Vaccine hesitancy: is it fed by the media?

MASTER’S THESIS PUBLIC

ADMINISTRATION

Vaccine hesitancy: is it fed by the media?

Name: Elles de Vogel

Student number: S2097117

Track: International and European governance

Date: January 24, 2019

Course: Master thesis

Supervisor: Dr. C. H. J. M. Braun

Second reader: Dr. S. N. Giest

Abstract

This study examines the extent to which media coverage affects vaccine hesitancy. The presence of vaccine preventable diseases has been increasing during the last years. This phenomenon has been studied extensively and a list of determinants has been composed. One of these determinants of vaccine hesitancy is the media, through the media messages and opinions can be dispersed which can influence a person’s opinion, emotions and behavior. The influence of media has been studied in different contexts, however, other topics have shown an effect of media while studies regarding vaccine hesitancy show less definitive results. For this reason, I first study the effects of media coverage on vaccine hesitancy in a large-N country comparison that focuses on newspapers articles. Both media coverage and vaccine hesitancy can be affected by a large range of factors which is why the analysis includes several potential moderating or mediating factors and confounding variables. Looking at the current developments, it is expected that higher media coverage will result in lower vaccine coverage. The second part of this thesis dives into one country to find an explanation for the outcomes of the statistical analysis or possibly find a mechanism that causes the correlation between media and vaccine coverage. The study reveals that instead of the expected negative correlation, there is a positive correlation between media and vaccine coverage. Moreover, the results show that the wealth of a country mediates the relationship between media and vaccine coverage and, the effect is stronger in countries with lower levels of education. The results of the second part reveal the tone of media articles regarding vaccines and immunization in Spain. It shows that the government is highly present in the newspaper articles in that through the media they inform the public about the vaccines and about important events, like casualties due to a vaccine preventable disease. In other words, in contrast to the expected increase of negative media attention, the articles are rather positive or neutral which means that, based on this study, the negative effect of media on vaccine hesitancy is disproven. An explanation for this could be the fact that the public now has a wide range of media sources to choose from which allows them to read only the sources that confirm their preexisting beliefs. Additionally, it could be said that the importance of Goldstein et al.’s (2015) theory on the effectiveness of health communication applies well to this topic.

Content page

List of figures and tables ...

1. Introduction ... 5 2. Theory ... 7 2.1 Literature review ... 7 2.2 Theoretical framework ... 11 3. Research design ... 18 3.1 Unit of analysis ... 18

3.2 Case selection: large-N analysis ... 18

3.3 Case selection: in-depth content analysis ... 19

3.4 Independent variable and dependent variable ... 20

3.5 Other variables ... 22 3.6 Method of analysis ... 24 3.7 Limitations ... 25 4. Large-N analysis ... 27 4.1 Empirical findings ... 27 4.2 Analysis ... 35

5. In-depth content analysis ... 38

5.1 Empirical findings ... 38 5.2 Analysis ... 41 6. Conclusion ... 43 7. References ... 48 8. Appendix A ... 52 9. Appendix B ... 55

List of figures and tables

Figure 1: SAGE WG, WHO report on vaccine hesitancy, 2014 ... 8

Figure 2: Dependency model (Ball-Rokeach & DeFleur, 1976)... 13

Figure 3: Conceptual model ... 17

Figure 4-8: Scatterplots of vaccine coverage and independent variables ... 27

Figure 9: Example of mediation ... 31

Figure 10: Example of moderation ... 32

Figure 11: Appendix B assumptions large-N; normal P-P Plot ... 56

Table 1: Pearson’s correlation ... 28

Table 2: Simple regression analysis; model summary ... 29

Table 3: Simple regression analysis; coefficients ... 30

Table 4: Multiple regression analysis; model summary ... 30

Table 5: Multiple regression analysis; coefficients ... 31

Table 6: Moderator analysis; coefficients LDI... 33

Table 7: Moderator analysis; coefficients GDP ... 34

Table 8: Moderator analysis; coefficients level of education... 34

Table 9: Moderator analysis; coefficients level of education (low) ... 34

Table 10: Media content analysis; amount of articles and statements ... 38

Table 11: Media content analysis; tonality ... 39

Table 12: Media content analysis; actor type ... 40

Table 13: Appendix A media content analysis; codes ... 54

Table 14: Appendix B assumptions large-N; model summary ... 55

Table 15: Appendix B assumptions large-N; coefficients ... 55

5

1. Introduction

In 2011 an expert working group from the World Health Organization (WHO) first realized the extent of reported reluctance regarding vaccination in both developed and developing countries (WHO, 2014). Since then, countless efforts have been made to come up with fitting strategies that would increase confidence in vaccination or immunization programs. The problem was defined as vaccine hesitancy: “.. [a] delay in acceptance or refusal of vaccines despite availability of vaccination services. Vaccine hesitancy is complex and context specific varying across time, place and vaccines. It includes factors such as complacency, convenience and confidence (WHO, 2018, para. 1)”. The importance of addressing vaccine hesitancy is high because to prevent an outbreak of vaccine preventable diseases a certain level of coverage through vaccination is crucial (Salmon, Dudley, Glanz and Omer, 2015). A study by the WHO (2014) reveals the many aspects that influence vaccine hesitancy, one of these is communication and media which refers to traditional media, social media and anti-vaccination activities. Media analyses regarding vaccine hesitancy have been carried out on a national level but a comparative study of countries with different levels of vaccine hesitancy has not been done yet. Moreover, despite all these studies, several countries in the EU are currently facing outbreaks of vaccine preventable diseases. Analyzing the main concerns expressed through the media regarding vaccines could provide input for strategies aimed to decrease vaccine hesitancy and can both test and enrich the field of research into the effects of media.

One possible reason for the ongoing increase could be negative publicity in the media. As the literature review will demonstrate later, there is an extensive list of factors that cause vaccine hesitancy. Within this list, communication and media are mentioned as one of these determinants. However, many other factors, like anti- or pro-vaccination lobbies, politics, the perception of the pharmaceutical industry, other people’s experiences, etcetera, can be dispersed through the media. This makes media both an influential factor and a mechanism through which the other factors reach the public. It would be interesting to know whether stakeholders decrease or increase vaccine hesitancy through expressing themselves in the media. In their article, Dubé et al. (2013) clearly emphasize the power of the media to keep vaccination fears and scares alive, this happens even when scientific research has disproven its arguments. Therefore, this thesis will focus on media coverage of vaccine related topics in newspaper articles. The main question to be answered is: to what extent does media coverage

6 This will be analyzed in a two-part study that starts with a large-N statistical analysis. The data for this first part will be collected from databases of reliable organizations like the WHO, UNICEF and the World Bank. This data will be used to conduct several tests; a Pearson’s correlation test, to find out to what extent the variables relate; a multiple linear regression, to test the predictive value of media coverage and other variables on vaccine coverage; and a moderator and mediator analysis, to see which other variables might affect the relationship between media and vaccine coverage. These other variables are included to find out how influential other country characteristics are. To measure this, several interaction-effects of other relevant variables will be studied in the analysis to reach a comprehensive conclusion on the effects of media on vaccine coverage. The second part of this study consists of a media content analysis of Spanish newspaper articles. Data will be collected from the Factiva database (n.d.) and coded in an excel document to enable analysis of the tone and content of the relevant articles. The data consists of newspaper articles from three politically different positioned newspapers. Within this in-depth analysis, the objective is to explain the results of the large-N analysis. The argument that is developed in this study suggests that there will be a negative correlation between media coverage and vaccine coverage because there is a chance that even positive or balanced media coverage can lead to vaccine hesitancy. It is expected that this effect will be specifically stronger in rich, highly educated and more democratic countries.

The first chapter of this thesis first lays out the different theoretical perspectives regarding vaccine hesitancy and the effects of media. Then, these two phenomena will be merged into a theoretical framework and hypotheses will be formulated. Second, the research design of both the large-N and in-depth analysis will be outlined. Third, the empirical findings and the analysis of the large-N statistical analysis will be displayed followed by the findings and analysis of the in-depth content analysis. Finally, conclusions will be drawn and the thesis process evaluated.

7

2. Theory

2.1 Literature review

2.1.1 Explaining vaccine hesitancy

As aforementioned the term vaccine hesitancy refers to the decreasing vaccine rates despite regular availability of vaccines. According to the Strategic Advisory Group of Experts Working Group on Immunization (SAGE WG), vaccination is beneficial on both the individual and community level (WHO, 2014). This means that if a vaccine is used throughout a community it could possibly eliminate or even eradicate vaccine preventable diseases. The SAGE WG also discovered that the problem of vaccine hesitancy is not restricted to one region or part of a community, however, areas in the world where availability of vaccines and adjacent services are low do not fall within the scope of this problem definition (WHO, 2014). Kaufman (2010) argues that in the United States, vaccine hesitancy can be traced back to the 1980s when a connection was made between vaccinations and autism. This connection has not been scientifically proven, however, this hesitance has not left since (Kaufman, 2010).

The SAGE WG has reviewed several models that describe the determinants of vaccine hesitancy and consequently designed a matrix of determinants of their own (WHO, 2014). This is a complex and extensive list of factors which consists of contextual influences, individual/social group influences and vaccine and vaccination-specific issues. Figure 1 below shows a more detailed description of these three factors, one specifically is ‘communication and media environment’, among these contextual influences it also includes a few influential actors that might express themselves through media (WHO, 2014). According to the SAGE WG (2014), this model is highly useful, globally applicable and moderately complex. The authors stress that these determinants are based on research, practice, surveys and discussions with experts (WHO, 2014).

8 Several factors that require an explanation will be discussed below. The contextual influences, as mentioned above, are most relevant to this study. The first factor is about both traditional and social media, where both negative and positive attitudes towards vaccines can easily be expressed and shared (WHO, 2014). This fits well with the second factor that is “Influential leaders, gatekeepers and anti- or pro-vaccination lobbies” (WHO, 2014, p.12), these actors can use media to convey their message. The third factor is the influence of history, this means that negative experiences with vaccines, trials or even less related topics can increase mistrust in the public (WHO, 2014). Fourth, religion, culture, socio-economic and gender issues can influence a person’s attitude towards vaccines. For instance, in some religions it is forbidden to receive vaccines and in other cultures men are not allowed to vaccinate children (WHO, 2014). The fifth factor ‘politics’ refers to the potentially negative effect of mandatory vaccines. Next, the geographical distance between the family and the health center can cause hesitation if the distance is too large. Lastly, the pharmaceutical industry is responsible for a large part of the provision of vaccines and the public might have doubts concerning the motivations of these

9 corporations being profits instead of public health. This can even spill over into mistrust in the government because they are being lobbied by the corporations (WHO, 2014).

The second main group of determinants, described by the SAGE WG, are the individual and group influences (WHO, 2014). The first factor – past experiences with vaccines – is quite self-explanatory, this may also include the experiences of acquaintances; second, beliefs and attitudes refer to, for instance, the belief that diseases are needed to build immunity or other types of health care are as efficient as vaccines; third, the level of knowledge and awareness on the topic is crucial, this can also be influenced by the media; fourth, experiences with and trust in the healthcare system and the government can affect hesitancy; next, perceived risks or benefits of vaccines are highly influential, if a parent does not estimate high chances of the child to become infected with a disease he or she will be less likely to let the child be vaccinated; finally, the presence of norms that see immunization as normal and self-explanatory or that vaccines are unnecessary or even damaging will influence vaccine hesitancy (WHO, 2014). Additionally, the third group of determinants is ‘vaccine and vaccination specific issues’ (WHO, 2014). This determinant also consists of a risk-benefit aspect, however, this is more focused on scientific evidence. For instance, vaccines that were suspended due to unforeseen consequences like narcolepsy following the 2011 vaccines against H1N1 (WHO, 2014). Second, the introduction of new vaccines might evoke hesitancy because it might be believed that it is not tested sufficiently or long term consequences cannot be visible yet. Next, the administration of the vaccine can affect hesitancy in that oral administration will be preferred over injection for people that are afraid of needles or do not trust the capability of health workers. Fourth, the type of program/campaign surrounding the vaccine and the mode of delivery of the vaccine plays a role. The fifth factor refers to the supply of the vaccines, this is related to trust in the source of the vaccine supply and in the health center where the vaccine is administered. Next, having to receive multiple vaccinations or receiving it at a particular age might have a negative effect on vaccine hesitancy because it requires people to adhere to a schedule. Seventh, costs vaccines or related costs might be too high for certain people, in turn, free vaccines might lead to diminishing trust in the effectiveness of the vaccine. Lastly, healthcare professionals play a large role in convincing people to trust or distrust a vaccine (WHO, 2014).

2.1.2 Why the relationship between media and vaccine hesitancy

Besides the mention of communication in the SAGE WG model above, it is also categorized as a tool that can both increase and decrease vaccine hesitancy (WHO, 2014). One type is focused

10 on communication from the government to the public. This type was described by Goldstein et al. (2015) as health communication which is a very broad concept regarding the outreach to the public and to communities to exchange health-related information to create change or adapt behavior of the public and other stakeholders. The authors refer to all types of media as tools to accomplish effective health communication but it is also a process that needs to be planned carefully and requires a proactive attitude from the initiators. They mention that a concern exists that creating media attention on vaccine hesitancy will legitimize it through familiarity which could worsen the situation. Another observation is that the presence of vaccine opposition is often accompanied by a political figure that functions as a leader. The article concludes that health communication can help to promote vaccine uptake if done properly, a question left unanswered is what the effect is of communication from stakeholders that are not part of a strategy to decrease hesitancy (Goldstein et al., 2015). As mentioned in the introduction, media attention can independently affect attitudes towards vaccination as well as it can be used as a tool for communication. This means that other determinants, for instance others’ experiences, risk/benefit analyses and politics, can be shared with the public through the media. The fact that vaccine coverage has decreased over the last years could then easily be due to increasing negative communication through the media. An empirical study was conducted into the determinants of vaccine hesitancy on a European scale by the WHO/UNICEF (WHO, 2014). They asked immunization managers on a national level to fill out a small survey, this showed that 31 out of 45 European countries had assessed vaccine hesitancy in their country. Furthermore, 16 of 45 countries responded to the question regarding reasons for vaccine hesitancy of which the top three answers were: “1) beliefs, attitudes, motivation about health and prevention, 2) risk/benefit of vaccines (perceived risks, experiences (heuristics)), and 3) communication and media environment (WHO, 2014, p.27)”. The first two reasons can be influenced by stakeholders through the media, this makes media a comprehensive source of influence from all kinds of stakeholders.

A more focused study by Smith et al. (2008) assessed the correlation between media coverage and vaccine hesitancy of one specific vaccine (measles-mumps-rubella vaccine) in the United States. The authors collected data on vaccine rates between 1995 and 2004, this showed an increase in hesitancy during the same time that the link between autism and vaccinations became prevalent in scientific literature. However, by the time the issue received increased media attention, the vaccination rate had already increased again. This led them to conclude that media coverage only has a limited influence on vaccine uptake (Smith et al., 2008). This

11 study was conducted in the time period when vaccine hesitancy was a novel concept in the United States, this puts the results in perspective and sparks interest in a study that analyzes this in current times. A study by Handy et al. (2017) looked into the effect of media on vaccination in three countries: the Dominican Republic, Botswana and Greece. Their main finding was that media does influence attitudes towards vaccination but citizens would consequently take their doubts and questions to healthcare workers (Handy et al., 2017). Another interesting study was carried out by Dixon and Clarke (2013), they concluded that newspaper articles that offered both arguments in favor and against the autism-vaccine link affected vaccine hesitancy negatively. In fact, balanced articles increased uncertainty in the readers because they believed that even medical experts were divided on the issue (Dixon & Clarke, 2013).

The determinants of vaccine hesitancy indicate that media can be quite influential in changing the opinion of the public regarding vaccines. However, there have not been many studies into this topic and the studies that have been conducted are often small-scale and do not show large effects. For this reason this study will examine the effect of media, specifically coverage in

newspapers, on a large scale to measure the significance of the effect.

2.2 Theoretical framework

As aforementioned, vaccine hesitancy is determined by a lot of factors. One of these is communication and the media environment but many of the other determinants can be influenced through the media as well. When looking at the literature regarding the effect of media, it shows that while scholars disagree, there is agreement on the fact that it does have, too some extent, an effect on public opinion. Scholars argue about the size of the effect and whether it also affects the behavior of people. According to Reeves and de Vries (2016), several scholars have concluded that the effects of media are minimal, it merely reinforces preexisting opinions. A recent development might even decrease the effect more since the public now has a wide variety of media outlets to choose from. Reeves and de Vries (2016) argue that this allows for citizens to avoid media that conflicts with their personal beliefs. According to Goidel, Shields and Peffley (1997), early studies in the 1960s into the effects of mass media showed the same result; media solely strengthens existing attitudes instead of changing them. However, critics in the 1980s disagree and argue that dominant media messages limit the scope of political debates and eventually might be accepted as common sense (Goidel, Shields & Peffley, 1997). Towards the end of the twentieth century the evidence supporting the influence of mass media became larger and several models were designed. The first model that Goidel et al. (1997)

12 describe is the Receive, Accept, and Sample (RAS) model designed by Zaller (1992). The authors describe the model as follows: “the extent of exposure to dominant and countervailing information sources as well as the ability to resist dominant messages varies dramatically over time and across individuals” (Goidel et al., 1997, p.290). This means that the amount of exposure to media and the amount of knowledge a person has on the topic in general will affect the extent of mass media influence. This leads to the conclusion that the less political awareness on the topic a person has, the more likely they are to be influenced by the media (Goidel et al., 1997). Moreover, clear political partisanship will make a person less likely to be conveyed by media messages of the opposing political party, these political parties will presumably also differ in presence in the media (Goidel et al., 1997).

Another theory presented by Goidel et al. (1997) is priming theory. This theory states that “citizens with limited motivation and cognitive capacities” often depend on the information that is easily accessible, which would mainly consist of media messages (Goidel et al., 1997, p.293). According to the author, media has the ability to focus the attention of citizens on specific topics which leads them to perceive these as more important than others, however, this is regarding political judgements. Not all scholars that studied priming in the media agreed with the statements above. For instance, another study argued that citizens that are more knowledgeable on the topic are more likely to be affected by priming because they are more capable of storing the messages from the media and recall them later while making judgements (Goidel et al., 1997). This statement also contradicts the assumption of the RAS model concerning the level of knowledge of the reader.

A third model, the dependency model, was introduced in 1976 by Ball-Rokeach and DeFleur. This model requires taking into account the relationship between audience, media and society to fully analyze the effects of mass media. The first relationship is the dependence of audiences in developed societies on media (Ball-Rokeach & DeFleur, 1976). Dependency is conceptualized as the fulfilment of ones needs or objectives by the resources of the other; in the case of dependence on media this may be, looking for the best discount in a product or maintaining a connection to the social world (Ball-Rokeach & DeFleur, 1976). “The greater the need and consequently the stronger the dependency in such matters, the greater the likelihood

13 that the information supplied will alter various forms of audience cognitions, feelings, and behaviour” (Ball-Rokeach &

DeFleur, 1976, p.6). The authors acknowledge the growing functions of the media through technological developments but the centrality of different media sources differ. They argue that the broader the functions and the greater the centrality, the higher the dependence on that media source is (Ball-Rokeach & DeFleur, 1976). Moreover, if there is a lot of change, conflict and uncertainty within society, dependency on media

will increase. These two propositions, written by Ball-Rokeach and DeFleur (1976), form the dependency model (figure 2). The first effect that the authors explain in the category cognitive impact is ambiguity, the media has the ability to maintain and resolve ambiguity after unexpected events if dependence is high. Additionally, it can also influence attitude formation in that the media can steer the input that reaches the audience, for example, which political actors to focus on (Ball-Rokeach & DeFleur, 1976). The third cognitive effect is agenda-setting, this effect can best be explained by the priming theory where media has the ability to focus attention on specific topics (Goidel et al., 1997). The fourth effect is the expansion of people’s systems of beliefs, Ball-Rokeach and DeFleur (1976) argue that beliefs can consist of all types of categories like religion, politics and family. The amount of categories and the number of beliefs within those categories can be broadened when a person gains more knowledge, this can happen through the media (Ball-Rokeach & DeFleur, 1976). Finally, the last effect regards values, this includes basic beliefs on how one should act in the world and can only be slightly affected by media but nonetheless it is possible (Ball-Rokeach & DeFleur, 1976). The next category of effects is affective impact, this means that media can influence the feelings and emotions of the audience towards objects, people becoming numb to violence through long exposure or changing feelings towards other people (Ball-Rokeach & DeFleur, 1976). At last, the final effect category is behavioral which is the most relevant in this study of vaccine hesitancy. Ball-Rokeach and DeFleur (1976) discuss activation and de-activation, where the audience does or does not perform an act which would have another outcome if there was no

14 exposure to the media. A media announcement concerning a protest can have an activation effect for instance or, in this case, negative messages regarding vaccines might deactivate the audience which will decrease vaccine uptake. This may be a consequence of affective and cognitive effects (Ball-Rokeach & DeFleur, 1976).

The provision of information through the media is used by badly informed citizens as a shortcut to still be able to make decisions that seem well informed (Reeves & de Vries, 2016). Several empirical studies have been conducted in which the influence of media coverage is measured in a single case. One empirical study into the effects of media was written by Nisbet and Myers (2011) and focused on the effect of media in anti-American sentiments in the Middle East. The outcome clearly shows one important mechanism; social/political identities, as only specific identities are affected by specific media outlets (Nisbet & Myers, 2011). This is in accordance with the RAS model where political attitudes make a person more or less susceptible to specific media sources (Goidel et al., 1997). Another empirical research studied the influence of media on the attitudes towards welfare recipients after the Riots in England in 2011 (Reeves & de Vries, 2016). After a man was shot by the police, riots broke out in many large British cities with rioters mainly consisting of the urban poor. Reeves and de Vries (2016) studied the attitudes of British citizens towards welfare recipients before and after the riots that sparked a lot of media attention. According to Reeves and de Vries (2016), in that period the focus of the media quickly turned towards welfare policies and when politicians started to connect the riots to welfare policies it turned into an extensive debate regarding welfare and poverty. The results showed that before the riots there were no large differences between newspaper readers and non-readers, however, after the riots the newspaper readers began to express more doubt regarding the extent to which welfare beneficiaries deserved help in finding a job etcetera (Reeves & de Vries, 2016). Lastly, in his dissertation, Kirkpatrick (2006) confirmed that political views of the public are often shaped by frames used in media, there is a positive correlation between positive media outings and positive public opinion.

Based on this literature, I thus expect that media attention will have a negative effect on vaccine

coverage (Hypothesis 1).

Remarkably, in the previous paragraph several empirical studies have been outlined that analyze the effect of media on vaccine hesitancy. These studies show a much lower impact of media on vaccine hesitancy than other empirical studies into the effects of media. Because of this and because there are many other determinants that can affect vaccine coverage, the effect will most likely not be all-encompassing. The negative effect of media on vaccine coverage is

15 expected because it is assumed that only organized governmental campaigns will be filled with positive messaging in the media. This is not expected to be present in most countries which is why a negative effect will be more likely. Also, it is clear that the presence of vaccine preventable diseases has been increasing which indicates higher hesitancy. This, in turn, could have been caused be a deactivation effect from the media through negative statements.

Besides the expected relationship between media and vaccine coverage, there are other determinants that can possibly affect this relationship and adding these will increase the reliability of this study. For this reason, several determinants of both media coverage and vaccine hesitancy will be included in the analysis. For instance, for media coverage to be influential, it is essential that there is freedom of press in a country, this depends on the type of political system. Moreover, both education levels and the GDP per capita can affect the access of the public to media sources. It can be expected that countries with higher freedom of press, level of education and GDP will demonstrate a larger effect between media coverage and vaccine coverage. The political system of a country is expected to be the most influential since accurate media coverage requires freedom of media. If there is no freedom of press and the government is in favor of immunization, it could mean that negative messages regarding vaccines will be censored.

In other words, it is expected that the effect of media on vaccine coverage is stronger for

countries with higher freedom of speech than for countries with lower freedom of speech value

(Hypothesis 2a).

The assumption made in the literature on priming theory is that higher educated people will be more affected by media than lower educated people. Additionally, a study carried out by Elvestad and Blekesaune (2008) on the influence of individual and national characteristics on newspaper reading behavior, shows that there is a positive correlation between education level and the amount of time spend reading newspapers.

Based on this information it can be expected that the effect of media on vaccine coverage is

stronger for countries with a higher level of education than for countries with a lower level of education (Hypothesis 2b).

This assumption is used for this thesis because it specifically focuses on newspapers articles that address vaccination. This hypothesis contradicts the argument of the RAS model which states that people with less knowledge on the topic are more easily influenced by media. However, due to the fact that the aforementioned authors state that lower educated citizens

16 spend less time reading the newspaper, it is expected that this relationship – from the RAS model – will not show when analyzing media coverage in newspapers only. Regarding the general effect of education, there could also be an effect of education on vaccine coverage itself. This is because it relates to several determinants from the SAGE WG model (2014) in that general knowledge and awareness about the topic is mentioned to have the ability to influence vaccine hesitancy and scientific risk and benefit analyses can have this effect as well.

Another determinant that I believe is capable of influencing the relationship between media and vaccine coverage is GDP per capita. This effect has also been researched by Elvestad and Blekesaune (2008) and it shows a positive correlation between income and time spend reading newspapers. As with level of education, there also is a possible correlation between GDP per capita and vaccine coverage. In countries with a low GDP there might not be enough availability of vaccines for hesitancy to be able. Additionally, as the SAGE WG model of determinants (2014) show, the costs of vaccines have an effect on the hesitancy of caretakers. Countries with a lower GDP therefore will probably show low numbers of vaccine coverage which is due to unavailability of vaccines or to high costs of immunization.

Thus, it is expected that the effect of media on vaccine coverage is stronger for countries with

a higher GDP per capita than for countries with a lower GDP per capita (Hypothesis 2c).

Moreover, the last two factors that are included, are potential confounding variables. As aforementioned, it was difficult to select influential determinants on a national level which is why only two have been included. Religion is a factor that the SAGE WG lists as a determinant of vaccine hesitancy. As mentioned in the literature review, some religions have rules and norms that concern immunization which might cause a lower vaccine uptake. Last, several factors refer to more practical issues that can influence hesitancy and some show the importance of healthcare workers in the decision of caretakers to let children be vaccinated. Trust in healthcare workers, supply of the vaccine, physical distance to the vaccine and the attitude and knowledge base of professionals are all influential. This will give an indication of how easy it is to get vaccinated and how much time and energy a ‘patient’ will receive. It will be measured on an abstract level by looking at the physician density of a country.

There also is a possibility that there is no relationship found between media coverage and vaccine hesitancy, even when controlling for the political system, level of education or GDP. This is a possibility that is quite likely when looking at the determinants from the SAGE WG model (2014), there are many determinants, and several can vary even on an individual level.

17 If this is the case, the second phase of this thesis will look into the media coverage of one country and study its contents and the stakeholders that are present. In terms of this content analysis there have been assumptions discussed in the literature. For example, Goldstein et al. (2015) argue that there is a chance that even increased positive media attention can lead to an increase in vaccine hesitancy due to familiarity. However, this statement is the opposite of what all scholarship regarding the influence of media argues since this would mean that, in some case of all positive media, citizens will do or believe the opposite of what is mentioned in the media. For this reason this particular assumption will not be studied in this thesis. Next, Dixon and Clarke (2013) state that publishing articles that show both the negative and positive arguments around immunizations might also increase hesitancy because readers will think that even scientists are divided on the issue.

Based on this I pose a third hypothesis; countries that show a high amount of balanced media

attention will have lower vaccine coverage (Hypothesis 3).

2.2.1 Conceptual model:

The figure below (figure 2) shows all the indicators that will be studied during this thesis. As aforementioned, these factors are partly based on the SAGE model (WHO, 2014) since it provides a comprehensive view of all determinants of vaccine hesitancy. This conceptual model shows that the political system can influence the amount of media coverage; that the level of education and the GDP per capita can affect both media coverage and vaccine hesitancy; and finally, that religion and physician density can both affect vaccine hesitancy.

18

3. Research design

As briefly mentioned before, the chosen research method is a comparative country analysis of two parts; first, several countries will be analyzed using a large-N analysis of all countries for which vaccine coverage can be determined; second, based on the first part one country will be selected for an in-depth qualitative media analysis to reveal the type of stakeholders involved and their statements. The first paragraph will describe the unit of analysis, followed by the case selection of both the large-N analysis and the media content analysis. The next paragraph focuses on the operationalization and the data collection method of the dependent and independent variables. This will be done for the other variables – GDP, LDI, level of education, physician density and religion – in the following paragraph. Next, the method of analysis will be outlined and finally, the limitations of this research design will be listed.

3.1 Unit of analysis

This thesis analyzes vaccine hesitancy on a national level, this level is chosen instead of the international or regional level because vaccination or immunization policies are a matter of national governments. Another possible unit of analysis would be to study vaccine hesitancy on an individual level, even though this would be quite interesting it is difficult to measure the impact of media coverage in this unit. Therefore, the focus lies on the national level where it is more feasible to collect data on vaccine coverage and media coverage. It takes the shape of a cross-sectional research design, which means that the cases will be analyzed at one moment in time.

3.2 Case selection: large-N analysis

For the large-N analysis the two main variables are media coverage and vaccine hesitancy. Additionally, two variables are included as possible confounding variables, they are chosen based on the SAGE WG model (WHO, 2014). Many of the determinants from that model are individual or vaccine specific which is why only religion and physician density were selected. Three other variables will be analyzed as potential mediating or moderating variables, GDP and education can be both but the political system can only be a moderator. I rely on information from several online databases to collect the data for this large-N analysis. In order to realize a high level of generalizability and prevent selection bias, I have decided to include all countries – of which data was available – in the sample. It has to be kept in mind that one of the first important aspects of vaccine hesitancy is the fact that it is only present in countries where availability of vaccines is not an issue. However, by using the variables ‘level of education’ and

19 ‘GDP per capita’ it is still possible to differentiate between developed and developing countries. With this type of case selection I guarantee maximum variation between cases in both the dependent, independent and other characteristics. The objective was to collect data as current as possible. If this was not possible, data was used from the period of 2014-2017.

Data was first collected on the dependent variable ‘vaccine coverage’, this resulted in a database of 186 cases. In other words, 186 countries were included in this analysis, out of a total of 195 countries. For these cases the amount of media coverage was collected which is the independent variable in this study. For the other variables it was more difficult to collect data for all cases, the political system could be collected for 163 cases; GDP for 183; level of education for 161; physician density for only 110; and finally, religion on 182. This means that the results from statistical tests including, for instance, physician density will not be as strong as variables with a higher number of observations.

3.3 Case selection: in-depth content analysis

As mentioned before, the second phase of this thesis project consists of a small-N analysis aimed at explaining and interpreting the outcome of the large-N analysis. There are different reasons for doing this: if a relationship found in the large-N analysis is strong, a small-N analysis can uncover the mechanism behind the relationship; or, if a relationship is not strong, small-N analysis can help to make sense of and interpret the outcome of the large-N analysis (Toshkov, 2016). An advantage of combining two types of analysis is that it helps solve the limitations that would occur if only one of these research designs was chosen. The results of the statistical analysis deviate from the hypothesized effects which makes it interesting to study the existing findings in order to find a mechanism or explain why the expected effects were not found. This is realized by a media analysis of one country/case in which several newspapers are chosen and analyzed based on their contents related to immunization. This might lead to new or known stronger determinants that are indirectly related to media and will help to understand the outcome of the large-N analysis. In terms of single case selection, there are six options according to Gerring (2008): first, select a typical case, this means that the case will be most representative of the population of the study; second, an extreme case exhibits extraordinary values of the dependent or independent variables; third, a deviant case would be an outlier with a value highly different from the mean; fourth, influential case selection where values of independent variables have a high effect on the outcome; fifth, a crucial case is one that is most likely or least likely to demonstrate the hypothesized outcome; and finally, pathway cases house an exceptional causal path between the independent and dependent variable

20 (Gerring, 2008). As aforementioned, the outcome of part one deviated from the hypotheses. For this reason it would be interesting to analyze a country that exhibits this outcome. This way, it is possible to look for new hypothesis to explain the outcome of the large-N analysis. This means that a typical case was selected, not a typical case in terms of the expected relationship but a typical case in the discovered relationship, this is typicality from a causal understanding. In terms of representativeness this is a fitting case in that this correlation has been detected in the statistical analysis of almost all countries which will make the results relatively generalizable for the whole population.

3.4 Independent variable and dependent variable

Vaccine coverage was chosen as the dependent variable instead of vaccine hesitancy because on a scale this size it was not possible to find reliable data on vaccine hesitancy. As aforementioned, coverage is caused by hesitancy and accessibility, the latter will be controlled for by only including vaccines that are accessible in most countries. The WHO and UNICEF, in cooperation, provide national estimates of vaccine coverage of 14 vaccines for all their member states which is reported annually. According to the WHO, the data collected on vaccine coverage originates from official reports of the member states, as well as data from published and grey literature. If it was possible, the WHO and UNICEF also consulted with local experts. This collection of data has first been published in 1980 and the latest update that was used for this analysis was published on the 15th of July 2018. It consists of an excel document with 16 sheets that contain information on: the Bacille Calmette Guérin vaccine; the first and third dose of diphtheria toxoid, tetanus toxoid and pertussis vaccine; Hepatitis B birth dose (within 24 hours of birth); third dose of hepatitis B vaccine; third dose of Haemophilus influenzae type B vaccine; first dose of inactivated polio vaccine; measles-containing vaccine; second dose of measles-containing vaccine; third dose of pneumococcal conjugate vaccine; third dose of polio vaccine; rubella containing vaccine; rotavirus last dose; and finally, yellow fever vaccine. The final sheet exhibits regional coverage data per year (WHO, 2018b) .

For this study I will select several of these vaccines/diseases that have been introduced in most countries to be able to include as many countries as possible in the analysis. Another selection of these vaccines was based on the availability of data. To make sure that countries do not miss data for more than two vaccines I have chosen the one with most data of two polio and hepatitis vaccines (HepB BD and IPV1). The data that is still missing after selection of the vaccines and

21 deletion of countries with more than two data slots missing will be disregarded and the average of the remaining data will be used. The first selection of the vaccines was based on the list of most important vaccines, written by the WHO (2018a). As aforementioned, the vaccines that have been introduced in most countries have been included which means that, for instance, the vaccines for HPV (human papilloma virus), Yellow fever, rotavirus and meningitis A have been excluded. Additionally, the BCG virus -which prevents TBC- was included because there has been an increase in TBC in Europe (WHO, 2018a).

The collected data on this variable, thus, contains 186 observations with estimates from the year 2017. The average vaccine coverage rate of seven vaccines was calculated which resulted in an average rate for each country. The observations have a minimum coverage of 43,3% and a maximum of 99%, with a mean of 88,8%.

3.4.2 Media coverage

The independent variable ‘media coverage’ in this analysis will be determined from newspaper articles. This choice has been made because of the relative ease with which this variable could be measured and analyzed. Media coverage is measured with the results from the Factiva database (n.d.), this database shows all newspaper articles from newspapers of each country. This will provide an indication of how many articles have been dedicated to vaccination. The Factiva database is produced by Dow Jones and combines more than 32.000 sources from 200 countries in 28 languages. It includes newspapers (national, international and regional), magazines, journals, newswires, podcasts, relevant news websites, company reports, photo agencies and material about European Union laws (ProQuest, 2018).

For this part of the analysis I have chosen to count all the newspaper articles that were about immunization – a category that could be selected in the Factiva search –, this means that some articles are not entirely relevant. The articles can be in all languages and the amount covers the year 2017. To create a realistic impression of media coverage the numbers have been scaled to the amount of newspaper articles per one million citizens. This will ensure that larger countries do not seem to have a higher media coverage due to the size of the country. The population totals have been taken from the CIA factbook and cover the years 2016 to 2017 (CIA, 2018). Media coverage was found for all 186 countries. As mentioned above, calculating the value for media coverage per one million citizens ensures that countries of all sizes can be compared. For the year 2017 this resulted in a minimum coverage of zero, a maximum coverage of 800 articles per one million citizens and an average of 29 articles.

22 For the second part of the analysis the variable media coverage was operationalized differently. Data was still retrieved from the Factiva database, however, a more detailed search was conducted in order to retrieve as many relevant articles as possible. This will be done by using the keywords: vacunación and inmunización. To include all conjugations these words will be written as vacuna* and inmuniza*. Next, a decision needed to be made between the many newspapers in Spain. One area of focus is the political position of the newspaper to assure that a broad political readership is represented. Moreover, the high popularity of the newspaper was important according to the ‘dependency model’ in that it increases the centrality of the newspaper. For this reason El País is chosen as the biggest left wing/liberal newspaper which originates from Madrid. ABC is another newspaper from Madrid with a right wing/conservative position and El Mundo is a newspaper with a moderate right wing/independent political position. These newspapers will be analyzed due to their different audiences and political attitudes and articles will be selected from the year 2017 because the vaccine coverage data also stems from this period. Using the Factiva database, these demarcations will be used to retrieve a list of all relevant newspaper articles, which will then be coded and analyzed. The articles and statements will be coded on actor type, level of mobilization, whether a statement is included, what the statement contained, what the statement refers to and what tone is used. An extensive explanation of the coding instructions can be found in appendix A.

3.5 Other variables

3.5.1 Liberal Democracy Index (LDI)

The third variable, the political system of a country or the freedom of speech, will be measured by the ‘LDI’ taken from the V-Dem database (Coppedge et al., 2018). V-Dem stands for Varieties of Democracies and covers 201 countries from 1789 till 2017. This indicator was chosen because free press is an essential element of a liberal democracy and will therefore reveal the relationship between media coverage and vaccine uptake. Data is collected by, among others, 170 country coordinators and 3.000 country experts. This makes it one of the largest social science data collection projects (V-Dem, n.d.). The scores are used from the year 2017 and are values that vary between 0.0 and 1, where 1 is a perfect liberal democracy. This variable is clarified as:

“The liberal principle of democracy emphasizes the importance of protecting individual and minority rights against the tyranny of the state and the tyranny of the majority. The liberal model takes a "negative" view of political power insofar as it judges the quality of democracy by the limits placed on government. This is achieved by constitutionally protected civil liberties,

23 strong rule of law, an independent judiciary, and effective checks and balances that, together, limit the exercise of executive power. To make this a measure of LDI also takes the level of electoral democracy into account” (Pemstein et al., 2018, p.40).

The LDI value could be collected for 163 countries. In this sample the minimum score is 0.011, the maximum 0.867 with a mean of 0.425.

3.5.2 GPD per capita

The variable GDP per capita is added as a potential mediator or moderator, this variable was retrieved from the World Bank database and varies between the year 2015 and 2017. This variation was chosen to be able to include as many countries as possible in the analysis while keeping the data relatively recent. The World Bank consists of five organizations that are all related to finance and development. They conduct a high amount of research and collect large amounts of data to help increase the understanding of large challenges, for this reason their data has open access (The World Bank, n.d.). As aforementioned, the GDP per capita was collected for 183 countries. The data showed that the minimum GDP is $483, the maximum is $80.190 and the mean is $18.077 (The World Bank, 2017).

3.5.3 Enrolment in lower secondary education.

This variable has also been retrieved from the World Bank database and equally varies from 2015 to 2017. It contains the number of enrolments for both men and women and will be scaled as the number of enrollments per one million citizens to make the data as comparable as possible. This variable was chosen due to the fact that data on education is often far from complete for all countries in the world. This variable was one with a high amount of available data and it will give an indication of education levels per country. The minimum of enrolments per one million citizens is 2336, the maximum is 65.107 and the mean is 23.043 (The World Bank, 2018).

3.5.4 Physician density

Additionally, physician density per 1000 citizens was chosen to measure the importance of the healthcare system and healthcare workers in vaccine uptake. However, this data was not available for all countries, therefore the results of tests with this variable will be less reliable than others. To include as much data as possible the data stems from the years 2014 to 2016. The data was retrieved from the CIA Factbook, this is a database with all types of information collected by the American Central Intelligence Agency, its data covers 267 world entities. The

24 countries for which physician density was collected showed a minimum of 0.22 physician per 1000 citizens, a maximum of 7.5 and a mean of 2.3 physicians per 1000 citizens (CIA, n.d.). 3.5.5 Religion

Finally, religion is one of the contextual determinants that influences vaccine hesitancy. The Pew Fact Tank has published a report containing the percentages of the largest religions per country in 2010. The Pew Fact Tank started in 2013 and uses other reliable sources and its own data for their database (Pew Research Center, n.d.). This data from 2010 was the most recent that could be found, it was decided to include these because it is not expected that the proportions of religions will have changed drastically in the last years.

The data was used to retrieve the largest religion present in a country, which will be used to test the relationship between vaccine coverage and religion. The religions will be coded with numbers from 1 to 7: 1=Christian; 2=Muslim; 3=Buddhist; 4=Unaffiliated; 5=Hindu; 6=Jewish; and, 7=Folk religion. The data shows that 61,4% of countries is mainly Christian; 22,8% is Muslim; 3,8% is Buddhist; 3,3% is Unaffiliated; 1,6% is Hindu; 0,5% is Jewish and 0,5% is mainly Folk religion (Pew Research Center, 2012).

3.6 Method of analysis

Now that the unit of analysis, the data collection method and the operationalization has been discussed, the next step that follows is the method of analysis. As aforementioned, this thesis consists of two parts with the first being a statistical analysis and for the second part a media content analysis will be carried out.

The first hypothesis assumes that media coverage will have a negative effect on vaccine hesitancy. This will be analyzed in several ways, first, a Pearson’s correlation test will show if there is a correlation between them. Next, the effect of media coverage will also be measured by conducting a simple and multiple linear regression analysis. This indicates the extent to which the independent variable and other chosen variables can predict the outcome of vaccine coverage. The multiple regression analysis also shows how important the other variables are in predicting vaccine coverage.

The next two tests will focus solely on the interaction between two independent variables in their effect on the dependent variable. First, the political system of a country influences the extent of media coverage because low freedom of speech will not allow all types of media content to be published. Therefore, the LDI would serve as a moderating variable that, for instance, reveals the relationship between media and vaccine coverage in countries with a low

25 LDI or with a high LDI. Moreover, the theoretical framework demonstrates that both GDP and level of education can affect media and vaccine coverage. For this reason the statistical analysis will both look for the correlation of education and media coverage, as well as education and vaccine coverage. This will be done as part of a mediation analysis that will determine if education levels influence both the independent and dependent variable. Additionally, a moderator analysis will look into the possibility of a relationship being revealed between media and vaccine coverage in countries with low levels of education or the opposite.

Next, the media analysis in this second phase of the study is based on the literature and theories that have been documented in the literature review. However, this analysis will be highly specific compared to the abstractness of the statistical analysis. The type of content analysis that will be conducted is a tonality analysis; “Tonality is an analysis that uses a subjective assessment to determine if the content of article is either favorable or unfavorable to the person, company, organization or product discussed in the text” (Michaelson & Griffin, 2005, p.4). This assessment will be done for the entire article or per statement that is made by different stakeholders, it will be classified as “positive”, “negative”, or “balanced/unclear”. Thus, newspaper articles of several main Spanish newspapers will be analyzed in terms of the mention of vaccine related topics and what type of statements and stakeholders can be found. The relevant articles will be coded and conclusions will be drawn from the results.

3.7 Limitations

Regarding the large-N analysis, the main limitation of this cross-sectional design is the level of abstraction in terms of data. In order to gather data on approximately 186 countries it is difficult to be highly specific. This means that the causal claims will most likely not be strong. Moreover, the collected data is likely to be an approximation of concepts. This is the case for variables as physician density and level of education because they only refer to the determinants from the SAGE WG model on an abstract level. It makes analyzing complex issues quite difficult which will probably be the case in this analysis. This limitation is also present in the variable of media coverage. As described in this chapter, the Factiva database (n.d.) retrieves data from many sources. However, it cannot be guaranteed that, for instance, all newspapers of each country are included. This potentially inhibits a comparison between countries. The level of abstractness also affects the equivalence of data in different country contexts. For instance, media coverage has different meanings in different countries due to the fact that freedom of press is not present in every country. Selection bias is only a slight limitation for this analysis because from the onset all countries were included and the only reason for excluding countries was a lack of data,

26 in particular, physician density data was not available for all countries. The risk of spuriousness is one that is important for this analysis because there are many determinants of vaccine hesitancy, this means that there is a large chance that influential variables are not included in the analysis.

Then, for the in-depth media analysis, selecting a typical case causes the probability of the case its representativeness to be high. This does not exclude the risk of unrepresentativeness, especially because only one case is analyzed. This is why it is a limitation of the second phase of the analysis. However, the objective of this analysis is to find explanations for the observed relationship between media and vaccine coverage that might or might not be related to media. Moreover, the fact that media analysis is highly time consuming is a limitation in that it inhibits inclusion of many coding frames and cases due to time constraints.

27

4. Large-N analysis

4.1 Empirical findings

Now the data has been collected, the hypotheses have been formulated and the research design has been outlined it is time to start the statistical analysis for the first part of this thesis.

4.1.1 Correlation

The first statistical test will focus on determining if there is a relationship between the independent and dependent variable. The first step is to create a visual that will show how the observations are scattered between two variables. This will already provide an indication of the type and size of the relationship between the variables:

Figure 4-8: Scatterplots of vaccine coverage and independent variables

28 Additionally, as described by the Kent Library (n.d.),

“[t]he bivariate Pearson Correlation test will produce a sample correlation coefficient, r, which measures the strength and direction of linear relationships between pairs of continuous variables. […]” (para. 1).

The first outcome of the test – r, that measures strength – will present a number between 0.00 and 1 where 1 is considered a perfect correlation. A score between .30 and .69 means that there is a moderate relationship between the variables. The significance of the relationship – ρ – shows “how unlikely a given correlation coefficient, r, will occur given no relationship in the population” (Janda, 2001, para. 2).

Table 1: Pearson’s correlation

The first scatterplot shows the distribution of countries in terms of media coverage and vaccine coverage. The red line shows the type of the relationship between the two variables, in this case it is a small positive relationship. Pearson’s correlation determines the strength of the correlation between the two variables, this is 0.109 with a significance of 0.146. It shows only a small correlation that is not statistically significant, which means that this outcome might just be a coincidence. The second scatterplot – vaccine coverage and the LDI – shows another small positive relationship between the variables. This is also visible in the small correlation from Pearson’s test. However, this correlation is statistically significant which means that it is highly likely that the outcome is no coincidence and a relationship is present. This is the same for the next variables: vaccine coverage and GDP per capita. The scatterplot already shows a larger positive relationship and with a value of 0.314 on the Pearson’s correlation test there is a moderate relationship. Moreover, the significance is zero which means that this correlation is

Pearson correlation or r Sig. or ρ Vaccine coverage and media

coverage

0.109 0.146

Vaccine coverage and the LDI

0.181 0.023

Vaccine coverage and GDP per capita

0.314 0.000

Vaccine coverage and education level

0.134 0.098

Vaccine coverage and physician density

0.235 0.015

Vaccine coverage and religion

29 undeniable. The fourth scatterplot exhibits the relationship between vaccine coverage and education levels. It is only a minor relationship which is not significant. Next, the scatterplot shows quite a substantial positive relationship between vaccine hesitancy and physician’s density. The Pearson’s correlation is 0.235 and is significant which means that this relationship most likely is not a coincidence. Sixth, due to the categorical type of the variable religion, it could not be presented in a scatterplot. However, by using crosstabs it became visible that the highest rates of vaccine coverage can be found in countries where the main religion is Christian and after that Muslim. Pearson’s correlation value is 0.137 but it is not significant, thus, the correlation might be a coincidence.

4.1.2 Linear regression analysis

With the second statistical test, linear regression analysis, the goal is to find out if the independent variables can predict the outcome of the dependent variable, which independent variables are most relevant and to what extent (Statistics Solutions, 2013). Simple regression analysis looks only at one variable while multiple regression analysis looks at the predictive capability of multiple variables together. In order to look at the prediction value of media on vaccine coverage, a simple linear regression analysis will be carried out first. However, to do this test there are several assumptions that the data must comply with. These assumptions will be discussed further in appendix B.

Now we can perform the regression analysis for the data to find out its predictive values. The first table is the summary table that shows ‘R’, this stands for the quality of the prediction between 0.000 and 1 (1=100%

prediction), in this case quality is not great; 0,163. The ‘R Square’ number shows how much of the outcome can be predicted by the independent variables

(again, 1=100%), this means that only 2,7% of vaccine coverage can be predicted by media coverage. The next table is the ANOVA table, this shows the overall suitability of this regression model for the data, this is positive when the ‘Sig.’ column is <.0005. For this test the value is 0.030 which means that this model is insignificant.

The third table – coefficients – shows the effect of the variable on vaccine coverage. The first column ‘B’ in unstandardized coefficients “indicate how much the dependent variable varies with an independent variable when all other independent variables are held constant” (Laerd

30 Statistics, n.d., para.

Estimated model coefficients). If the number is lower than zero, it means that for every one increase of the independent

variable, the dependent variable decreases with the mentioned number. This is the other way around for numbers above zero. Moreover, the column ‘Sig.’ shows the statistical significance of the coefficients, this is true when the number is < .05 (Laerd Statistics, n.d.). This means that for every unit of increase in media coverage, vaccine coverage increases by 0.74. This predictive value is significant and exhibits the same positive correlation as Pearson’s correlation.

The second test is the multiple linear regression analysis. Again, the ‘R’ column indicates the quality of the prediction. In

this case the independent variables do not largely predict vaccine coverage. The ‘R Square’ number of this test is only 10,1%, this is quite low for an analysis

with multiple variables. The ANOVA table should show a significance of <.0005 if the collection of variables is suitable for the model. The significance of this test is 0.094, this means that these variables are not statistically relevant in predicting vaccine coverage. This is not a highly surprising outcome since research shows that vaccine coverage is affected by many variables and only some are included in this analysis.

Table 4: Multiple regression analysis; model summary Table 3: Simple regression analysis; coefficients

31 The coefficients table shows the effect of each variable. In this case, this means that an increase in media coverage causes a decrease of 0.011 in vaccine coverage. A unit of increase in the LDI decreases vaccine coverage by 2,806. Both variables GDP and level of education have an extremely low predicting effect on vaccine coverage. And finally, an increase of one unit in physician density increases vaccine coverage by 1,296. Unfortunately the only statistical significant variable is physician density. The other variables are not significant which means that these outcomes might just be a coincidence.

4.1.3 Mediation

A specific type of regression analysis is called mediation analysis and can be used to find out if another variable is the actual reason for one of the outcomes. In this case, the third hypothesis indicates GDP can both affect media coverage and vaccine coverage. This could mean that GDP acts a mediating variable in this instance. This analysis will also be done for the variable ‘level of education’. To test this, the first step is to determine if there is a correlation between the GDP and vaccine hesitancy, however this is not obligatory. The significance is 0.000 which means that the relationship is highly significant.

Next, a linear regression analysis needs to be carried out for the independent variable (media coverage) and the mediator variable (GDP), this correlation needs to be significant (coefficients table- Sig. <.05). The significance of the relationship here is lower than 0.000 as well which means it is

Table 5: Multiple regression analysis; coefficients