Effect of the state of the economy on health : a study of the effect of the economic crisis, the Great Recession, using the level of unemployment as the main indicating factor, on people’s health in the Netherlands usin

(1)

EFFECT OF THE STATE OF THE ECONOMY ON HEALTH

A study of the effect of the economic crisis, the Great Recession, using the

level of unemployment as the main indicating factor, on people’s health in the

Netherlands using data from 2000-2016.

UNIVERSITY OF AMSTERDAM MSc Economics

Master Specialisation Behavioural Economics and Game Theory 15 ECTS

Author: N.M. Weijers

Student number: 10589872

(2)

PREFACE

This thesis is written to complete my Masters Economics with the specialisation Behavioural Economics and Game Theory at the University of Amsterdam. Writing this thesis is a mandatory part of the Master’s. Thus, my main motivation to write this thesis is to graduate. However, my choice for this thesis’ topic interrelates with my choice for the Master’s. I have always been interested in people and the way they think or make choices. The connection between the economy and how individuals make choices is extensively examined in the specialisation. Because of this interest, I was curious to the influence of the state of the economy on people, and in this case people’s health. Which of course, does not have one main effect, but consists of various health effects, which are not all that obvious at first sight. The choice for a topic I am interested in, made the process of writing the thesis that much easier. Because I did not experience any major setbacks, I review this period of writing my thesis mainly as insightful. Nienke Weijers, Amsterdam, 6 July 2018, 20:29.

Statement of Originality

This document is written by Student Nienke Weijers who declares to take full responsibility for the contents of this document.

(3)

ABSTRACT

This thesis investigates the influence of the Great Recession, and the rise in unemployment resulting from it, on people’s health in the Netherlands in the short- and long-term, using data ranging from 2000 until 2016 derived from Statistics Netherlands (CBS). An innovative approach is used by considering both the short- and long-term and by dividing health in three composite health variables: mental health, healthy behaviour and physical health. Unemployment is used as indicator for the state of the economy. OLS regressions with robust standard errors are used to test the effect of unemployment on health. A negative, significant effect of unemployment on mental and physical health is proven, both in the short- and long-term. Poorer mental health is explained by stress and uncertainty about the future and physical health by lower standards of living and less health expenditures. For healthy behaviour a positive, significant effect of unemployment is found, mainly explained by lower opportunity costs of leisure time and a decline in the use of damaging goods. The long-term effect of unemployment on health behaviour appears to be significantly lower than the short-term effect. Additionally, significantly opposite effects of unemployment on mental health between men, negative, and women, positive, are found in both the short- and long-term. Explained by higher career expectations, potentially larger income losses, and scaring effects of job loss for men. Moreover, unemployment negatively influences the mental and physical health of people younger than 55 years in the short-term. While people who are older than 55 years show a negative influence of unemployment on healthy behaviour. Explained by increased social security for people who are older than 55 years and worse perspective to find a job again.

Keywords: macroeconomic conditions, economic crises, health, mortality, unemployment. JEL Classification: I01

(4)

LIST OF TABLES

Table 1 - Descriptives statistics, 2000-2016 ... 15

Table 2 - Results OLS regressions of mental health, healthy behaviour, and physical health ... 28

Table 3 - Difference between the short- and long-term ... 31

Table 4 - Correlation unemployment and social security expenses ... 34

Table 5 - Results of prior studies ... 39

Table 6 - Correlations indicators economy ... 41

Table 7 - Skewness and Kurtosis test for Normality ... 43

Table 8 - Correlations all variables ... 44

Table 9 - Correlations main variables ... 47

Table 10 - Multicollinearity ... 47

Table 11 - Shapiro-Wilk normality test ... 48

Table 12 - Durbin-Watson statistics ... 48

Table 13 - Breusch-Godfrey test ... 48

Table 14 - Breusch-Pagan test ... 49

Table 15 - DF-GLS test ... 49

Table 16 - Results regression including lagged variables ... 50

Table 17 - Durbin-Watson statistics lagged variables ... 50

Table 18 - Breusch-Godfrey test lagged variables ... 51

Table 19 - Results regression including interaction variable men ... 51

Table 20 - Difference short- and long-term interaction variables ... 52

Table 21 - Results regression including interaction variable age ... 52

Table 22 - Results regression including interaction variable education low ... 53

(7)

LIST OF FIGURES

Figure 1 - Movements of unemployment in the Netherlands, 2000-2016 ... 19

Figure 2 - Movements of unemployment and mental health, 2000-2016 ... 23

Figure 3 - Movements of unemployment and healthy behaviour, 2000-2016 ... 24

Figure 4 - Movements of unemployment and physical health, 2000-2016 ... 24

Figure 5 - Movement of unemployment and social security expenses, 2000-2016 ... 34

Figure 6 - Movement of unemployment and wage per hour, 2000-2016 ... 42

Figure 7 - Movement of unemployment and public debt, 2000-2016 ... 42

Figure 8 - Movement of unemployment and GDP growth, 2000-2016 ... 42

(8)

CHAPTER 1 Introduction

1.1 Motivation

The Great Recession, a worldwide economic crisis that took place from 2008 until 2013, has been the most severe crisis since the Great Depression, a worldwide economic crisis that started in 1929 and ended in 1932 (Bacigalupe, Shahidi, Muntaner, Martín, & Borrell, 2016). As a result of this recession, unemployment increased sharply, leading to deteriorating social and welfare conditions. Economic conditions influence the unemployment rate, income, and leisure time (Gerdtham & Ruhm, 2006) and have a fundamental influence in shaping the distribution of health and disease within and across populations (Bacigalupe et al., 2016). Which in their turn influences people’s health through harm to mental health, lower standards of living, poor dietary habits, and reduced health expenditure (Bender, Economou, & Theodossiou, 2013)

The effect of recessions and changing economic conditions, such as a rise in unemployment, on health has already been studied in the past. Durkheim (1897) was one of the first to study the link between economic activity and health. He showed that economic crises and booms result in more suicides among jobless people. However, studies have led to contradicting findings, even recently. Ruhm (2016) found that unemployment, often used as indicator of a recession, decreased the rate of mortality. While, Charles & DeCicca (2008) found that unemployment lowers physical health. One of the main limitations of prior studies, however, is the lack of research into the influence of recessions, and the rise of unemployment as a result of it, on health outcomes, other than mortality rates (Bacigalupe et al., 2016). Besides, the impact of recessions through unemployment and other socioeconomic experiences on health can become evident many years after the onset of the economic crisis (Bacigalupe et al., 2016). Bender et al. (2013) are one of the few who studied both the short- and long term effects. They found that the temporary effect of unemployment is that it lowers the mortality rate, while the long-term effect is to increase it. Furthermore, most studies, discussed in the literature review, have investigated the effects in the United States, Greece, Finland, or Germany, while nothing specific is found about the effects of the Great Recession, and the rise of unemployment resulting from it, in the Netherlands. The Netherlands is often studied in combination with other European countries, which does not lead to specific results for the Netherlands, but only for Europe.

(9)

Although the economic conditions influence people’s health, it is not precisely known which health factors are influenced and in which direction. For that reason, the influence of the last recession, the Great Recession, on the people in the Netherlands should be examined much more thoroughly.

1.2 Research question

The research question to be investigated is: Did the Great Recession, from 2008 until 2013, influence the health of people in the Netherlands positive or negative, both in the short- and long-term? To that end health is divided into three different categories: mental health, healthy behaviour, and physical health, which are composite variable groups. The Great Recession is defined by unemployment, because unemployment reflects economic fluctuations and is the indicator most often used by other studies (Neumayer, 2004). Data are derived from Statistics Netherlands (CBS) ranging from 2000 until 2016.

The main goal of this study is to take into account as much factors possible that influence health and to make a distinction between these different factors. In prior studies mainly the mortality component of physical health is considered, while this health factor should be composed by much more physical components. Next to this, the second health component, mental health, is studied before but not in combination with other health factors. Moreover, the third health component, healthy behaviour, is not studied before as health component itself. Despite, healthy behaviour may influence the other two health factors. Because, for instance, when people consume a lot of alcohol, which is a behaviour, their physical health is expected to get worse. Thus, it is also considered as a separate category because it has an important, direct influence on people’s health, through for instance cigarette smoking, alcohol consumption, obesity, and dietary habits.

Besides, most studies about the recent economic crisis only focused on the short-term because the research period was not extended long-term after the end of the crisis. In this study both the short, up to and including 2011, and long-term, up to and including 2016, will be considered. With the combination of the three composed health factors, a clear overview of the effects of the state of the economy, using the level of unemployment as indicator, on peoples’ health can be given.

(10)

In addition, we take a look at the social security expenses of the government, because social security for unemployed people leads to diminished effects of unemployment on health (Stuckler, Basu, Suhrcke, Coutts, & McKee, 2009). Besides that, during recessions, unemployed people are confronted with even more disadvantages when the total social security expenses do not increase (Gerdtham & Ruhm, 2006).

1.3 Findings

The thesis shows a negative, significant influence of unemployment on mental and physical health, both in the short- and long-term. A positive, significant effect of unemployment on health behaviour in the short- and long-term is found, with the long-term effect being significantly smaller than the short-term effect. Additionally, significantly opposite effects of unemployment on mental health between men and women is shown in both the short- and long-term. Moreover, unemployment negatively influences the mental and physical health of people younger than 55 years in the short-term. While this age group shows a positive influence of unemployment on healthy behaviour in the long-term. Furthermore, a strong correlation of social security expenses by the government with unemployment is shown. 1.4 Set up

The structure of the thesis is as follows. Chapter 2 discusses existing literature about the influence of recessions on people’s health. Chapter 3 outlines the theoretical framework. Chapter 4 clarifies the data and the used methodology. In Chapter 5 the empirical results will be presented. Concluding remarks are contained in Chapter 6.

(11)

CHAPTER 2 Literature review

Studies about the effect of a recession on people’s health date back to the nineteenth century. Multiple studies about this topic are conducted in the past forty years. In this chapter the most important studies are discussed to give a clear overview of what is already known and what is still unclear.1

2.1 Earlier work

In 1973, Brenner showed that the unemployment rate and rapid economic growth increase stress-related mortality and morbidity and thus decrease physical health in the short-term. The negative relationship between economic growth and health was notable, because the general assumption back then, was that mainly unemployment negatively affects health through stress, associated with financial pressure and the loss of psychosocial assets (Eliason & Storrie, 2009). Furthermore, in 1979, Brenner found an opposite result for the long-term trend in economic growth, where growth reduces the mortality rates, thus leads to an increase in physical health. This indicates different short- and long-term effects for, at least, economic growth. However, in 2004, Laporte evaluated studies done by Brenner and argues that the failure to account for the time-series properties2 is a key omission in his study.

Until 2000, studies usually focused on psychological determinants like stress and risk-taking. However, in economic models since 2000 health is seen as being influenced by factors such as income and the relative price of medical care (Ruhm, 2000). The assumption in these models is that a permanent higher income is associated with health improvements since the budget constraint is shifted out, as is found by Brenner (1973). But unexpectedly Ruhm (2000) found a strong inverse relationship between macroeconomic conditions and physical health, with external sources of mortality as main cause. A possible explanation for this inversion is an increase of obesity and cigarette smoking, and a decrease in diet and exercise. The advantage of this analysis, in comparison to Brenner (1973), is the use of panel data. Panel data are data derived over multiple time periods for multiple entities. Time and entity fixed effects can be included to control for effects that vary over time or vary over entities. In this way, it is possible to control for macroeconomic conditions, such as national policies, and heterogeneity between individuals.

1

(12)

Hereafter, Neumayer (2004) tried to extent and improve Ruhm’s original analysis by using robust standard errors, using a dynamic model and testing for gender differences. One of the drawbacks of Neumayer (2004) and others, who only use data on mortality, is that it would be better to use morbidity, as mortality captures only the extreme consequences of bad health. The main effect found by Neumayer, as found by Ruhm (2000), is a decrease in mortality during recessions. Notably this seems to occur in the short run.

Additionally, using data from 23 countries, Gerdtham & Ruhm (2006) extended the study by Ruhm (2000) and Neumayer (2004). They examined whether deaths rise when labour market conditions improve and if it is generalizable across industrialized countries. They found that unemployment and the mortality rate move in opposite directions. Just like, Ruhm (2000) and Neumayer (2004), unemployment seems to have a positive influence on physical health. Gerdtham & Ruhm (2006) emphasized that it is important to keep in mind that reductions in mortality during bad times need not be restricted to those becoming newly unemployed because unemployment is used as main proxy for the economic condition and that not all types of health respond in the same way.

Somewhat later, Stuckler et al. (2009) argued that existing studies are incomplete because they do not analyze the effects per se on health or used GDP as measurement of the state of the economy. Unemployment is a more widely available indicator than GDP, and GDP relates to the average income and does not account for inequality within countries. Stuckler et al. (2009) used age-standardized and age-specific mortality data for their study, because earlier studies show that effects vary substantially for different age groups. They found that a rise in unemployment leads to a short-term decrease in road-traffic deaths. They did not find other significant effects, which leads to an insignificant, negative effect on mortality overall. Thus again, unemployment improves physical health.

A few years later, Ariizumi & Schirle (2012) studied the relationship in Canada, because until then most of the studies involve the United States. Although the labour markets of the countries are comparable, they expected different results because the health care institutions are not comparable. The Canadian working age population benefits more from universal health care coverage, than people in the United States. In general they found a significant negative relationship between unemployment and mortality, thus a positive effect of unemployment on physical health. Only when age groups are examined, they find a difference

(13)

This suggests important cross-country differences in the quality of health care provided to seniors and expectant mothers, but not for the working age population.

A positive effect of unemployment on health is not always found. Economou, Nokolaou, & Theodossiou (2008) used the same approach as Ruhm (2000), but innovated by taking into account behaviour such as cigarette smoking and dietary habits which could influence the relationship. They found, in contradiction to what is discussed above, a negative relationship between unemployment and physical health.

At the same time, the relationship found by Economou et al. (2008) is also found by Charles & DeCicca (2008). They examined the relationship between labour market conditions and several measures of health and health behaviour instead of only mortality. They expected, based on literature, that physical health moves countercyclical, but on the contrary they found a procyclical relationship. Countercyclical means a negative correlation with the overall economic cycle, while procyclical means a positive correlation. An explanation is the observation that people tend to gain weight during recessions. Notable for their study is that they only included men ranging from 24 until 59 years old.

Coile, Levine, & McKnight (2014) found a positive effect of unemployment on physical health as well. However, they focused on a specific group of workers, close to their date of retirement. During recessions older unemployed workers are less likely to be re-employed, than younger workers. Coile et al. (2014) showed that workers between an age of fifty through sixty experience only short-term health benefits from recession. These benefits are immediately offset by subsequent health deterioration. This is not observed for people older than sixty years. Notable is that this study is executed in the United States, where people who lose their job may also lose their health insurance. This is not the case for the Netherlands. In addition, unemployed people born before 1 January 1965, receive extra social protection in the Netherlands.

Avendano, Moustgaard, & Martikainen (2017) focused as well on social protection and seek to find an explanation for the unobserved reduction in mortality during recessions in Nordic countries. This positive effect of unemployment on health is also found by Ruhm (2000), Neumayer (2004), Gerdtham & Ruhm (2006), Stuckler et al. (2009), and Ariizumi & Schirle (2012). Avendano et al. (2017) presumed that social benefit programs and benefits in those countries protect populations against the effects of economic downturns. Instead of aggregate

(14)

This enables to control for individual characteristics and examine heterogeneity. Their main finding suggests that recessionsdo not affect mortality.

As written above, the use of panel data allows to control for macroeconomic conditions and heterogeneity between individuals, but without decomposing the effects into temporary and permanent effects in these fixed effect models, panel data only allows an estimation of the short-term effects (Bender et al., 2013). Bender et al. (2013) innovated by distinguishing between the temporary and permanent effects of unemployment on mortality. For the temporary effect, in line with, amongst others, Ruhm (2000) and Gerdtham & Ruhm (2006), they found that unemployment has a negative, but insignificant, influence on mortality for all causes, and thus a positive effect on physical health. However, when they looked at the permanent effects they found that a 1 percent increase in unemployment leads to a 1.5 percent significant increase in mortality rates. The explanation given by Bender et al. (2013) for this variation in direction is that the permanent effect may take some time to appear. Besides this, they also showed that the permanent effect is much stronger, than the temporary effect. Besides the influence of unemployment on physical health, certain studies also investigate the influence on mental health. This is mainly studied by taking into account the number of suicides. Gerdtham & Ruhm (2006) found some evidence that unemployment is positively related to suicides. The main contribution of Charles & DeCicca (2008) is that they explore a more detailed measure of mental health. They expected, from literature, that mental health moves in the same direction as labour market conditions. The same effect is found by Stuckler et al. (2009) who show that increased suicide is mostly attributed to economically constrained families in lower socioeconomic groups. However, Avendano et al. (2017) only observed a rise in suicides because of increased unemployment for higher educated men. No evidence is found for women.

As already done by Economou et al. (2008), Stuckler et al. (2009) took into account certain types of behaviour. They looked at especially large rises in unemployment and found a significant increase of deaths from alcohol abuse. This indicates that short-term negative effects of unemployment mainly affect psychological distress. It is remarkable that Economou et al. (2008) and Charles & DeCicca (2008), as one of the few to consider behaviour, found different effects than many other studies. This emphasizes the importance of not only taking into account mortality.

(15)

Furthermore, it is remarkable that some studies only focus on short-term effects, such as Ruhm (2000) and Stuckler et al. (2009), while Coile et al. (2014) and Bender et al. (2013) showed different effects in the short- and long-term.

2.2 Great recession

In 2013, physical health in Spain is studied by Regidor, Barrio, Bravo, & de la Fuente (2013) during the Great Recession by examining fifteen physical health indicators. Almost all indicators of mortality showed a significant downward trend during the recession period in the short-term. Thus, the physical health improved, as found in most studies before the Great Recession, during the first four years of the economic crisis in Spain. Unlike most studies, not unemployment, but the gross domestic product adjusted for purchasing power parity is used as indicator of the state of the economy. According to Stuckler et al. (2009) this is not the right indicator for a recession, because it relates to the average income of a country. They argue that instead measures such as unemployment and consumer confidence are more relevant, because these capture economic turmoil and insecurity faced by individuals.

Again, Ruhm studied this topic but extended his research period to the Great Recession and tried to find whether the relationship between economic conditions and mortality changes over time. Ruhm (2015) demonstrated that mortality shifted from a strong procyclical relationship with economic conditions, to a weak or unrelated relationship. This is mainly proven for males. Furthermore, results derived from short time periods, show instability and are unreliable.

A year later, Ruhm (2016) studied the same topic again, but this time he tried to examine whether health effects of economic crises differ, compared to those of more standard economic downturns. The findings suggest an inverse relationship between unemployment and mortality rates. Ruhm showed a more noticeable decline in mortality during the fierce recessions, compared to mild recessions, and showed that unemployment is at the highest level at the end of the recessions.

Until 2017, only the short-term impact of the Great Recession has been investigated. This is mainly due to the fact that the economic crisis happened recently. Filippidis, Gerovasili, Millett, & Tountas (2017) examined the impact for a longer time period for Greece. Also, they used a comprehensive set of health indicators including mortality, mental health,

(16)

They observed a decline in mortality rates during the crisis, but in the last year observed, 2015, the mortality rate increased. Cigarette smoking and physical inactivity declined during the crisis, which is possibly correlated with less observed traffic accidents.

The long-term decrease in health because of unemployment found by Filippidis et al. (2017) is also found in the short-term by Bacigalupe et al. (2016) and Ruhm (2015). Ruhm (2015) showed that cancer and some external causes of mortality appear to move countercyclical. Bacigalupe et al. (2016) studied the influence of the Great Recession on three European countries, which have experienced and responded in different ways. The economic crisis has been severe in Greece and Spain, while Ireland only suffered a short period. The results showed a negative influence of the crisis on health in all three countries, but in particular for Greece. Especially the effects on mental health, physical well-being and increased suicide-related mortality are strong. No clear increase in total mortality rates exist for Greece and Spain during the crisis. The latter is explained as the aftershocks are still unfolding and the permanent effect of an increase in unemployment could be that the mortality rates only increase in the longer term.

Gili, Roca, Basu, McKee, & Stuckler (2012) studied the effect of unemployment on mental disorders in Spain during the Great Recession. They found an increase in mental disorders during the recession in the short-term. The rise in unemployment has spread effects. It does not only impact the unemployed, but also their family members and members of communities. Next to the effect of unemployment, Gili et al. (2012) found that mortgage payment difficulties also influence the risk of mental health disorders.

Filippidis et al. (2017) also showed increased suicides during the economic crisis. Quality of life scores and unmet healthcare needs decreased significantly. This increase in suicides and depressive symptoms reflect an overall decrease in mental health.

The general effects found on health during the Great Recession are the same as found for prior recessions. The short-term effect of unemployment on physical health is positive most of the time, while negative effects are found as well. The long-term effect on physical health is negative, shown by Bender et al. (2013) and Filippidis et al. (2017). Unemployment always shows a negative effect on mental health.

(17)

CHAPTER 3 Theoretical framework

In this chapter the most important literature and findings from the literature review is used to develop our theoretical framework and to derive our methodology.

3.1 Health

The state of economic affairs influences health in both the short- and long-term (Bender et al, 2013). But the extent and direction can be different for the short- and long-term. Besides, the various health factors respond differently (Charles & DeCicca, 2008).

Mortality rates are often used as indicator for people’s health because they represent the most severe negative health outcome and are well measured (Ruhm, 2016). But, Charles & DeCicca (2008) and Gili et al. (2012), for example, showed that mortality is not the only health factor influenced by the state of the economy, and by the level of unemployment, as being regarded as the main contributing factor. Mental health, healthy behaviour, like obesity, cigarette smoking and health care spending, as well as other physical health factors, than mortality rates, are also important to be analyzed (Charles & DeCicca, 2008).

Not only unemployment influences health. Also individual characteristics, in combination with unemployment, such as age, gender and education, play a role. Education is clarified below, age and gender are clarified in combination with the three different health factors. Because financial difficulties are often linked to losing one’s job or becoming unemployed, those who lose jobs are also vulnerable to the adverse effects of the economic crisis (van Hal, 2015). Edwards (2008) found that the least well-educated workers are at greater risk for ill health when jobs are lost during times of economic contraction. Those with a higher education, who presumably have some buffer-stock savings and decent prospects of avoiding long-term unemployment, actually seem to benefit during economic hard times, perhaps from working less hard or being exposed to less pollution. However, Avendano et al. (2017) showed that during a recession, mortality among tertiary educated people increases, driven by an increase in mortality from cardiovascular disease and suicide, while no relation is found for less educated people. But, Bacigalupe et al. (2016) note that this is different for women. The self-rated health increased among the most educated women, while it deteriorates among the most disadvantaged during a recession.

(18)

3.1.1 Mental health

The mental health component does not only consist of mortality caused by suicides, because suicides are often a delayed consequence of an underlying mental health disorder, rather than an immediate response to stressful life events (van Hal, 2015). Brenner (1973) mentions that stress related to unemployment is caused by uncertainty about the future. Unemployed people not only lose in material respect, but potentially they also lose social networks, self-esteem, self-confidence, and a purpose for life (Neumayer, 2004). However, Avendano et al. (2017) argues that this is mainly the case for highly educated men, because of higher career expectations, potentially larger income losses, and scaring effects of job loss. Another explanation is the presence of family responsibilities. Marriage increases the risk of poor mental health among men in the manual group, whereas, among women, this acts as a buffer (Artazcoz, Benach, Borrell, & Cortès, 2004). Additionally, van Hal (2015) proves improved mental health among employed women during a recession, because they may experience additional recognition and greater self-esteem in a new breadwinner role.

Stuckler et al. (2009) show that short-term negative effects of unemployment mainly affect psychological distress. Bender et al. (2013) distinguish temporary and permanent effects of a recession. They found a positive, significant effect of unemployment on mental disorders as cause for mortality, both temporary and permanent, the latter being stronger. Filippidis et al. (2017) found a significant increase in suicides during and after the economic crisis compared to pre-existing trends. Thus, a recession negatively influences mental health because of poor access to mental health services, untreated mental disorders, unemployment and financial losses (Filippidis et al., 2017).

3.1.2 Healthy behaviour

Ruhm (2000) already considers that some causes of mortality are sensitive to behavioural responses, such as cigarette smoking, alcohol consumption, and physical exercise (Charles & DeCicca, 2008). However, Economou et al. (2008) are one of the first to take such behaviour into account in their regression model. Behaviour of people could change differently according to two opposite perspectives (Neumayer, 2004). The first perspective focuses on the material losses associated with unemployment and the material insecurity for the employed, which leads to lower personal health-related expenditures and possibly unhealthy diets because people purchase less expensive food (van Hal, 2015).

(19)

The other perspective focuses on utility maximization, where economic upturns have a negative effect on healthy behaviour, and the other way around.

Four reasons are mentioned why health declines because of behaviour during an economic upturn. First, the opportunity costs of leisure time increase during upturns, meaning less time is spent on health-preserving activities like cooking and medical check-ups. Second, work-related accidents increase during an upturn because car driving increases. Third, temporary increases in income increase the use of goods which are damaging for the health, like alcohol and tobacco. Fourth, health is an input into the production of goods and services. Hazardous working conditions, the physical exertion of employment, and job-related stress have negative effects on health (Gerdtham & Ruhm, 2006). If one perspective is observed, it does not mean that the other does not exist, because it is possible that one of the effects is stronger (Neumayer, 2004) or takes some time to appear (Bender et al., 2013).

Bender et al. (2013) found the second perspective in the short-term, while the first perspective is observed in the long-term. Thus, healthy behaviour increases in the short-term, but then decreases because of material losses in the long-term. The latter, permanent, effect is eventually stronger.

3.1.3 Physical health

As mentioned above, most studies focus on mortality rates, which is the most severe physical component of health. But physical health consists, among others, of weight (Charles & DeCicca, 2008) and life expectancy (Bezruchka, 2009). In the short-term, physical health might increase due to less cigarette smoking, increased physical exercise, and healthier diets on behalf of unemployment (Ruhm, 2016). However, in the long-term it might decrease because of reduced health expenditures and lower standards of living (Bender et al., 2013). The health of females is in general less harmed by unemployment, than the health of males, especially in the long-term (Bender et al., 2013).

It is also possible to analyse different mortality causes as indicator for physical health. This is already extensively studied by others. A distinction between external and internal sources of death is often made. External sources of death consist, for example, of transport accidents, other non-transport accidents, suicides and homicides, while internal sources, mainly diseases, can be grouped in cardiovascular, cancer and other diseases.

(20)

Transport fatalities move procyclical to economic activity, because during economic contractions people travel less frequently and thus have a lower probability of traffic accidents (Bender & Theodossiou, 2009). Other accidents, like accidental poisonings have changed over the years from being strongly countercyclical to procyclical (Ruhm, 2015).

An explanation given by Ruhm (2015) is a greater availability of certain poisoning drugs, resulting in self-injury and accidental death during bad economic times. This increased procyclicality to unemployment may be a physical manifestation of an original mental health problem. Suicides are discussed before, and they increase with a rise in unemployment. Cancer appears to be strongly countercyclical to the economy, because cancer mortality is sensitive to the availability of financial resources and access to health care (Ruhm, 2015). Whereas, cardiovascular disease appears to be strongly procyclical, this suggests sensitivity to changes in lifestyles, environmental factors, and medical interventions (Gerdtham & Ruhm, 2006).

3.1.4 Short- and long-term

Ruhm (2000) gives two basic reasons why the short- and long-term may differ. In the long term, people have greater flexibility in making consumption, time allocation, and production decisions to improve health. Besides that, even small negative shocks, can cause fragile individuals to die sooner, whereas having a small effect on life expectancy or population health.

3.1.5 Differences now and in the past

As mentioned in the literature review, Ruhm (2015) shows that mortality shifted from being strongly procyclical to being weakly related or unrelated to economic conditions. He gives two possible explanations for this observed variation. First, it may be part of a change in the effects of short-term changes in economic performance, or in the role of unemployment as indicator of the economy. Second, the higher death rates of accidental poisoning reflect greater availability of these drugs, as mentioned in the physical health part. Another possible explanation is given by Bacigalupe et al. (2016) who argue that policy responses to the recent economic crisis differ significantly from those to past crises. Governments responded to the recent crisis by imposing severe austerity measures, while before they adopted expansionary policy agendas.

(21)

3.2 Heterogeneity countries

3.2.1 Social uncertainty/labour protection

Gerdtham & Ruhm (2006) are one of the first to include more countries in their regression model. They observe heterogeneity between the different countries and note that this is caused by differences in social insurance systems or labour protection laws.

The mortality rate fluctuates more for countries with relatively weak social insurance protections. Individuals in those countries work particularly hard during good economic times to offset the effects of reduced income during downturns. Stuckler et al. (2009) add to this that investments in active labour market programs lead to a lower effect of the rise in unemployment on suicide rates. Avendano et al. (2017) relate the unobserved reduction in mortality in some countries during economic downturns to two explanations. First, during downturns the real wages remain largely unchanged because of the large role of unions in collective agreements related to the real wages in those countries. Second, the people who lose their job could benefit from the generous unemployment benefit system. Thus, it seems that the effects on health of recessions are largely dependent on the fiscal austerity policy and social protection measures in a country (van Hal, 2015). Gerdtham & Ruhm (2006) showed that the Netherlands belong to the top category of countries that spend an average of 28.7 percent of GDP on public social programs.

3.2.2 Health care services

Budget cuts during economic recessions affect health care services adversely, just when the levels of need and demand are rising (Cooper, 2011). Economic health care could be best achieved not by decimating services but by planning and deploying these services to meet the needs of defined area populations. This applies more for low-income and transitional societies, in which even small reductions in health service budgets can be very damaging. It is important to investigate if cost-cutting measures in the health sector do lead to an increase in social inequities (van Hal, 2015). The Netherlands decreased the extent of coverage by instituting or increasing user charges for some health services in response to the economic crisis (Karanikolos, et al., 2013). This suggests possible consequences because rises in user charges are a particular cause of concern. They might increase the financial burden on households and probably reduce the use of care, especially by people with low incomes and high users of health care, even when user charges are low (Karanikolos, et al., 2013).

(22)

CHAPTER 4 Methodology and Data

4.1 Data

The data used for the study are derived from Statistics Netherlands (CBS) located in the Netherlands. The CBS is a Dutch governmental institution with the goal to publish reliable and consistent statistical information which responds to the needs of society. All the data can be found in StatLine, the database of the CBS. The program Stata is used to summarize the data, run the regressions, and perform the tests.

Time-series data from 2000 until 2016 are gathered on the three dependent composite health variables: mental health, healthy behaviour and physical health. Table 1 summarizes the descriptives of the data and shows by which variables the three composite health variables are defined. The first column, Variable, shows all the variables. In the second column, Observations, the number of observations of each variable are given. Most variables consist of seventeen observations, one for each year from 2000 until 2016. Four variables, sum score mental health, unhappy, depression and anxiety, and exercise, show nine observations. No data are available for these variables for the remaining six years. The variable medical services misses two years of data. In the third and fourth column, Mean and Standard Deviation, the mean and standard deviation are presented. In the fifth and sixth column, Minimum and Maximum, the minimum and maximum value of all the variables is shown. The last column, Composite Health Variable, shows which variables are used to define mental health, healthy behaviour, and physical health.

Table 1 - Descriptives statistics, 2000-2016

(1) (2) (3) (4) (5) (6)

VARIABLES Observations Mean Standard

Deviation

Minimum Maximum Composite

Health Variable

Years 17 2008 5.0498 2000 2016

Unemployed 17 .0517 .0128 .0334 .0744

Wage per hour 17 19.8018 1.6186 16.44 22.13

Public debt 17 55.6529 8.2672 42.7 68

GDP 17 592125.4 78021.37 448061 702641

GDP growth 17 .0297 .0238 -.0338 .0639

(23)

(1) (2) (3) (4) (5) (6)

VARIABLES Observations Mean Standard

Deviation

Minimum Maximum Composite

Health Variable

Psychic mortality 17 7779.588 2023.077 5133 11957 MH

Sum score mental health 9 79.0111 .3920 78.5 79.7 MH Unhappy 9 15.6778 .8043 14.4 16.9 MH Depression or anxious 9 16.3222 .9731 15.1 17.8 MH Experience health less than good

17 19.6 .5244 18.6 20.6 MH Antidepressant 11 5.6 .2646 5.1 5.9 MH Health expenditures 17 76999.76 16334.18 46452 96711 HB Cigarette smoking 17 28.8824 3.4634 23.9 34.7 HB Alcohol 17 83.1118 1.7360 80 85.9 HB Total sport 9 133 4.2720 128 139 HB Medical service expenses 14 2.5143 3.7874 -8.4 6.6 HB Mental health expenditures 17 5.2 1.3747 2.7 6.7 HB Prescribed medicines 17 37.6529 2.0664 33.8 41.2 HB Overweight 17 40.5177 1.7576 37.7 43.3 PH

Life expectancy men 17 78.0053 1.4961 75.54 79.88 PH

Life expectancy women 17 82.1212 .9548 80.58 83.29 PH Mortality internal 17 125522.7 3464.709 121338 130585 PH Mortality external 17 4005.412 398.7217 3395 4974 PH Mental health 17 -.0614 .5426 -.9184 .9062 Healthy behaviour 17 .0487 .5824 -.9145 .8254

Physical health 17 -8.59e-08 .3432 -.5354 .4663

Age 17 .1211 .0328 .0685 .1719

Men 17 .5422 .0099 .5291 .5619

Education low 17 .2600 .0219 .2221 .2951

Education middle 17 .4292 .0101 .4175 .4497

Education high 17 .3022 .0279 .2598 .3481

Notes: Years is the time component in years, unemployed is the percentage unemployed of the labour force, wager per hour is the average wage in euros, public debt is the percentage public debt of GDP, with GDP in million euros, GDP growth is the percentage growth of GDP each year, social security expenses is in million euros, suicides is the amount of suicides, psychic mortality is the amount of people who died with a psychic cause, sum score mental health is in the percentage of the

(24)

average person, medical service expenses are volume mutations in medical service expenditures, mental health expenditures is in million euros, use of prescribed medicines is the percentage of total use of medicines that is prescribed, overweight is in the percentage of the population with overweight, life expectancy of men and women is in years of age, mortality internal and external is in the amount of deaths, mental health, healthy behaviour and physical health are the standardized health variables, age is in the percentage of the labour force which is 55 years of age or older, men is in the percentage men of the labour force, and education low, middle and high is in the percentage of the labour force for which it is the highest achieved education.MH means mental health, HB means healthy behaviour, and PH means physical health.

As can be seen in Table 1, also data of non-health variables are gathered. Unemployment, wage per hour, public debt, and GDP growth are used as indicators of the state of the economy. The variable social security expenses is included to check whether the government of the Netherlands has spent more money on unemployed people. Because individuals are heterogeneous in characteristics such as age, gender, and education.3 These characteristics are included as control variables. Age is representing the percentage of the labour force which is 55 years of age or older. A distinction between younger and older than 55 is made because of the social protection, the IOAW benefit, exists for unemployed people born before January 1st 1965 in the Netherlands.4 Because Gerdtham & Ruhm (2006) mention that social protection influence the effects of the business cycle on health, we control for this specific age category. Moreover, age independently affects death rates (Gerdtham & Ruhm, 2006). Men represents the percentage of men in the labour force and education is divided in three different levels, low, middle or high. Each level, defined by the CBS, represents the percentage of the labour force for which it is the highest achieved education. Two levels (low and middle) are included in the regression to prevent multicollinearity.

4.2 Methodology

As mentioned above, the three general health variables are created, mental health, healthy behaviour, and physical health, from all the health variables gathered. First, all different health variables are standardized using the formula:

Standardized variables are used because in this way absolute differences are deleted and only relative differences are left. All standardized variables contain the same average of zero and standard deviation of one. In this way, we are able to compare the variables with each other and take them together to create one general variable.

3_{To control for gender the variable men is included.}

(25)

These standardized variables are divided over the three different health variables – mental health, healthy behaviour, and physical health – sum them up per group and divide them by the number of variables in that specific group, for each year.

No prior studies created composite health variables. In prior studies mental health was measured by suicides, physical health by mortality and healthy behaviour was not used in this way. Because the standardized variables contain only relative differences, all variables are equally weighted and taken into account.

Some of these standardized variables are multiplied by , so they move in the same direction as the general health variable. For example, if overweight decreases, it means that our physical health increases. While, when life expectancy increases, it means that our physical health also increases. Thus, the first variable of the example, overweight, is multiplied by .

These steps result in a standardized variable for mental health, healthy behaviour, and physical health. We regress these three health variables separately on the crisis variable. In this regression four control variables are added: age, men, low education, and middle education.

To get the short-term effects of the crisis on the dependent variables, the following regression equations are used, using data until 2011:

(26)

Unemploymentt is the percentage unemployed of the labour force in year , Aget is the percentage of people who are older than 55 years in the labour force in year , Ment is the percentage of men in the labour force in year , EducationLowt is the percentage of the labour force which highest education is level low in year , EducationMiddlet is the percentage of the labour force which highest education is level middle in year , and is the error term. These regression equations are approximately the same as used by Economou et al. (2008) and Bender et al. (2013), except for the inclusion of fixed effects in their regression models and for taking into account gender in the mortality rates. This difference is caused because no panel data are used, thus we are not able to include fixed effects, and do not only use mortality rates as dependent variable.

To make a distinction between the short- and long-term effects of the economic crisis, we regress these same equations until 2016 to get the long-term effects.

4.3 Descriptives of the variables

While almost all prior studies take unemployment as indicator of the economy (Neumayer, 2004), which gives sufficient reason to use it as well, we check if the unemployment rate in the Netherlands represents the Great Recession as well. This is done by verifying if unemployment increased during the Great Recession and by examining how unemployment correlates to other indicators of the economy.

In Figure 1, the movement of unemployment over time from 2000 until 2016 is shown. Most authors agree that the Great Recession started in 2008 and ended in 2013 (van Hal, 2015). In the figure a change in the direction of unemployment is observed, from decreasing to increasing in 2008. Ruhm (2016) shows that unemployment often reaches a maximum after recessions officially end. As observed, unemployment in the Netherlands indeed reached a maximum in 2014, a year after the Great Recession ended and started decreasing afterwards.

.0 5 .0 6 .0 7 .0 8 U n e m p lo y m e n t

(27)

Other possible indicators of the state of the economy are GDP growth, public debt and the wage per hour.5 Public debt is another good indicator of the state of the economy during the Great Recession, because of the negative correlation between debt and growth (Panizza & Presbitero, 2014). This applies to the Netherlands as well. However, because GDP growth shows high variations during the economic crisis while the wage per hour increased, we cannot rely on these two variables as indicators of the economy.6 GDP growth continues to increase during the Great Recession because of discretionary fiscal policies (Coenen, Straub, & Trabandt, 2012). Moreover, GDP is not a very precise measure of the economic circumstances (Hayo & Seifert, 2003). Wage per hour is not a good indicator of the economy for the Netherlands as well, because it is not directly influenced by macroeconomicconditions because of the large role by the labour unions (Avendano et al., 2017).

By using the skewness and kurtosis test for normality we test whether the variables are normally distributed.7 The null thypothesis for the variable medical services is rejected (0.0023, 0.0079, and 0.0024). The mortality internal cause and the healthy behaviour variable are significantly different from the normal distribution for the kurtosis (0.0151 and 0.0264) and mortality internal cause is significantly different from the normal distribution for the overall test statistic (0.0498). This means that for the other variables no significant departure from normality is found.

It is important that the three health variables are not strong correlated, because we assume that those are influenced differently by unemployment.8

5

In Table 6 in the Appendix, the correlations of these three variables with unemployment are given. Unemployment and wage per hour have a strong, positive correlation. The correlation with public debt is somewhat lower, but still above 0.5 and positive and the correlation with GDP growth is low.

6_{The three other indicators of the economy are put in the same graph as unemployment to observe how they}

move together over time, see Figures 6, 7, and 8 in the Appendix. From these figures a positive relationship between unemployment and wage per hour is observed. Wage per hour is increasing over time, including during the crisis. When we take a look at public debt, we see that it moves together with unemployment. A weak relationship between GDP growth and unemployment was not expected. When we take a look at the figure, they decrease and increase at the same time, only at other rates. The Figure 4 shows a high variation of GDP growth.

7

See Table 7 in the Appendix. With a probability of 0.05 or lower for the skewness, we reject the null hypothesis and assume that the variable is significantly different from the skewness of a normal distribution at the 5 percent significant level. With a probability of 0.05 or lower for the kurtosis, we assume that the variable is significantly different from the kurtosis of a normal distribution, thus reject the null hypothesis at a 5 percent significance level. With a probability of 0.05 or lower for the overall test of normality we assume that the distribution of the variable is significantly different from a normal distribution for a 5 percent significance level.

8

(28)

When the correlation for unemployment is checked, for the 3 health variables and for the control variables, we see that some of those are correlated, see Table 9 in the Appendix. However, the 3 health variables are weak or moderately correlated with each other.

The moderate correlation between healthy behaviour and physical health is unexpected, because when people for example smoke cigarettes, which is a behaviour, their physical health is expected to get worse. The moderate correlation can be explained by the delayed influence of healthy behaviour on physical health. To control for this, a lag variable of two years of healthy behaviour is created, using the command "gen Healthy behaviour 2 = Healthy behaviour [_n-2]”. The correlation between the lagged variable of healthy behaviour and physical health is 0.3591. This is even lower than the correlation presented in Table 9. Many of the strong correlation relations are found between the control variables. This could mean that some multicollinearity between the control variables exists. This represents a strong correlation between two or more independent variables and creates redundant information. The variance inflation factor command, “vif”, is used after running a regression, to check for multicollinearity. A variable with a VIF value greater than ten may merit further investigation. See Table 10, for the results of this test for the independent and control variables.9 The average VIF is 15.52, which actual means multicollinearity. This high average VIF value is caused by high VIF values for the control variables age and education middle. Because these are control variables, no concern is needed about the VIF value of these variables.

4.4 Hypothesis

The hypotheses are tested by ordinary least squares (OLS) regressions, using robust standard errors, with the equations above. Two opposite effects at the beginning and during the Great Recession are expected. The mental health immediately decreases, because of stress about being unemployed and the increased risk of losing a job. Thus unemployment has a negative effect on mental health in the short-term, resulting in the following first hypothesis:

(29)

We presume that healthy behaviour and physical health increases, caused by, for example, more leisure time, increased physical activity, and reduced cigarette smoking. Thus a positive effect of unemployment on healthy behaviour and physical health in the short-term is expected, resulting in the following second hypothesis:

At the end of the Great Recession, the long-term effects of unemployment come into effect, resulting in a total negative effect on a person’s health. We presume that lower standards of living and reduced health expenditures will result in a decrease in physical health at the end of the Great Recession. Thus a negative effect of unemployment on all three health variables in the long-term is expected, resulting in the following third hypothesis:

These three hypotheses are tested with a t-test on the coefficient of unemployment. To run an OLS regression a few assumptions about the sample have to be made. In this Chapter the test for multicollinearity is already performed. This test showed that multicollinearity exists in the control variables. However, it should not be concerning. Furthermore, some tests about the normality of residuals, the serial correlation, and the heteroscedasticity need to be performed (de Souza & Junqueira, 2005). These will be discussed in the next Chapter, Results.

(30)

CHAPTER 5 Results

We start with providing a first indication of the relationship between health and unemployment in a graph. Then some tests about the assumptions of an OLS regression are performed. Finally, the results of the OLS regressions and some additional regressions are discussed.

5.1 Figures

A first indication of the relationship between health and unemployment is provided in Figures 2, 3, and 4. In Figure 2, we observe that the mental health started to decrease around 2008, at the exact same moment when unemployment started to increase. And the other way around, mental health started to increase, when the unemployment, after a period of increasing, started to decrease at the end of the economic crisis. From the graph we conclude that unemployment has a direct negative influence on mental health in the short- and long-term. In Figure 3, an increase of healthy behaviour is observed at the same time when unemployment increased. Furthermore, identical movements of unemployment and healthy behaviour are observed. From this, we conclude that unemployment has a positive effect on healthy behaviour. In Figure 4, a stable line of physical health during the start of the economic crisis is observed, which starts to decrease just when unemployment also starts to decrease. This indicates that the negative physical health effects of unemployment appear some years later. From this figure we conclude that unemployment has a delayed, negative effect on physical health.

-1 -. 5 0 .5 1 M e n ta l H e a lt h .0 3 .0 4 .0 5 .0 6 .0 7 .0 8 U n e m p lo y m e n t 2000 2005 2010 2015 Years

Unemployment Mental Health

(31)

5.2 Robustness Checks 5.2.1 Normality of residuals

Normality of residuals is required to assure the p-values for the t-tests and F-test to be valid. The swilk test is used which performs the Shapiro-Wilk W test for normality (Shapiro & Wilk, 1965). First the residuals are created, using the predict command “predict r, resid”. Then, the command “swilk r” is used to perform the test. The results of this test are presented in Table 11 in the Appendix.1011 We conclude that all the residuals are normal distributed.

10 -1 -. 5 0 .5 1 H e a lt h y B e h a v io u r .0 3 .0 4 .0 5 .0 6 .0 7 .0 8 U n e m p lo y m e n t 2000 2005 2010 2015 Years

Unemployment Healthy Behaviour

-. 5 0 .5 P h y s ic a l H e a lt h .0 3 .0 4 .0 5 .0 6 .0 7 .0 8 U n e m p lo y m e n t 2000 2005 2010 2015 Years

Unemployment Physical Health

Figure 3 - Movements of unemployment and healthy behaviour, 2000-2016

(32)

5.2.2 Serial correlation

Serial correlation, also known as autocorrelation, is the similarity between observations as a function of the time lag between them. This is tested with a Durbin-Watson test (de Souza & Junqueira, 2005), using the command “dwstat”. The results of this test are presented in Table 12 in the Appendix.12 The value of healthy behaviour indicates negative, and physical health positive autocorrelation, thus an extra test for serial correlation is performed.

The Breusch-Godfrey test tests for higher-order serial correlation in the error term (Johnston, 1984), using the command “estat bgodfrey”. The results are presented in Table 13 in the Appendix.13 This indicates autocorrelation for physical health. This problem is solved by using robust standard errors in the OLS regressions.14

5.2.3 Homoscedasticity

Another assumption we test for is homoscedasticity, which implies that the variance of the residuals is constant. The Breusch-Pagan test is used (Breusch & Pagan, 1979), with command “estat hettest”. The results of this test are presented in Table 14 in the Appendix and indicate that the assumption of homoscedasticity is correct.1516

5.2.4 Stationarity

Stationarity is tested, which means that the probability distribution does not change over time, because time-series data are used (Laporte, 2004). The DF-GLS test is used (Elliott,

Rothenberg, & Stock, 1992), with command “dfgls variable”.17 The results of this test are presented in Table 15 in the Appendix. The null hypothesis of a unit root is only rejected at the 10 percent level for healthy behaviour for lags three and four.18 These results indicate stationarity.

11

The results of this test for the data ranging from 2000-2011 leads to the same conclusion.

12

The Durbin-Watson value of mental health is close to 2, whereas the value of healthy behaviour is above 2 and of physical health below 2.

13

The Breusch-Godfrey test shows a p-value of 0.1235 for healthy behaviour and 0.0279 for physical health. This means that the null hypothesis of no serial correlation is rejected for physical health. Indicating

autocorrelation for physical health.

14

The results of the Durbin-Watson and Breusch-Godfrey test for the data ranging from 2000-2011 results in autocorrelation for healthy behaviour. By using robust standard errors this is corrected.

15

P-values of 0.6324, 0.5847, and 0.3866 are found, which means that the null hypothesis is not rejected for any of the three health variables. The null hypothesis is homoscedasticity.

16

The results of the Breusch-Pagan test for the data ranging from 2000 until 2011 leads to heteroscedasticity of healthy behaviour. By using robust standard errors this is corrected (Ettner, 1996).

(33)

5.3 Hypotheses 5.3.1 First Hypothesis

The first hypothesis is a negative effect of unemployment on mental health in the short-term:

The following equation is estimated by OLS for this hypothesis:

where ranges from 2000 until 2011. To test this hypothesis, we regress mental health as dependent variable on unemployment as independent variable and age, men, education low, and education middle as control variables, using robust standard errors. The results of this regression are presented in Table 2 below.

This table shows a negative, significant effect of unemployment on mental health. This confirms our first hypothesis that an increase in unemployment, leads to a decrease in mental health in the short-term. Thus, the null hypothesis is rejected. Also a positive, significant effect for men is observed, which means that mental health of men is better than of women. Furthermore an F-value of 8.44 with probability 0.0109 is observed, which means that the null hypothesis is rejected because all the coefficients are zero. The R2 of the regression is 0.6907, which indicates that 69.07 percent of mental health is explained by the model. The root MSE is 0.2937, which represents the standard deviation of the residual in this model. This is an absolute measure of fit, with closer values to zero meaning a better fit, so this model predicts the response well.

5.3.2 Second Hypothesis

The second hypothesis is that unemployment has a positive effect on healthy behaviour and physical health in the short-term:

(34)

The following equations are estimated by OLS for this hypothesis:

where ranges from 2000 until 2011. To test this, we regress healthy behaviour and physical health as dependent variables on unemployment as independent variable and age, men, education low, and education middle as control variables, using robust standard errors. Table 2 gives the output of the effects of unemployment on healthy behaviour and physical health in the short-term.

A positive and significant effect of unemployment on healthy behaviour is observed. This confirms the hypothesis that an increase in unemployment, leads to an increase in healthy behaviour in the short-term. Thus, our null hypothesis is rejected. Besides, a positive, significant effect for age is observed. This means that people who are older than 55 years behave healthier in the short-term. Also a significant, but negative, effect is observed for men. This indicates that men show less healthy behaviour than women. Furthermore, we observe an F-value of 17.17, which means that the null hypothesis that all the coefficients are zero is rejected. The R2 of the regression is 0.8877, which means that 88.77 percent of mental health is explained by the model. The root MSE is 0.2247, which represents the standard deviation of the residual in this model. This is an absolute measure of fit, with closer values to zero meaning a better fit. Thus, this model predicts the response well.

A negative, significant effect of unemployment on physical health is observed. This does not confirm the hypothesis that an increase in unemployment, leads to an increase in physical health in the short-term. Thus, the null hypothesis cannot be rejected. Besides, a negative, significant effect for education low is observed, which means that physical health is worse for people with a low level education in the short-term. Furthermore, we observe an F-value of 31.14, which means that the null hypothesis that all the coefficients are zero is rejected. The R2 of the regression is 0.9085, which means that 90.85 percent of physical health is explained by the model. The root MSE is 0.1582, which represents the standard deviation of the residual

Effect of the state of the economy on health : a study of the effect of the economic crisis, the Great Recession, using the level of unemployment as the main indicating factor, on people’s health in the Netherlands usin