the primary findings remained unchanged.
Interaction Effects
Results from regression analyses including political affiliation interaction terms are shown in Table 3. Models 1-3b included the same predictors as Models 1-3a plus an interaction term to explore effect modification by political affiliation. In addition to the significant main effects previously described, Model 1b revealed a significant interaction term for Republicans, such that the positive effect of known social network infections on perceived risk of infection was elevated by 0.036 units (SE 0.014) among Republicans, p < .05. Knowing a close friend or family member who had been infected with COVID-19 was more impactful on infection risk perception among Republicans compared to Democrats. In contrast, the effect was significantly attenuated among those of Other political affiliation. There was no significant interaction effect for Independents. Predicted margins plots illustrating these interaction effects are presented in Appendix B (Figures B1-B3).
Table 3.
Interaction Effects between Social Network COVID-19 Exposures and Political Affiliation
| Variable | Est. | SE |
|---|---|---|
| Model 1b: Perceived risk of infection | ||
| Social network infections | 0.030 | 0.007*** |
| Political affiliation (Democrat) | ||
| Republican | −1.369 | 0.392*** |
| Independent | −0.952 | 0.359** |
| Other political affiliation | −1.083 | 0.670 |
| Social network infections * Democrat | ||
| Social network infections * Republican | 0.036 | 0.014* |
| Social network infections * Independent | 0.009 | 0.020 |
| Social network infections * Other political affiliation | −0.030 | 0.013* |
| Model 2b: Perceived risk of hospitalization | ||
| Social network hospitalizations | 0.092 | 0.042* |
| Political affiliation (Democrat) | ||
| Republican | −1.443 | 0.476** |
| Independent | −0.905 | 0.398* |
| Other political affiliation | −0.156 | 0.630 |
| Social network hospitalizations * Democrat | ||
| Social network hospitalizations * Republican | 0.010 | 0.053 |
| Social network hospitalizations * Independent | 0.010 | 0.071 |
| Social network hospitalizations * Other political affiliation | −0.313 | 0.164 |
| Model 3b: Perceived risk of death | ||
| Social network deaths | 0.209 | 0.087* |
| Political affiliation (Democrat) | ||
| Republican | −0.855 | 0.429* |
| Independent | −0.517 | 0.352 |
| Other political affiliation | −0.107 | 0.569 |
| Social network deaths * Democrat | ||
| Social network deaths * Republican | 0.043 | 0.098 |
| Social network deaths * Independent | −0.086 | 0.100 |
| Social network deaths * Other political affiliation | −0.272 | 0.107* |
Source: Understanding America Study (UAS) COVID-19 National Sample Longitudinal File, 2020-2021. Note: Model controls for state and month fixed effects and individual random effects, including all covariates from Models 1-3a. The final month was omitted from Models 2b and 3b due to collinearity.
p < .05;
p < .01;
p < .001. SE = robust standard error.
Model 2b yielded a similar pattern of main effects to those observed in Model 1b. However, there were no significant interactions between social network hospitalizations and political affiliations; although interaction effect directions were similar to Model 1b for each term, interaction terms for Republican and Other political affiliation were non-significant.
Model 3b retained the significant negative main effect from Model 3a of Republican affiliation on perceived risk of death. Results from Model 3b also revealed a significant interaction effect for those of Other political affiliation; i.e., known social network deaths had significantly less effect on death risk perception among this group of respondents compared to Democrats. In contrast to Model 1b, there were no significant interaction effects for Republicans in this model’s results.
Phases of the Pandemic
The pandemic context changed substantially over time, with shifts in disease prevalence, public health guidance, vaccine availability, and political leadership. We therefore examined whether associations differed across phases of the pandemic (Appendix C). The association between social network infections and perceived risk of infection was strongest during Phase 1 (March through May 2020) and decreased though remained statistically significant across subsequent phases (June 2020 onward). In contrast, associations between social network hospitalizations and deaths and their corresponding risk perceptions were most pronounced during Phase 2 (June 2020 through December 2020) and Phase 3 (January 2021 and onward).
Political affiliation was also more strongly associated with perceived risk during earlier phases of the pandemic and generally weakened over time, with few significant differences remaining by Phase 3.
DISCUSSION
From a social networks and health perspective, our findings illustrate the role of close ties as carriers of experiential information that recalibrates health risk perceptions over time. We extend sociological accounts of risk by showing how networked information (availability-by-recall) and partisan identity (motivated reasoning under polarization) jointly structure perceived risk during a prolonged public health crisis. This complements prior work on network clustering and diffusion by demonstrating how network experiences shape individual cognition even in the absence of observable behavioral contagion.
Results from Models 1-3a and 1-3b supported our confirmatory hypotheses, while interaction and phase-specific analyses provided insight into how these processes operate jointly over time. First, known social network COVID-19 infections, hospitalizations, and deaths each positively predicted corresponding risk perceptions over time (Models 1-3a), consistent with theory and prior evidence on the availability heuristic and risk perceptions.
Political affiliation was also a significant predictor of perceived risk. Republicans and Independents reported lower average perceived risks over time compared to Democrats. These main effects were more consistent than the interactions observed in Models 1-3b, suggesting that political affiliation may have influenced baseline perceptions of COVID-19 more strongly than responsiveness to social network experiences. Partisan identity may shape perceived health threat through differences in informational environments and institutional trust, whereas personal experiences within one’s social network may represent a more direct source of information that affects risk perceptions across political groups, albeit not uniformly. As COVID-19 became more politicized in news media and other communication outlets in the months following March 2020, political partisanship emerged as a risk factor for COVID-related outcomes, including engagement in risk-mitigating health behaviors like social distancing, quarantining, and getting vaccinated (Allcott et al., 2020; Barrios & Hochberg, 2020; Bolsen & Palm, 2022; Gadarian et al., 2021; Wallace et al., 2022). Notably, the association between known network infections and perceived infection risk was stronger among Republicans (Model 1b); that is, knowing more people who had been infected with the virus increased risk perception more steeply for Republicans compared to Democrats. This pattern aligns with motivated reasoning in a polarized media environment: Republicans, on average, report lower trust in governmental or scientific institutions and greater receptivity to in-group cues, making personally proximate, networked information especially consequential for updating perceived risk (Iyengar & Massey, 2019; Pink et al., 2021; Stein et al., 2021). These interactions should indicate caution for future pandemic preparations. We have presented evidence that information and identity, in the form of social network exposures and political affiliation, shape risk perceptions during a pandemic with great uncertainty and devastating consequences.
Phase-specific analyses provided insight into how information and identity influences change across pandemic phases. During initial national lockdowns of Phase 1, infections among social networks influenced risk perception of infection, but not hospitalization or death; this may be due to lack of information within social networks at the time. The pandemic was only just beginning. However, when experiential information was lacking, political affiliation significantly predicted risk perceptions during this early phase. Interaction effects became more prominent during Phase 2, as social network exposures accumulated and political parties adopted more established perspectives during uncertain times. By Phase 3, more network hospitalizations and deaths had occurred, which significantly predicted corresponding risk perceptions, and the effects of political affiliations waned. This pattern is consistent with evidence that direct experience and information transmitted through close social ties are important determinants of risk perception (Dryhurst et al., 2020) and with availability-by-recall accounts of risk judgment, which posit that risks become more salient when individuals can recall relevant experiences from their social networks (Pachur et al., 2012). Our results suggest that identity may become less potent as information increases, and the relative weight of proximate experiential information and partisan identity is not fixed; it can shift as epidemiologic conditions and political context evolve. This study clarifies how established informational and identity-based processes unfold over an extended crisis using repeated measures from a large national panel and provides evidence that can inform more adaptive risk communication and preparedness efforts.
Although the estimated effects were small in absolute magnitude, they were consistent across outcomes, model specifications, and sensitivity analyses, suggesting that social network experiences were associated with risk perceptions even after accounting for political affiliation and numerous demographic characteristics. Rather than producing large shifts in perceived risk, social network exposures appear to function as one of many informational cues contributing to risk assessment during a prolonged and highly politicized public health crisis.
Limitations
This study is not without limitations. First, although the longitudinal design facilitates temporal ordering and reduces concerns about reverse causation, the observational nature of the study precludes definitive causal conclusions. Both social network exposures and perceived risk are self-reported, raising the possibility of correlated measurement error or common underlying traits. For example, stable dispositions (e.g., pessimism, risk aversion) or broader ideological and worldview differences could influence both the reporting of social network experiences and perceptions of health risk. While our models adjust for observed covariates, and robustness checks support consistency, we cannot rule out such alternative explanations. Future work should incorporate independent measures of infection and objective health risks where available. We were also unable to estimate individual fixed-effects models because political affiliation was measured only once. While our random-effects and lagged outcome models yielded consistent results, the possibility of unobserved heterogeneity remains. Further, there was missing data across waves. Although missingness was generally modest after variables were fielded, attrition and item nonresponse may have introduced bias. Retained respondents were younger and more educated at baseline, though income, employment type, citizenship, and disability status did not differ. Additionally, measurement of social network predictors could have been improved. Rather than ask respondents to report a number, the item could have prompted respondents to list the names of individuals who had been infected or died due to COVID-19 for more accurate estimates.
We also analyzed counts of infected, hospitalized, or deceased ties rather than proportions standardized by network size. This choice may have introduced skew and greater variance, since larger networks naturally allow higher counts, potentially obscuring relative differences in risk of network exposures. We also relied on political affiliation as our measure of group identity. The UAS survey did not include a repeated measure of political ideology (e.g., liberal-conservative self-placement), which limits our ability to assess whether ideology would have produced similar moderation patterns. While correlated, partisanship and ideology can diverge in meaningful ways, and future research should incorporate repeated measures of both constructs to disentangle their independent and joint roles. Finally, we explored political affiliation but did not probe specific reasons for observed differences between parties. Future studies could directly examine trust in news, media, and institutions and exposure to different types of information sources to better understand the role of political affiliation in risk perception.
Conclusions
We found evidence that known social network COVID-19 infections, hospitalizations, and deaths predict corresponding risk perceptions over time. Experiences within close social networks appear to serve as salient informational cues that shape risk perceptions, consistent with the availability-by-recall mechanism of risk perception. From a Health Belief Model perspective, these changes in perceived susceptibility and severity may ultimately affect engagement in preventive health behaviors.
These and future findings will be essential to developing a more effective pandemic response with specific consideration of information via availability heuristics among different identity groups and resulting risk perceptions across partisan lines. Lessons can be learned from understanding risk perceptions based on knowledge of network exposures with implications for mitigating future pandemics. Findings from this study may inform population education and dissemination of health information, including the roles of social network information and political identity in shaping risk perception amid viral outbreaks with uncertainty.
Supplementary Material
Highlights.
Longitudinal panel data linked social network exposures to perceived COVID-19 risk
Known infections, hospitalizations, and deaths predicted higher perceived risk
Republicans reported lower perceived COVID-19 risks than Democrats
Political affiliation modified some network exposure-risk relationships
Social network infections had stronger effects among Republicans
An ethics approval was not required because the study uses fully anonymized public data and does not directly engage with human subjects.
Acknowledgements:
This research is funded by an NSF grant (2125530) for Weihua An, “SCC-IRG JST: Hyperlocal Risk Monitoring and Pandemic Preparedness through Privacy-Enhanced Mobility and Social Interactions Analysis,” and the NIH/NIDA training grants (T32DA037202 and T32DA050552) for Caroline Barry. The results and opinions expressed therein represent those of the authors and do not necessarily reflect those of NSF, NIH, or NIDA.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Declaration of Interest Statement:
The authors declare no conflicts of interest related to this research.
Declaration of AI and AI-assisted technologies in the manuscript preparation process
During the preparation of this work the authors used ChatGPT (OpenAI) to assist in organizing and structuring reviewer comments. The tool was not used to generate scientific content, analysis, or interpretations. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
Contributor Information
Caroline M. Barry, Rollins School of Public Health, Emory University
Weihua An, Department of Sociology and Department of Data and Decision Sciences, Emory University.
Data Availability Statement
The project described in this paper relies on data from survey(s) administered by the Understanding America Study, which is maintained by the Center for Economic and Social Research (CESR) at the University of Southern California. The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of USC or UAS. The collection of the UAS COVID-19 tracking data is supported in part by the Bill & Melinda Gates Foundation and by grant U01AG054580 from the National Institute on Aging, and many others. The data may be requested by following directions on this webpage https://uasdata.usc.edu/index.php. Computer code for producing the empirical results of the paper will be shared on reasonable request to the corresponding author.
References
- Achen CH. (2000). Why lagged dependent variables can suppress the explanatory power of other independent variables. 20(22), 7–2000. [Google Scholar]
- Allcott H, Boxell L, Conway J, Gentzkow M, Thaler M, & Yang D. (2020). Polarization and public health: Partisan differences in social distancing during the coronavirus pandemic. Journal of Public Economics, 191, 104254. 10.1016/j.jpubeco.2020.104254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Angrist JD, & Pischke J-S. (2009). Mostly harmless econometrics: An empiricist’s companion. Princeton university press. [Google Scholar]
- Barrios JM, & Hochberg Y. (2020). Risk perception through the lens of politics in the time of the COVID-19 pandemic. National Bureau of Economic Research. [Google Scholar]
- Bolsen T, & Palm R. (2022). Politicization and COVID-19 vaccine resistance in the U.S. Progress in Molecular Biology and Translational Science, 188(1), 81–100. 10.1016/bs.pmbts.2021.10.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Botzen WJW, Duijndam SJ, Robinson PJ, & van Beukering P. (2022). Behavioral biases and heuristics in perceptions of COVID-19 risks and prevention decisions. Risk Analysis, n/a(n/a). 10.1111/risa.13882 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Brewer DD. (2000). Forgetting in the recall-based elicitation of personal and social networks. Social Networks, 22(1), 29–43. 10.1016/S0378-8733(99)00017-9 [DOI] [Google Scholar]
- Carpenter CJ. (2010). A meta-analysis of the effectiveness of health belief model variables in predicting behavior. Health Communication, 25(8), 661–669. 10.1080/10410236.2010.521906 [DOI] [PubMed] [Google Scholar]
- CDC. (2024, July 8). CDC Museum COVID-19 Timeline. Centers for Disease Control and Prevention. https://www.cdc.gov/museum/timeline/covid19.html [Google Scholar]
- CDC. (2026, June 5). Surveillance and Data Analytics. Covid. https://www.cdc.gov/covid/php/surveillance/index.html [Google Scholar]
- Dryhurst S, Schneider CR, Kerr J, Freeman AL, Recchia G, Van Der Bles AM, Spiegelhalter D, & Van Der Linden S. (2020). Risk perceptions of COVID-19 around the world. Journal of Risk Research, 23(7–8), 994–1006. [Google Scholar]
- Entradas M. (2022). In Science We Trust: The Effects of Information Sources on COVID-19 Risk Perceptions. Health Communication, 37(14), 1715–1723. 10.1080/10410236.2021.1914915 [DOI] [PubMed] [Google Scholar]
- Gachabayov M, Sharun K, Felsenreich DM, Nainu F, Anwar S, Yufika A, Ophinni Y, Yamada C, Fahriani M, Husnah M, Raad R, Khiri NM, Abdalla RY, Adam RY, Ismaeil MI, Ismail AY, Kacem W, Teyeb Z, Aloui K, … Harapan H. (2022). Perceived risk of infection and death from COVID-19 among community members of low- and middle-income countries: A cross-sectional study (11:345). F1000Research. 10.12688/f1000research.109575.2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gadarian SK, Goodman SW, & Pepinsky TB. (2021). Partisanship, health behavior, and policy attitudes in the early stages of the COVID-19 pandemic. PLoS ONE, 16(4), e0249596. 10.1371/journal.pone.0249596 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Iyengar S, & Massey DS. (2019). Scientific communication in a post-truth society. Proceedings of the National Academy of Sciences, 116(16), 7656–7661. 10.1073/pnas.1805868115 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Janz NK, & Becker MH. (1984). The health belief model: A decade later. Health Education Quarterly, 11(1), 1–47. [DOI] [PubMed] [Google Scholar]
- Kapteyn A, Angrisani M, Bennett D, Bruin W. B. de, Darling J, Gutsche T, Liu Y, Meijer E, Perez-Arce F, Schaner S, Thomas K, & Weerman B. (2020). Tracking the Effect of the COVID-19 Pandemic on the Lives of American Households. Survey Research Methods, 14(2), Article 2. 10.18148/srm/2020.v14i2.7737 [DOI] [Google Scholar]
- Kiviniemi MT, Orom H, Hay JL, & Waters EA. (2022). Prevention is political: Political party affiliation predicts perceived risk and prevention behaviors for COVID-19. BMC Public Health, 22(1), 298. 10.1186/s12889-022-12649-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kunda Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. 10.1037/0033-2909.108.3.480 [DOI] [PubMed] [Google Scholar]
- Lichtenstein S, Slovic P, Fischhoff B, Layman M, & Combs B. (1978). Judged frequency of lethal events. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 551. [PubMed] [Google Scholar]
- Lorenzo-Redondo R, Ozer EA, & Hultquist JF. (2022). Covid-19: Is omicron less lethal than delta? BMJ, 378, o1806. 10.1136/bmj.o1806 [DOI] [PubMed] [Google Scholar]
- Marin A. (2004). Are respondents more likely to list alters with certain characteristics?: Implications for name generator data. Social Networks, 26(4), 289–307. 10.1016/j.socnet.2004.06.001 [DOI] [Google Scholar]
- Marsden PV. (1990). Network data and measurement. Annual Review of Sociology, 16(1), 435–463. [Google Scholar]
- Morgan SL, & Winship C. (2014). Counterfactuals and causal inference: Methods and principles for social research. Cambridge University Press. [Google Scholar]
- Pachur T, Hertwig R, & Steinmann F. (2012). How Do People Judge Risks: Availability Heuristic, Affect Heuristic, or Both? Journal of Experimental Psychology. Applied, 18, 314–330. 10.1037/a0028279 [DOI] [PubMed] [Google Scholar]
- Pink SL, Chu J, Druckman JN, Rand DG, & Willer R. (2021). Elite party cues increase vaccination intentions among Republicans. Proceedings of the National Academy of Sciences, 118(32), e2106559118. 10.1073/pnas.2106559118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Prior M. (2013). Media and Political Polarization. Annual Review of Political Science, 16(Volume 16, 2013), 101–127. 10.1146/annurev-polisci-100711-135242 [DOI] [Google Scholar]
- Reynolds RM, Weaver SR, Nyman AL, & Eriksen MP. (2022). Trust in COVID-19 information sources and perceived risk among smokers: A nationally representative survey. PLOS ONE, 17(1), e0262097. 10.1371/journal.pone.0262097 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rosenstock IM. (1974). Historical origins of the health belief model. Health Education Monographs, 2(4), 328–335. [DOI] [PubMed] [Google Scholar]
- Siegrist M, Luchsinger L, & Bearth A. (2021). The Impact of Trust and Risk Perception on the Acceptance of Measures to Reduce COVID-19 Cases. Risk Analysis, 41(5), 787–800. 10.1111/risa.13675 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sigal A, Milo R, & Jassat W. (2022). Estimating disease severity of Omicron and Delta SARS-CoV-2 infections. Nature Reviews Immunology, 22(5), Article 5. 10.1038/s41577-022-00720-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Simonov A, Sacher SK, Dubé J-PH, & Biswas S. (2020). The persuasive effect of fox news: Non-compliance with social distancing during the covid-19 pandemic. National Bureau of Economic Research. [Google Scholar]
- StataCorp. (2021). Stata Statistical Software: Release 17 [Computer software]. StataCorp LLC. [Google Scholar]
- Stein R, Swan AB, & Sarraf M. (2021). Hearing From Both Sides: Differences Between Liberal and Conservative Attitudes Toward Scientific and Experiential Evidence. Political Psychology, 42(3), 443–461. 10.1111/pops.12706 [DOI] [Google Scholar]
- Stone D. (2014). Risk Perception. In Wexler P (Ed.), Encyclopedia of Toxicology (Third Edition) (p. 184). Academic Press. 10.1016/B978-0-12-386454-3.00981-7 [DOI] [Google Scholar]
- Stroebe W, vanDellen MR, Abakoumkin G, E. P. L. Jr, Schiavone WM, Agostini M, Bélanger JJ, Gützkow B, Kreienkamp J, Reitsema AM, Khaiyom JHA, Ahmedi V, Akkas H, Almenara CA, Atta M, Bagci SC, Basel S, Kida EB, Bernardo ABI, … Leander NP. (2021). Politicization of COVID-19 health-protective behaviors in the United States: Longitudinal and cross-national evidence. PLOS ONE, 16(10), e0256740. 10.1371/journal.pone.0256740 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sugg Skinner C, Tiro J, & Champion VL. (2015). Chapter 5: The Health Belief Model. In Glanz K, Rimer BK, & Viswanath K (Eds.), Health Behavior (pp. 75–94). [Google Scholar]
- Sulat JS, Prabandari YS, Sanusi R, Hapsari ED, & Santoso B. (2018). The validity of health belief model variables in predicting behavioral change: A scoping review. Health Education, 118(6), 499–512. 10.1108/HE-05-2018-0027 [DOI] [Google Scholar]
- Taber CS, & Lodge M. (2006). Motivated Skepticism in the Evaluation of Political Beliefs. American Journal of Political Science, 50(3), 755–769. 10.1111/j.1540-5907.2006.00214.x [DOI] [Google Scholar]
- Tversky A, & Kahneman D. (1974). Judgment under Uncertainty: Heuristics and Biases: Biases in judgments reveal some heuristics of thinking under uncertainty. Science, 185(4157), 1124–1131. [DOI] [PubMed] [Google Scholar]
- Wallace J, Goldsmith-Pinkham P, & Schwartz JL. (2022). Excess Death Rates for Republicans and Democrats During the COVID-19 Pandemic (Working Paper No. 30512). National Bureau of Economic Research. 10.3386/w30512 [DOI] [Google Scholar]
- Wong CML, & Jensen O. (2020). The paradox of trust: Perceived risk and public compliance during the COVID-19 pandemic in Singapore. Journal of Risk Research, 23(7–8), 1021–1030. 10.1080/13669877.2020.1756386 [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The project described in this paper relies on data from survey(s) administered by the Understanding America Study, which is maintained by the Center for Economic and Social Research (CESR) at the University of Southern California. The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of USC or UAS. The collection of the UAS COVID-19 tracking data is supported in part by the Bill & Melinda Gates Foundation and by grant U01AG054580 from the National Institute on Aging, and many others. The data may be requested by following directions on this webpage https://uasdata.usc.edu/index.php. Computer code for producing the empirical results of the paper will be shared on reasonable request to the corresponding author.