Epidemic context effects
Across participants, we varied the context (i.e., the epidemic development) in which individuals decided whether to accept a given vaccine or not. In Figure S.2, we utilize this to test whether the epidemic context matters by regressing vaccine acceptance on the context variable. The figure shows only marginal differences in acceptance and these differences remain statistically indistinguishable from zero. In conclusion, the imagined epidemic development in Denmark does not affect participants’ likelihood of accepting a vaccine independently.
Average attribute effects
The average attribute effects on the probability of accepting a vaccine are plotted in Fig. 1 (Table S.5 in the SM provides the supporting regression table). On the vaccine characteristics, the results show that vaccines that provide 70% protection are about 13 percentage points more likely to be accepted compared to vaccines with 50% protection, while vaccines with 90% protection are about 24 percentage points more likely to be accepted. We find similar but smaller effects for vaccines with lower risks of serious side-effects. Vaccines where 1 in 10,000 and 1 in 100,000 are hit by serious side-effects, are about 8 and 15 percentage points more likely to be accepted, respectively, compared to vaccines where the risk is 1 in 1,000.
On the characteristics related to vaccine development, we observe that participants attach high importance to an increased testing period. Test periods of 6 and 12 months as compared to 3 months, increase vaccine acceptance by about 5 and 8 percentage points, respectively. We find a similar but smaller premium of about 3.5 and 7 percentage points, respectively, for vaccines that 100,000 and 1,000,000 as compared to 10,000 have already received. Vaccines that are produced abroad face a penalty of about 2.5 percentage points (UK) and 6.5 percentage points (USA).
On the vaccination strategy characteristics, we find that compared to vaccines recommended by health authorities, a recommendation from your own doctor or the government face penalties of about 3 percentage points each, while vaccines recommended by a researcher are about 6 percentage points less likely to be accepted. Similarly, we observe some difference with respect to vaccination place with a penalty of about 3 and 3.5 percentage points, respectively, for getting the vaccination at the pharmacy and regional hospital compared to your own doctor. Moving to the two remaining factors, we find no difference heterogeneity in vaccine acceptance with respect to vaccination time or with respect to making appointments.
Taken together, the results demonstrate that vaccine acceptance indeed depends on characteristics related to the vaccine itself, its development, and the vaccination strategy. Specifically, preferences over vaccines are structured by seven main factors: vaccine efficacy, side-effects, number of previously vaccinated, testing period, production country, endorsements, and place of vaccination administration.
Effect heterogeneity
Do vaccine preferences vary across people with different psychological dispositions? As discussed above, we might expect that the attribute effects are driven by those who given their background characteristics are already more inclined to be vaccinated. For instance, we know from previous studies that those who feel personally at risk are more likely to accept vaccines [5, 7, 8, 9]. This could suggest an interaction, where the attribute effects would be strongest for those who perceive a high degree of personal risk. In order to test for interactions between participants’ psychological dispositions and the vaccine attribute effects, we split the main analysis according to four recognized psychological vaccination dispositions, including 1) institutional trust, 2) personal COVID-19 worry, 3) vaccine motivations, and 4) vaccine worry (see Table S.4 in the SM for details on the measurements of these psychological factors).
Figure 2 shows the results of these subgroup analyses (Table S.6 in the SM reports the supporting regression table). Overall, we observe a striking homogeneity in the attribute effects across the subgroups that are below the median (blue circles) and above the median (yellow triangles) on these psychological dispositions, respectively. Additional analyses using more fine-grained subgroups show that results are similarly homogenous (see Figure S.5-Figure S.8).
Do the vaccine attribute effects vary across demographic subgroups? In order to test for interactions between participant demographic and the effects of vaccine attributes, we stratify the main analyses by participants’ sex, age, education, and geographical location (see Figure S.9 and Figure S.10 in the SM). Overall, we find that the attribute effects are broadly similar across different subgroups. This suggests that there is a consensus—among females and males, young and old, low and high educated, and across geographic regions—on which vaccines are preferred over others. However, there is one notable exception to this homogeneity. Hence, the age panel shows that the young drive the estimated impact of side-effects. This aligns with studies making the argument that SARS-COV-2 has an asymmetric age-profile [31], implying that the elderly would be more likely to accept any vaccine.
Although the epidemic context did not independently affect vaccine acceptance (see Figure S.2), the attribute effects might still vary across contexts. For example, vaccine safety might matter less in the context of an increasing epidemic and more in the context of a decreasing epidemic. However, we observe that the estimated effects remain fundamentally similar across contexts (see Figure S.11). Relatedly, we ask if the attribute effects vary across different vaccine profiles. Consistent with psychological research on decision-making under uncertainty, we might, for instance, expect that the influence of vaccine safety characteristics (e.g., side-effects) is largest when vaccine efficacy is relatively low. However, we find no substantively meaningful first-order interaction among any of the conjoint attributes. Altogether, the absence of such interaction effects suggest that participants consider the different characteristics independently from each other (see Figure S.12-Figure S.20).
Predictions of acceptance
To better understand the substantive meaning of these results, Fig. 3 illustrates the predicted probability of vaccine acceptance based on the model estimates from our baseline model in Fig. 1. In particular, we extract the probability of acceptance at the minimum, 25th percentile, median, 75th percentile, and the maximum. The acceptance rate of the least likely vaccine is about 23 percent (95% CI: 19–28), while the rate of the most likely vaccine is about 95 percent (95% CI: 90–99). The intermediate vaccines at the 25th percentile, the median, and the 75th percentile are predicted to be accepted by about 50 percent (95% CI: 46–55), 60 percent (95% CI: 55–64), and 69 percent (95% CI: 65–74), respectively.
Mirroring the results from Fig. 1, we see a large jump in the expected acceptance from the minimum (about 23 percent acceptance) to the 25th percentile (about 50 percent acceptance) driven by changes in the vaccine safety attributes. Moving from the minimum to the 25th percentile, we most notably see a decrease in side-effects, an increase in the number of previously vaccinated, and a change in production country from the US to the UK. Moving further up the distribution to the median (about 60 percent acceptance), we see that the jump in acceptance is driven by a change in vaccine efficacy from 50 to 90 percent. Comparing the 75th percentile (about 69 percent acceptance) to the median, we see an increase in vaccine acceptance of about 9 percentage points driven by increases in vaccine safety in terms of a lower risk of serious side-effects and an increased test period. Finally, we see that the increase from the 75th percentile to the maximum (about 95 percent acceptance) is driven by: 1) an increased test period, 2) an increasing number of vaccinated, 3) that the vaccine is produced in Denmark, 3) that the authorities recommend the vaccine, 4) that vaccination is at your own doctor, 5) that you do not have to book an appointment.