Kerem Cerit, Azim Selvi, Lea Cohausz
Does the percentage of the population already vaccinated and the infection rate of a virus influence people regarding their decision towards vaccination?
The Covid-19-pandemic and the vaccine to overcome it are on everyone's mind. Though the news of successful vaccine trials and its subsequent admission in the United States and then in the European Union were celebrated, a quite substantial percentage of the population seems to feel wary regarding the decision to get vaccinated themselves with the Süddeutsche Zeitung even dubbing it "Die Vertrauensfrage", i.e., "The question of trust" (Becker 2020, Charisius 2020). The question is, how willing people are to get vaccinated with a vaccine with limited long-term experience. Even people who are generally advocating vaccination seem uncertain regarding this specific situation (Becker 2020).
Previous vaccination research often assumes free-riding - when many people have been vaccinated already, a person does not have to get vaccinated him-/herself and will, nonetheless, be protected (Ibuka et al. 2014). However, the theory of free-riding is usually employed in cases where a large percentage of the population is vaccinated and has been for a while. Nonetheless, some people currently argue in a similar manner when saying that they will wait for others to be vaccinated first (Charisius 2020). Their reason for this decision is not explicitly to wait until it is no longer necessary for them to receive the vaccine - though implicitly they do assume that the current crisis gets resolved if others get vaccinated - but rather to be certain that there are no negative consequences of receiving it. This, we argue, is another form of free-riding at play (where the risks are unequally distributed), though, not the one traditionally researched.
Other scholars showed and argued for more complex processes relating to the decision in favor of receiving a vaccine (Streefland et al.1999, Streefland 2001). For example, people are influenced by other people in their social circle and are more likely to get vaccinated when they do (bandwagon effect). They are also more likely to get vaccinated if they are educated both on the disease's risks as well as on the vaccine. Again, the current situation only allows the latter in a limited manner and thus makes it very different from anything previously studied. The former, however, is probably very much given currently. Connected to this, an important aspect is the perceived risk for oneself and others and with the a new and potentially even more infectious Covid-19-variant discovered (Zeit 2020), people should be more willing to receive the vaccine, if this theory holds true.
Thus, given the previous research results and the current ever-changing, new situation, we ask the following question: Does the percentage of the population already vaccinated and the infection rate of a virus influence people regarding their decision towards vaccination?
As infection rate we define the number of people an infected person infects on average. The easier a disease spreads, the higher the risk of being affected by it oneself (and by spreading it to others).
Given the two kinds of free-riding introduced earlier, we argue that people are at first waiting for other people’s experiences with the vaccine before getting vaccinated themselves and that when many have been vaccinated already, they again do not vaccinate because of the traditional free-riding phenomenon. Let's compare two fictional people A and B. Person A who is not free-riding will always vaccinate, even if there is a low infection rate (population vaccinated is rather irrelevant for this Person). Person B on the other hand (who is a free-rider) will not get vaccinated if the percentage of the population vaccinated is so high, that there is essentially herd immunity. Person B will also not take the vaccine if the percentage of the population vaccinated is low, because in this situation Person B will also try to avoid potential negative consequences of the vaccine. The only scenario in which Person B would receive the vaccine is, if the vaccination is deployed to enough people to see, that there is no consequences, but it is still not enough to achieve herd immunity. Thus, our first hypothesis is:
H1: If the percentage of the vaccinated population increases up to a certain point, more individuals decide in favor of vaccination. From then on, the relationship turns negative (i.e. the more people are vaccinated, the fewer individuals decide in favor of vaccination).
Furthermore, in accordance with Streefland (2001) we assume that an increased risk awareness drives people towards vaccination. If we think about Person B again, we would assume, that, ceteris paribus, with higher infection rates the likelihood that he/she would decide in favor of the vaccine will get higher. Person B would at some point perceive the risk of getting the disease as higher than potential costs of the vaccine. Thus:
H2: If the infection rate is high, more people will get vaccinated.
To test the above hypotheses, we will conduct a vignette online experiment varying the two independent variables.
Thus, the dependent variable will be a dummy variable that shows whether a person says that he/she would get vaccinated or not. The independent variables will be the infection rate and the percentage of the population which is already vaccinated. There are three different conditions of the infection rate (1, 5, 10), and five different conditions for the percentage of the population already vaccinated (5%, 23%, 51%, 72%, 95%). The respondents will always receive a combination of the two treatment variables. First, the respondent will be assigned to one of the four treatment groups, three groups for the infection rates and one for control. The assignment of the infection rate which mentioned as part of a scenario in the treatment groups, is random. However, every respondent will be asked about if he/she would get vaccinated for every five conditions of the population already vaccinated. The order of these questions that related to the percentage of the population which is already vaccinated is also random. Since we always have a combination of those two treatment variables, the prediction about how many people would decide to take the vaccine is a result of a combination of those two variables. According to our hypothesis, most people would be in favor of the vaccine, if the infection rate is at 10 and the percentage of the population already vaccinated is at 51%. In this case, nearly everybody would say yes to the vaccine. If the infection rate is changed to 5, maybe only half of the people who would say yes to the vaccine in the ladder scenario, would say yes to the vaccine. The same process would be if the percentage of the population already vaccinated is reduced to 23% or increased to 72%. Furthermore, we predict that only a fifth of people would say yes to the vaccine if the infection rate would change from 5 to 1, or the population vaccinated goes from 23% to 5% or from 72% to 95% respectively.
Furthermore, we will control for certain aspects. Essentially, we try to control for variables which could directly influence the outcome variable or which could potentially have a mediating effect between the treatment and the outcome variable. Among the latter kind are measures for how involved people are in the lives of family members and whether they spend time volunteering. We assume that both of these variables will increase people's willingness to get vaccinated either to protect others (family) or because of a sense of community duty (volunteering) - and that this happens in particular, if the infection risk is high or few people have been vaccinated already as their own risk might be less salient to them. We also tentatively ask about their medical history. Regularly going to the doctor or having a chronic disease should increase people's willingness to vaccinate as should a generally health-aware lifestyle. Thus, it could be that they are less affected by the treatment variables than other participants. As medical information is a very sensitive topic and could lead to people dropping out of the survey, we only ask participants how they would rate their own health within a 1 to 10 scale and how regularly they exercise. Additionally, we ask for respondent's trust in the government and large companies. We believe that people distrustful to those entities generally are less likely to trust a new vaccine admitted by those entities. Thus, they may be even less willing to get vaccinated, when few others have been vaccinated than the average participant.
Demographic information (such as age, gender, etc.) might influence willingness to get vaccinated in general. Two of the variables we are especially interested in are the respondent's level of education and income. In particular the first one is often assumed to be important with a higher level of education is usually associated with a higher willingness to get vaccinated. Income is asked for by giving people categories to choose from. There, they can say where they would place themselves regarding their income in relation to that of others in their country (e.g., "middle upper-class income level"). This is done in that manner, because this is also a traditionally sensitive topic and because we aim to ask people in many different countries making a comparison otherwise difficult.
As already mentioned, we decided to test out our hypotheses using a vignette experiment. The questionnaire is split up in several parts. After an introduction to the survey where we mention the topic and expected duration, we explain important terms - such as infection rate - to them and test their understanding to make sure that the respondents will give informed answers. Afterwards, we explain the specific scenario to them which includes an infection rate as our treatment and, again, check their understanding. The participants will then be given a percentage of the population vaccinated in a random order and will be asked whether they would get vaccinated. After the participants answered all five questions with different percentages of the population, we asked them to give us a short explanation about the reasons behind their decisions about the vaccination in an open ended question. After collecting the necessary information for the main variables, in the second part , we asked respondents further questions related to control variables in four different sections. First section consists of socio-demographic information, we asked respondents about their gender, age, income, occupation, education level, country of origin and finally their involvement in the lives of their family members. In the politics section, we asked respondents their self positioning on the liberal-conservative continuum, their participation in any community service and their level of trust on different organs of the government and large companies. Finally, in the last two sections, we collect the information related to the participants' perceived health by asking them about their overall health, diet and exercise routine and information related to their risk behavior with a simple game. Overall we try to keep the survey short and clear to make the survey easy to understand and easy to answer for all respondents.
Our vignettes look like this:
"In your country, a virus breaks out. The virus is the main cause of disease B which can lead to several symptoms such as fever, coughing, nausea, severe headache. In later stages, Disease B proves to be deadly with a death rate of 2%.
According to the experts, the virus that causes Disease B has an infection rate of 1. This means that every infected person on average infects 1 other person.
There exists a vaccine that has already proven to be effective against the vaccine. This vaccine has already been given to 5% of the population."
The first paragraph of the vignette stays the same for all treatments. Only the name of the disease varies depending on the infection rate. In the middle and last paragraph, the infection rate and the vaccinated percentage of the population is altered.
Once we have a sufficient amount of data, we can start with the analysis. As we have a dichotomous dependent variable (vaccinate - 1, not vaccinate - 0), we choose logistic regression as our means of analysis. Every model must include our two treatments - infection rate and percentage of the population vaccinated - as independent variables. As these are categorical variables with ordering, they will be recoded to values 1 to 3 and 1 to 5 (1 representing the smallest and 5 the largest number) respectively. Furthermore, H1 requires a quadratic term of the percentage of vaccinated population to be able to observe the effect we expect. The control variables will also be preprocessed and added. For hypothesis 1 to be true, the variable percentage of the population vaccinated needs to be positive and significant and its squared term needs to be negative and significant on average and considering all control variables. If that is the case, it means that up to a certain point, people become more likely to vaccinate as the percentage of the vaccinated population rises. But after that point - as the percentage continues to rise - people become less willing to get vaccinated again, because they do not believe that it is necessary for them to do so as others have done it. Hence, they assume protection due to others not being able to carry the disease. For H2 to be true, the variable infection rate needs to be positive and significant. This means that when more people get infected on average, people become more willing to get vaccinated as. they perceive the risk to contract the disease as higher.
However, our vignette study also is limited in its informative value and there are some unresolved issues. First, in contrast to a classical experimental design, we simply ask respondents if they would get vaccinated. This means that we can't be really sure, if the respondents would act in accordance with their answer in the real-world. One reason for this might be, that the respondents could be influenced by social desirability effects and therefore alter their answers. This could go in either direction, meaning that some individuals may want to appear altruistic and tell us that they will get vaccinated even though they would not; Or that alternatively, some people could tell us that they would not get vaccinated, even though they would in real-life. The latter could be possible in situations were the percentage of the population vaccinated is very high and the infection rate very low. Respondents may not want to appear fearful and report to us that they would not get the vaccine. Either way, this would lead to biased results. Additionally, the vignette design does not consider all the various information and experiences the respondents have in a real-life setting, especially if we think about the current Covid-19 pandemic and the media's coverage (Gozzi et al., 2020).
Second, we can't confidently deduce the motivation behind the answers which the respondents give us about their vaccination decisions. Even though we include an option to explain why the respondents decided the way they did, it is highly unrealistic that we get a satisfying answer from most of the respondents in a way that we understand if the person acted – in his/her mind – selfishly or altruistic. And for those answers which are well explained, we also have the danger of social desirability. This means, that the interpretation of potential results is limited insofar as one can't interpret the underlying mechanisms of these results.
Third, we do not know where the exact "cutoff" point for every respondent is. We define the cutoff point as a certain value of the variables population vaccinated and infection rate at which a respondent changes his/her behavior. For example, in our first hypothesis, we state that at a certain percentage point of the vaccinated population, respondents would theoretically change their vaccination behavior. This would be a cutoff point. Since population vaccinated and infection rate can theoretically take on infinite numbers, we just hardly can predict where certain cutoff points are. We only ask for Y infection rate and Z percentages of the population vaccinated. Moreover, the interaction between our two independent variables makes it even harder to find the true cutoff point. This issue is even more exacerbated, when we think about, that every respondents has an individual cutoff point.
While we are aware of these problems, we nonetheless believe that our study is a good starting point to address limits of previous research that we highlighted in the introduction, e.g., we consider the - realistic yet barely researched - scenario of having a very low rate of the population previously vaccinated.
In this section, we will reflect on the process leading to the finished questionnaire and report. A large challenge was to construct the questionnaire out of the research question, to find appropriate questions, and to use the benefits while dealing with the disadvantages of online experiments such as quality of the answers and drop outs. Related to these disadvantages we took several measures. For example, we decided to limit the number of our questions and decided against using more than five treatments per participant so that people do not get bored and exit the survey or fast track the questions with random answers. We try to make our treatments such as infection rate and vaccinated percentage of the population as basic and clear as possible. Furthermore, we try to make sure that we only receive serious and informed responses by including two understanding checks. These understanding checks appear after we explain some basic terms and after we introduce the scenario. A participant cannot continue unless he or she answered the understanding checks correctly. Understanding checks may have the disadvantage that people feel annoyed by them, but as the quality of the answers is of great importance to us, we take this risk. Finally, we also aim to prevent people from dropping out by asking potentially sensitive questions (such as the questions regarding income and health) in a hopefully also sensitive way. We hope that this is sufficient to allow people to share their information. However, by asking the questions in the way we do, the information we receive back is less detailed and clear. We believe that the risk of too many people dropping out or them feeling uncomfortable is much greater than the risk of receiving lower quality information.
The technical implementation of the questionnaire itself was less of a challenge given the template provided by the nodegame and the material we had (such as projects from past years). Some issues such as allocating the treatments and saving the data take more time to figure out than the minor ones such as to make sure that everything looks fine or to implement the risk game, but not too long with an extensive reading of the nodegame wiki. It may have been helpful to have some more slides to look up fast (instead of having to rewatch the videos). This may have also helped us to structure our work process some more as we were not really certain that we were done after we had completed programming the questionnaire. Something like a checklist may have been of help.
Becker, K.B. (2020). "Die andere Hälfte überzeugen". Frankfurter Allgemeine Zeitung. [Last Accessed: 22.12.2020.] https://www.faz.net/aktuell/politik/inland/aerztin-heidrun-thais-will-zur-corona-impfung-motivieren-17109316.html.
Charisius, H. (2020). Die Vertrauensfrage. Süddeutsche Zeitung. [Last Accessed: 22.12.2020.] https://www.sueddeutsche.de/wissen/coronavirus-impfung-deutschland-1.5151552?reduced=true.
Gozzi, N., Tizzani, M., Starnini, M., Ciulla, F., Paolotti, D., Panisson, A., & Perra, N. (2020). Collective Response to Media Coverage of the COVID-19 Pandemic on Reddit and Wikipedia: Mixed-Methods Analysis. Journal of medical Internet research, 22(10), e21597.
Ibuka, Y., Li, M., Vietri, J., Chapman, G. B., & Galvani, A. P. (2014). Free-riding behavior in vaccination decisions: an experimental study. PloS one, 9(1), e87164.
Streefland, P. H. (2001). Public doubts about vaccination safety and resistance against vaccination. Health policy, 55(3), 159-172.
Streefland, P., Chowdhury, A. M. R., & Ramos-Jimenez, P. (1999). Patterns of vaccination acceptance. Social science & medicine, 49(12), 1705-1716.
Zeit Online. (2020). Britische Corona-Mutation wohl deutlich ansteckender. Zeit. [Last Accessed: 22.12.2020.] https://www.zeit.de/wissen/gesundheit/2020-12/christian-drosten-coronavirus-mutation-grossbritannien-ansteckung.