The effects of Meat Consumption on Health and the role of Socio-economic status

Master’s Program in

Public Administration

Economics and Governance

Leiden University

Master’s Thesis

The effects of Meat Consumption on Health and the role of

Socio-economic status

Student: Kata Majeczki

Supervisor: Eduard Suari Andreu

The effects of Meat Consumption on Health and the role of Socio-economic status Kata Majeczki

Leiden University

Author Note

Kata Majeczki is an MSc. student in the Public Administration specialization

Economics & Governance program at the Faculty of Governance and Global Affairs; Leiden University. Student number 2378388

The supervisor concerning this paper is Eduard Suari Andreu, Post-doctoral researcher at the Department of Economics, Leiden University.

Correspondence for this article should be addressed to Kata Majeczki. E-mail: katamajeczki@gmail.com

In this paper we make use of data of the LISS (Longitudinal Internet Studies for the Social sciences) panel. The LISS panel data were collected by CentERdata (Tilburg

University, The Netherlands) through its MESS project funded by the Netherlands Organization for Scientific Research.

Acknowledgments

I would like to thank my supervisor, Eduard Suari Andreu, for his enthusiastic support in researching a topic which is of my high interest. I am grateful that he provided me with prompt advice throughout the whole research process. My gratitude to Servio Kloeth, for supporting me in achieving my dream to have a master’s degree; for being a helpful, critical, challenging, yet pragmatist sparring partner. My gratitude for CentERdata, for providing valuable survey data in the form of the LISS dataset, which made it possible that this research could be conducted.

Abstract

This study finds adverse effects of meat consumption on health. The association between socio-economic status (SES) and health suggest that income, education and occupation lead to different dietary patterns and in turn to health inequalities. When SES variables are controlled for, the effect of meat consumption decreases yet remains to exert a statistically significant adverse effect on health, allowing for a higher validity of this relation. By using a sample of 5076 individuals from the 2018 LISS survey in The Netherlands, this study first examines the relation between SES variables of income, education, occupation, and meat consumption. Income is found to be positively associated with meat consumption frequency, while education and occupation negatively correlates with meat consumption. It appears that older age cohorts; men; larger households and married respondents consume meat at a higher frequency. Investigation into the association between meat consumption and health shows that meat consumption increases the likelihood of high blood cholesterol, high blood pressure and diabetes, and decreases the likelihood that one feels healthy.

Table of Contents

1. Introduction ... 6

2. Institutional context ... 8

2.1. Consumer trends ... 9

2.2. Environmental concerns ... 10

2.3. Policy on curbing meat consumption ... 10

3. Literature review ... 13

3.1. Socio-Economic Status (SES) ... 13

3.2. Income and Meat consumption ... 15

3.3. Education and Meat Consumption ... 17

3.4. Occupation and Meat Consumption ... 19

3.5. Meat consumption and Health ... 20

Mortality ... 20 Cardiovascular diseases ... 21 Diabetes ... 24 Self-reported health ... 24 3.6. Control variables ... 25 Gender ... 25 Age. ... 25 Household size. ... 26 Marital status... 26

4. Theoretical framework & Hypotheses ... 26

4.1. Income... 27 4.2. Education ... 28 4.3. Occupation ... 28 4.4. Health ... 28 5. Research method ... 29 5.1. Data collection ... 29 5.2. Data preparation ... 30 Missing data. ... 30 Outlier test... 31

Conversion into new variables ... 31

Assumption tests ... 32

5.3. Descriptive statistics of the data... 33

SES and Meat Consumption ... 33

Demographic control variables ... 34

5.4. Measurement instrument development ... 36

5.5. Model ... 38

Part 1. ... 38

Part 2. ... 39

6. Data analysis & Results ... 40

6.1. SES and Meat Consumption ... 40

Step 1 results ... 41

Step 2 results ... 44

6.2. Meat Consumption and Health ... 47

High blood cholesterol ... 47

High blood pressure ... 50

Diabetes ... 51 Self-reported health ... 53 7. Discussion... 56 7.1. Policy implications ... 56 7.2. Limitations ... 58 7.3. Future research ... 58 8. Conclusion ... 59 9. Bibliography ... 61 10. Appendix A ... 73 10.1. Missing data ... 73 10.2. Outlier test ... 74 10.3. Assumption tests ... 74

OLS linear regression with multiple explanatory variables ... 74

Linear probability model. ... 77

10.4. Model fit... 77

SES and Meat Consumption. ... 77

11. Appendix B ... 79

12. Appendix C ... 82

The effects of Meat Consumption on Health and the role of Socio-economic status Master Thesis Public Administration – Economics and Governance

1. Introduction

The average amount of per capita yearly meat consumption has increased substantially in the past fifty years, from around 23kg in 1961 to 43kg in 2014 (Devlin, 2018). The large increase in global meat consumption calls attention for two major concerns: it plays a role in the rising carbon emissions; and has potential adverse health effects (Giovannucci et al., 1994; Drewnowski & Sepcter, 2004; Hu et al., 2000; Rose, Boyar, & Wynder, 1986; James, Nelson, Ralph, & Leather, 1997). Increased meat consumption leads to intensive livestock production, which is one of the causes of greenhouse gas emissions (Garnett, 2009). The production of one kg beef in the U.S. leads to 14.8 kg carbon-dioxide emission (Fiala, 2008). Tukker et al. (2006) find that meat production accounts for up to 12% impact on global warming. In order to cut emissions and to fight climate change, action is needed from the governments to step in and create behaviour-changing “nudge” strategies on cutting meat consumption (McLoughlin et al., 2019). Currently, no policy aims at curbing meat consumption in The Netherlands.

Regarding health effects, a large amount of medical studies provide evidence that meat consumption is correlated with increased risks for cardiovascular diseases (Fraser, 1999; Appleby, Davey, & Key, 1999; 2002; Tonstad et al., 2013; Spencer, Appleby, Davey, & Key, 2003; Campbell & Campbell, 2006), increases mortality levels (Sinha, Cross, Graubard, Leitzmann, & Schatzkin, 2009; Pan et al., 2012; Fraser, 2003; Orlich et al. 2013), and induces diabetes (Tonstad et al., 2013). Increased prevalence of diseases for meat consumers lead to substantially higher health care costs compared to vegetarians, as pointed out by a research in the U.S. (Barnard, Nicholson, & Howard, 1995).

The motivation of the paper is the environmental- and health related benefits that can be achieved by lower meat consumption. Lowering the risk of diseases related to meat consumption leads to lower medical costs, and reduces CO2 emissions, therefore beneficial from society’s perspective. Insight into individual level characteristics can shape policies designed to nudge meat consumption behaviour of individuals in The Netherlands. The results of this research can add value to policy makers to create both environmental and health policies that curb meat consumption.

This paper investigates how meat consumption affects health. Meat consumption increases the risk of various diseases, such as high blood cholesterol, high blood pressure and diabetes. In an attempt to get closer to a causal effect of meat consumption and health, this research first estimates the effect of socio-economic status (SES) on meat consumption. To date the impact of SES variables on meat consumption has not yet been researched in the context of The Netherlands. One related academic journal analyses the relation between socio-economic status and nutritional intakes in The Netherlands using 1988-1998 data (Hulshof, 2013).

Research shows that individuals with a low SES are more likely to suffer from diseases given that their diet consists of energy-dense, less healthy food products (Geurts, van Bakel, van Rossum, de Boer, & Ocke, 2017). Hulshof et al. (2013) argues that in The Netherlands, the main disparities in health inequalities are caused by SES, a conceptual construct composed of the indicators of income, education and occupation (Gossard & York, 2003). A large number of studies suggest that SES indicators influence meat consumption (Leahy, Lyons, & Tol, 2010; Leahy, Lyons & Tol, 2011; Gossard & York, 2003; Kearney, 2010; Wang, Beydoun, Caballero, Gary, & Lawrence, 2010, Cutler & Lleras-Muney, 2006). In the first part of this paper, this research focuses on investigating the independent effect of SES components, being income, education and occupation on the frequency of meat consumption of individuals living in The

Netherlands by using linear ordinary-least-squares regression with multiple explanatory variables.

This leads to the first research question of this paper, RQ1: “How does socio-economic status determine the extent of meat consumption of individuals in The Netherlands?”.

As the relationship between health and education (Cutler & Lleras-Muney, 2006), income (Kawachi & Kennedy, 1999; Kawachi, Kennedy, Glass, & Prothrow-Stith, 1998; Ettner, 1996) and occupation (Law, Leclair, & Steinwender, 1998) is well-established, the second part of this study will explore how meat consumption affects health, while controlling for the confounding effect of SES variables. By doing so, the research attempts to solve the potential omitted variable bias and measure causal effect of meat consumption on health outcomes. The second part of this study uses a linear probability model. This leads to the second research question of this paper, RQ2: “How does Meat Consumption predict Health outcomes?”

Findings of this research revealed that the consumption of meat increases the probability for high blood cholesterol, high blood pressure and diabetes; and is associated with feeling less healthy. While higher levels of education and occupations are predictors of less frequent meat consumption, higher income appears to be associated with more frequent meat consumption. Higher education is consistently associated with decreased probability of high blood cholesterol, high blood pressure and diabetes. Findings on the adverse health effects in this paper underpin the need for public policies in order to reduce potential health risks induced by meat consumption.

2. Institutional context

For the past decades there has been a rapid rise in meat consumption around the world (Geurts et al., 2017), such that the amount of meat that is made available through production has increased from 30 kg in 1980 to about 41 kg per capita per year in 2005 (Food and

Agriculture Organization [FAO], 2010). Meat production involves an inefficient use of water, land and natural resources (van der Goot & Matser, 2017).

The extent of meat consumption in developed countries can be a health concern (Giovannucci et al., 1994; Drewnowski & Sepcter, 2004; Hu et al., 2000; Rose, Boyar & Wynder, 1986; James et al., 1997). In 2015, the World Health Organization (WHO) stated that processed meat is carcinogenic to humans, and classified it to the same category as tobacco and asbestos (World Health Organization, 2015). Scientific awareness on the adverse effect of meat has led to a current revision of the Harvard Medical School Guide dietary recommendations on meat consumption (Willett, Skerrett, & Giovannucci, 2017), pointing out that one should limit consumption of red meat, and abstain completely from processed meat. Inequalities in health are expanding in The Netherlands, as diet patterns of people in low socioeconomic groups lead to overweight and chronic diseases (Ocké et al., 2017).

This chapter on “Institutional context” focuses on The Netherlands, and examines consumer trends, public and private initiatives, existing, and proposed policies regarding meat consumption and meat production.

2.1. Consumer trends

In the Dutch context, Dagevos, Voordouw, van der Weele, and de Bakker (2012) describe meat consumption frequency, reporting that about 35% of the Dutch population consume meat five or six times a week as part of their main meal. About 27% consumes meat three or four times a week, 18% every day, 15% once or twice per week, and 5% are vegetarian. On average, Dutch residents consume 100 grams of meat per day (Geurts et al., 2017).

De Bakker and Dagevos (2010) identify an emerging new group of consumers regarding meat consumption in The Netherlands, called “flexitarians”. Flexitarians amount for one third of meat consumers and alternate meat and vegetarian meals, motivated by concerns on the environment, animal welfare and health. Despite the large number of consumers who

regularly abstain from meat, de Bakker and Dagevos (2010) conclude that the volume of meat consumption remains constant.

2.2. Environmental concerns

Next to the adverse effects on health, increased meat consumption leads to intensive livestock production, which is one of the causes of greenhouse gas emissions (Garnett, 2009). Tukker et al. (2006) find that meat production accounts for up to 12% impact on global warming. The FAO (2010) reports that the livestock sector alone accounts for 18% of the greenhouse gas emissions and for 80% of land use. The 2015 Programma Aanpak Stikstof (PAS) is a policy framework developed by the Dutch government, which tackles the problem of the increasing nitrogen gas emissions by livestock (“Programma Aanpak”, n.d.). The government strives to reduce greenhouse gas emissions that harm the environment by imposing tight standards on the maximum emissions of ammonia from stalls. Also, agricultural activities that involve the emissions of nitrogen, ammonia, or nitric oxide require parties to apply for an agricultural permit.

The meat industry is attracting widespread attention of researchers, consumers and policy makers due to increasing interest in sustainable production and climate change.

2.3. Policy on curbing meat consumption

Numerous policy advisory reports (Ocké et al., 2017) recognize that meat consumption is harmful for the environment and health, and call for governmental intervention and an active role of the government in curbing food consumption. However, there is an absence of a policy that aims at curbing meat consumption thus far (Ocké et al., 2017). An example of an advisory report in The Netherlands was conducted by Stehfest et al. (2008) in the assignment of the Dutch National Institute for Public Health and the Environment (RIVM) for global climate policy development. The following paragraphs will illustrate the highlights of this report,

which provides a picture about the developments in the policy area regarding meat consumption and production.

The research of Stehfest et al. (2008) examines the impact of dietary changes (reduced, and no meat consumption) on future climate change mitigation policies. Stehfest et al. (2008) refer to and base meat consumption levels on the adjusted recommendations of the Harvard Medical School (Willett et al., 2001). Environmental scenario analyses in this research measure the effects of varying levels of reduced meat consumption and zero meat consumption on the climate. The authors find a significant impact of dietary changes in meat consumption levels on the climate and land use, such that a global food transition to low-meat diets, or a complete switch to no meat consumption would reduce climate mitigation costs by half for achieving the target for CO2 emissions (450 parts per million). Additionally, methane and nitrous oxide emission would be reduced substantially. The authors note that a main impediment during the implementation of a policy targeted on meat consumption would be the income loss meat producers would have to face (Stehfest et al., 2008). It has not yet been established which policy tool is best suitable for the strategical development towards a low meat, egg and milk consumption diet. A few advocate an increased tax on meat products, while others favour the implementation of price mechanisms which they believe would lead to substantial changes in consumption patterns (Smil, 2002). The advisory report highlights that health considerations, as advocated by medical scientists, can work as a supplementary argument in reducing meat consumption. Thus, the message of Stehfest et al. (2008) to the Dutch Ministry of Public Health and the Environment is that the health factor would be the most realistic central argument of a policy on curbing meat consumption.

A very recent policy advisory document prepared for the RIVM by Ocké et al. (2017) investigates the health factors and sustainability of food in The Netherlands, and describes the dilemmas and the potential for a prospective integrated food policy. The report acknowledges

that the diet of an average Dutch person has a two-sided problem. One is the adverse effects of the diet on health of individuals, mostly in low socioeconomic groups; and the significant burden of food production on the environment (Ocké et al., 2017).

A diet with more plant-based and less animal-based products is considered an opportunity for reducing chronic illnesses, health inequalities, and as a sustainable way of food production in the country (Ocké et al., 2017). The report highlights the importance of the active role the government should take in promoting sustainable and healthy food. An integrated policy would manifest in the cooperation of the government, the agricultural sector, producers, businesses (supermarkets, retailers, wholesalers), citizens, and social organizations (Ocké et al., 2017). Discrepancies exist between the long-term goals of establishing a sustainable, healthy diet and the concrete choices at the point of consumer purchase determined by price and convenience (Ocké et al., 2017). Proposed regulations of the report are clear labelling on food products to increase awareness of the consumer on which product is healthy and sustainable, and an increased tax rate on meat.

In addition, Dutch entrepreneurship is characterized by innovation capacity. Companies are likely to develop smart, innovative solutions that address the growing concern of citizens on “responsible food” and provide them with eligible profit. Private firms, such as the ‘Vegetarische Slager’, recognize that consumers are most likely to switch meat to plant-based products when their texture resembles the original. The development of direct alternatives to meat require evolving technologies. A public-private partnership (PPP) called “Plant Meat Matters” is an initiative of research universities and private firms that jointly develop plant-based meat replacements (van der Goot & Matser, 2017). The five-year long PPP is financed by the Dutch Ministry of Economic Affairs and Climate and started in 2017. The PPP highlights the joint interest of scientists, policy-makers and private firms towards the curbing of meat consumption, an important step towards sustainability.

Ocké et al. (2017) envision a facilitating role of the government, whereby it assists the progress of innovative developments and citizens’ initiatives. Market-based incentives, such as subsidizing firms to develop plant-based products manifests itself in the “Plant Meat Matters” PPP, and in a recently developed programme called the “New Food Challenge”. Under this programme, The Netherlands Enterprise Agency, a governmental body, funds the development of new food products based on plant-based protein, that are alternatives for meat. This initiative that started in 2017 shows that the government takes steps towards the reduction on meat consumption (“Rijksoverheid stimuleert”, n.d.). Besides the initiatives of the government, the past few years have shown an increase of self-regulation of the industry itself, such as the emergence of certificates like “Beter Leven”, promoting sustainability and animal welfare (Geurts et al., 2017).

3. Literature review

This section shall outline the main findings from the literature on the main concepts as well as the relation between these concepts and excerpts of the relevant control variables. To clarify, this study’s interest is in the relation between socio-economic status variables income, education, and occupation and meat consumption, which is controlled for general demographics age, gender, household size and marital status. The second main relation of interest for this study is between meat consumption and health, which is controlled for by the same demographic factors in addition to the three socio-economic status variables. Socio-economic status variables are added as control variables due to the expected mediating effects, thus allowing a more reliable view on the relation between meat consumption and health. 3.1. Socio-Economic Status (SES)

SES is defined in various manners throughout the literature. Wang et al. (2010) measures SES by education and income solely, and Hulshof et al. (2003) measures SES by education, occupation and occupational level. Consistent with the studies of Leahy et al. (2010;

2011), Maguire & Monsivais (2015), Clonan et al. (2016), and Gossard and York (2003), this study measures the conceptual construct SES by the indicators of income, education and occupation.

Vegetarians overall are found to display higher SES (Pollard, Greenwood, & Kirk, 2001), while differences in SES are identified as considerable contributors of disparities in diet and a potential contributor to health inequalities (Maguire & Monsivais, 2015; Hulshof, Brussaard, Kruizinga, Telman, & Löwik, 2003; Clonan Roberts, & Holdsworth,2016). Hence SES is relevant both in relation to meat consumption and health.

In a broader context, literature on the drivers of SES on nutritional intake of individuals focuses on measuring the consumption of fruits, sugar, fat, meat, cereals etc. of individuals (Maguire & Monsivais, 2015; Hulshof et al., 2003; Fraser, Welch, Luben, Bingham, & Day, 2000; Geurts et al., 2017). Of these consumer products this study focuses on meat consumption in relation to a healthy diet. Moreover, research conducted on the perception of people from varying SES groups on what constitutes a healthy diet (Margetts et al., 1997), and findings into meat consumption choices driven by animal welfare (Hoogland, de Boer, & Boersma, 2005), are taken into consideration for this study.

The importance of SES variables is evident when investigating the relation between meat consumption and health. SES can potentially affect health through various channels, such as the ability to purchase inputs that lead to health outcomes (e.g. nutrition, medical services); through greater stress caused by income uncertainty of the poor (Jensen, 2004, p. 315) or through potential other factors such as attitudes towards risk and time preference which relate to both low SES and poor health. Leather (1996) argues that chronic diseases such as obesity, diabetes, heart disease, stroke, and cancer particularly affect poor people in Britain, leading to inequalities in health induced by income. Furthermore, Cutler and Lleras-Muney (2006) state that health is largely associated with education. As put forward by Cutler and Lleras-Muney

(2006), education is important as it leads to different thinking and decision-making patterns. Therefore, this paper will include SES variables of income, education and occupation as explanatory variables for meat consumption in the first part of this study, and as control variables when studying the relation between meat consumption and health in the second part of this study.

3.2. Income and Meat consumption

Low income may limit the access to healthy food (Köhler, 1997; James et al., 1997), as high energy food products have a relatively lower price compared to healthier food items, such as vegetables and whole grains (Jetter & Cassady, 2006).As a consequence of that, low income households tend to buy high energy food products that contain high fat and sugar. Also, low income households tend to purchase food products that are cheaper in units compared to what are considered healthy products, such as vegetables (James et al., 1997). James et al. (1997) investigated social class differences in health in Britain by employing an annual British food survey containing 7000 households. The study focuses on the nutritional pattern of different income groups. Results reflect that low-income households consume more meat products (especially higher fat containing meat products) compared to high income households.

Leahy et al. (2010) investigated how SES and other demographic factors are associated with vegetarianism and meat consumption frequency among 22,623 adults and 7,521 children by using the 2008 Health Survey in England. Leahy et al. (2010) use a two-step logit regression model, where the first step measures the binary state of vegetarianism and the second step the varying levels of meat consumption. Results indicate that vegetarianism increases with income, and most vegetarians earn above the national average income level. However, at extremely high-income levels, vegetarianism decreases. The authors find that the binary function of being a vegetarian displays higher variation across income levels than meat consumption frequency. The most salient result to emerge from the data is that meat consumption frequency is quite

stable across various income levels. Hence, the analysis did not confirm any significant deviations in the frequency of meat intakes between various income levels.

Leahy et al. (2011) use the 2007 Survey of Lifestyles, Attitudes and Nutrition (SLÁN) to explain meat and fish consumption among 9,223 adults in Ireland. Findings of the two-step logit regression model suggest that vegetarianism is the highest at the lowest and the highest levels of income. Leahy et al. (2011) find that at the lowest levels of income meat consumption frequency is low, which increases with income. However, at the highest levels of income, meat consumption levels are even lower compared to the portions consumed by the poorest adults. Interestingly, the observation that meat consumption decreases at the highest levels of income opposes to the findings of the authors in the UK, therefore constituting a gap in literature. The observation in Ireland regarding meat intake at high income levels is in agreement with the study of Vranken, Avermaete, Petalios and Mathijs (2014), which investigates the relation between income and meat consumption in 120 countries. Vranken et al. (2014) used panel data over the time period of 1970–2007 and pointed out that meat consumption increases with income, but then, over time, higher levels of income are associated with decreasing levels of meat consumption.

Variables used in the paper of Clonan et al. (2016) bear a close resemblance to this research, as Clonan et al. (2016) investigated how consumption of red meat varies with age, gender, and SES variables in the UK by using data from the National Diet and Nutrition Survey from 2008 to 2011. Next to analysing patterns in the UK, Clonan et al. (2016) examined how meat consumption differs across countries. The findings of their study reflect that in a high-income context people in low SES groups are associated with higher meat intakes. The authors argue that people’s decisions about meat consumption reflect characteristics of Bourdieu’s theory of distinction (Bourdieu, 1984). Accordingly, meat intakes are reflective of social status in society, so when meat is initially an expensive food item then people in higher SES groups

purchase meat in order to distinguish themselves from the ‘masses’. Over time, as meat become more accessible to the wider population it then loses its appeal as it is no longer associated with the ‘taste of luxury’, which corresponds to declining meat consumption (Clonan et al., 2016). Maguire and Monsivais (2015) conducted an analysis of socio-economic dietary inequalities in the UK by using data from the National Diet and Nutrition Survey between 2008–2011 and demonstrate that respondents in the lowest-earning households consumed 15.7 grams/day more red and processed meat than the highest-earning households (Maguire & Monsivais, 2015, pp. 185).

The study of Gossard and York (2003) use ordinary-least-squares (OLS) regression to assess the effects of SES on the total quantity of meat consumption. Based on data on 15,028 United States of America (U.S.) residents from the 1996 Continuing Survey of Food Intakes by Individuals, the authors conclude that income does not influence total meat consumption. On the other side, beef consumption, as part of total meat consumption does rise with higher levels of income. Gossard and York (2003) note that the effect of income on meat consumption is highly context dependent.

In sum, there are mixed results in literature on the association between income and meat consumption, with various studies suggesting that the lowest income households consume more meat (James et al.,1997; Leahy et al., 2011; Clonan et al., 2016; Maguire & Monsivais, 2015), while Leahy et al. (2010) find stable meat consumption levels across all income groups, and Gossard & York (2013) find no effect of income.

3.3. Education and Meat Consumption

The study of Leahy et al. (2011) conducted in Ireland indicates that individuals with higher education levels than upper secondary level (reference category) eat significantly less meat. This is consistent with the findings of Leahy et al. (2010). In addition, Leahy et al. (2010)

put forward that the chance that an individual with higher education is a vegetarian is 46% higher compared to those holding a secondary school certificate.

In their research studying the effects of SES on meat consumption, Gossard and York (2003) find that education is inversely related to meat consumption, which means that people with more education consume less meat (Gossard & York, 2003). The results are significant with a 99% confidence level, and reflect that the coefficient of education decreases meat consumption by 3.7 grams/day (Gossard & York, 2003). This corresponds to the results demonstrated by Maguire & Monsivais (2015), who found that people with no qualifications consume 21.9 grams/day more red and processed meat than individuals holding a degree.

Fraser et al. (2000) examine the relationships between dietary patterns of 1,968 middle-aged (44-75 years) individuals and demographic variables indicated by the variables of age, gender, education, and marital status. Based on data from the European Prospective Investigation of Cancer (EPIC), findings of the study suggest that people with low education consumed more meat compared to lower-educated individuals (measured in standardized mean servings of meat per day, 0.36 for low-educated, 0.32 for higher-educated).

In a cross-sectional study, Margetts et al. (1997) investigated the socio-economic factors that influence perceptions of what constitutes a healthy diet across 15 EU countries. Through a questionnaire completed by 14,331 individuals, results of the study reveal that educational level has the strongest influence on perceptions of a healthy diet. Margetts et al. (1997) explain that higher educated people might be more aware of the benefits of a healthy diet and possess more knowledge about the types of food which are healthier. Education therefore partly explains the different food consumption patterns between SES classes (Hulshof et al., 2003).

Overall, research into the correlation between educational level and meat consumption support the observation that education is negatively correlated with meat consumption.

3.4. Occupation and Meat Consumption

A number of studies investigating the correlation between SES and meat consumption have consistently found that skilled workers in higher occupations consume less meat compared to manual workers (Clonan et al., 2016; Hulshof et al., 2003; Leahy et al., 2010; Leahy et al., 2011).

Clonan et al. (2016) report that people in higher occupations, referring to managerial and professional positions consume about 37.24 grams of meat (measured as per 1000 calories intake). The above-mentioned quantity is significantly less red meat compared to individuals in lower technical (47.35 grams) and routine occupations (47.65 grams). Hence, Clonan et al. (2016) conclude that people in routine occupations consume the most meat. The same conclusion is reached by Maguire & Monsivais (2015), whereby individuals in higher managerial and professional occupations consume on average 25.5 grams less red and processed meat than those in routine occupations.

Employing a cross-sectional study with three years of data from the Dutch National Food Consumption Survey, Hulshof et al. (2003) studied the differences in dietary intake between adults in different SES groups. The authors measured SES based on the indicators of education, occupation and occupational level. A sample of 12,965 individuals were categorized into low, medium and high SES groups; low referring to unskilled work force, and high SES groups referring to university graduated professions and executive managers. Consistent with the findings of Clonan et al. (2016) and those of Maguire & Monsivais (2015), individuals in the low SES group reported the highest meat intake.

In their investigation into the effects of SES on the total quantity of meat consumption, Gossard and York (2003) categorize occupation into four categories, being professional, service worker, laborer, and not working. Their research demonstrates that people in laborer occupations consume both more beef and total meat than those in either service or professional

occupations. With a confidence level of 99%, the study highlights that professionals consume 36.448 grams less meat than those in laborer occupations (reference category).

The evidence of the study of Leahy et al. (2011) indicates that those employed in a professional or managerial position consume meat less frequently than people occupied in the skilled manual category (reference group). The research affirms that individuals in semi-skilled and unskilled manual occupations consume meat more often than those in the skilled manual occupations.

3.5. Meat consumption and Health

Studies on Meat Consumption and mortality rates find that increased red and processed meat intakes lead to higher mortality levels (Sinha, Cross, Graubard, Leitzmann, & Schatzkin, 2009; Pan et al., 2012; Fraser, 2003). A vegetarian diet is correlated with less risk for cardiovascular diseases (Fraser, 1999; Appleby, Davey, Key, 1999; 2002; Tonstad et al., 2013; Spencer, Appleby, Davey, & Key, 2003), diabetes (Barnard, Lattimore, Holly, Cherny & Pritkin, 1982; Story et al., 1985; Tonstad et al., 2013), hypertension, and metabolic syndrome (Pettersen, Anousheh, Fan, Jaceldo-Siegl, & Fraser, 2012; Rizzo, Sabaté, Jaceldo-Siegl, & Fraser, 2011; Tonstad, Butler, Yan, & Fraser, 2009).

The studies reviewed in this section are either based on randomized controlled dietary experiments or focus on social and religious groups that are vegetarians by acculturation. Hence, results of these studies are the closest to measuring causal relations between diet and diseases.

Mortality. Orlich et al. (2013) evaluate the association between vegetarian dietary patterns and mortality by studying the religious population of Seventh-day Adventist men and women who consume various types of vegetarian diets. The study of Orlich et al. (2013) is considered a “natural experiment” conducted between 2002 and 2007. The experimental design of this paper allows for measuring causality between meat consumption and health. The authors

conclude that a vegetarian diet significantly reduces mortality risk induced by cancer, stroke, diabetes, and cardiovascular, infectious and neurologic diseases etc., which is consistent with the findings of previous studies on the topic. The authors find reduced mortality in the study population of 73,308 U.S. residents. By collecting mortality data from the National Death Index, and reported dietary data on intakes of foods of animal origin, Orlich et al. (2013) performed a mortality analysis of the respondents using Cox proportional hazards regression. Generalizability of the findings regarding the effects of a vegetarian diet on health is assured by the large number of participants consuming various vegetarian diets, categorized into vegans, lacto-ovo-vegetarians, pesco-vegetarians and semi-vegetarians. The most remarkable result to emerge from the statistical analysis is that all types of vegetarians combined had 0.88 times the risk of all-cause mortality of nonvegetarians (a hazard ratio of 1 means that vegetarians and non-vegetarians face equal mortality risks, a hazard ratio of less than 1 means that vegetarians have smaller mortality risk). Surprisingly, the hazard ratio of men was lower compared to women (0.82 for men and 0.93 for women). There is a significant difference between the effects on men and women, whereby reduced mortality risks are stronger and more significant for men than for women. Orlich et al. (2013) contribute this fact to possible gender-specific mechanisms, which adds to the importance of including gender as a control variable in this research. In their analysis, Orlich et al. (2013) differentiate between the mortality rates of different cohorts of vegetarians, whereby compared to non-vegetarians, vegans have the lowest mortality rates (0.74), followed by pesco-vegetarians (0.81) and lacto-vegetarians (0.91).

Cardiovascular diseases. Before the start of the long-term research project called the Framingham Study in 1948, most scientists have considered heart diseases as a natural course that comes with the “wearing down” of the body (Campbell & Campbell, 2006). The Framingham Study was designed to explore the risk factors that correlate with cardiovascular

diseases, by using medical records of about 5000 residents of Framingham in the U.S. (Campbell & Campbell, 2006). As part of the of this study, Kannel Dawber, Kagan, Revotskie and Stokes (1961) followed up the medical records of 5127 men and women from different age groups for six years, and establish the concept of “risk factors” for heart diseases being cholesterol, and high blood pressure. The strong correlation of blood cholesterol and heart disease is a breakthrough finding in medicine, as preventive programs proved necessary to lower blood pressure and blood cholesterol in order to prevent heart diseases.

Developments in science identified fat intakes and cholesterol as major attributing factors of heart diseases. As fat and cholesterol are considered indicators of animal food intake in the last decades scientists have shown widespread interest in exploring the effects of a plant-based diet on heart diseases (Campbell & Campbell, 2006). In an attempt to study the relationship between fat intake and the development of atherosclerosis (a disease that refers to the build-up of plaque inside the arteries), Morrison (1960) developed an experiment for 100 patients who have suffered from a heart attack. In the treatment group, individuals consumed reduced amounts animal protein, while in the control group people could maintain their normal diet. Eight years after the experiment, only 24% of the 50 people were still alive in the control group, while in the treatment group this ratio was considerably higher, 56%.

The study of Jolliffe & Archer (1959) focused on the relationship between animal protein consumption and heart disease occurrences in men aged fifty-five to fifty-nine across twenty countries. Given that individuals in the treatment group who consumed less animal protein had a lower rate for heart diseases, the study concludes that the more animal protein intake the higher the rate for heart diseases.

The growing importance of plant-based diets in preventing heart and coronary diseases have been demonstrated by the experiments of Ornish et al. (1990) and Esselstyn, Ellis, Crowe

and Medendorp (1995). Once again, the experimental design of these medical studies allows to get closest possible to causality.

Ornish et al. (1990) tested the effect of a low-fat plant-based diet on the cholesterol levels and heart diseases for 28 patients for one year. A remarkable finding of the study of Ornish et al. (1990) is that 82% of the heart patients had a retreat of the heart disease over the course of merely a year.

The research of Esselstyn et al. (1995) tested the effect of plant-based diets on people with established heart and coronary diseases by dividing patients into treatment and control groups. The most remarkable finding of this research is that abstaining from meat products and animal protein can lower cholesterol levels in the blood, and has the potential to reverse the event of a coronary operation (in 70% of the treatment subjects). The studies of Li, Siriamornpun, Wahlqvist, Mann, and Sinclair (2005), and Ding (2006) confirm that meat consumption increase the probability of heart diseases.

Another attempt to infer causality as close as possible is the study of Beilin et al. (2011), who compared the blood pressure of Seventh-day Adventist to Mormons. These two accultured religious groups are similar in many ways, except that while Seventh-day Adventists are vegetarians, Mormons are omnivores. Beilin et al. (2011) conclude that the comparison provide evidence for the blood pressure-lowering effect of a vegetarian diet. Meanwhile, Beilin et al. (2011) also highlight that cross-sectional population studies cannot prove cause-and-effect relationships as the possibility exists that other unidentified factors were contributing to the blood pressure differences. In addressing the issue, this study will control for the possible confounding effect of SES variables and demographic characteristics when measuring the effect of Meat Consumption on Health.

However, because of the consistent pattern between a plant-based diet and health outcomes there is a reduced controversy surrounding the power of diet on treating heart diseases (Campbell & Campbell, 2006).

Diabetes. Tests have been carried out on the effect of a plant-based diet on diabetes. In an experiment, Anderson (1986) provides a plant-based food diet and reduced meat intake (max. two cold cuts of meat a day) to patients in the treatment group. After four weeks of the treatment only one Type 2 diabetes patient had to continue the insulin medication, Type 1 patients could lower their insulin medication by an average of 40% (Anderson, 1986).

Similar results have been reached by Story, Anderson, Chen, Karounos, and Jefferson (1985), where the cholesterol of Type 1 diabetes patients on plant-based food was reduced by 32% in two weeks of time. Equally remarkable results have been demonstrated in an experiment (Barnard et al., 1982), where out of the forty non-insulin diabetic patients, thirty-four could stop taking medication after twenty-six days on a plant-based diet.

The long-term benefits of a plant-based diet have been demonstrated (Barnard, Massey, Cherny, O'Brien, & Pritikin, 1983; Story et al., 1985) and highlight that a continuation of a plant-based diet is associated with the benefits of a lower cholesterol level and discontinuation of medication usage.

Self-reported health. Literature that measures health identifies self-rated health as a suitable variable that measures overall health of individuals (Ferrie, Shipley, Marmot, Stansfeld & Smith, 1995; Caroli & Godard, 2016). In line with the study of Fraser et al. (2000), this research measures how meat consumption predicts self-rated health. The study of Fraser et al. (2000) divided the population into various cohorts based on similar characteristics and showed that cohorts consuming less meat feel generally better. Pollard, Greenwood, Kirk and Cade (2001) found no statistically significant differences in self-reported health among vegetarians. In a cross-sectional study, Burkert, Muckenhuber, Großschädl, Rásky, and Freidl (2014)

analyse differences in diet based on SES, age and gender and demonstrate that vegetarians report poorer health more frequently. Bukert et al. (2014) argue that individuals might follow a vegetarian lifestyle as a consequence of diseases they suffer from, given that a vegetarian diet is often recommended to lose weight (Farmer, Larson, Fulgoni, Rainville, & Liepa, 2011) and a way to improve health (Leitzmann, 2005).

3.6. Control variables

Gender. Several studies conclude that there exists gender-specific differences in meat consumption, whereby men consume more meat than women (Clonan et al., 2016; Maguire & Monsivais, 2014; Wang et al., 2010; Leahy et al., 2010; 2011; Fraser et al., 2000; Geurts et al., 2017). Gender-specific differences in meat consumption are explained in literature by connotations attached to meat, such as virility and masculinity (Beardsworth & Keil, 1997; Fiddes, 1991; Roos, Prättälä, & Koski, 2001; Rozin, Hormes, Faith, & Wansink, 2012; Ruby & Heine, 2011; Ruby, 2012; Sobal, 2005). Through the feminist-vegetarian theory, Adams (2000) explains the differences in meat consumption between men and women by a perception of men regarding meat, namely meat promoting masculinity.

Age. Most studies (Leahy et al., 2011; Clonan et al., 2016; Dibb & Fitzpatrick, 2014) find that the elderly consumes the least meat among all age cohorts, while people in their middle ages consume more meat than younger adults. A few studies find contradictory evidence for this, suggesting that the elderly consume more meat than any other age cohort (Fraser et al., 2000; Papp et al., 1997), which can be explained by the relative stability of dietary habits people tend to keep over time and the difficulty elderly people face when making the recommended changes towards integrating less fat and more vegetables in the diet (Fraser et al., 2000; Papp, Lakner, Komáromi, & Lehota, 1997). Interestingly, even within one study, Wang et al. (2010) find contradictory evidence for age when comparing different surveys.

Therefore, one can conclude that there is a gap in literature regarding meat consumption patterns of different age cohorts.

Household size. Studies on both aggregate and micro-level demonstrate that as the number of members of the household increase, the less likely that the individual in the household will be a vegetarian (Leahy et al., 2010; 2010-b; 2011), furthermore, meat consumption frequency increases with household size (Leahy et al., 2010; 2010b; 2011; Schmid et al., 2017). Leahy et al. (2011) explain this by the “economies of scale” in preparing meat and the tendency of individuals to consume meat when they have company. In sum, there is merely an agreement in literature about the positive relation between meat consumption and household size.

Marital status. Marital status appears to influence meat consumption, and Leahy et al. (2010; 2011) found that divorced and cohabiting yet unmarried respondents consumed less meat. Fraser et al. (2000) investigated the eating habits of individuals in the UK and found that married subjects consumed more red meat and poultry than single ones. Pollard et al. (2001) use control variable marital status for investigating the demographic factors that influence vegetarianism.

4. Theoretical framework & Hypotheses

SES expresses individuals’ “access to collectively desired resources”, such as material goods, money, and opportunities for education (Oakes & Andrade, 2017, pp. 23). The conceptual construct SES is based on three explanatory variables measuring Income, Education, and Occupation. In this paper, when estimating the association between Meat Consumption and Health, the researcher will control for the confounding effects of SES variables in an attempt to get closer to measuring a causal effect of Meat Consumption on Health in a sample of 5076 individuals in The Netherlands.

Figure 1. Graphical representation of the conceptual model.

This section provides a summary of the findings in literature, from which the hypotheses will be drawn. The alternative hypotheses to be tested are grounded in the research questions.

4.1. Income

The first hypothesis on the association between income and meat consumption is based on evidence put forward by James et al. (1997), Clonan et al. (2016) and Maguire and Monsivais (2015), indicating that individuals with low levels of income consume more, while high-income households consume less meat.

Hypothesis 1. (H1): Independent variable Income has a negative relation to dependent variable Meat Consumption.

Given that various studies found contradictory evidence to the negative relation between income and meat consumption (Leahy et al., 2010; Leahy et al., 2011; Vranken et al.,

2010), the nature of the relation can be said to be context dependent (Gossard & York, 2013). Income as an indicator in this research adds value to the gap in literature by exploring the nature of the association in the context of The Netherlands.

4.2. Education

The second hypothesis can be drawn from the research of Leahy et al. (2010; 2011), Gossard & York (2003), Maguire & Monsivais (2015), Fraser et al. (2000) and Hulshof et al. (2003), which demonstrate an inverse relationship between education and meat consumption. Hypothesis 2. (H2): Independent variable Education has a negative relation to dependent variable Meat Consumption.

4.3. Occupation

The third hypothesis is drawn from the observations in literature whereby people in lower occupational positions consume more meat, while people in higher occupations display lower meat consumption (Leahy et al., 2011; Clonan et al., 2016; Maguire & Monsivais, 2015; Gossard & York, 2003; Hulshof et al., 2003).

Hypothesis 3. (H3): Independent variable Occupation has a negative relation to dependent variable Meat Consumption.

4.4. Health

Hypotheses on the link between Meat Consumption and Health are drawn on literature. Randomized controlled experiments (Campbell & Campbell, 2006; Fraser, 1999; Appleby et al., 1999; Tonstad et al., 2013; Kannel et al., 1961; Morrison, 1960; Joliffe & Archer, 1959; Esselstyn et al., 1995) and natural experiments (Orlich et al., 2013; Beilin et al., 2011) demonstrate that a non-vegetarian diet induces high blood cholesterol, high blood pressure, and contributes to cardiovascular diseases. This leads to the fourth hypothesis, which states that higher Meat Consumption frequency is connected to higher chance of High blood cholesterol;

and the fifth hypothesis, which refers to the positive relationship between Meat Consumption and High blood pressure.

Hypothesis 4. (H4): Independent variable Meat Consumption has a positive relation to dependent variable of High blood cholesterol.

Hypothesis 5. (H5): Independent variable Meat Consumption has a positive relation to dependent variable of High blood pressure.

Blood cholesterol is also correlated with diabetes. The sixth hypothesis is based on the results of controlled experiments, which highlight that a plant-based diet leads to lower insulin dependence (Anderson, 1986), to reduced levels of cholesterol in Type 1 diabetes patients (Barnard et al., 1983; Pritkin, 1983; Story et al., 1985), such that patients can even stop taking medication (Barnard et al., 1982; Story et al., 1985).

Hypothesis 6. (H6): Independent variable Meat Consumption has a positive relation to dependent variable of Diabetes.

As an overall estimate of individuals’ health, the paper also measures self-rated health. People with higher meat consumption perceive that their health is worse compared to those who consume more meat (Fraser et al., 2000), which leads to the seventh hypothesis of this research paper.

Hypothesis 7. (H7): Independent variable Meat Consumption has a negative relation to dependent variable of Self-reported Health.

5. Research method 5.1. Data collection

The data of this study is collected through the use of secondary individual-level data from the Longitudinal Internet Studies for the Social sciences (LISS) panel, which consists of 4500 households in The Netherlands, in total 7000 individuals. The panel represents a true representative sample of the Dutch population, as participating households are randomly drawn

from the population register by Statistics Netherlands (CentERdata, n.d-a.). Panel members above the age 16 provide answer to online questionnaires about their income, health, and education etc. (CentERdata, n.d-a.). For this study the LISS datasets on Health, Work and Schooling, and the Background Variables will be used.

The LISS dataset called “Background Variables” have been collected since 2007, and each participating household is required to complete the questionnaire on Background Variables when joining the panel. Monthly update on changes in the information included in the dataset ensures up-to-date information of the respondents (CentERdata, n.d.-b).The latest, 2018 data collection of the Health dataset will be used for this study, where out of the sample of 6,466 household members 5,455 provided a complete answer to the questionnaire (CentERdata, n.d.-c). Similarly, the Work and Schooling dataset collected in 2018 provides the latest information on the 7,256 respondents who were included in the survey. With a 77.6% response rate, 5,633 surveys provide complete answers (CentERdata, n.d.-d).

Setting up the databases for the LISS dataset on Health, Work and Education and Background variables where data on respondents can be matched on each spectrum yields a sample size of 5,076 respondents between the age between 16 to 101.

5.2. Data preparation

Data preparation consists of three steps. In the first step the variables in their original form were assessed for outliers and missing data. In the second step of the preparation several variables were converted into dummy-coded variables. In the last step the suitability of the variables employed in the OLS regression were tested to the extent necessary to the underlying assumptions of the OLS linear regression. This section provides a summary overview, the full data preparation method is found in Appendix A.

Missing data. Variables Monthly net household income and Occupation featured missing data to the extent of respectively 8.8% and 6.25%. Data was missing completely at

random (MCAR). Mean replacement was applied, in accordance with suggestions by Bartlett (2012), by imputing the mean value for any missing data points in Monthly net household income, and the category closest to the mean was chosen to impute for missing data points in Occupation. All other variables have less than 1% missing data, and are deemed sufficiently complete for further analysis.

Outlier test. Extreme multivariate outliers were analysed by the Mahanalobis distance metric in accordance with the procedures described by Tabachnick & Fidell (2007) and Univariate (n.d.). Seven cases were deemed multivariate outliers, however, further analysis through Cook’s distance test suggests that these outliers are of no significance influence, and thus the outliers are retained for the integrity of the dataset.

Conversion into new variables. This section describes how dummy-coded variables of Income, Education, Occupation and Age are created from the original variables in the LISS dataset. Dummy-coded variables provide more insight into the in-between group differences in the regressions.

In creating variable Income, the original variable Net monthly household income in Euros is multiplied by twelve. Secondly, the net yearly household income is divided by the square root of the number of household members of the household to correct for the returns to scale of having an additional household member. Next, the observations of Income are divided into four defined intervals. By dividing the sample into the four percentiles (25%, 50%, 75%, 100%) of income groups, the first cohort will represent households with the lowest income, followed by households in the second quadrille, the third, and the fourth, highest income households. Variable Income in its final form thus represents the four quartiles of net yearly household income adjusted for the number of household members in the sample.

Variable Education in its original form includes the six categories of the Dutch educational system, in line with the Statistics Netherlands (CBS) classifications. Given the

nature of the Dutch secondary education, the three different secondary educational levels are all distinct, grouping these school categories together does not provide useful insight. The researcher believes that retaining all six categories as dummy-coded categories on their own reflect the Dutch population best. Education is therefore a dummy-coded variable where each category represents an item scale in the survey.

Based on the literature review, skill level is a good measure of Occupation. The nine scales provided in the original data item include the occupational categories used in literature. In line with the studies of Leahy et al. (2010; 2011), the researcher constructed a dummy-coded variable for Occupation, measured on a three-item scale with ascending skill levels. The first category consists of “agrarian workers”, “unskilled and trained manual work”, and “semi-skilled manual work”; followed by the second category including ““semi-skilled and supervisory manual work”, “other mental work” and “intermediate supervisory or commercial profession”. The third category for Occupation consists of occupations requiring the highest levels of skills, being “intermediate academic or independent profession”, “higher supervisory profession” and “higher academic or independent professions”. This categorization corresponds to Leahy et al. (2010; 2011), where occupational categories refer to skills, such as “semi-skilled and unskilled” and “non-manual and skilled manual”. Nominal categorizations used by other studies such as Clonan et al. (2016) and Maguire & Monsivais (2015), who range occupations from routine to lower supervisory and technical, to higher managerial and professional occupations are not available in the LISS dataset. The same applies to categorizations used by Gossard and York (2003), who differentiate between labourer, service, professional and never worked categories.

Assumption tests. Data to be processed by OLS regression is recommended to meet several criteria in order to yield reliable results, the respective requirements are set out by Keller (2005). Meat Consumption, as the only interval variable in the OLS regression, is normally

distributed based on Q-Q plots. Only minor deviations of homogeneity of variance are observed, sufficiently satisfying the requirement of homoscedasticity. No significant multicollinearity is observed by variance inflation factors, in line with recommendations. Furthermore, the Durbin-Watson test implies increased risk of a type I error, yet the extent is minor and acceptable. Do note that most variables are entered into the regression as dummy variables and as such do not have to meet the requirements for a normal distribution. Through these tests the data has been deemed appropriate for OLS regression.In significance testing this paper shall employ the adjusted R-squared in assessing the explanatory power in order to adjust for the number of explanatory variables.

Given the common violations of the underlying assumptions of a linear probability model (LPM), it is recommended to use alternative models (i.e. probit or logit) in order to impose additional controls on LPM outcomes, and thus establish validity of the results. In accordance with the recommendations (Stewart, 2016), and similarly to Caroli & Godard (2016) who use probit estimates for health outcomes, this paper uses a probit model to control for the consistency of the outcomes provided by the LPM. The data was tested for the assumptions, of which a further description is provided in Appendix A.

5.3. Descriptive statistics of the data

SES and Meat Consumption. Table 1 below shows descriptive statistics for Meat Consumption, Income, Education, and Occupation.

Table 1

The cohort of vegetarians (i.e. those who never eat meat) is the smallest in the sample, consisting of 204 people. The largest group of individuals consume meat 5 to 6 times per week (1483 people) and 2 to 4 times a week (1466 people). The median for income is €25,200. Most of the respondents have a higher vocational (HBO) educational level and work in non-manual skilled positions.

Demographic control variables. Descriptive statistics of the demographic control variables are presented in Table 2. The median of Age falls on category 55-64 years. There is an almost equal distribution between men and women, yet there are somewhat more women in the sample. Regarding the differences in Meat Consumption of men and women in the LISS

dataset, the number of vegetarians among women is almost threefold of men (54 males, 150 female). Household size composition in the sample suggests that most participants on average live in two-persons households. More than half of the respondents are married.

Table 2

Health. Table 3 shows the descriptive statistics of the Health variables. Out of the 5064 respondents, 15% takes medicine for High blood cholesterol; 21.05% for High blood pressure, and 5% of the sample has Diabetes. Respondents perceive their health as “good” on average. Self-reported health’s mean is 3.11 and its standard deviation is 0.786.

Table 3

5.4. Measurement instrument development

Meat consumption is the dependent variable of the first part of this study and the explanatory variable in the second part of the study. Meat consumption is measured as the frequency a person consumes meat. Meat consumption is an interval variable operationalized as the frequency of meat consumption. Meat consumption frequency will be measured by a single survey item containing six scales, being “Do you eat meat or meat products?” from the Health LISS database.

Income is operationalized as the yearly net total household income adjusted to the

number of household members, and will be measured through a computed variable based on an

existing continuous data item in the LISS Background variables dataset where respondents indicate their “Net household income in Euros” (CentERdata, 2012).

Education can be defined as the highest level of schooling a person attained, conceptualized as the level of knowledge one has in making health related decisions. Education

is operationalized by an ordinal variable referring to the highest level of education, in categories corresponding to the Statistics Netherlands (CBS) from the LISS dataset (CentERdata, 2012). Education levels are measured by the survey item that ranges from 1 to 6, corresponding with “primary school” to “university”.

Occupation refers to one’s work activity; it is a categorical variable that indicates characteristics of manual or skilled work. Occupation is operationalized in this study as the

extent to which a work requires manual effort. This research will measure Occupation through

an ordinal data item in the Work and Schooling LISS dataset, being “What is your current profession / What profession did you exercise in your last job?”.

Health, the dependent variable in the second part of the study, is conceptualized in medical research as “the absence of disease or infirmity” (Larson, 1999). Self-reported health, operationalized as the extent to which a person feels healthy will be measured by a five-point Likert scale item ranging from poor to excellent in the LISS Health survey, referring to the question “How would you describe your health, generally speaking?”. Self-rated health is a subjective, perceived health measure, which might include individual heterogeneity (Etilé & Milcent, 2006; Tubeuf et al., 2008; Caroli & Godard, 2016). However, similarly to Caroli and Godard (2016), Fraser et al. (2000) and Burkert et al. (2014), this paper also measures Self-reported health.

Objective health, operationalized as the use of medications, is observed in the form of dichotomous variables. For each health disorder, a corresponding dummy variable takes value 1 if the individual suffered from it, 0 otherwise.

Diseases induced by meat intakes were identified in the literature review, which will be the focus of the objective health measures. The survey question “Are you currently taking medicine at least once a week for…” in the LISS Health database measures dichotomous

outcomes on the three selected variables, being: (1) high blood cholesterol, (2) high blood pressure, and (3) diabetes.

5.5. Model

The model for this study consists out of two parts. The first part of the model then consists out of two steps. The second part of the model contains three steps.

Part 1. The main interest of the first part is the relation between Income, Education, Occupation as explanatory variables and Meat Consumption as dependent variable, and thus hypotheses 1, 2, and 3 will be assessed. The method is linear OLS regression with a minimum acceptable significance of 95%. The two-step nature of the model suggests that in the first step the effect of Income, Education and Occupation will be assessed on Meat Consumption, while in the second step demographic control variables will be added in the form of Age, Gender, Household size and Marital status.

(1) 𝑦_𝑖 = 𝛼₀ + 𝛽₁𝑥_1𝑖+ 𝛽₂𝑥_2𝑖+ 𝛽₃𝑥_3𝑖+ 𝛾₁𝐴_1𝑖+𝜀_𝑖 𝛼0= intercept term 𝑦𝑖= Meat consumption 𝛽_{1𝑖,2𝑖,3𝑖}= coefficients of interest 𝑥1𝑖= Income 𝑥2𝑖= Education 𝑥_3𝑖= Occupation

𝐴_1𝑖= Demographic control variables 𝜀_𝑖= Error term

The first part of the study is similar to that of Leahy, Lyons and Tol (2010; 2011) and the one of Gossard and York (2003), whereby the varying levels of meat consumption are

analysed in a linear OLS regression model while controlling for SES variables and demographic control variables. However, Gossard and York (2003) use two dependent variables in their research design, with one indicating average meat consumption by individuals measured in grams, and another dependent variable is beef consumption only.

In contrast to Leahy et al. (2010; 2011), this research does not add fish to the consumption levels but focuses on meat consumption as a whole. The 2010 study in the UK separately measures the effects for children in different SES households. However, this paper will include individuals from the age of 15 up to the age of 65 and older, which is similar to the 2011 study of Leahy et al. (2011) conducted in Ireland.

Part 2. In investigating how Meat Consumption predicts Health, the second part of the study uses linear probability regression (LPM). Discovering a causality relation on the effect of Meat Consumption on Health would be best possible through a randomized controlled experiment, which is however not possible with observable data. Evaluating the causal impact of Meat Consumption on Health raises a challenge, which will be addressed by the stepwise nature of the model. In getting towards causality, the researcher will control for possible confounders in the form of SES variables and demographic control variables. In the first step, the effect of Meat Consumption on Health will be measured; in the second step SES variables will be controlled for; in the third step demographic control variables will be added as well.

The objective health variables are based on dichotomous data (i.e. the absence or the presence of medication use for an individual), therefore, linear probability regression is suitable to measure the relation between Meat Consumption and the presence of medications taken for diseases that are connected to meat consumption-related risk factors (i.e. high blood cholesterol; high blood pressure; diabetes). Self-reported health is likely to be endogenous, and older people are expected to report worse health. Endogeneity of this variable will be addressed once again by controlling for SES variables and demographic control variables. In the second

part of the study, Hypothesis 4, 5, 6, and 7 will be tested with a minimum acceptable significance level of 95%. (2) 𝑦_𝑖 = 𝛼₁+ 𝛽₁𝑄_1𝑖+ 𝛾₁𝐴_1𝑖+ 𝛾₂𝐴_2𝑖+ 𝜀_𝑖 𝑦𝑖= Health 𝛽1= coefficient of interest 𝑄_1𝑖= Meat Consumption 𝐴_1𝑖= SES control variables

𝐴_2𝑖= Demographic control variables 𝜀𝑖= Error term

6. Data analysis & Results

This section is structured by the order of the two parts of the study, so first the results of the effect of SES variables on Meat Consumption will be demonstrated, followed by the results of the relationship between Meat Consumption and Health.

6.1. SES and Meat Consumption

Table 4 below demonstrates the results of the OLS regression on Meat Consumption and SES variables (Step 1).