The effect of vaccine attributes on vaccine acceptance

Several studies examine individual-level predictors of (COVID-19) vaccine acceptance. Women, conservatives, lower-income people, and the highly religious, for example, tend to reject vaccines more [22, 23]. Less work has been dedicated to examining the effects of vaccine characteristics themselves. Much of this research has been undertaken in the context of COVID-19 as competing vaccines have been developed simultaneously, and such has mostly focused on the US (with the exception of work in Brazil, see [24]). People tend to prefer Western-origin vaccines over those developed in Russia or China, and often prefer their own country’s vaccine when available (that is, US citizens have preferred US-developed vaccines in these experiments) [12, 25]. Not surprisingly, people also prefer vaccines with greater effectiveness and lower risk of side effects [12, 25, 26]. People also prefer vaccines with full FDA approval over those with full FDA emergency use authorization [10]. Motta (2021) found that US citizens were indifferent to whether the vaccine required one or two doses, time taken for development, or mRNA vs. live virus vaccines. Regarding the vaccine technology, Owen et al. (2021) found a similar pattern, namely that people are indifferent between mRNA or a weakened virus vaccines. Participants in the Motta (2021) and Owen et al. (2021) study were not given information about vaccine types prior to making a choice (whereas our study does provide this information to respondents).

We identify at least three further vaccine attributes that might affect vaccine preferences. When evaluating a specific vaccine, people may prefer a vaccine that has already been administered to a large (vs. small) number of people as this can be interpreted as heuristic about safety [11]. People might also prefer a vaccine that allows faster large-scale vaccination. People might be sensitive to how expensive vaccines are to the government, particularly as COVID-19 has challenged government finances. In fact, assessing the impact of the cost of vaccines is important given that the EU emphasized per-dose price when negotiating contracts with manufacturers (e.g., Guardian 2021; Reuters 2020). We note, though, in our three countries, vaccines are offered at no charge to the recipient. Given the research gap on the effect of these vaccine attributes, we aim to explore to what extent the number of people who received a vaccine, the duration for large-scale coverage, and the costs a vaccine imposes on the society affect vaccine acceptance.

Vaccine hesitancy, risk preferences, and vaccine acceptance

A rich literature has examined individual-level characteristics correlated with immunization uptake. For instance, vaccine acceptance is relatively high among people who score low in conspiratorial thinking, high in cognitive reflection, and who trust the state, scientists and the healthcare system [23, 27, 28]. In this study, we focus on general vaccine hesitancy and risk preferences.

Vaccine hesitancy, defined as a delay in acceptance or refusal of vaccines despite their availability [29–32], is a major barrier of vaccine acceptance. Vaccine hesitancy is not randomly distributed in the population, a phenomenon that is highly problematic from an epidemiological perspective [33]. Urged by the need to tackle vaccine hesitancy, research has identified various driving factors, including contextual factors (e.g., media, politics), individual and group factors (e.g., knowledge, social norms), and vaccine-specific factors (e.g., mode of administration, costs) [30, 32, 34].

We expect that people scoring low compared to high on vaccine hesitancy are more favorable towards vaccines in general. However, we also expect vaccine hesitancy to moderate responsiveness to vaccine attributes in terms of overall vaccine preferences, though we lack strong priors on whether higher vaccine hesitancy enhances or attenuates the influence of vaccine attributes on vaccine acceptance. The stronger people with varying levels of vaccine hesitancy differ in their favorability towards (attributes of) vaccines, the higher seems the need for targeted messaging [17]. The role of individual differences in shaping the effect of vaccine attributes on vaccine acceptance is relatively unexplored (see [24, 25] for exceptions), and we hope our work invites further subgroup analyses.

Another body of work looks at the relationship between risk preferences and vaccination. At least in some cases, vaccine hesitancy and general risk aversion are distinct and unassociated with one another [35]. General risk preferences capture tendency to engage in behaviors or activities that are rewarding yet involve some potential for loss [36]. High risk tolerance predicts a series of risky health behaviors, ranging from smoking to avoiding dental care to non-use of seat belts [37]. Recent results suggest that lower risk tolerance is linked to greater vaccine uptake in the case of influenza, as these people may recognize the greater risk of non-vaccination in the face of highly effective vaccines [38]. A complementary perspective argues that vaccination decisions are often made by weighing the probability of getting a disease as well as its severity against the risks that come with inoculation. In particular, people scoring high on general vaccine hesitancy might be more likely to endorse getting a vaccine when the benefit of it is clear and salient, but likely reject vaccines if its benefit is unclear and not salient [39].

So far, it is relatively unexplored whether and how individual differences shape the influence of vaccine attributes on vaccine acceptance. Studies examining conditional effects are essentially a subset of the vaccine attribute studies described above. Gramacho et al. (2021) find that rejection of a Chinese-developed vaccine is particularly strong among those who support President Jair Bolsonaro, for example. Others [25] also show that vaccine efficacy was more influential on acceptance for some subgroups than others in the US (whites, Democrats). Finally, a Canadian study provides initial evidence highlighting that people who intend to get a COVID-19 vaccine are more responsive to vaccine attributes (e.g., country of origin, effectiveness) than people who intend not to get a COVID-19 vaccine [15]. Given that people varying in terms of individual differences such as vaccine hesitancy and risk preferences may need to be targeted differently by vaccination campaigns [17], we highlight the importance of further subgroup analyses.

Overall, we extend research on vaccine acceptance by surveying citizens of France, Germany, and Sweden. With our pre-registered research (link: https://osf.io/esmdt/), we aim to replicate prior work on the effect of vaccine attributes. Specifically, we expect to replicate the strong positive effect of high vaccine effectiveness, few side effects, and a preference for vaccines of German, American and British origin over Chinese and Russian origin. We also expect respondents to prefer less expensive vaccines, vaccines with a larger number of already-injected people, vaccines that take less time to administer at a large scale, and vaccines that are lower-cost. We lack strong priors on responsiveness to vaccine technology used. We also conduct exploratory analyses addressing the question whether vaccine hesitancy and risk preference influence (the effect of vaccine attributes on) vaccine acceptance.

Sample

We conducted online experiments with 5,432 respondents recruited from Dynata in France (N = 1,802; April 7–22, 2021), Germany (N = 1,799 March 29-April 24, 2021), and Sweden (N = 1,831; April 8–23, 2021). Sampling implemented nationally representative quotas for gender, age, and region. Supplementary Table SI1-3 in S1 File provide an overview of the demographics of the three samples. Respondents provided informed consent to participate by clicking "I agree to participate" in the online survey, and their data were anonymized. Respondents were paid a local fee for participating by Dynata. Note that we excluded people that did not fill in any conjoint question.

Note that we considered France, Germany and Sweden as suitable European countries as they had comparable timelines for their vaccine rollouts. In fact, all countries received their first vaccine delivery on the same day (December 27, 2020) as well as administered first vaccinations on the same day (an overview of the development of vaccine uptake of the three countries (in comparison with the EU average vaccine uptake) can be found on the European Center for Prevention and Control’s website (https://vaccinetracker.ecdc.europa.eu/public/extensions/COVID-19/vaccine-tracker.html#uptake-tab). Importantly, it is not this research’s major aim to examine country-specific differences. Rather, our aim is to gain initial insight into the generalizability across different countries.

Prior to fielding the survey, we conducted a power analysis using the DeclareDesign R package (Version 0.26.0) [40] to determine optimal sample size per country given an α of .05 and 80% power. We used Motta (2021) to inform expected effect sizes for attribute levels. That is, we used an effect size of .05 for 75%, and .12 for 95% effectiveness, .21 for China (country), .06 for UK (country), and .18 for Russia (country). Given our lack of strong priors on the effect sizes of other attribute levels, we assumed the need to detect a small effect (.03). The power analysis with 1,000 iterations revealed that a conjoint experiment with eight trials requires a minimum sample of N = 1,500 per country.

Procedure and conjoint experimental design

Prior to the experiment, respondents filled out demographic information and a battery on baseline vaccine attitudes. Then, respondents were given information about four vaccine types: mRNA vaccines, subunit vaccines, live virus vaccines, and viral vector vaccines. While a previous conjoint experiment [12] did not find an effect of vaccine type, it is unclear whether this is because people are truly indifferent, or whether they lack sufficient information to make a judgement. We gave respondents information to establish that null effects can be attributed to indifference. The exact wording of the provided set of information can be found in the Supplementary Material (see Fig SI8 in S1 File).

Next, respondents were taken to a conjoint choice task. In each of eight scenarios, respondents were presented with two different vaccines. These vaccines differ along seven attributes: (1) side effects, (2) effectiveness, (3) country of origin, (4) vaccine type, (5) vaccinated people, (6) vaccination coverage, and (7) costs (see Table 1 for more information). In each scenario, respondents indicated on a 0–10 scale the raw likelihood of taking such a vaccine (likelihood of uptake) and indicated which of two vaccines they preferred (vaccine choice). Fig SI8 in S1 File depicts a screenshot from the online experiment showing a tabular description of hypothetical vaccines A and B, and the likelihood of uptake and vaccine choice questions. This resulted in 16 self-reported likelihoods of uptake and 8 vaccine choices per respondent, totaling 84,816 self-reported likelihoods of uptake and 41,755 vaccine choices after listwise deletion of missing responses. Missingness was negligible (2.41% for likelihood of uptake, 3.91% for vaccine choice).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Conjoint experimental design.

https://doi.org/10.1371/journal.pone.0266003.t001

Measures

Likelihoods of uptake and vaccine choice.

The main response variables were self-reported likelihoods of uptake and vaccine choice. Likelihoods of uptake captured how likely respondents think they would choose to receive vaccine A or vaccine B. Specifically, respondents were asked “How likely would you be to choose to receive each of the vaccines?” and could indicate for both vaccines how likely they were to choose them on a scale from 0 (‘not at all likely’) to 10 (‘extremely likely’). Vaccine choice captured which vaccine respondents chose in hypothetical and randomly assigned A-B choice scenarios. Specifically, respondent could see a tabular description of vaccines A and B and were asked “Which vaccine do you prefer to receive?”. Note that we did not force respondents to evaluate the likelihoods of uptake and choice of vaccines. Fig 1 is a screenshot of the vaccine decision scenario and outcomes questions that respondents received.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Screen shot of a vaccine decision scenario (translated into English).

Before being exposed to eight vaccine decision scenarios, respondents were exposed to (1) general information about vaccines, and (2) descriptions of the different vaccine types (see Supporting information).

https://doi.org/10.1371/journal.pone.0266003.g001

Identifying favorable vaccine attribute levels

We computed marginal means (MMs) for vaccine attribute levels. MMs describe the average self-reported likelihood of a vaccine being accepted, on a 0–10 scale, when that vaccine has an attribute at a particular level (e.g., 95% effectiveness). We report MMs as they illustrate the baseline level of vaccine uptake, and allow for the comparison of vaccine uptake between the full sample and different subgroups. We preregistered our analyses as average marginal component effects (AMCEs), which estimate treatment effects when vaccine attributes are altered relative to a reference level (e.g., from 55% effectiveness to 95% effectiveness), and are available in the Supplementary Material (Fig SI1 in S1 File), as are full numerical estimates of both MMs and AMCEs (Tables SI4-5 in S1 File) [42–44]. We present the MMs in the main text to facilitate interpretation that is not predicated on a specific reference category [44]; similarly, we choose to present the vaccination likelihood as the primary outcome (versus the binary choice measure) to more clearly demonstrate levels of support for different vaccine profiles. Substantive results are identical across model choice (MMs vs. AMCEs) and outcome variable (vaccine uptake likelihood vs. binary vaccine choice).

Fig 2 plots MMs from the full sample of self-reports of how likely they would be to receive a vaccine with particular attribute levels across all respondents in all three countries, on the original 0–10 scale. Note that the dashed line represents the grand mean (i.e. overall mean across all attributes) and is useful for interpretation: Values above the line indicate attribute levels that increase profile favorability and values below indicate attribute levels that decrease favorability (compared to the average of all profiles). Here, we focus our attention on vaccine attributes with the strongest effects on respondents’ likelihood of accepting a vaccine: country of origin of a vaccine, effectiveness, and side effect frequency. Our conjoint experiments also ask respondents to make a binary choice between the vaccines they are shown. Results for this outcome variable are substantively similar to the results for our main outcome (Fig SI2-3 in S1 File, Tables SI6-7 in S1 File).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. MMs for self-reported likelihood of uptake.

The figure reports the marginal mean point estimates are plotted with 95% CIs, representing the average likelihood of uptake at each vaccine attribute level. The dashed line represents the grand mean (5.11).

https://doi.org/10.1371/journal.pone.0266003.g002

The MMs for country of origin show that uptake was more likely than not when vaccines originated from the UK (MM = 5.33, 95% CI = [5.25–5.40]), the USA (5.35, [5.28–5.43]), or especially Germany (5.52, [5.44–5.60]), whereas this was not the case for Russian (4.82, [4.74–4.90]) or Chinese vaccines (4.64, [4.56–4.72]). Similarly, respondents were less likely to accept a vaccine with an effectiveness rate of 55% (4.68, [4.61–4.75]), but more likely to accept one with 95% effectiveness (5.57, [5.50–5.64]). The likelihood of uptake for a vaccine with an effectiveness rate of 75% (5.15, [5.08–5.22]) was average. In both cases, going from the least favorable (China/55% effectiveness) to the most favorable attribute level (Germany/95% effectiveness) brought about a similar change in the average likelihood of accepting the vaccine, of just under 0.9 points (a 19% increase). Side effect prevalence shows a similar pattern of favorability, but this is less pronounced, with rates of ‘1 in 10,000’ slightly unfavorable (4.94, [4.87–5.01]), ‘1 in 100,000’ (5.16, [5.09–5.23]) average, and ‘1 in 1,000,000’ (5.30, [5.22–5.37]) more favorable. Going from the least favorable (‘1 in 10,000’) to the most favorable attribute level (‘1,000,000’) brought a 0.4 point change in the average likelihood of accepting the vaccine (a 8% increase). Again, note that AMCEs (i.e. effect sizes) estimate treatment effects when the MMs for vaccine attributes are altered relative to a reference level (see Fig SI1 in S1 File).

Compared to these strong effects of country of origin, side effect frequency and effectiveness, the remaining vaccine attributes affect the likelihood of uptake very little. Respondents were slightly more favorable towards vaccines that provide coverage of the population more quickly, that cost less, and that have already been used on more people. Notably, far from being skeptical of the newly innovated mRNA vaccines, respondents actually slightly preferred these relative to more established types.

Our findings are robust to restricting the sample to respondents who passed attention checks (see Fig SI4 in S1 File and Table SI8-11 in S1 File). There were minimal cross-country differences in the effects of vaccine attributes, with the exception of German-developed vaccines being particularly well received in Germany. MMs revealed, however, that French (compared to German and Swedish) respondents reported relatively low overall favorability towards vaccines. That is, French respondents’ reported likelihood of uptake was estimated to be 4.70 (4.57–4.82) for vaccines with ‘1 in 10,000 side effects’, 4.92 (4.80–5.05) for vaccines with ‘1 in 1,000,000 side effects’, 4.40 (4.27–4.52) for vaccines with ‘55% effectiveness’, 5.21 (5.08–5.34) for vaccines with ‘95% effectiveness’, 5.13 (4.99–5.27) for vaccines from Germany, and 4.31 (4.17–4.45) for vaccines from China. In contrast, German / Swedish respondents’ reported likelihood of uptake was estimated to be 5.00 (4.88–5.12) / 5.12 (5.00–5.24) for vaccines with ‘1 in 10,000 side effects’, 5.42 (5.30–5.54) / 5.54 (5.42–5.66) for vaccines with ‘1 in 1,000,000 side effects’, 4.71 (4.59–4.83) / 4.93 (4.81–5.05) for vaccines with ‘55% effectiveness’, 5.73 (5.61–5.85) / 5.78 (5.66–5.90) for vaccines with ‘95% effectiveness’, 5.79 (5.66–5.92) / 5.64 (5.51–5.77) for vaccines from Germany, and, 4.71 (4.57–4.85) / 4.91 (4.77–5.04) for vaccines from China (see Fig SI5 in S1 File and Table SI12-15 in S1 File).

Subgroup analyses: Vaccine hesitancy, and risk preference

To examine the moderating role of vaccine hesitancy and risk preferences on the effects of vaccine attribute on vaccine preferences, we estimated AMCEs and MMs for different subgroups on these dimensions. Just as with the main results, we depict MMs here, but subgroup AMCEs can be found in the Supporting information. We do note, though, that AMCEs, while useful for estimating the causal effect of attributes within subgroups, are subject to inferential errors when considering relative preferences across groups, a problem not encountered with MMs [44]. Unless otherwise indicated, we only describe differences of MMs where we also found a corresponding statistically significant AMCE of the attribute within subgroups (see Table SI17 in S1 File).

For vaccine hesitancy, we separated the sample through conducting a LCA on the 7-item battery of vaccine attitudes to identify groups of people with similar profiles of vaccine attitudes. Our LCA produced a three-group solution: strong vaccine supporters (38%), reserved (or concerned) vaccine acceptors (37%), and the vaccine hesitant (25%). MMs for each of these groups are plotted in Fig 3. In absolute terms, vaccine supporters are consistently more favorable to vaccines than reserved vaccine acceptors and the vaccine hesitant. In fact, vaccine supporters were substantially more likely to accept the vaccines with objectively the worst attributes–‘1 in 10,000’ side effects (5.64, [5.53–5.74]) or 55% effectiveness (5.30, [5.19–5.40])–than the vaccine hesitant were to accept the vaccines with objectively the best attributes–‘1 in 1,000,000’ side effects (4.39, [4.23–4.56]) or 95% effectiveness (4.62, [4.46–4.79]). This pattern is also observed across the other vaccine attributes: type, number already vaccinated, speed of population coverage, and cost. (Note that full numerical estimates of MMs and AMCEs (effect size) for likelihood of uptake and vaccine choice as well as a subgroup MM plot for vaccine choice are provided in the Supplementary Material (Fig SI6 in S1 File, Tables SI16-19 in S1 File).)

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Subgroup analysis: Differences across different vaccine hesitant groups.

Each dot and error bar represents the MM (and its 95% CI) of vaccine attributes on self-reported likelihood of uptake for the three groups of people with varying vaccine attitudes. The dashed line represents the grand mean.

https://doi.org/10.1371/journal.pone.0266003.g003

Additionally, we find that the preferences of subgroups higher in vaccine hesitancy are less responsive to vaccine attributes. For example, vaccine supporters are much more favorable towards vaccines of German origin (6.56, [6.45–6.66]) than they are towards those of Chinese origin (5.22, [5.10–5.34]) (a 26% increase), but this difference is smaller for vaccine acceptors (a 19% increase), and it is comparatively negligible for the vaccine hesitant (a 7% increase). Similarly, vaccine supporters are substantially more likely to take a vaccine with 95% (6.51, [6.41–6.62]) as opposed to 75% effectiveness (5.97, [5.86–6.10]), which is in turn more favorable than 55% effectiveness (5.29, [5.19–5.40]). This patterns holds for vaccine acceptors, though to a slightly lesser extent. For the vaccine hesitant, these differences are again much smaller, as is evident from comparing the variance within subgroups in Fig 3. Going from the least favorable (55% effectiveness) to the most favorable attribute level (95% effectiveness) brought a 1.2 point change in the average likelihood of accepting the vaccine for vaccine supporters (a 23% increase), a 0.8 point change for vaccine acceptors (a 18% increase), and a 0.5 point change in for vaccine hesitants (a 11% increase). Finally, vaccine supporters clearly favor vaccines with rarer side effects, and there is a smaller effect in the same direction for vaccine acceptors, but the vaccine hesitant show an inconsistent pattern of favorability regarding side effects (see Table SI16-17 in S1 File). Such differences in effects are far less apparent across the remaining attributes because the effects of these attributes are small to begin with.

We also examined whether the effects of the vaccine attributes vary across people with different levels of risk preferences. To capture risk preferences, respondents reported on a scale from 0 to 10 to what extent they avoid (0), or seek risks (10). We used the scale midpoint to split people into a risk-averse and risk-seeking group. In absolute terms, risk-seeking respondents are consistently more favorable to vaccines than risk-averse respondents. To a large extent, risk-seeking respondents are as likely to accept even the least optimal vaccines than risk-averse respondence are to accept the most optimal vaccines (see Fig 4). For example, risk seekers are more likely to accept a vaccine with `1 in 10,000’ chance of severe side effects (5.49, [5.39–5.60]) than the risk-averse are to accept a vaccine with an order magnitude decrease (`1 in 1,000,000’) chance for a severe side effect (4.91, [4.81–5.01]). Risk seekers are also approximately as likely to accept a vaccine with 55% effectiveness (5.19, [5.09–5.30]) as the risk-averse are to accept a vaccine with 95% effectiveness (5.17, [5.07–5.27]), and approximately as likely to accept a Chinese vaccine (5.16, [5.04–5.28]) as the risk-averse are to accept a German vaccine (5.15, [5.04–5.26]). (Full numerical estimates of MMs and AMCEs (effect sizes) for likelihood of uptake and vaccine choice as well as a subgroup MM plot for vaccine choice are provided in the Supplementary Material (Fig SI7 in S1 File, Tables SI20-23 in S1 File).)

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Subgroup analysis: Differences across different risk preference groups.

Each dot and error bar represents the MM (and its 95% CI) of vaccine attributes on self-reported likelihood of uptake for the two groups of people with varying risk preferences. The dashed line represents the grand mean.

https://doi.org/10.1371/journal.pone.0266003.g004

Unlike vaccine hesitancy, however, risk perceptions do not meaningfully moderate the effect of vaccine attributes on vaccine acceptance. For example, people scoring low and high on risk preferences are similarly more favorable towards vaccines of German origin than they are towards those of Chinese origin. The difference in MMs is approximately 0.9 points for both risk-averse and risk-seeking people (0.93 and 0.86) (Again, we only describe differences of MMs where we also found statistically significant AMCEs for these differences (see Table SI21 in S1 File)). Similarly, risk-averse and risk-seeking people are roughly equally more likely to take a vaccine with 95% as opposed to 75% (risk-averse difference approx. 0.44, risk-seeking 0.43), and 55% effectiveness (risk-averse difference approx. 0.92, risk-seeking 0.89). Finally, risk-averse and risk-seeking people similarly favor vaccines with rarer side effects. Similar tendencies hold, both absolutely and relatively, for the other vaccine attributes. These findings indicate that risk acceptance affects baseline favorability towards vaccines, but not responsiveness to vaccine attributes.