Vaccine-hesitant people misperceive the social norm of vaccination

Abstract Vaccine hesitancy is one of the main threats to global health, as became clear once more during the COVID-19 pandemic. Vaccination campaigns could benefit from appeals to social norms to promote vaccination, but without awareness of the social norm in place any intervention relying on social norms may backfire. We present a two-step approach of social norm diagnosis and intervention that identifies both whether a vaccination norm exists or develops over time and corrects misperceptions. In two studies (N=887 and N=412) conducted in Rome, Italy from June to August 2021 (during the first COVID-19 vaccination campaign), we show that vaccine-hesitant people strongly underestimated vaccine acceptance rates for COVID-19 despite increases in region-wide vaccination rates. This suggests a false consensus bias on the social norm of vaccination. We presented a subgroup of vaccine-hesitant people with the accurate vaccine acceptance rates (both planned uptake and vaccine approval) and tested if this social information would lower their vaccine hesitancy. We do not find clear effects, most likely because of the introduction of the COVID-19 health certificate (the “green pass”) that was implemented during our data collection. The green pass reduced both misperceptions in the social norm and vaccine hesitancy, thus undermining our treatment effect. We conclude that to alleviate misperceptions on the social norm of vaccination in early stages of the vaccination campaign governments and media should report not just the current vaccination rate, but also about vaccination intentions and approval.

For Study 1, the same survey was conducted among a new stratified sample of respondents living in Rome (Italy) every two weeks between June 6 and August 1, 2021. To minimize differences between respondents in the extent to which they were informed about recent Covid-19 developments, we provided information about recent Covid-19 developments (infections, dates, and vaccinations) at the beginning of the survey (see Table S1). A dataset with the dates and numbers that were used in each wave is available on OSF at https://osf.io/xp2h6. To diagnose norms, we asked about individual behavior, empirical expectations, personal normative beliefs, and normative expectations (see Table S2). The questions about empirical and normative expectations were incentivized. The person whose estimate to all of the social expectations questions in the survey was closest to the reported behaviors and beliefs earned an additional 25 Euro. We ask you to make an estimate of how many people have answered Yes to the previous question (Yes, I already got vaccinated or Yes, I will get vaccinated when it is my turn).
The estimate -if accurate -allows you to earn more money from participating in this survey. The participant that is most accurate in his/her estimates to this and the other questions obtains an additional 25 Euro.
1 0-10% 2 10-20% 3 20-30% 4 30-40% 5 40-50% 6 50-60% 7 60-70% 8 70-80% 9 80-90% 10 90-100% According to you, what is the most frequent answer given by people in Rome participating in this survey like you have given to the question "How appropriate do you find the decision of Person A not to get vaccinated?" We ask you to make an estimate that -if accurate -allows you to earn more money from participating in this survey. The participant that is most accurate in his/her estimates to this and the other questions obtains an additional 25 Euro.
1 Extremely inappropriate 2 Rather inappropriate 3 Slightly inappropriate 4 Slightly appropriate 5 Rather appropriate 6 Extremely appropriate Note: 1 'I prefer not to answer' was coded as missing.
In the analyses we controlled for gender (female or not), age, nationality (Italian or not), perceived health, whether the respondent has been infected with Covid-19 in the past, the perceived risk of Covid-19, and the change in the number of infections in Lazio compared to the week before (as reported at the start of the survey). Perceived health was measured asking respondents to rate their personal health on a scale from 1 ('Very poor') to 6 ('Very healthy'). The perceived risk was measured through four items: 'The Coronavirus (Covid-19) will affect very many people in Rome', 'I will probably get sick with the Coronavirus', 'The Coronavirus is dangerous', and 'The current infection rate of the Coronavirus scares me'. Respondents could indicate their answer on a five-point Likert scale from 1 'Strongly disagree' to 5 'Strongly agree'. The four items had an acceptable Cronbach's α reliability (α = 0.695) and we took the average to create a single perceived risk scale. Finally, we controlled for the difference in the number of infections in that week compared to the week before as communicated to respondents at the beginning of the survey (Table S1). See Table S3 for the summary statistics of all variables used in the analyses.

S1.2 Generalizability
To test the external validity and generalizability of our results, we compared the findings to those of two other cross-country data collections. Data from the MIT Covid-19 beliefs survey conducted in 67 countries 1 demonstrate that while both vaccine acceptance and empirical expectations about vaccine acceptance vary widely across countries, in all countries worldwide people on average underestimate the degree of misperception by at least 10% between October 2020 and March 2021 (see [1]). Using their aggregate data for Italy 2 , we estimated the misperception between estimated and reported vaccine acceptance in Italy for this period (see Figure S1). Before the vaccination program started (from October 20 to December 7), empirical expectations correctly estimated nation-wide vaccine acceptance (EE = 63%, acceptance = 62%). As soon as the vaccination program starts, empirical expectations underestimate vaccine acceptance. On December 27 2020, at the start of the vaccination program, acceptance jumped to 72.4% while empirical expectations were still 63.8% (i.e., an average misperception of 8.6%). By March 15 2021, acceptance was 81.4% (similar to the 82% on June 8 2021 in our results) and expectations 74.5% (reflecting a misperception of 6.9%). While the empirical expectations are higher than the 63% reported by people in Rome on June 8 2021 in our survey, the misperception is substantial over the full study period also using this data source.  The second data source is the Periscope Survey [3], that collected information about vaccine acceptance and empirical expectations of acceptance in Bulgaria, France, Italy, Poland, Spain, and Sweden in June 2021 (around the same time as Waves 1 and 2 of our Study 1). We estimated the misperception between aggregate acceptance and individual empirical expectations for each of the six countries (see Figure S2. We find significant underestimation in the empirical expectations of vaccine acceptance for Poland (% accepting = 56.0%, EE = 35.1%, misperception = 20.9, p < 0.001), France (% accepting = 66.1%, EE = 50.3%, misperception = 15.8, p < 0.001), Spain (% accepting = 86.2, EE = 70.8, misperception = 15.3, p < 0.001), Italy (% accepting = 74.2%, EE = 67.8%, misperception = 6.4, p < 0.001), and Sweden (% accepting = 75.7, EE = 73.2, misperception = 2.5, p = 0.014). For Bulgaria, on the other hand, we find a significant overestimation, which is due to the low overall vaccine acceptance (% accepting = 26.5%, EE = 47.6%, misperception = -21.1, p < 0.001).

S2 Supplementary materials to Study 2 S2.1 Methods
In Study 2, N = 192 vaccine refusing and N = 220 undecided people were randomly assigned to one of four treatment conditions. After reading one of four norm-based messages, they had to answer five questions of the Oxford Covid-19 Vaccine Hesitancy scale (Table S3). [2] The 'Don't Know' category was coded as missing. The five items had a very high Cronbach's α reliability (α = 0.951). We used the average of the five items to create a vaccine hesitancy scale. In testing for treatment differences we used the same control variables as in Study 1 (gender, age, nationality, perceived health, history of Covid-19 infection, perceived risk, and change in the number of infections. For the latter, we used information of new Covid-19 infections during the four weeks of the data collection and took the number of cases in week 1 of Study 2 as a baseline (Table S4).

S2.2 Results
Before testing for treatment differences, we checked whether the estimation bias still existed by repeating the same OLS regression analyses that we ran also for Study 1 (Table S5). This is the case for both empirical and normative expectations, but non of the control variables explain this bias for the vaccine hesitant sample. Note: * p < 0.05, * * p < 0.01, * * * p < 0.001; 1 Reference category: undecided people.
As a robustness check, finally, we checked whether the treatment effects were consistent regardless of the hesitancy item used (see Figure S4. The effect was replicated for items 3, 4, and 5 of the vaccine hesitancy scale, but the difference between treatment 1 and 2 was not significant for items 1 and 2.