Unraveling attributes of COVID-19 vaccine acceptance and uptake in the U.S.: a large nationwide study

A total of 36,711 users responded to the vaccine acceptance question. The largest number of respondents came from Connecticut and California with 8697 and 4668, respectively (Supplementary Fig. 1a). HWF’s user base is approximately 79% female (Supplementary Fig. 1b) and 83% white (Supplementary Fig. 1c). Users are 18 years of age or older and are equally distributed by age groups (Supplementary Fig. 1d). More than 68% of respondents were non-essential workers, and users cover a diverse range of income groups. All descriptive statistics of the study participants are available in Supplementary Table 1.

In total, 30,618 (83%) were willing (“Likely” or “Very Likely”) to be vaccinated (Fig. 1a). After applying a census-based post-stratification weight (see Methods), Vermont (92%) and Washington D.C. (88%) had the highest rates of vaccine reluctance while South Dakota (27%) and Louisiana (23%) had the highest rates of undecided users (Fig. 1b). Weighted bar plots of vaccine reluctance across demographic characteristics revealed that “Undecided” users represented the largest proportion of non-willing users across all demographic groups (Fig. 2a, Supplementary Table 2).

COVID-19 Acceptance rates: (a) (Left) Number of responses and (Right) unweighted and weighted percentages. (b) Weighted average willingness and undecided rates by state.

Demographic Acceptance Rates: (a) Weighted percentages of reluctant responses by race/ethnicity, profession, location, age, income, and use of protective measures. State level weighted reluctance rates by (b) cumulative case rates (/100 individuals), (c) cumulative death rates (/1000 individuals), (d) and average number of users practicing protective behavior.

State level reluctance (“Undecided”, “Unlikely”, or “Very Unlikely”) rates were negatively associated with the average number of users that practiced transmission mitigating behaviors and were positively associated with cumulative COVID-19 case and death rates by January 10, 2021 (Fig. 2b–d). Unweighted plots are available in Supplementary Fig. 2.

To assess demographic associations with vaccine acceptance, we fit a univariate logistic regression with socio-demographic, occupation, preexisting medical conditions, geographical and COVID-19 related predictors (Supplementary Table 3) and a multivariable logistic regression model to adjust for potential relationships between the predictors (Fig. 3, Supplementary Table 4). We implemented post-stratification weights using census estimates of sex, age, race, and census location (see Methods). People of color reported higher rates of vaccine reluctance compared to white non-Hispanic users (African American OR, 3.94; CI, 3.47, 4.48; p = 1.26e−96). Vaccine reluctance was more likely among females than males (OR,1.67; CI, 1.51, 1.83; p = 4.09e−25); younger users than those over 65 years old (18–30 OR, 2.17; CI, 1.86, 2.53; p = 1.03e−22); those with three or more preexisting conditions than those with zero (OR, 1.19; CI, 1.06, 1.34; p = 0.0036); and parents than non-parents (OR, 1.26; CI, 1.15, 1.38; p = 9.61e−7). Individuals that were furloughed or job-seeking were also more vaccine reluctant compared to those working full- or part-time (OR, 1.48; CI, 1.29, 1.70; p = 4.04e−8). Respondents from the South (OR, 1.25; CI, 1.05, 1.48; p = 0.0105), from less densely populated areas, or with lower incomes were all more likely to be vaccine reluctant. Users that responded before the Pfizer Emergency Use Authorization (EUA) on December 11, 2020 were more vaccine reluctant than users who responded after the Pfizer EUA (OR, 1.48; CI, 1.37, 1.60; p = 9.96e−23), users who practiced behavior protective against COVID-19 such as mask-wearing or social distancing were less vaccine reluctant (OR, 0.78; CI, 0.72, 0.85; p = 6.12e−9), and users that received a COVID test were less vaccine reluctant (OR, 0.79; CI, 0.71, 0.89, p = 5.88e-5).

Logistic regression-based association analysis results of vaccine acceptance: Forest plots for (Left) unweighted and (Right) weighted multivariable logistic regression analyses for vaccine reluctance with 95% confidence intervals. Non-significant variables at the 0.05 level (white), significant positive associations (red), and significant negative associations (blue).

Nominal logistic regression (see Methods) evaluated whether vaccine reluctance was driven by “Undecided” vs. “Unlikely/Very Unlikely” responses (Supplementary Table 5) and was also conducted with a weighted analysis (Supplementary Table 6). Reluctance in healthcare workers, those aged 55–64, Asian users, and those in locations with a median income between $70,000 and $100,000 was driven by the “Undecided” group, whereas reluctance in the unemployed, those with 3 + preexisting conditions, and southern users was driven by the “Unlikely” group. Sensitivity analyses were performed for the weighted multivariable and nominal regression analyses with a less restrictive threshold for the trimming weights (Supplementary Table 7–8) and found similar results. We conducted a sensitivity analysis to assess differences in reluctance in individuals that tested positive for COVID-19 and found no difference in intention based on testing results (see Methods, Supplementary Table 9–10).

Of the 36,711 users who responded to the vaccine acceptance question, 23,429 also responded to the vaccine uptake question and its distribution is provided in Fig. 4a. Demographic distributions remained similar to those of respondents of the vaccine acceptance question with a slight increase in the proportion of users ages 65 + (Supplementary Fig. 3). Vaccination rates by state are shown in Fig. 4b for all users that responded to the vaccine uptake question and subset to respondents who were offered a vaccine. Users with lower levels of vaccine acceptance had lower rates of vaccination (Fig. 4c), and Black and Hispanic/Latinx users reported lower rates of vaccination than White, Non-Hispanic users (Fig. 4d). Plots of weighted and unweighted vaccination rates across all demographic features are available in Supplementary Figs. 4–5.

Vaccine Uptake Rates: (a) Vaccine uptake question responses for all users. (b) Weighted vaccination rates by state of (Left) all users that responded to the vaccine uptake question and (right) users that were offered a vaccine. (c) Weighted vaccination uptake of users that were offered a vaccine by vaccine acceptance and (d) race/ethnicity.

To formally identify demographic features associated with differences in vaccination rates, we conducted an unweighted and weighted multiple logistic regression analysis (see Methods, Fig. 5, Supplementary Tables 11–12). All age groups reported lower rates of vaccinations compared to users over 65 (18–30 OR: 0.10; CI, 0.06, 0.18; p = 1.43e−16); Black users reported lower rates of vaccinations (OR, 0.58; CI, 0.38, 0.91; p = 0.0165) compared to White non-Hispanic users; essential workers outside of healthcare reported lower rates of vaccinations (OR, 0.64; CI, 0.44, 0.92; p = 0.0162) compared to non-essential workers; parents reported lower rates of vaccination (OR, 0.63; CI, 0.45, 0.89; p = 0.0086) compared to users who are not parents; users in areas with a median household income (MHI) of $40–70 K (OR, 0.56; CI, 0.37, 0.85; p = 0.0066) and $70–100 K (OR, 0.63; CI, 0.42, 0.96; p = 0.0316) reported lower rates of vaccinations compared to those in areas with a MHI $100 K + ; users logging in from areas with 0–149 people/sq. mi reported lower rates of vaccinations (OR, 0.53; CI, 0.34, 0.82; p = 0.0049) compared to users in high population density areas; and users that responded “Unlikely/Very Unlikely” (OR, 0.02; CI, 0.01, 0.03; p = 2.07e−114) and “Undecided” (OR, 0.08; CI, 0.06, 0.12; p = 1.06e−39) to the vaccine acceptance question reported lower rates of vaccinations compared to willing users.

Logistic regression-based association analysis results of vaccine uptake: Forest plots for (Left) unweighted and (Right) weighted multivariable logistic regression analyses for vaccination uptake with 95% confidence intervals. Non-significant variables at the 0.05 level (white), significant positive associations (red), and significant negative associations (blue).

While vaccination rates were lower in the reluctant group compared to the acceptant group, 86% (2157/2520) of reluctant users were vaccinated. In a formal multiple regression analysis looking at demographic associations with vaccine uptake among reluctant users, similar associations were found (see Methods, Supplementary Table 13). Younger age groups, healthcare workers, people from lower income households, and residents of areas with lower population density had lower vaccination rates. Users who responded to the vaccine acceptance question as “Undecided” reported higher rates of vaccination compared to those that responded “Unlikely/Very Unlikely” (OR, 4.57; CI, 3.47, 6.03; p = 2.26e−26).