key: cord-0729622-oviloku5
authors: de Bruin, Wändi Bruine; Bennett, Daniel
title: Relationships Between Initial COVID-19 Risk Perceptions and Protective Health Behaviors: A National Survey
date: 2020-05-22
journal: Am J Prev Med
DOI: 10.1016/j.amepre.2020.05.001
sha: 58efdc6574fe1923c9111e5377aa5bdbcab83f23
doc_id: 729622
cord_uid: oviloku5

INTRODUCTION: Perceptions of health risks inform decisions about protective behaviors, but coronavirus disease 2019 (COVID-19) was an unfamiliar risk as it began to spread across the U.S. In the initial stage of the epidemic, authors examined perceived risks for COVID-19 infection and infection fatality, and whether these risk perceptions are associated with protective behaviors. They also examined whether findings differed between later versus earlier responders. METHODS: Between March 10 and 31, 2020, investigators conducted a cross-sectional online survey with a nationally representative U.S. sample (N=6,684). Half responded before March 13, 2020 (versus later). Participants assessed their risks of COVID-19 infection and infection fatality (0%–100%), and were transformed into quartiles (1–4). They reported their implementation of protective behaviors like handwashing and social distancing (yes/no). Analyses were conducted in April‒May 2020. RESULTS: Median perceived risk was 10.00% for COVID-19 infection and 5.00% for infection fatality, but respondents showed large disagreement. An increase of one quartile in perceived infection risk was associated with being 1.45 (95% CI=1.33, 1.58) more likely to report handwashing, with perceived infection fatality risk showing no significant association. When predicting social distancing behaviors such as avoiding crowds, both quartile-based risk perceptions were significant (OR=1.24, 95% CI=1.17, 1.30 for infection and OR=1.19, 95% CI=1.13, 1.26 for infection fatality). Perceived COVID-19 infection risk, protective behaviors, and their relationship increased among later (versus earlier) responders. CONCLUSIONS: Despite disagreements about the risks, people perceiving greater risks were more likely to implement protective behaviors—especially later (versus earlier) in March 2020. These findings have implications for risk communication.

As COVID-19 spreads across the U.S., people have faced a new and unfamiliar health threat about which information is still limited and changing. Though the outlook remains uncertain, COVID-19 has already caused far greater morbidity and mortality than other human coronaviruses such as Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS). 1 To limit disease transmission, the Centers of Disease Control and Prevention recommended protective behaviors such as hand hygiene and social distancing. 2 Mass adoption of these behaviors is especially important when pharmacological interventions are not yet available. 3 According to theories of decisions about health behavior, people who perceive greater risks are more motivated to implement protective behaviors. [4] [5] [6] Hence, assessing associations between risk perceptions and protective behaviors has practical and theoretical relevance.

Links between perceived risks and protective behaviors have traditionally been studied for familiar risks like seasonal influenza. 7 With emerging diseases like COVID-19, objective risk information is typically scarce, characterized by uncertainty, and subject to change. It is unclear how people perceive the risks, or whether their initial risk perceptions inform their decisions about protective actions. 8 Therefore, the study objective is to examine people's initial risk perceptions for COVID-19 infection and infection fatality, as well as associations of these initial risk perceptions with self-reported protective behaviors. Because the survey was completed between mid-to-late March 2020, the authors were able to add exploratory analyses to examine how risk perceptions, protective behaviors, and their associations varied as the initial stage of the epidemic unfolded.

Data collection was approved by the University of Southern California's IRB, as part of the Understanding America Study (UAS). Since 2014, the UAS has been recruiting U.S. adults aged ≥18 years by mailing invitations to randomly selected U.S. addresses. If needed, interested individuals received Internet access and a computer or tablet. UAS members are regularly invited to participate in online surveys, and receive about $20 per 30 minutes of survey time. Between

March 10 and March 31, 2020 an online survey was conducted with 6,684 of 8,489 invited UAS participants (response rate=79%). 9 Half completed the survey before (versus on or after) March 13, 2020 (Figure 1 ), when the White House issued a national emergency, the ban on European travel went into effect at midnight EDT, and several states announced school closures and bans of large gatherings. [10] [11] [12] Table 1 shows how the current sample compares to invited UAS participants who did not complete the survey, and to the U.S. population.

To assess experiences with COVID-19, participants were first asked: Has a doctor or another healthcare professional diagnosed you with the coronavirus 

Analyses were conducted with SPSS, version 26 in April-May 2020. Post-stratification weights, generated through a raking algorithm, were used in all analyses to align the sample to the U.S. adult population, in terms of distributions for sex, race/ethnicity, age, education, and geographic location (more information is available at uasdata.usc.edu/page/Weights).

Before conducting the main analyses, the authors examined the percentage of participants who had been diagnosed with COVID-19, and the percentage who thought they had it. Descriptive statistics were computed, including medians and means (95% CIs) of respondents' perceived COVID-19 infection and COVID-19 infection fatality risk. To examine disagreement between respondents, box plots were created for perceived risks (Figure 2) , and t-tests were conducted to compare perceived risks between demographic groups (Table 2) .

To facilitate interpretation of the main analyses, investigators first examined the percentage of participants who engaged in each protective behavior . To answer the main research question  about the relationship between risk perceptions and protective behaviors, authors conducted   separate logistic regressions predicting each of the protective behaviors, including handwashing, avoiding crowds, avoiding contact with people who could be high risk, and canceling or postponing air travel. Because ORs for continuous scales with small units (such as the 0%-100% risk perception scales) can be difficult to interpret, these logistic regressions were conducted with the quartile-based risk perceptions as predictors (Tables 3 and 4 ). Specifically, respondents were divided into quartiles by perceived risk (or four similar-sized groups of respondents giving the lowest 25% to the highest 25% of responses). The authors also created graphs presenting the percentage of participants who reported protective behaviors, by quartile of risk perception ( Figure 3 ). Analogous logistic regressions with continuous risk perceptions yielded similar results (Table 5) , as did Pearson correlations that did not include control variables (Table 6 ). In each of the reported logistic regressions, predictor variables were perceived infection risk and perceived infection fatality risk for COVID-19, while accounting for whether or not participants responded later (between March 13 and 31, 2020=1; between March 10 and 12, 2020=0); were in the at-risk group aged ≥65 years (yes=1, no=0); identified as male (yes=1, no=0); identified as African American (yes=1, no=0), Hispanic/Latinx (yes=1, no=0), or another minority (yes=1, no=0) versus white; had a college degree (yes=1, no=0); and lived in one of the states that was worst hit at the time of the survey (California, Massachusetts, New Jersey, New York, or Washington=1; other state=0).

In a final set of exploratory analyses, authors examined whether the findings differed between the 50% of respondents who completed the survey on or after (versus before) March 13, 2020, when the national emergency and the European travel ban were taking effect, and school closures were being announced. 10-12 This study used this dichotomous variable rather than a continuous variable for survey day because the number of respondents across survey days was highly skewed and dropped off substantially for later survey days ( Figure 1 ). Authors computed t-tests to compare risk perceptions of later versus earlier responders. Chi-square tests compared reported protective behaviors of later versus earlier responders. This study also examined contributions of later versus earlier responding in the logistic regression models that predicted each protective behavior (Table 3 ). To examine whether relationships between risk perceptions and protective behavior differed between later versus earlier responders, Figure 4 added interactions between later versus earlier responding and each risk perception to logistic regressions predicting each protective behavior (Tables 3, 4, and 5), and logistic regressions were run separately for later responders and earlier responders (Tables 7 and 8 ).

None of participants had yet been diagnosed with COVID-19, though 0.3% were unsure. None of participants thought they had been infected, though 6.9% were unsure. Median risk perceptions were 10.00% for perceived COVID-19 infection (mean=21.25, SD=22.90, 95% CI=20.70, 21.80) and 5.00% for perceived COVID-19 infection fatality (mean=15.17, SD=22.45, 95% CI=14.63, 15.71). Box plots revealed large disagreement between respondents about the emerging risks, although most perceived risks to be toward the lower end of the scale ( Figure 2 ). Some of the disagreement in perceived risks likely reflected respondent characteristics ( Table 2) . As noted, this study controlled for these characteristics in subsequent analyses that answered the main research question about the relationship between risk perceptions and protective behaviors.

Of the 6,684 respondents, 90% reported handwashing, 58% avoiding high-risk individuals, 57% avoiding crowds, and 37% canceling or postponing travel. Although respondents showed large disagreement between the risks of COVID-19 infection (Figure 2 ), risk perceptions were generally associated with protective behaviors. In logistic regressions that took into account quartile-based risk perceptions as well as respondent characteristics, as their perceived risks of COVID-19 infection increased by 1 quartile, participants were 1.45 times (95% CI=1.33, 1.58) more likely to report handwashing ( Table 3 , Model 1). Reports of handwashing increased from 83% to 94%, between the quartile of respondents perceiving the lowest risk for COVID-19 infection and the quartile reporting the highest risk for COVID-19 infection ( Figure 3A ).

In the logistic regressions that accounted for quartile-based risk perceptions and respondent characteristics, perceived COVID-19 infection fatality risk added little to predictions of handwashing (Table 3 , Model 1). Handwashing only increased from 87% to 90% between the quartile of respondents perceiving the lowest risk for COVID-19 infection fatality and the quartile reporting the highest risk for COVID-19 infection fatality ( Figure 3B ).

When predicting social distancing behaviors such as avoiding public spaces or crowds, both risk perceptions were significant (OR=1.24, 95% CI=1.17, 1.30 for infection risk and OR=1.19, 95% CI=1.13, 1.26 for infection fatality risk) ( Table 3 , Model 2). Avoiding public spaces or crowds increased from 45% to 67% between the quartile of respondents perceiving the lowest risk for COVID-19 infection and the quartile reporting the highest risk for COVID-19 infection ( Figure   3A ). For infection fatality, the increase was from 46% to 63% ( Figure 3B ).

Median perceived risk for COVID-19 infection was 10.00% for the 50% of participants responding later (between March 13 and 31, 2020) and 5.00% for the 50% of participants responding earlier (between March 10 and 12, 2020 that took into account quartile-based risk perceptions and respondent characteristics found that later responders were approximately two to three times more likely to implement protective behaviors than earlier responders (Table 3) . ORs in models with quartile-based risk perceptions (Table 3 ) varied between 2.02 (95% CI=1.70, 2.39) for handwashing and 3.33 (95% CI=3.00, 3.71) for avoiding public spaces or crowds.

Adding interactions between responding later versus earlier and risk perceptions to the logistic regressions (Table 3 ) revealed that the association between perceived COVID-19 infection risk and protective behaviors was stronger for later responders. Possibly, earlier responders were still hesitating to act on their risk perceptions than earlier responders. This relationship held for all behaviors, except for canceling or postponing travel. For example, Figure 4 shows that handwashing increased with 10 percentage points (from 86% to 96%) for later respondents perceiving COVID-19 infection risk in the lowest (versus highest) quartile, but with 9 percentage points (from 82% to 91%) for earlier respondents perceiving COVID-19 infection risk in the lowest (versus highest) quartile. Avoiding public spaces or crowds increased with 20 percentage points (from 58% to 78%) for later respondents reporting infection risk perceptions in the lowest (versus highest) quartile, but with 13 percentage points (from 36% to 49%) for earlier respondents reporting COVID-19 infection risk perceptions in the lowest (versus highest) quartile. The interaction effect for handwashing may have been relatively smaller, owing to a potential ceiling effect in reported handwashing.

Interactions between perceived COVID-19 infection fatality risk and later versus earlier responding were not significant in most of the logistic regression models predicting protective behaviors (Table 3) . Only in models predicting handwashing was this interaction significant independent of whether risk perceptions were entered as continuous or quartile-based predictors.

That is, risk perceptions of COVID-19 infection fatality risk were somewhat less strongly related to handwashing, among participants who completed the survey later. Handwashing increased with 4 percentage points (from 83% to 87%) for the earlier respondents perceiving COVID-19 infection risk in the lowest (versus highest) quartile, and with 1 percentage point (from 91% to 92%) for the later respondents perceiving COVID-19 infection risk in the lowest (versus highest) quartile. It is possible that this finding also reflected the aforementioned ceiling effect in the uptake of handwashing. A cross-sectional study conducted during the initial stages of the SARS outbreak in the Netherlands reported similar effect sizes as the ones reported here, for correlations between perceived SARS infection risk and reports of protective behaviors. 14 By comparison, crosssectional and longitudinal reports of correlations between perceived risk of infection with seasonal influenza and getting the influenza vaccine appeared to be somewhat higher than the correlations between risk perceptions and protective behaviors reported here. 7,15 However, a cross-sectional study of risk perceptions for hypothetical pandemic influenza showed no significant correlations with protective behaviors in Asian or European regions that were not experiencing outbreaks at the time. 16 These findings are in line with construal level theory of psychological distance, which posits that risks that are perceived as uncertain or further in the future may reduce willingness to act. 17 At the early stages of the COVID-19 epidemic, it is possible that many people were still hesitant to act on their risk perceptions and preferred taking a "wait and see" approach-perhaps especially because social distancing can be perceived as difficult and costly.

This interpretation is also supported by the exploratory analyses of differences between later and earlier responders. Those analyses suggested that, as the epidemic started to unfold, reported risk perceptions and protective behaviors increased, and the relationship between perceived COVID-19 infection risk and most protective behaviors became stronger. Because survey days were not randomly assigned, it is possible that different people responded earlier or later. However, these findings held when accounting for respondent characteristics (Tables 3, 4, and 5). A longitudinal study of H1N1 risk perceptions and intentions to vaccinate for H1N1 also found initial increases over the first few months of the epidemic but did not report on changes in the relationship between risk perceptions and protective behavior. 13

Like any study, this study had limitations. One limitation is that this cross-sectional study yielded only correlational findings, and that survey days were not assigned randomly, which precludes causal conclusions. For example, correlations between risk perceptions and reported actions could reflect the effect of a third variable, such as risk messages that increased both. Another limitation is that behaviors were self-reported, and may have reflected social desirability bias. 18 Longitudinal studies are needed to understand how risk perceptions, protective behaviors, and their associations changed over time, beyond the early stages of the epidemic. 13,15 Over time, it is possible that people who take actions to protect themselves may subsequently perceive less COVID-19 risk. As an example, people who have received a seasonal influenza vaccination have been found to lower their perceived risk afterward. 15 These analyses included no measures of perceived uncertainty or timing of the risk, which would have been needed to further understand the role of psychological distance in hesitancy to act on initial risk perceptions. The analyses also included no measures of other factors that could have potentially motivated protective behaviors, such as perceptions of the chance of infecting others, social norms, the ability to implement protective behaviors and bear any associated costs, or the need to follow policymakers' recommendations and stay-at-home orders. [4] [5] [6] CONCLUSIONS Even though people in the U.S. seemed to disagree about the risks associated with COVID-19 in the early stages of the epidemic, those perceiving greater risks were more likely to report that they implemented protective behaviors-and more so later (versus earlier) in March 2020.

Research on psychological distance has suggested that people may be more willing to act if risks are presented as happening in the here and now, and as real rather than hypothetical. 17 To promote protective behaviors, communications may need to address risks, as well as other factors that (as noted above) have been deemed relevant to behavior change, 4-6 such as such as the perceived chance of infecting others, social norms, ability to implement protective behaviors and bear any associated costs, or the perceived need to follow policymakers' recommendations and stay-at-home orders. Note: 2018 U.S. population statistics as reported by the U.S. Census Bureau (www.census.gov/quickfacts/fact/table/US/PST045218), with percent with college degree for population over age 25 years. Worst-hit states at the time of the survey were California, Massachusetts, New Jersey, New York, and Washington. There were 6,684 responders and 1,805 non-responders among the 8,489 invitees, but a few non-responders had missing data for specific characteristics. ; t-tests were conducted to examine differences between means. Significant differences are indicated on the first row for each group (***p<0.001; **p<0.01; *p<0.05). For race/ethnicity, the first three groups were compared to the white group. Median income was $50,000-$59,999. Worst-hit states at the time of the survey were California, Massachusetts, New Jersey, New York, and Washington. Boldface indicates statistical significance (***p<0.001; **p<0.01; *p<0.05). Later responders completed the survey on March 13-31, 2020, and earlier responders on March 10-12, 2020. Logistic regressions used post-stratification weights and controlled for at-risk age group ≥65 years (yes=1; no=0); sex (male=1; female=0), race/ethnicity including African American (yes=1; no=0), Hispanic/Latinx (yes=1; no=0), and other minority (yes=1; no=0), income below median of $50,000-$59,999 (yes=1; no=0), college degree (yes=1; no=0), living in a state that were worst hit at the time, including California, New Jersey, New York, Massachusetts, and Washington (yes=1; no=0). Interactions were added to the reported main effects, in a separate model. Tables 4 and  5 show full models with quartile-based and continuous risk perceptions. Table 6 provides associated Pearson correlations. Table 5 shows full models with continuous risk perceptions. Table 6 provides associated Pearson correlations. Table 3 's Models 1-4A. ORs for continuous risk perceptions are associated with only 1 unit change on the 0-100% scale. Later responders completed the survey on March 13-31, 2020, and earlier responders on March 10-12, 2020. At-risk age group was aged 65 years or older. Median income was $50,000-$59,999. Worst-hit states were California, Massachusetts, New Jersey, New York, and Washington. Interactions were added to the reported main effects, in a separate model. Logistic regressions used post-stratification weights. Table 4 shows full models with quartile-based risk perceptions. Table 6 provides associated Pearson correlations. 

COVID-19: Lessons from SARS and MERS

Take steps to prevent getting sick

Expert judgments of pandemic influenza risks

Risk Perception and Communication

The health belief model and preventive health behavior

A protection motivation theory of fear appeals and attitude change

Measuring subjective probabilities: the effect of response mode on the use of focal responses, validity, and respondents' evaluations

Learning during a crisis: the SARS epidemic in Taiwan

USC's Understanding America Study

Precautionary behavior in response to perceived threat of pandemic influenza

Construel-level theory of psychological distance

The social desirability or preventive health behavior

Note: N=6,684 for later responders and N=6,684 for earlier responders. Boldface indicates statistical significance (***p<0.001; **p<0.01; *p<0.05). ORs for continuous risk perceptions are associated with only 1 unit change on the 0-100% scale. At-risk age group was aged 65 years or older. Median income was $50,000-$59,999. Worst-hit states were California, Massachusetts, New Jersey, New York, and Washington. Logistic regressions used post-stratification weights.