Communicating doctors’ consensus persistently increases COVID-19 vaccinations
Supplementary survey among doctors
To gather the views of doctors on COVID-19 vaccines, we implemented a survey in partnership with the CMC, to maximize coverage of the medical community. The survey was implemented online in February 2021. Because membership in the CMC is compulsory, the CMC has a list of contacts for the whole population of doctors in the country. The CMC approached all doctors who communicate with the CMC electronically (70%) and asked them to participate in a short survey, using the Qualtrics platform. Of doctors contacted, 9,650 (24%) answered the survey. The doctors in our sample work in all regions of the country, are on average 52 years of age, 64% are female individuals and 62% have more than 20 years of experience. A comparison of characteristics of doctors in our sample and of all doctors in the Czech Republic is presented in Supplementary Table 1.
Main experiment
Sample
Our main sample consisted of 2,101 participants of the longitudinal online data collection ‘Life during the pandemic’, organized by the authors in cooperation with PAQ Research and the NMS survey agency. In March 2020, the panel began to provide real-time data on developments in economic, health and social conditions during the COVID-19 pandemic. We used data from 12 consecutive waves of data collection conducted at 3–4-week intervals between mid-March and the end of November 2021.
The information intervention was implemented on 15 March 2021, which we labelled as wave 0. The sample from wave 0 is the ‘base sample’ (n = 2,101, 1,052 female participants and 1,049 male participants, mean age of 52.9 years (s.d. = 15.98), youngest 18 years of age, oldest 92 years of age). The base sample is broadly representative of the adult Czech population in terms of sex, age, education, region, municipality size, employment status before the COVID-19 pandemic, age × sex, and age × education. Prague and municipalities with more than 50,000 inhabitants are oversampled (boost 200%). Sample statistics are presented in Extended Data Table 1. The sample is close to being representative of the adult Czech population in terms of attitudes to COVID-19 vaccines. The development of the proportion of people getting vaccinated in the Control condition very closely mimics the actual vaccination rates in the Czech Republic (Extended Data Fig. 1), when we weighted the observations in our sample to be representative in terms of observable characteristics.
An important feature of the panel is that participants agreed to be interviewed regularly, and the response rate is high throughout the study: it ranges between 76% and 92% in individual follow-up waves, and is 86% for the last wave, implemented at the end of November 2021. Of participants, 1,212 (58%) took part in all 12 waves of data collection: they form the ‘fixed sample’. Consequently, in the analysis, we report the main results for (1) all participants from the base sample who responded in a given wave, which we denote ‘full sample’, and for (2) the ‘fixed sample’, composed of individuals who participated in all 12 waves, eliminating the potential role of differences in samples across waves and making it easier to gauge the dynamics of treatment effects.
Information intervention
In wave 0, the participants were randomly assigned to either the Consensus condition (n = 1,050) or the Control condition (n = 1,051). In the Consensus condition, they were informed that the CMC conducted a large survey of almost 10,000 doctors from all parts of the country to collect their views on COVID-19 vaccines. They were also informed that the views were similar for doctors of different genders, ages and regions. Then, the participants were shown three charts displaying the distribution of responses of doctors regarding their trust in the vaccines, willingness to get vaccinated themselves and intentions to recommend the vaccine to their patients. Each of the charts was supplemented by a short written summary. The exact wording and the charts are provided in Section 3.3 of the Supplementary Information. In the Control condition, the participants did not receive any information about the survey of medical doctors.
Data
Before the information intervention in wave 0, we elicited prior beliefs about doctor’s views to quantify misperceptions about doctors’ opinions. Specifically, the participants were asked to estimate (1) the percentage of doctors in the Czech Republic who trust the approved vaccines, and (2) the percentage of doctors who are either vaccinated or intend to get vaccinated themselves. Later, in wave 1, we elicited posterior beliefs to estimate whether people in the Consensus condition actually updated their beliefs about doctors’ views based on the information provided. In each of the 12 waves, we asked respondents to report whether they got vaccinated against COVID-19. The main outcome variable ‘vaccinated’ is equal to one if the respondent reported having obtained at least one dose of a vaccine against COVID-19.
In the analysis, we report two main regression specifications: (1) a linear probability regression controlling for pre-registered covariates: gender, age (6 categories), household size, number of children, region (14 regions), town size (7 categories), education (4 categories), economic status (7 categories), household income (11 categories) and prior vaccination intentions, and (2) a double-selection LASSO linear regression selecting from a wider set of controls in Extended Data Table 1, including prior vaccine uptake and beliefs about the views of doctors.
Additional data to verify vaccination status
We collected two sets of additional data to verify the reported vaccination status in the main dataset.
Third-party verification
First, we used data collected for us by a third, independent party. We took advantage of the fact that different survey agencies have access to the panel our respondents are sampled from (the Czech National Panel). Although the main data collection was implemented by one agency (NMS), we partnered with another agency (STEM/MARK) to include a question on vaccination status in a survey implemented on its behalf among the same sample. As the survey agency, graphical interface and topic of the survey were different from our main data collection, we believe that respondents considered the two surveys to be completely independent of each other, and thus experimenter demand unlikely had a role in the second survey. The response rate was high (92.8%) and independent of the treatment (Extended Data Table 6). Out of 1,801 participants in wave 11, 1,672 also took part in the third-party verification survey implemented 2 weeks later. This allowed us to compare reported vaccination status at the individual level for a vast majority of our sample, and to test whether Consensus affects the level of consistency in reporting of being vaccinated across surveys.
Certificate verification
The second verification links the reported vaccination status with an official proof of vaccination. We exploited the fact that all vaccinated people receive an EU Digital COVID certificate issued by the Czech Ministry of Health, which was often used as a screening tool at the time of data collection. We collected the data on vaccination certificates among respondents from our full sample who (1) participated in wave 11, and (2) reported to have at least one dose of the COVID-19 vaccine in wave 11 (n = 1,414). We asked respondents whether they had the certificate with them. Of participants, 96% confirmed that they had the certificate with them, and this proportion is very similar across the Consensus and Control conditions (χ2(1,n = 1,414) = 0.999, P = 0.318). Those with a certificate were asked to type in several specific pieces of information about the applied vaccine that are unlikely to be known by someone without a certificate (for example, the correct answer for those who got a vaccine from Pfizer/Biontech is ‘SARS-CoV-2 mRNA’). Assessment of the typed text by independent raters suggests that, conditional on their having the certificate, more than 94% of respondents actually looked at the certificate when responding to our detailed questions. This rate is again very similar across conditions (χ2(1,n = 1,364) = 0.473, P = 0.492).
More details about both verification procedures and results are in the Supplementary Information.
Ethics approval
The research study was approved by the Commission for Ethics in Research of the Faculty of Social Sciences of Charles University. Participation was voluntary and all respondents provided their consent to participate in the survey.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.
