European healthcare systems, health outcomes and socio-economic inequalities in health

(1)

European healthcare systems, health outcomes

and socio-economic inequalities in health

Master’s Thesis

Philippe Meyer (11126868) philippe.meyer89@gmail.com University of Amsterdam

MSc Sociology – Social Problems and Social Policy June, 2016

Supervisor: Dr. Agnieszka Kanas Second Reader: Dr. Johan De Deken

(2)

1

Abstract

This study examines the effect of healthcare systems on self-reported health and socio-economic inequalities in health in Europe. Earlier research with the same focus often used a welfare-state typology approach, which is not adequate for measuring the effect of policies on health outcomes. Policies in welfare states are not necessarily consistent in the different policy areas, and the healthcare system is not included in the most used welfare state typology by Esping-Andersen. We decided to use Reibling’s approach to a healthcare typology, focusing on patient access by including measurements on gatekeeping, cost sharing and availability of medical personnel and technology. We combine this typology with existing research on those measurements to formulate our hypotheses. We then use this framework in a multilevel model using European Social Survey data from two waves, 2004 and 2010, looking at the 16 countries that Reibling included in her typology. The findings show that there are significant differences in self-reported health between the different healthcare systems. Two approaches seem to have the best results: Countries with no or moderate gatekeeping and moderate or high cost sharing, but a relatively high availability of medical personnel and technology fare best in both self-reported health, with the highest average value, and socio-economic inequalities, where those countries have the lowest inequalities. A close second are countries organizing their healthcare system the opposite way, by having high gatekeeping and no cost sharing, but also a lower availability of medical personnel and technology. These countries come second in self-reported health and in socio-economic inequalities in health.

(3)

2

1. Introduction

Health is important for every individual; Men, women, poor or rich: Everyone wants to be healthy. This is reflected in the research community, where a lot of papers have been written on the determinants of individual health. The most comprehensive approach to date can be found in a study by Aldabe et al., 2010. The authors use over 20 individual level variables, from different material factors like housing conditions and financial problems over occupational factors like dangerous working conditions and economic activity to psychosocial factors like social networks and the level of trust in other people (Aldabe et al., 2010). Instead of following this beaten path, this paper focuses on the role of the context: The institutions that play a role in an individual’s health. Different institutions can be found in different countries: We decided to use European countries, because there is a variety of healthcare policies to be found in the European countries and there is also a lot of data available.

In the last few years, healthcare systems in Europe have increasingly become the focus of reforms, trying to make the different approaches fit for the 21st_century.

Several challenges lie ahead, the most important ones among those are the looming demographic changes in Europe, new and increasingly expensive innovations in medical technology and the increasing expectations of patients that the utmost is done to restore their health. Spending on healthcare is often one of the biggest expenses nation states and their population have1, which also

leads to an increased focus on the system in place to allocate those funds. To improve the systems already in place, one has to know what the advantages and disadvantages of the different approaches are and which path to reforms is the most efficient one to both save money and increase the general health of the population. As with every other policy as well, healthcare systems differ vastly in

1 http://ec.europa.eu/eurostat/statistics-explained/index.php/Healthcare_statistics, 18.05.2016

(5)

4 the European Countries: There is no common European healthcare system. This makes a comparative approach to European healthcare systems fruitful, since the multitude of approaches can be explored and compared to one another.

But while doing comparative research on healthcare systems, the general health of the population is not the only measurement that should matter. One of the most important functions of a state is to make life better for all its citizens, which also includes people that are not very well-off in a socio-economic sense. The link between socio-economic status and health has long been established in the social sciences: People with a lower socio-economic status are generally in worse health than people with a higher socio-economic status (Adler et al., 1994). Socio-economic inequalities in health are reported all over Europe (Schütte, Chastang, Parent-Thirion, Vermeylen, & Niedhammer, 2013) and are, thus, important for policy makers in all European countries. This research therefore tries to find ways for policy makers to analyse and ultimately reduce socio-economic inequalities in health in their respective countries by comparing different healthcare systems and their impact on socio-economic inequalities in self-reported health.

In academia, while there is substantive research on welfare state typologies, with Esping-Andersen’s three worlds of welfare capitalism still being the most influential one, the research on the different healthcare systems in Europe is way less systematic. Often, researchers simply use the existing welfare state typologies and base their research on health on them (Chung & Muntaner, 2007; Eikemo, Bambra, Judge, & Ringdal, 2008). While it could be that the different welfare states (social democratic, liberal, conservative, sometimes extended with Eastern and Southern European types) also do have similarities in their respective healthcare systems, this is not a given: Welfare states are not necessarily consistent in their policy in different policy areas. It could mean that an otherwise very market-oriented country might have a very decommodifying

(6)

5 healthcare system (meaning their healthcare system uses very few if any market mechanisms), which could lead to better health outcomes especially for people that are in lower socio-economical strata. An example of this is the UK, where the welfare state as defined by Esping-Andersen is not very decommodifying, but their National Health Service is actually one of the most decommodifying healthcare systems in Europe, because there is no cost sharing: The service is free for every resident of the United Kingdom. This has various implication for previous research. A possibility is that, since the UK belongs to the liberal regimes type, the health outcomes of the liberal regimes might be overestimated (Clare Bambra, 2005). Others do not use typologies at all and look at health expenditure per capita (Missinne, Meuleman, & Bracke, 2013). Looking at health expenditure per capita is a rather flawed method, since its biggest determinant is Gross Domestic Product (GDP) per capita, thus it does not tell us how this money is spent (Hitiris & Posnett, 1992, p. 177).

We propose another typology, a typology that is more related to individual health and policies concerning individual health than the approach by Esping-Andersen. Researchers in the healthcare field have realized the shortcomings of using a welfare-state approach and there are efforts in creating a healthcare systems typology, which could be used for further research (C. Bambra, 2007; Moran, 2000; OECD, 1987; Reibling, 2010; Wendt, Frisina, & Rothgang, 2009). As of today, no paper has systematically used one of those typologies to look at individual health outcomes across Europe. This can also be said when looking at socio-economic inequalities in self-reported health, since most research in this area concentrates on welfare-state typologies as well (Clare Bambra, Lunau, Wel, Eikemo, & Dragano, 2014; Eikemo, Bambra, Judge, et al., 2008).

This research consequently tries to improve on the existing research by including a healthcare-system typology instead of a welfare states one when looking at self-reported health outcomes and socio-economic inequalities in self-self-reported health

(7)

6 in European countries. We have included an overview over the most recent healthcare system typologies in the paper. The most unique and promising healthcare systems typology is the one by Reibling, because it focusses on the access of patients to healthcare. She looks at gatekeeping as well as cost sharing and the availability of medical services to the patient. These features differentiates this typology from the other healthcare system typologies, because those mostly focus on the public/private split in financing the system and delivering care. Thus, using Reibling’s typology as a basis, we formulate two research questions to look at both self-reported health and socio-economic inequalities in self-reported health.

Do certain healthcare systems result in better self-reported health of the population in different European countries?

Is the positive relationship between socio-economic status and self-reported health weaker in some healthcare systems than in others?

2. Theory

2.1. Esping-Andersen’s welfare state typology and its critique

Research on the social determinants of health and inequalities in health has increasingly integrated a welfare states regimes approach. The most used typology is the one by Esping-Andersen. It is mainly based on a so called decommodification index, which measures how reliant an individual’s well-being is on the market. The less a citizen is reliant on the market, the higher the decommodification score. This score includes the pension system, the unemployment benefits and the sickness insurance. Additionally, Esping-Andersen looks at the role of the welfare state in social stratification, mainly if it distributes benefits according to socio-economic status (e.g. if the amount of unemployment benefits depend on the earnings of a person’s last job) or with a more egalitarian concept, where the benefits do not depend on a person’s last

(8)

7 income. He also takes the private-public mix into account, where he measures the relative role of the state versus non-state actors like family, voluntary sector or market in providing welfare benefits. According to the decommodification score and the additional measures, Esping-Andersen then divides the welfare states into three ideal regime types. In an ideal liberal regime, there is only a minimal amount of state provided welfare, benefits are modest and only accessible to a small part of the population. The conservative welfare state is used to cement the status of its recipient, benefits are often related to earnings. In general, benefits are higher than those in a liberal regime. The third type is the social democratic regime, characterised by universal and generous welfare benefits as well as a general trend towards equality and redistribution (C. Bambra, 2007, p. 1098). As mentioned in the introduction, previous research on individual health outcomes have mostly used this welfare-state typology, often an extended version of it with southern and eastern European regimes. The best outcomes were usually found in the social democratic and the liberal regime, while the conservative welfare states were found in the middle. The worst self-reported health can be found in the southern and eastern European welfare state regimes (Eikemo, Bambra, Judge, et al., 2008).

The general critique is mostly focussed on three fronts: Theoretical, methodological and empirical. Theoretically, the range of countries and regimes is criticised, with different researchers arguing for additional welfare state regimes (Southern European, Eastern European, and Confucian). Another weak point is the gender-blindness of Esping-Andersen’s work. It is argued that the decommodification index is gender-blind, the role of the woman and family in the provision of welfare is ignored. Lastly, and most fundamentally, Esping-Andersen’s typology assumes that the key social policy areas within a specific welfare regime type are similar across the board. Many scholars argue that this is not the case, and that different countries use different approaches in different

(9)

8 social policy areas. Most importantly for this paper, the provision of essential welfare state services like education, social services and healthcare are not included in Esping-Andersen’s decommodification index (C. Bambra, 2007, p. 1100). Especially this last point is very important in healthcare research, but overlooked by many researchers in the field. In the last decade, this specific shortcoming of Esping-Andersen’s typology has been addressed by different scholars and approaches.

2.2. Alternative approaches

The OECD developed a healthcare typology in 1987, which incorporates three dimensions. First, it looks at access to healthcare, measured by the degree of coverage. Second, the OECD describes how the healthcare system is funded, mostly by taxation or public and private insurance. Third, the provision of healthcare is examined. Does the emphasis lie on the public or the private sector for care provision? According to these three dimensions, the OECD finds three types of healthcare systems. The first on is a national health service, like it exists in the United Kingdom. Coverage is universal, funded through taxation and care is provided by mostly public providers. The private insurance system is found for example in the United States, but currently no European country has such a healthcare system. Coverage is not universal, because insurance is voluntary and quite expensive if not offered by the employer. Thus funding is private, as is the care provision, which is mostly done by non-state actors. The third system is somewhat in between, the OECD calls it the social insurance system. The prime example of it is Germany. Health insurance is compulsory, which makes coverage nearly universal. Responsible for the funding are private health insurers, care is provided by a mix between public and private institutions (OECD, 1987). Since the categories somewhat overlap with the welfare state typology of Esping-Andersen, we would expect the national health service to

(10)

9 have the most healthy individuals, the worst health can be found in the private insurance system and the social insurance system should be found in the middle. The second reviewed typology in this chapter is the one by Moran. His focus lies on the state involvement in funding and provision as well as health technology. The four types of his typology resemble closely the ones from the OECD. His first type is the entrenched command and control state, which is similar to the national health service type: Universal access, strong state control over funding and care providers as well as moderate state control over the available health technology. The second type is the supply state, meaning non-universal access and a low degree of state control in funding, provision and health technology. This is similar to the private insurance system of the OECD typology. The equivalent to the social insurance system is Moran’s third type, the corporatist states. Coverage is quasi-universal, control of the state over funding, provision and health technology is moderate. The fourth type does not have an equivalent in the OECD typology, but resembles the southern European welfare states. Moran calls it the insecure command and control state. They have a national health service, but state control in the three dimensions is not strong enough to guarantee the rights to healthcare for every citizen (Moran, 2000). This typology does not include any Eastern European states and it does not provide a comprehensive overview over which countries belong to which type of healthcare system, thus its use for research is limited.

Bambra tries to enhance the typology of Esping-Andersen with a healthcare index, where she measures private health expenditure as a percentage of GDP, private hospital beds as a percentage of total bed stock and the percentage of the population covered by the health care system. By plotting her health care index against the decommodification index, she finds similarities between the decommodification index and the healthcare index in a lot of countries. The social democratic countries are the most decommodifying in both welfare and

(11)

10 healthcare, there is a middle group with most of the conservative welfare states, which have a similar score. Some conservative welfare states (Germany, Switzerland and the Netherlands) have a lower score on the healthcare services index than their position on the cash benefit index, meaning market mechanisms play a bigger role in healthcare than in the other welfare state policies. The same can be found in the group of the liberal countries: While the US, Australia and Japan have both a low decommodification in welfare benefits and healthcare services, Ireland, New Zealand and the UK have a high score on the healthcare decommodification index, the UK even scoring as high as Norway. Thus, in addition to the original three regime types, Bambra proposes a subgroup of liberal and a subgroup of conservative welfare states that do have stark differences between their decommodification and their healthcare index (Clare Bambra, 2005, pp. 207–209). Again, no eastern European countries have been included in Bambra’s typology. Another problem is that her index does not look directly at the policies themselves (which is the case with Esping-Andersen typology), but by counting the amount of private and public beds and the percentage of private health expenditure, her index mostly measures the outcomes of healthcare policies. This can be useful as well, but for this paper we want to have a typology that consists mostly of direct measurements of policies. Wendt, Frisina and Rothgang establish their own ideal-types of healthcare types in their 2009 paper. They concentrate on three dimensions as well, which are financing, service provision and regulation. Within these dimensions they then look if they are state run, organised on a societal level or privately organised.

“Essentially, there are three responsibilities in healthcare: first, the financing of health services through taxation, social insurance contributions or private means; second, the provision of healthcare which can be carried out in state-run facilities respectively by state-based actors, in societal-based facilities, or in private for-profit facilities respectively by private actors; and third, the regulation by these actors of the various aspects of

(12)

11

financing and provision.”

(Wendt et al., 2009, p. 77)

With this typology, they identify three ideal-types and 24 mixed-types. The ideal types are a State Healthcare system, a societal healthcare system and a private healthcare system, but the authors conclude that no such ideal type exists in the empirical world (Wendt et al., 2009). They also only provide some examples of where countries might fit in, without using their typology for a comprehensive overview of different countries and their health care systems. Another problem is the huge amount of types available, 27 in total. Compared to at most 5 different types of the previous typologies, this is a huge difference and, in our eyes, renders this typology unfit for our research purpose.

As we can see, the reviewed healthcare typologies concentrate mostly on three dimensions: Funding, provision and coverage. Within those dimensions, they look at the involvement of the state versus non-state actors like the market or voluntary organisations. Coverage is mostly universal or quasi-universal in Europe at least, because there is either a national health service system in place or it is mandatory to buy health insurance. Each of the presented typologies has drawbacks, the most glaring one is the often missing inclusion of eastern European countries.

2.3. Reibling’s healthcare systems typology: Focus on patient access

As opposed to the typologies in the chapter above, Reibling tries to incorporate the patient’s perspective on the healthcare system into her theory. She moves the focus away from the private-public angle towards a patient access angle. The base of her new typology is Esping-Andersen’s decommodification concept:

(13)

12

“De-commodification in a more general sense refers to the access to welfare programmes defined by the benefit levels, and the conditions under which benefits can be accessed. Thus, when we apply this concept to healthcare services, we need to look at the institutional regulations that define the conditions for access to care and the potential services that can be received.”

(Reibling, 2010, p. 8)

Reibling then lists the four indicators that make up Esping-Andersen’s decommodification index and how they can be related to the healthcare field.

Conditions is the first indicator, meaning the conditions under which benefits can

be claimed. In welfare, these conditions are eligibility criteria such as a needs test, e.g. the government looks at a person’s income and wealth before handing out welfare benefits. Reibling’s equivalent in the healthcare field is gatekeeping. In certain countries, access to a specialist is only possible if referred to one by your general practitioner (GP), while other countries allow free access to as many specialists and GP’s. These gatekeeping systems introduce conditions, under which patients have access to specialist care, thus fitting in with Esping-Andersen’s definition of conditions for welfare benefits (Reibling, 2010, p. 8). The second indicator according to Esping-Andersen are disincentives. They are implemented in welfare programmes to reduce the amount of benefits that can be claimed. In welfare programmes, these disincentives are mostly a waiting period, before benefits can be claimed. Waiting periods in the healthcare field are not institutionalized, they are merely an undesired effect of certain ineffective healthcare systems. The disincentives in the healthcare field are financial in nature: Most countries have adopted cost sharing as a measure to reduce healthcare costs, meaning patients have to pay a certain amount for their treatment out of pocket (Reibling, 2010, p. 8).

(14)

13

Benefit levels define the amount of benefits that can be claimed depending on

income replacement ratios. Since this has no direct equivalent in the healthcare field, Reibling decided to measure the availability of medical providers and health

technology. She argues that this represent the range and intensity of care services

available to the patient and that this is a more precise measurement than health expenditure, because the amount of money spent on healthcare does not necessarily lead to a higher level of healthcare provision (Reibling, 2010, pp. 8–9). The last dimension on Esping-Andersen’s decommodification index is

universality, which describes the degree of the population that is covered by a

welfare programme. Since all European countries have a universal or near-universal (over 99%) coverage of the population for healthcare, this indicator cannot be used to further distinguish European healthcare systems and is thus left out of Reibling’s typology (Reibling, 2010, p. 9).

Reibling then uses different sources, ranging from the OECD to MISSOC, WHO and country experts to collect data for her three main dimensions. Gatekeeping is measured by four indicators. GP registration describes whether it is necessary to choose a GP, geographic restriction refers to the possible area from which that GP can be chosen. The third indicator is if GPs are paid by capitation, which means they are paid a certain amount of money for each patient that is registered with them. This should motivate doctors to good practice, because they do not get paid for (potentially unnecessary) treatments. Lastly, the access to specialist care. There are three possibilities here: A referral from the GP is needed to access specialist care (otherwise the specialist has to be paid out of pocket), a referral is usual but can be skipped by a higher co-payment or the access to specialists is free of a referral system (Reibling, 2010, pp. 9–10).

(15)

14 The next dimension is cost sharing, for which Reibling identifies three different systems. There is the co-payment, where the client pays a fixed fee for each GP visit. Another system is the co-insurance, where people pay a certain percentage of the fee for a doctor visit. The last possibility is called deductible, where people pay all costs for their healthcare up to a certain limit per year. Reibling then calculated a single score for each country, also taking purchasing power into account (Reibling, 2010, pp. 11–12).

The last dimension, availability of medical providers and health technology, is measured by the number of physicians, GPs, specialists and nurses per 1000 population as well as the number of MRI and CT scanners per 1 million population (Reibling, 2010, p. 13). With the help of a cluster analysis, which groups cases that are most similar together, she then arrives at four different healthcare system clusters. They are presented in table 1:

Table 1: The four different healthcare clusters (Reibling, 2010, p. 15).

Cluster Gatekeeping Cost Sharing Providers Technology Cluster 1

AT, BE, CH, FR, SE

None/moderate Moderate/high Higher availability

Higher availability

Cluster 2 CZ, DE, GR

None None Higher

availability Lower availability Cluster 3 DK, NL, PL, ES, GB

High None Lower

availability

Lower availability

Cluster 4 FI, IT, PT

High Moderate Medium

availability

Higher availability

(16)

15 The first cluster consists of Austria, Belgium, France, Sweden and Switzerland. Those countries have none or only moderate gatekeeping mechanisms, but moderate to high cost sharing. Care personnel and medical technology are more highly available. Czech Republic, Germany and Greece make up the second cluster, where neither gatekeeping nor cost sharing mechanisms are in place. Health providers are more highly available, medical technology is less so. Cluster 3 includes Denmark, the Netherlands, Poland, Spain and the United Kingdom. All of those countries have strong gatekeeping mechanisms in place, but do not rely on cost sharing. Care providers and technology is heavily regulated by the government to reduce costs, thus this group has the lowest availability of both supply indicators. The last cluster is made up from Finland, Italy and Portugal. High gatekeeping is combined with moderate cost sharing and a medium availability of providers. This group has the highest availability of medical technology of all the clusters (Reibling, 2010, pp. 14–15).

Problems may arise because the clusters are very different from the usual typologies used in comparative health research. This makes it hard to incorporate previous comparative healthcare studies with a similar focus, because they mostly use Esping-Andersen’s typology: The clusters in Reibling’s typology do not have a lot in common with the regimes of Esping-Andersen. Other problems might be the underrepresentation of eastern European states, with Poland and Czech Republic as the only countries in this group, but as we have seen most of the alternative approaches have not one single country from Eastern Europe included, which makes Reibling’s typology the most valuable for an evaluation of European countries. The data collected by Reibling is also dated, it reflects the regulations that were in place in 2003, despite the article being from 2010 (Reibling, 2010, p. 9). Newer developments, like the introduction (and later abandonment) of a cost sharing mechanism in Germany are not accounted for (Augurzky, Bauer, & Schaffner, 2006).

(17)

16 In conclusion we can say that Reibling’s approach differs vastly from the other alternative healthcare system typologies in chapter 2.2., but is nonetheless built on the foundation laid by Esping-Andersen. The focus on patient access and availability provides an interesting template to look at health outcomes and socio-economic inequalities in health.

2.4. The link between Reibling’s typology, self-reported health and

socio-economic inequalities in health

Since no other study has used Reibling’s typology and its influence on self-reported health and socio-economic inequalities in health yet, we have to take into account healthcare research that looked at similar characteristics, namely gatekeeping, cost sharing and medical personnel and technology.

According to Reibling and Wendt in their 2012 paper, gatekeeping can have two outcomes when it comes to quality in healthcare: Either quality is improved, because the GPs coordination leads to a more effective treatment of the illness. Or quality deteriorates, because compared to specialists, GPs provide lower quality of healthcare (Reibling & Wendt, 2012, p. 491). Garrido, Zentner and Busse review the literature for the effects of gatekeeping in their 2011 paper. They include 26 studies, but rate only two of them as being of good quality, three being of fair quality and the rest, 21 studies, being of poor quality (Garrido, Zentner, & Busse, 2011, p. 30). While none of the studies looked at self-reported health, we will take those into account that researched the effect of gatekeeping on health symptoms and morbidity as well as those looking at quality of life as an outcome variable. With both outcome variables, evidence suggests that there is no relevant difference between gatekeeping and a free access system and its effects on health, though some studies have non-significant results either in favour of gatekeeping or against it (Garrido et al., 2011, p. 32).

(18)

17 The influence of gatekeeping on socio-economic inequalities in health may theoretically also have two sides: On the one hand, inequalities might be reduced, because the GP gives support in decision-making to people from a lower socio-economic status and it reduces the use of specialists by people with higher socio-economic status. On the other hand, advantaged groups could pressure their GPs into giving them access to specialist care, thus maintaining existing socio-economic inequalities in health (Reibling & Wendt, 2012, p. 491). Here the evidence is conflicting, though more in favour of gatekeeping reducing socio-economic inequalities in health. Van Doorslaer and Koolman find that people with a higher income use specialist healthcare more than people with lower income, regardless of the existence of gatekeeping in the country’s healthcare system (Doorslaer & Koolman, 2004). Reibling and Wendt look at educational groups in their 2013 paper. They find that the positive relationship between education and the use of specialist healthcare is more pronounced in countries with a free access to specialists, meaning gatekeeping mechanisms reduce socio-economic inequalities in health (Reibling & Wendt, 2013). Le Fur and Yilmaz look at France, where a gatekeeping system was introduced in 2006. Before the introduction, people with higher educational background were more likely to access specialists. This changed after the introduction of a gatekeeping system, where the relationship between education and specialist access was not significant anymore (Le Fur & Yilmaz, 2008). The last reviewed study is by Or et al., where they also found a significant reduction in educational inequalities in health if a gatekeeping healthcare system was present (Or, Jusot, Yilmaz, & others, 2008). We conclude that a gatekeeping system reduces socio-economic inequalities in health.

The literature on cost sharing is a bit clearer when it comes to health outcomes. A huge literature overview with 160 articles by Eaddy, Cook, O’Day, Burch and Cantrell finds that out of 25 studies which looked at outcomes, 19 papers showed

(19)

18 an adverse health outcome if cost sharing is in place. 6 papers showed no effect of cost sharing on health outcomes. These results have to be taken with a grain of salt, since some of the studies looked at very specific subgroups of people, for example Medicaid-patients or veterans in the US (Eaddy, Cook, O’Day, Burch, & Cantrell, 2012, p. 50). Another literature review, from 2005, found only one study that looked directly at the health status from the patients. In this study, an increase in cost sharing lead to a decrease in health status (Gibson, Ozminkowski, & Goetzel, 2005, p. 735). Thus, although the research is somewhat fragmented into research on different special groups, we can conclude that cost sharing has a negative effect on self-reported health.

The evidence on the effect of cost sharing on socio-economic inequalities in health seems to be quite strong. The most cited study is an experiment by Manning, Newhouse, Duan, Keeler and Leibowitz from 1987. They offered over 5000 people in different American cities different healthcare plans, some of them with cost sharing, some of them without. The positive relationship between income and use of medical services was strongest in those plans with cost sharing and weakest in plans without cost sharing (Manning, Newhouse, Duan, Keeler, & Leibowitz, 1987). In their short literature overview, Drummond and Towse also find that the evidence is in favour of cost sharing being inequitable (Drummond & Towse, 2011, p. 1). We therefore conclude that cost sharing mechanisms reinforce socio-economic inequalities in health outcomes.

The per-capita number of medical personnel and technology seems to have a positive influence on health outcomes. A study by Anand and Bärninghausen looks at the provider density and finds that a higher density of human resources in the healthcare field leads to significantly lower maternal, infant and under-five mortality rate (Anand & Bärnighausen, 2004). A study by Robinson and Wharrad shows that higher density of physicians leads to better health outcomes (Robinson & Wharrad, 2001). Sarma and Peddigrew looked at Canada and

(20)

19 physician density, finding a positive influence of phyisican and medical technology density on self-reported health (Sarma & Peddigrew, 2007). In conclusion we can say that a higher density of medical personnel and technology leads to better health outcomes. We were unable to find papers that concern themselves with the availability of medical personnel and technology and its influence on socio-economic inequality in health. Since we don’t have any data or theory on that, we suggest that a broad availability of both do lower socio-economic inequalities in health. This is because if there is a wider availability, people from a lower socio-economic background may have a higher chance of getting adequate care from personnel and getting tests done with the aid of technology, simply because there are more resources available. If resources are constraint, people with higher influence or higher social status might be able to pressure their GP into giving them access to more personnel and technology than they actually need, or if there is no gatekeeping involved, people from higher social strata might be able to afford the more scarce resources more than people from lower socio-economic background. An overview over the dimensions and their suspected outcomes can be found in table 2.

Table 2: Reibling’s three dimensions and their effect on self-reported health and socio-economic inequality in health

Dimension Self-reported health Socio-economic inequality in health Gatekeeping No influence Reduces inequality Cost sharing Negative influence Reinforces inequality Providers and technology Positive influence Reduces inequality

As evident in table 2, gatekeeping is supposed to have no influence on self-reported health, but should reduce socio-economic inequality in health. Cost sharing is predicted to have a negative influence on self-reported health and is

(21)

20 also expected to reinforce socio-economic inequalities in health. The availability of healthcare providers and medical technology has an expected positive influence on self-reported health and we assume that a wide availability of providers and technology reduces socio-economic inequalities in health.

To make our model statistically sound, we also have to account for other variables that could influence self-reported health and socio-economic inequalities in health. In the literature, theories can be found for a number of explanatory variables, that can be used jointly to explain most of the social inequalities in self-reported health (Aldabe et al., 2010, p. 1). To ensure comparability, we will model our control variables similar to previous studies (Aldabe et al., 2010; Eikemo, Bambra, Joyce, & Dahl, 2008; Eikemo, Bambra, Judge, et al., 2008; Eikemo, Huisman, Bambra, & Kunst, 2008; Missinne et al., 2013). The included variables are gender, age, financial problems, health risks at work, social exclusion and general trust levels.

In previous studies, men have reported higher levels of health than women. The exact reasons for this are still not very clear, but a lot of the difference seems to be socially constructed, meaning that women are more likely to voice health complaints than men. The differences in self-reported between men and women is smaller in countries where gender norms are weaker (Torsheim et al., 2006). In all studies, increasing age results in worse self-reported health (Aldabe et al., 2010; Eikemo, Bambra, Joyce, et al., 2008; Eikemo, Bambra, Judge, et al., 2008; Eikemo, Huisman, et al., 2008). Problems to cope with one’s income also lead to lower self-reported health, because of a significant increase in stress levels (Wilkinson & Marmot, 2003). People working do have higher self-reported health than people that are unemployed, but certain jobs in hazardous working conditions are worse than jobs in safe working environments. People lifting heavy loads, performing repetitive tasks or operating heavy machines are at risk

(22)

21 of accidents and of accumulating work-related diseases like musculoskeletal disease or carpal tunnel syndrome (Clare Bambra et al., 2014).

Another well-known mechanism that leads to lower self-reported health is social exclusion. People are being excluded from participating in daily social life for various reasons. Poverty, racism, discrimination, stigmatization and unemployment are all contributors to social exclusion, and they have a very negative influence on an individual’s general health (Wilkinson & Marmot, 2003). The last included individual variable is general social trust. Higher trust in neighbours or people in general leads to higher self-reported health. The mechanisms are not quite clear, though Subramanian et al. hypothesize that higher social trust may result in better information about new illnesses or medical procedures that could prevent illnesses (e.g. screening for cancer), or that in communities with higher trust level people are more likely to help each other with health problems (Subramanian, Kim, & Kawachi, 2002).

3. Hypotheses

3.1. Healthcare systems and self-reported health

To predict which of the four healthcare clusters will have the highest self-reported health, we will look at the three dimensions mentioned in the chapter above. Since gatekeeping is most likely not influencing self-reported health, we will focus on cost sharing, which is supposed to have a negative influence, and the availability of medical personnel and technology, which is supposed to have a positive influence on self-reported health. As mentioned, the negative influence of cost sharing on self-reported health or health status is proclaimed in numerous studies, but they often look at specific groups of people (Gibson et al., 2005). For hypothesis 1a we thus consider the availability of medical personnel and technology the stronger indicator for a higher self-reported health. Countries in cluster 1 have moderate or high cost sharing, but also a higher availability of both

(23)

22 medical personnel and technology, while all the other clusters either have lower availability in one of those categories. Cluster 4, which has a medium availability of providers and a higher availability of technology, does also have cost sharing, while the countries in cluster 2 have a higher availability of medical personnel, but a lower availability of technology. Cluster 3 has a low availability of both personnel and technology, thus our first hypothesis is the following:

Hypothesis 1a: Countries in cluster 1 have significantly higher self-reported health than countries in the other clusters.

Conversely, if we focus on cost sharing instead of the availability of medical personnel and technology, there are two health clusters that do not have cost sharing, cluster 2 and cluster 3. The countries in cluster 2 do not have gatekeeping, while the countries in cluster 3 do. Countries in cluster 2 do have a higher availability of medical providers than countries in cluster 3, so if we assume that gatekeeping has no influence on self-reported health, our hypothesis 1b would be:

Hypothesis 1b: Countries in cluster 2 have significantly higher self-reported health than countries in the other clusters.

But, as Reibling and Wendt write in their 2012 paper, gatekeeping could have a positive influence on self-reported health: The coordination of the GP could lead to a more effective treatment of the illness (Reibling & Wendt, 2012, p. 491). With this in mind we create hypothesis 1c, where we focus on the countries in cluster 3, because they have both gatekeeping and no cost sharing:

Hypothesis 1c: Countries in cluster 3 have significantly higher self-reported health than countries in the other clusters.

(24)

23

3.2. Healthcare systems and socio-economic inequalities in health

The interaction between the different clusters and socio-economic inequalities are a bit more difficult to predict. Gatekeeping is supposed to lower socio-economic inequalities in health, while cost sharing is supposed to reinforce them, but we don’t have any studies regarding the effect of availability of medical personnel and technology. This does not mean that this availability has no effect on socio-economic inequalities in health. We will thus consider two conflicting hypotheses that we will test against each other. The first one contains only gatekeeping and cost sharing as indicators, while the second hypothesis also takes the availability of medical personnel and technology into account. If we only look at gatekeeping and cost sharing, cluster 3 will have the lowest socio-economic inequalities in health. Countries in cluster 3 have both high gatekeeping and no cost sharing, making them the ideal candidate according to theory and previous research. Therefore, the hypothesis is the following:

Hypothesis 2a: Countries in cluster 3 have a significantly lower socio-economic inequality than countries in the other clusters.

Hypothesis 2b also takes availability of personnel and technology into account. The highest availability of medical personnel and technology can be found in cluster 1. Some countries in this cluster also have a moderate amount of gatekeeping and a moderate amount of cost sharing. We thus consider cluster 1 to have the lowest amount of socio-economic inequalities compared to the other clusters, if the availability of personnel and technology has a negative influence on the positive relationship between socio-economic background and self-reported health.

Hypothesis 2b: Countries in cluster 1 have a significantly lower socio-economic inequality than countries in the other clusters.

(25)

24

4. Data and Methods

The Data used will be taken from two European Social Survey waves: Wave 2, conducted in 2004 and wave 5, conducted in 2010. We chose those waves because certain variables concerning work were only available in those two years. Countries included are Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Greece, Italy, the Netherlands, Poland, Portugal, Spain, Sweden, Switzerland and the United Kingdom. The sample is representative of all persons aged 15 and over for residents in each country, regardless of nationality, citizenship or language. The individuals are selected by random probability methods at all stages, quota sampling is not allowed. Each country uses the method that is expected to yield the best response rates, so the methodologies slightly differ from country to country (European Social Survey, 2014b). The minimum amount of interviews is 1500, in smaller countries with up to 2 million inhabitants the minimum is 800 interviews. In our country sample for 2010, Sweden has the least interviews with 1497, while Germany has the highest number with 3031. Response rates range from a low 30.5% in Germany to 70.3% in Poland. Poland and the Czech Republic are the only countries included in this study that reach the target response rate of 70%.2 For the sample in 2004, Spain

has the lowest amount of interviews with 1663 and, surprisingly, the Czech Republic has the most with 3026 interviews. Response rates range from 43.6% in France to, again, Poland with 73.7%.3

The method will be a multilevel analysis, using a logistic regression model. A logistic model is needed because in accordance with previous research, the dependent variable of self-reported health is made dichotomous. The multilevel part is important because we want to examine the self-reported health of individuals nested within their respective countries with their respective

2 http://www.europeansocialsurvey.org/data/deviations_5.html, 21.05.2016 3_{http://www.europeansocialsurvey.org/data/deviations_2.html, 21.05.2016}

(26)

25 healthcare system. Because we use two waves of the European Social Survey, the individuals will be nested within a country*year combination. Both waves together include 59’681 individuals. With missing values omitted from the analysis and people younger than 25 years filtered out, 49’481 individuals nested within 16 countries and 2 waves remain. To find out the variance in self-reported health that can be explained on the country-level, we will first compute a model without independent variables. This will give us the Variance Partition Coefficient, needed to see on which level the variance lies. The second model will be a full individual model, where all the individual variables are accounted for. In the third model, we will add the healthcare clusters and in the fourth the interactions between the clusters and socio-economic status will be added. In the fifth model, we will change the reference category for the healthcare clusters and the interaction, to test which of the hypotheses 2a and 2b can be rejected.

5. Operationalization

5.1. Dependent variable

The dependent variable (self-reported health) is an ordinal variable with five possible answers: „very good‟, „good‟, „fair‟, „bad‟, and „very bad‟. The question asked in the European Social Survey is simply “How is your health in general?” In accordance with previous research (Aldabe et al., 2010; Eikemo, Bambra, Joyce, et al., 2008; Eikemo, Bambra, Judge, et al., 2008; Eikemo, Huisman, et al., 2008), the dependent variable will be made dichotomous, resulting in good health (“very good”, “good”), coded as 1, and not good health (“Fair”, “bad”, “very bad”), coded as 0. This results in a variable where roughly two thirds of the individuals fall into the good health category, while one third ends up in the not good health category.

(27)

26

5.2. Main variables

There are two main variables in this study. The first one is a country level variable, namely the different healthcare system clusters according to Reibling’s typology. The 16 countries were assigned their respective healthcare cluster: Austria, Belgium, Switzerland, France and Sweden belong to cluster 1. The Czech Republic, Germany and Greece belong to cluster 2, Denmark, the Netherlands, Poland, Spain and Great Britain to cluster 3. The last cluster consists of Finland, Italy and Portugal. Since this is a nominal variable, dummy variables had to be created for each cluster, where 1 represents the countries that belong to the respective cluster and 0 are the countries in the other three remaining clusters. The other main variable is socio-economic status. As a proxy for socio-economic status we decided to use education in years, for different reasons. The first choice was household income to represent socio-economic status, but the variable for household income was only available for 44’818 individuals, with over 10’000 individuals missing, mostly because people did not want to respond to the question. A similar situation was encountered when looking at the educational level according to ISCED (International Standard Classification of Education). While there were not a lot of missing values, the education of 15’760 respondents could not be harmonised into the ES-ISCED categorisation. This leaves the variable for education in years as the one with the most respondents. This variable is not without flaws, especially because there are some extreme values on the amount of years people were in education. The most extreme value was for an individual with 50 years of education, but 15 people had 40 years of education or more. Since we did not want to include those outliers in our analysis, we decided to filter out respondents with more than 25 years of education, which resulted in a bit more than 100 cases that were filtered out.

(28)

27

5.2. Control variables

The included control variables are gender, age, financial problems, health risks at work, social exclusion and general trust levels. The first control variable is gender. Because men are expected to report higher health, it was coded 1 for male respondents and 0 for female respondents (Eikemo, Bambra, Judge, et al., 2008, p. 2287). The next variable is age. Since it is important that most people in the sample have finished their education, we decided to exclude people younger than 25 from the analysis. The assumption is, that older people report worse health status (Eikemo, Bambra, Judge, et al., 2008, p. 2287).

Financial problems are also supposed to lead to worse health, mediated through higher stress levels. To account for financial problems, we used the following question: “Which of the descriptions on this card comes closest to how you feel about your household's income nowadays?”(European Social Survey, 2014a). The answer categories are “living comfortably on present income”, “coping on present income”, “difficulty on present income” and “Very difficult on present income”. Because the scale cannot be considered continuous and we expected similar results for categories 1 and 2, we decided to dichotomize the variable into people that live comfortably or cope on the current income and people that find it difficult or very difficult to live.

Another important variable is the possibility of health hazards because of work. The European Social Survey uses the following statement: “My health or safety is at risk because of my work”. The answers are “not at all true”, “a little true”, “quite true” and “very true”. Because of the people currently not working, a high amount of “not applicable” values are present. We decided to form three dummy variables. One for people that answer “not at all true”, these are people that do not have any health risks at work. The second dummy variable was made from the answers “ a little true”, “quite true” and “very true”, these are the people

(29)

28 working in hazardous conditions. And lastly, a dummy for the “not applicable” respondents was formed, to represent people that currently do not hold a job. Social exclusion can also lead to worse health. The European Social Survey asks: “Compared to other people of your age, how often would you say you take part in social activities?” Five answers are possible: “Much less than most”, “less than most”, “about the same”, “more than most” and “much more than most”. Since the 5-point scale can hardly be considered continuous, we decided to form three dummy variables for low social inclusion (“much less than most”, “less than most”), medium social inclusion (“about the same”) and high social inclusion (“More than most”, “much more than most”). This resulted in 21’699 people which had low social inclusion, 27’877 which had medium social inclusion and 10’960 which had high social inclusion.

The last variable is the general trust level, where the assumed relationship is that higher levels of trust in people leads to higher self-reported health. Three items of the European Social Survey were used to build an index for this variable, all of them use a scale from 0-10: “Most people can be trusted (10) or you can’t be too careful (0)”, “Most people try to take advantage of you (0) or try to be fair (10)” and “Most of the time people are helpful (10) or mostly looking out for themselves (0)”. A factor analysis implies that all three statements load on the same factor with at least 0.67, which is a strong correlation. A reliability test using Cronbach’s Alpha shows a value of 0.77, also showing a strong correlation between the three items.

To see if there are differences between wave 2 and wave 5, a wave-variable was included, but since it was highly insignificant (Sig. 0.883) it was removed again from the model.

(30)

29

7. Results

7.1. Descriptive statistics

First we will have a look at the country averages for the years 2004 and 2010 to see how big the differences are that we can explain with our model. If we look at table 3, we see that in both years, Switzerland has the highest self-reported health with a mean of over 0.8, while Portugal has the lowest self-reported health with a mean of about 0.5. This shows that there is quite a bit of variance between countries. As can be seen in the 4th row, changes within countries between the

two measurement years are very small. Most countries change within +/- .01 or +/- .02 points, thus basically remain stable. The countries with slightly higher positive changes are Greece (+ 0.05) and Poland (+ 0.06), while Germans report the highest negative changes at - 0.05. Therefore, while we still control for the change over time by nesting the individuals within a country*year combination, we do not expect that changes over time account for a significant amount of variance.

(31)

30 Table 3: Country-means for self-reported health for 2004 and 2010 and changes between the years

Country 2004 2010 Change

Austria .77 No data No data

Belgium .77 .76 - .01 Czech Republic .56 .58 + .02 Denmark .77 .76 - .01 Finland .65 .64 - .01 France .62 .63 + .01 Germany .62 .57 - .05 Great Britain .70 .69 - .01 Greece .73 .78 + .05

Italy .61 No data No data

Netherlands .69 .70 + .01 Poland .56 .62 + .06 Portugal .50 .51 + .01 Spain .63 .63 + 0.003 Sweden .74 .76 + .02 Switzerland .83 .81 - .02

Table 4 shows us the means of self-reported health for the healthcare clusters. The differences between the years are minuscule, only cluster 1 and cluster 3 have slightly different values. The highest values can be found in cluster 1, followed by cluster 3, then 2 and countries in cluster 4 have the worst self-reported health on average.

(32)

31 Table 4: Cluster-means for self-reported health for 2004 and 2010

In table 5, the descriptive statistics for the individual variables included in the model are presented. Since a lot of dichotomous variables were used, the range is often 1.

Table 5: Descriptive statistics of the individual variables included in the model

Variable Range Mean Std. Deviation

Good subjective health 1.00 .67 .470

Male 1.00 .47 .499

Age of respondents older than 25 77 51.86 16.199 Years of full-time education completed 25 11.82 4.113 Comfortably/coping on current income 1.00 .75 .431

Currently not working 1.00 .60 .489

Risks at work 1.00 .19 .393

No risks at work 1.00 .21 .404

Low social inclusion 1.00 .36 .479

Medium social inclusion 1.00 .46 .498

High social inclusion 1.00 .18 .385

General Trust Level 10.00 5.14 1.923

Cluster Year 2004 Year 2010 Cluster 1 0.75 0.74 Cluster 2 0.64 0.64

Cluster 3 0.67 0.68 Cluster 4 0.58 0.58

(33)

32 To test whether these individual variables do have a significant influence on our dependent variable, we conducted different tests. Table 6 contains the t-test of the variables male, comfortably living or coping on current income and the three dummy variables concerning social inclusion. As we can see in the last column, the influence of all these variables is highly significant with very low standard errors of the means. The variables also behave like predicted: Men have higher self-reported health than women on average, people living comfortably or coping on current income have a much higher self-reported health than people struggling with their financial situation. People that do not work have a significantly lower self-reported health than people that do work, and people that do not have any health risks at work have the highest mean of self-reported health, while people that do work but experience health risks at their job are in the middle. Social inclusion has also the predicted outcome, with people that feel not very included having a lower mean than people feeling either medium or highly included.

(34)

33 Table 6: T-tests for various individual variables

Variable Value Mean Std. Error

Mean Significance Male 0 .648 .003 .000 1 .697 .003 .000 Comfortably/coping on current income 0 .529 .004 .000 1 .718 .002 .000

Currently not working 0 .790 .003 .000

1 .592 .003 .000

Risks at work 0 .650 .002 .000

1 .754 .004 .000

No risks at work 0 .631 .002 .000

1 .823 .003 .000

Low social inclusion 0 .729 .002 .000

1 .566 .003 .000

Medium social inclusion 0 .621 .002 .000

1 .728 .003 .000

High social inclusion 0 .657 .002 .000

1 .734 .004 .000

For variables that are not dichotomous, correlations can be used to measure their influence on self-reported health. They can be found in table 7. We can see that the age of the respondent correlates negatively with self-reported health. The older an individual gets, the worse their reported health. The inverse can be seen in education and general trust level. Additional years of full-time education add to ones self-reported health, while a higher level of trust corresponds with higher self-reported health. All the correlations are highly significant.

(35)

34 Table 7: Correlations of individual variables with self-reported health

Variable Correlation Significance

Age of respondents older than 25 -.338 .000 Years of full-time education completed .248 .000

General trust level .148 .000

7.2. Regression analysis

We estimated 5 different models, as can be seen in table 4. The first model is the null model, where only the intercept is included. This model is here to find out how much of the variance lies on the country-level, the so called intra-class-correlation (ICC). The following formula was used:

𝐼𝐼𝐼𝐼𝐼𝐼 =_{𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑜𝑜𝑜𝑜 𝑅𝑅𝐸𝐸𝐸𝐸𝐸𝐸𝑅𝑅𝑅𝑅𝐸𝐸𝑅𝑅 + 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑜𝑜𝑜𝑜 𝐼𝐼𝐼𝐼𝐸𝐸𝐸𝐸𝐼𝐼𝐼𝐼𝐸𝐸𝐼𝐼𝐸𝐸}𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑜𝑜𝑜𝑜 𝐼𝐼𝐼𝐼𝐸𝐸𝐸𝐸𝐼𝐼𝐼𝐼𝐸𝐸𝐼𝐼𝐸𝐸

Which resulted in the following equation with our data:

𝐼𝐼𝐼𝐼𝐼𝐼 = . 010251

. 218886 + .010251 = 0.045

Thus our ICC is 0.045. This means that 4.5% of the variance in self-reported health can be explained by variables on the country level. This is a bit lower than expected, Eikemo et al. found that 8% of the variance in self-reported health can be explained by country-level variables (Eikemo, Huisman, et al., 2008, p. 2288), while Missinne et al. reported an ICC of 6.5% (Missinne et al., 2013, p. 238).

(36)

35

Model 1 Model 2 Model 3 Model 4 Model 5 Estimate (Std. Error) Estimate (Std. Error) Estimate (Std. Error) Estimate (Std. Error) Estimate (Std. Error) Intercept .65 (.0189)*** .62 (.0206)*** .68 (.0299)*** .62 (.0288)*** .73 (.0322)*** Male .03 (.0039)*** .03 (.0039)*** .03 (.0039)*** .03 (.0039)*** Age -.008 (.0001)*** -.008 (.0001)*** -.008 (.0001)*** -.008 (.0001)*** Years of education .009 (.0005)*** .009 (.0005)*** .007 (.0008)*** .007 (.0008)*** No financial problems .13 (.0049)*** .13 (.0049)*** .13 (.0049)*** .13 (.0049)*** No risks at work .10 (.0053)*** .10 (.0053)*** .10 (.0053)*** .10 (.0053)*** Risks at work .05 (.0054)*** .05 (.0054)*** .05 (.0054)*** .05 (.0054)*** Medium social inclusion .09 (.0043)*** .09 (.0043)*** .09 (.0043)*** .09 (.0043)*** High social inclusion .11 (.0056)*** .11 (.0056)*** .11 (.0056)*** .11 (.0056)*** Trust in people .02 (.0011)*** .02 (.0011)*** .02 (.0011)*** .02 (.0011)*** Healthcluster 1 Ref. category .12 (.0401)*** Ref. category Healthcluster 2 -.06 (.0435) -.02 (.0429) -.14 (.0453)*** Healthcluster 3 -.09 (.0382)** Ref. category -.12 (.0401)*** Healthcluster 4 -.12 (.0460)*** -.11 (.0479)*** -.23 (.0470)*** education* healthcluster 1 -.003 (.0013)** Ref. category education* healthcluster 2 .003 (.0013)*** .006 (.0014)***

education* healthcluster 3 Ref. category .003 (.0013)**

education* healthcluster 4 .007 (.0013)*** .01 (.0014)*** AIC 65334.597 55641.941 55639.666 55589.791 55589.791

Table 8: Estimation of the multilevel logistic models

Significance: * p<.1, ** p<.05, *** p<.01 Missings are omitted using listwise deletion

(37)

36 Model 2 is the full individual model, where we can see the influence of our control variables and of education on self-reported health. First, all of our individual variables are highly significant on the 99% level. The variables also behave according to theory. Being male instead of female increases the reported health by 0.03 points, while each year aging decreases your self-reported health by 0.008. If a person has no financial problems, their self-self-reported health increases by 0.13 points compared to someone that does have financial problems. Compared to people that do not work, people that have a job without any health risks report 0.1 point higher health, while people with a dangerous job report 0.05 points higher health than people that are not working. This is most likely the case because people not working consists of a lot of older people and sick people, which would report lower health. Social inclusion is also important. Compared to someone with low social inclusion, medium (+ 0.09 points) and high (+ 0.11) social inclusion increase health significantly. The last of our control variables is general trust in people. An increase of one point on the 0-10 scale of this index leads to an increase of 0.02 points on self-reported health. One of our main variables, education in years, is also already included in this model. Each additional year of education leads to a 0.009 points higher score on self-reported health, which was to be expected. Important to note is that the Akaike’s Information Criterion (AIC) changed from the first model’s 65334.597 to 55641.941 in the second model, which is a large decrease, showing that including the individual variables was a necessary step.

The next step in the modelling process is to include the country-level variables, in this case our four different healthcare clusters. This model can be used to test our first group of hypotheses:

Hypothesis 1a: Countries in cluster 1 have significantly higher self-reported health than countries in the other clusters.

(38)

37

Corresponding to hypothesis 1a, we decided to make cluster 1 the reference category. With the individual variables virtually unchanged, we can see that hypothesis 1a seems to be confirmed: Compared to cluster 1, people in cluster 2 have a decrease of 0.06 points in self-reported health, a decrease of 0.09 points in cluster 3 and a decrease of 0.12 in cluster 4. But the estimate for cluster 2 is not significant with a significance of p=0.139, while the other two clusters are significant on at least the 95% level. This means that we cannot yet state that there is a difference between cluster 1 and cluster 2, thus we cannot confirm hypothesis 1a, 1b or 1c at this point, however hypothesis 1a seems to be the most likely candidate to be confirmed. The AIC also only changed to 55639.666, which is a very small improvement.

In model 4 we also included the interactions between the different healthcare clusters and our other main variable, education in years. To test hypothesis 2a, we decided to take healthcare cluster 3 as the reference category, because we want to see how the countries in healthcare cluster 3 behave compared to the other clusters.

Hypothesis 2a: Countries in cluster 3 have a significantly lower socio-economic inequality than countries in the other clusters.

All the cluster*education interactions are significant at least on the 95% level. The interactions between cluster 2 and years in education as well as cluster 4 and years in education are in fact significantly higher, meaning inequalities with regards to education (our proxy for socio-economic status) are higher in those two cluster than in cluster 3. But the interaction in cluster 1 is slightly negative,

(39)

38 which means that the educational inequalities in self-reported health are lower in the countries in cluster 1 than in the countries in cluster 3. Thus, we have to reject hypothesis 2a. Remarkably, the AIC decreases from 55639.666 in model 3 to 55589.791 in model 4. The addition of the interactions helped to build a significantly better model.

To test hypothesis 2b, the same model as before was used, but the reference category changed from cluster 3 to cluster 1. Interestingly, the addition of the interactions also made all the estimates for the different health clusters significant, on the 99% level, while in model 3 the estimate for cluster 2 was not significant. Compared to cluster 1, respondents in cluster 2 have a 0.14 point lower reported health. Respondents in cluster 3 have a 0.13 point lower reported health and respondents in cluster 4 even have a 0.23 point lower self-reported health. Thus with the help of model 5, we can confirm hypothesis 1a. Countries in healthcare cluster 1 do have significantly higher self-reported health than the countries in the other three healthcare clusters. To recall, hypothesis 2b was:

Hypothesis 2b: Countries in cluster 1 have a significantly lower socio-economic inequality than countries in the other clusters.

As we can see in table 8, all the interactions are significant, with the interactions between cluster 2 and education and cluster 4 and education being significant on the 99% level, while the interaction between cluster 3 and education is at least on the 95% significance level. The estimates for all the interactions are positive, meaning the socio-economic inequality in the three clusters are higher than the socio-economic inequalities in cluster 1. Thus, we can confirm our hypothesis 2b. To test the robustness of our model, we looked at a QQ-plot of the residuals, which can be found in the appendix. The distribution deviates from the expected normal distribution, but it is still within acceptable parameters.

(40)

39

8. Discussion

For the discussion we will start with our two research questions:

Do certain healthcare systems result in better self-reported health of the population in different European countries?

As we have seen in chapter 5.2., there are certainly differences in the mean of self-reported health (Cluster 1: 0.75, cluster 3: 0.68, cluster 2: 0.64, cluster 4: 0.58). These differences remain, even if we control for individual variables in our multilevel models. Thus we can state that certain healthcare systems result in better self-reported health of the population. The population in cluster 1 is significantly healthier than the population in the three other clusters, with the clusters 2 and 3 being very close to each other, while the countries in cluster 4 have the worst self-reported health.

Is the positive relationship between socio-economic status and self-reported health weaker in some healthcare systems than in others?

This question can also be answered with yes. The results for model 5 show that there is a significant difference between the different healthcare clusters concerning socio-economic inequalities in health: They are lowest in the countries of cluster 1. To have a closer look at the data with regards to our theory we will turn to our hypotheses. First, an overview over our five hypotheses and if they were accepted or not in table 9:

European healthcare systems, health outcomes and socio-economic inequalities in health