Hesitancy Toward a COVID-19 Vaccine

The threshold for herd immunity in the population (i.e., the proportion of the population that needs to be immune, either from a vaccine, previous infections or both, to ensure herd immunity) is typically inferred from the basic reproduction number for COVID-19, R0. R0 estimates vary across multiple dimensions, such as data availability, geographical location and methods used to produce the estimates (Liu et al., 2020). In the beginning of the pandemic, Sanche et al. (2020) used data from outbreaks in China and estimated R0 to be as high as 5.7, implying herd immunity may be reached first when 82.5% of the population is immune (Keeling and Rohani, 2008). Most estimates, across countries and regions, range between 2.5 and 4 (Fontanet and Gauchemez, 2020). Based on that range, the classical formula for herd immunity (1–1/R0) generates a herd immunity threshold within 60 to 75 percent of the population being immune. However, local variations can be large. Sy et al. (2020) estimated R0 for COVID-19 at the county level within the U.S. and found estimates ranging from 0.4 to 12.4, with a median county level R0 of 1.66. Further, the classic formula for herd immunity may be too simplistic to generate a good approximation of the herd immunity threshold (Aguas et al., 2020).

Our study relates to a rich body of literature on consumer responses to conflicting information, spanning multiple scientific disciplines. While not an exhaustive list, examples of important work in this area are Viscusi and Magat (1992), Magat and Viscusi (1992), Viscusi (1997), Viscusi et al. (1999), Rodgers (1999), Fox et al. (2002), Hoehn and Randall (2002), Cameron (2005), Rousu and Shogren (2006), Kelly et al. (2012), Carpenter et al. (2014) Hämeen-Anttila et al. (2014), Pushkarskaya et al. (2015), and Binder et al. (2016).

For instance, the few early estimates of the IFR available around the time of data collection for this study (i.e., in March 2020) were based on non-US data, primarily from Asia (see the meta analysis by Meyerowitz-Katz and Merone, 2020), which may not be representative of the USA, given local differences in factors such as public health and healthcare access and quality.

We kept the disparity in risk communicated by the CDC and the White House constant (at 15 percentage points) across both high and low infection risk treatments. Viscusi (1997) shows that the disparity in the risks communicated by different information sources may affect trust in all information sources, such that a change in the disparity in percentage point probabilities across high and low infection risk treatments could affect trust in both the CDC and the White House.

While the IFR in our study is high, our data imply that both the 1.5% and 10% IFR indicated in our survey were viewed as plausible by most study participants. After presenting respondents with our chosen design levels for the IFR for the average American (i.e., 1.5% or 10%), participants were asked whether they believed their own risk was lower, about the same, or higher than the indicated level. In all treatments, 65–70% of participants stated that they either considered themselves to be at “about the same” or “higher” risk, and we find no statistically significant (or substantive) difference in those beliefs across respondents in the low- and high-mortality treatments.

Note that we asked participants to suppose that a vaccine that was available today, although we expected participants to understand that a vaccine was in fact not yet available. An alternative would have been to ask about intentions to vaccinate at a future point in time, when a vaccine is more likely to be available. Our choice is based on control over the study environment. Participants may differ in their beliefs about when a vaccine will be available and how the risks of infection and death may evolve over the course of the outbreak—they might expect the pandemic to have concluded before a vaccine is available, herd immunity to be near, or that they personally will already have been infected. The recent polls that have measured COVID-19 vaccine hesitancy (see discussion in Sect. 4) vary in how they have dealt with the timing of the vaccine when asking about the willingness to vaccinate. Like our study, the poll by Pew Research Center (2020) asks about vaccine intentions if the vaccine was available today, while the polls by ABC news/Ipsos (2020) and LX/Morning Consult (2020) ask about willingness to vaccinate when a vaccine becomes available without specifying when that might be.

All results are adjusted for multiple hypothesis testing using the Bonferroni correction (Bonferroni, 1935) with a family-wise type I error rate of 0.05. Adjusted p-values represent Bonferroni corrected p-values.

The results reported in Fig. 2 (a) remain highly similar even if we remove subjects stating 50 percent (see figure included in Supplemental online appendix).

The results from the probit regression are reported in Supplemental online appendix. It is the same model that generates the result in Fig. 4, but with the treatment variables only (i.e., only the top three variables in Fig. 3), given the isolation of the treatment effects relies on the randomization of participants across treatments. The observed treatment effects are very similar to those generated by the model in Fig. 4. For space saving reasons, we therefore refrain from including a separate figure in the main text for the probit regression that has the treatment variables only.

The particularly low vaccine uptake in the group that received the “low-probability (communicated by both the CDC and the White House) and high-mortality” treatment can be partially explained by our finding that the IFR for the average American (as communicated in our study) has little (if any) influence on participants’ decision to vaccinate, as shown in Fig. 4. The probability of infection and the mixed messages from the White House and the CDC have larger effects on vaccine uptake (as suggested by the results in Fig. 4). Therefore, we should expect the vaccine uptake in this treatment to be lower than that in other treatments. Another factor that could contribute to the low vaccine uptake in this group is differences in participant characteristics between this group and the rest of the sample. As discussed in Sect. 2, we found no “meaningful” differences (Imbens and Rubin, 2015) in participant characteristics and attitudes across any treatment groups. However, there are small differences between this group and the rest of the sample (statistically significant at conventional levels, but uncorrected for multiple hypotheses testing). Specifically, the share of women, share of participants with low trust in government agencies, and share of participants ascribing to neither Democrats nor Republicans are all higher in this group than all other groups in the sample. These differences are in the direction that would suggest a lower than average vaccine uptake for this group, as suggested by our analysis of vaccine uptake for our sample as a whole.

The dummy variables Vaccine confidence, Vaccine complacency, Vaccine calculation, Vaccine collective responsibility, and Vaccine constraint represent the five key components of the vaccine hesitancy scale developed by Betsch et al. (2018).

While we measured trust in government agencies as a continuous variable, it is included in the regression as a categorical variable. The reason for this is to avoid problematic multicollinearity. There is high bivariate correlation between the continuous trust measure and the vaccine confidence and the flu shot variables, respectively. If the continuous trust variable is included in the regression, trust (misleadingly) appears to have no effect on vaccine uptake. However, if either vaccine confidence or the flu shot variable is removed from the regression, the continuous trust variable is highly statistically significant. Binning the trust variable into low, medium, and high trust eliminates the problematic multicollinearity.

We do not know the causality for this relationship—while people might be less inclined to take a vaccine if they perceive the risk of the virus to be lower, it is also possible that people who are hesitant to vaccines are motivated to downplay the risk of the disease (for motivated risk beliefs, see, e.g., Kopetz and Woerner, 2021).