We used data from the Franklin Templeton-Gallup Economics of Recovery Study, a self-administered web survey from an opt-in sample provided by Dynata of 24,649 US adults, aged 18 years and older, of whom 10,419 participated during more than one survey time point. The survey was conducted during the following time points: July 2 to 14, August 3 to 11, September 4 to 13, October 1 to 9, November 2 to 6, and December 1 to 7, 2020. Gallup weighted the obtained sample to correct for nonresponse and construct a nationally representative population. Nonresponse adjustments were made by adjusting the sample to match the national demographics of gender, age, race and ethnicity, region, educational level, marital status, and employment status. Demographic weighting targets were based on the Census Bureau’s 2018 data release of the American Community Survey and the Current Population Survey (February 2020).

We also supplemented this survey with data from the Gallup Panel, a research panel that is representative of the US adult population and includes approximately 100,000 members. Gallup fielded the COVID-19 tracking survey on March 13, 2020, and collected approximately 1000 responses daily until April 26, 2020, when the sample declined to approximately 500 responses daily. The Gallup Panel’s COVID-19 Tracking Survey includes information about self-reported household isolation and loneliness, with 123,516 observations from March 24, 2020, to February 21, 2021. This study was exempt from institutional review board review according to policies of the UCLA (University of California, Los Angeles) Office of the Human Research Protection Program.

We assessed self-reported household isolation by asking participants about the degree of in-person contact outside their household that they had over the past 24 hours. We considered a participant to be isolated if they reported being completely isolated (no contact) or mostly isolated (very little contact) from people outside their household. We assessed loneliness in the Gallup Panel survey by asking participants, “Did you experience the following feelings during a lot of the day yesterday?” Loneliness and other emotional experiences were included as response options. The specific questions and respondent options are provided in Multimedia Appendix 1.

We evaluated perceptions of COVID-19 health risks using multiple questions. In July and August, respondents were asked about the percentage of all US COVID-19 deaths that fell into the following age strata: age 24 years and below, age 25 to 34 years, age 35 to 44 years, age 45 to 54 years, age 55 to 64 years, and age 65 years and older. We assessed misperceptions using the reported proportion of deaths attributable to persons under the age of 55 years because most deaths from COVID-19 have occurred in persons older than 55 years. Perceptions about age-related COVID-19 health risks were assessed in September and October 2020 with an analogous question about the age distribution of COVID-19 hospitalizations.

In the November 2020 survey, respondents were asked what percentage of patients hospitalized with COVID-19 died. In the December 2020 survey, respondents were asked what percentage of patients infected with COVID-19 required hospitalization.

CDC data were used to estimate the proportion of deaths from COVID-19 by age. Data from the COVID-19–Associated Hospitalization Surveillance Network (COVID-NET) was used to estimate hospitalizations by age [12]. Because COVID-NET data on hospitalizations were reported using different age strata than those provided to survey respondents, we adjusted these data using simple proportional methods. Specifically, we multiplied hospitalizations reported in COVID-NET by the proportion of years in a corresponding age stratum in order to recategorize hospitalizations into different age strata. The likelihood of hospitalization after infection was estimated to be approximately 5% as reported by Reese and colleagues from the CDC estimate that there were 52.9 million infections from February to September 2020 and 2.4 million hospitalizations, implying a hospitalization rate of 4.5% [13]. Their method accounted for underreporting. We estimated the likelihood of death among patients hospitalized for COVID-19 to be 12% based on an analysis of 38,517 hospitalized patients from January 1, 2020, to June 30, 2020 [14].

In each wave, survey respondents were asked whether they or a household member had a comorbidity that increased the risk of severe COVID-19 illness. Respondents also reported sociodemographic characteristics, household income, and preferences for political parties. Per capita deaths from COVID-19 in each US county from March 1 until December 1 were assessed using CDC data [15].

Descriptive analyses of respondents’ characteristics were performed using data from July, August, and December 2020. The July and August surveys were combined for analyses because questions about risk perception were identical between those two time points. The September and October surveys were similarly combined. We used multivariable logistic regression analyses to examine the relationship between misperceptions about COVID-19 health risks and social isolation. Results from these models were presented as predictive margins, in which the regression models were used to estimate the marginal effect of risk overestimation, expressed as a proportion, while holding the distribution of all other covariates constant [16]. The adjusted association of respondent characteristics with social isolation was also presented using data from July and August as well as December in order to examine how behavioral patterns may have shifted over the course of the pandemic. In a secondary analysis, we used Gallup Panel data to assess the relationship between self-reported household isolation and loneliness.

Perceptions about risk were characterized as being overestimates, underestimates, or accurate estimates. To provide respondents with a reasonable degree of latitude and to account for any uncertainty in our reference estimates, we considered responses that were within 5 percentage points above or below the correct estimate as being accurate (eg, a response of 15% for estimated hospital mortality would be considered accurate because it fell within 5 percentage points of the actual rate of 12%) [14]. A range of 10 percentage points above or below was used for respondents’ estimates of the proportion of hospitalizations occurring in persons younger than 55 years old, due to the larger proportion.

We performed mean imputation from the overall sample for education (missing <1%), income (missing <1%), and whether the respondent or their family had a serious medical condition (missing <1%). We used this method instead of a more robust multiple imputation model because of the low rate of missingness. We did not report percentages as n/N values because all reported percentages were estimated using analytic weights. All analyses were performed using Stata 14 (StataCorp LP) and incorporated analytic weights to account for the effects of nonresponse.

The data used in this study are available from the corresponding authors upon request and with the permission of Franklin Templeton and Gallup.

We present descriptive characteristics of the respondents in Table 1. The mean age of the respondents was 47 years (SD 18), and 52.3% were female. The largest proportion of respondents had a household income that ranged from US $48,000 to US $89,999. Half of the respondents reported that they or a household member had a serious medical condition that increased their risk of serious illness from COVID-19.

Most participants held beliefs about COVID-19 health risks that were inaccurate (Figure 1). Overestimation of health risk, rather than underestimation, was the most common type of inaccuracy at each survey time point. For example, while persons younger than 55 years old accounted for 7% of total US deaths at the time of the July and August surveys, respondents estimated that they accounted for 43% of total deaths. In addition, while the proportion of COVID-19 hospitalizations that occurred in persons younger than 55 years old was 38%, respondents in the September and October surveys reported that this population accounted for 46%. The mortality rate of patients hospitalized with COVID-19 was estimated by respondents in the November survey to be 25% compared to an actual rate of 12%. The proportion of patients hospitalized after being infected with COVID-19 was estimated by respondents in the December survey to be 34% compared to an actual proportion of 12%.

The proportion of respondents living in self-reported household isolation did not vary substantially over time, ranging from 53% to 57% (Figure 2). Overestimating the proportion of death or hospitalizations from COVID-19 occurring in people under 55 years old was associated with a significantly increased likelihood of self-reported household isolation (Table 2). Overestimating the likelihood of death or hospitalization was also associated with a significantly increased likelihood of self-reported household isolation. Excluding respondents who underestimated risk modestly attenuated the results, but all associations between misperceptions and social isolation remained significant (Table S1 in Multimedia Appendix 2). In the Gallup Panel, adults living in self-reported household isolation reported higher rates of loneliness (Figure S1 in Multimedia Appendix 2).

Characteristics associated with self-reported household isolation from the July and August 2020 surveys and persisting in the December 2020 survey included younger age (18 to 39 years), having a serious medical condition, having a household member with a serious medical condition, and identifying as a Democrat (Table 3). Being Black or Hispanic was associated with a higher likelihood of social isolation in July and August, but this relationship was not present in December. Reporting a higher income (>US $48,000) was associated with a lower likelihood of social isolation in July and August, but this relationship had largely waned by December.