FOOD FOR THOUGHT

0

FOOD FOR THOUGHT

Rethinking foreign direct investment, nutritional risk factors and

the prevalence of diseases of affluence in developing countries

MSc thesis International Economics and Business

Abstract

This thesis examines the effect of unhealthy commodity consumption on diseases of affluence in developing countries, and to what extent the consumption of these unhealthy commodities has been influenced by foreign direct investment. This was analyzed by a number of cross-country and panel data regressions of 39 countries from 1999 to 2008. The model was adjusted for more traditional hypothesized causes such as urbanization, GDP per capita, inequality and population ageing. Unhealthy commodity consumption is found to be a significant determinant of changes in BMI-ratios whereas inward FDI stocks were significantly associated with the consumption of ultra-processed foods. Although changing dietary patterns are high on the research agenda, no previous study has taken together these variables and examined their interrelationship. These findings contribute to explaining the role of nutritional risk factors in the rising burden of disease in developing countries and proves that FDI is a subject worth of greater attention from the perspective of changing food environments.

Key words: FDI, diseases of affluence, BMI, cholesterol, blood pressure, diabetes type 2, unhealthy commodities, developing countries

1

Contents

2.1 Food environments and health ... 7

2.2 Trade and investment liberalization and food environments ... 9

2.3 The role of the United States of America ... 11

3| Data and methodology ... 13

3.1 Study design ... 13

3.2 Trends in the data ... 15

3.3 Statistical Analyses ... 17

3.3.1 Cross-country approach ... 17

3.3.2 Panel data approach ... 18

4| Empirical results... 23

4.1 Cross section approach ... 23

4.1.1 Health model ... 23

4.1.2. Robustness checks health model ... 25

4.2 Panel data approach ... 26

4.2.1 Health model ... 26

4.2.2. Foreign corporate presence model ... 27

5| Discussion ... 32

7| Conclusion ... 35

8| References ... 37

2

List of tables

Table 1. Sargan statistics health model ... 20

Table 2. Unhealthy commodity consumption and BMI-ratio (kg/m2) ... 23

Table 3. Robustness unhealthy commodity consumption and BMI-ratio (kg/m2) ... 25

Table 4. Second robustness test unhealthy commodity consumption and BMI-ratio (kg/m2) ... 25

Table 5. Unhealthy commodity consumption and diseases of affluence indicators ... 26

Table 6. Foreign direct investment and unhealthy commodity consumption ... 28

Table 7. Association FDI and specific food categories ... 29

Table 8. Free trade agreements and unhealthy commodity consumption ... 30

Table 9. Association free trade agreement dummy and the specific food categories ... 31

Table C.1. Descriptive statistics cross-country models ... 43

Table C.2: Data sources ... 46

Table C.3. Correlation matrix cross-country approach health model ... 46

Table C.4. VIF scores cross-country approach health model ... 47

Table C.6. VIF scores panel data approach health model ... 47

Table C.7. VIF scores panel data approach foreign corporate presence model ... 47

Table C.9. Descriptive statistics panel data models ... 48

Table C.10. Correlation matrix panel data approach health model ... 51

Table C.11. Correlation matrix panel data approach foreign corporate presence model ... 51

Table D.1. Unhealthy commodity consumption and TC-ratios (mg/dL) ... 52

Table D.2. Unhealthy commodity consumption and SBP-ratios (mmHg) ... 52

Table D.3. Unhealthy commodity consumption and diabetes-ratios (%) ... 52

Table D.4 . Free trade agreements and BMI-ratios (kg/m2) _{... 52}

Table D.5. Free trade agreements and TC-ratios (mmol/L) _{... 52}

Table D.6. Free trade agreements and diabetes-ratios (%) ... 52

Table D.7. Free trade agreements and SBP (mmHg) ... 52

Table E.1. The association between UC and BMI-ratio and TC-ratio ... 52

Table E.2. The association between UC and SBP-ratio and diabetes-ratio ... 52

3

List of figures

Figure 1: Unadjusted correlation between FDI and unhealthy commodity consumption ... 15

Figure 2: Association ultra-processed, processed foods and soft drink consumption ... 15

Figure 3: Trends over time of avarage diseases of affluence indicators ... 17

Figure 4: Trend over time in average FDI/GDP, 39 countries, 1999-2008 ... 17

Figure B.1. Trends in per capita retail sales of unhealthy commodities from 1999-2008 ... 422

4

List of abbreviations

AGE Life expectancy

BAK Bakery consumption

BMI Body mass index

CF Canned/preserved food consumption CON Confectionary consumption

CPF Chilled processed food consumpton

DA Dairy consumption

DPF Dried processed food consumption GDP Gross national product

FDI Foreign direct investment FTA Free trade agreement

GINI Gini index

IC Icecream consumption

NAFTA North American free trade agreement NCDs Noncommunicable diseases

OF Oils and fats consumption PF Processed food consumption UC Unhealthy commodity consumption UPF Ultra-processed food consumption

URB Urbanization

USA United States of America RM Ready-to-eat meals consumption

SB Snack bars consumption

SBP Systolic blood pressure

SD Soft drink consumption

SSS Sweet and savoury consumption

TC Total cholesterol

5

1| Introduction

on-communicable diseases (NCDs) are currently a major source of death and disease, causing 60% of all deaths worldwide of which 80% in the developing world (WHO global status report on NCDs, 2011). Classic epidemiological transition theory states that as countries achieve a certain level of development, the emergence of chronic diseases substitutes for the frequency of infectious diseases (Omran, 1971; Mathers and Loncar., 2006). The consumption of soft drinks, ultra-processed foods and processed foods, collectively referred to as unhealthy commodities, have been scientifically linked to a higher risk of NCDs such as obesity, diabetes type 2 and cardiovascular diseases. The consumption of these unhealthy commodities increased tremendously in low-and middle-income countries during the past years, however no significant increase is expected in high-income countries (Stuckler 2012). In a trend known as the nutrition transition, populations of developing countries are now consuming diets closer to those of developed countries with more processed foods and fewer whole grains (Popkin BM, 1998; Popkin BM, 2002). The increased consumption of these foods is either driven by shifting demand-side factors or shifting supply-side factors. On the demand side, population-level factors like increased income, rising inequality and rapid urbanization are widely researched and found to cause a demand for easily preparable consumption (Haddad, 2003.; Ruel et al., 1999).

On the other hand, the effects of the supply-side determinants are mainly caused by the expansion of multinational corporations. Changing government policies, including the opening of markets to trade and foreign investment create environments that are conducive to the distribution of various commodities by multinational corporations (Stuckler, 2012). Economic growth is associated with development and goes hand in hand with changing incomes and lifestyles. As Kennedy et al. (2004) state: “globalization is having a major impact on food systems around the world...which affect availability and access to food through changes to food production, procurement and distribution... in turn bringing about a gradual shift in food culture, with consequent changes in dietary consumption patterns and nutritional status that vary with the socio-economic strata”.

Previous work states that Foreign Direct Investment (FDI) has been a key mechanism in shaping the market for processed food (Hawkes., 2004). It has played a key role in the nutrition transition by enabling and promoting the consumption of these foods in developing nations. However, from the perspective of NCDs the subject of FDI deserves greater attention. Public health attention has only recently turned to links between trade agreements, food environments, diets and health (Raynerl et al., 2006; Hawkes et al., 2006)

6

7

Box 1: Diseases of affluence Diseases of affluence is a term given to selected diseases and other health conditions which are commonly thought to be a result of increasing wealth in a society. Examples of diseases of affluence include NCDs as diabetes type 2, cardiovascular diseases and obesity. A NCD, is a disease, which by definition is non-infectious and non-transmissible among people.

Blood pressure (mmHg) and cholesterol (mmol/L) are taken as indicators for cardiovascular

diseases whereas BMI (kg/m2)

measures the changes in weight. Diabetes prevalence (%) measures the incidence of diabetes type 2. Throughout this thesis NCDs and diseases of affluence are used interchangeably.

2 | Literature review

2.1 Food environments and health

At the 2011 United Nations high-level meeting on NCDs, a political declaration presented the case for the prevention of NCDs in low- and middle-income countries (Moodie et al., 2013). Since the early nineties the most globally pervasive health trend has been the rising burden of chronic diseases in these countries. The capacity to effectively deal with the existing burden of NCDs is inadequate and threatens to overwhelm already over-stretched health services and absorb substantial amounts of resources. Progress towards prevention has not kept pace it is therefore important to identify the causes and put the issue on the political agenda (Alwan et al., 2010).

Classic epidemiological transition theory states that as countries achieve a certain level of development, the emergence of chronic diseases substitutes for the frequency of infectious diseases (Omran, 1971; Mathers and Loncar., 2006). These chronic diseases were primarily apparent in western societies and therefore called diseases of affluence (box1). This development is associated with rising urbanization rates and population ageing (Popkin., 1999; Chow et al., 2009; Subramanian & Smith., 2006; Agyemang, 2006). Ageing of the world’s population is seen as a key driver since NCDs tend to occur in middle to later stages of life. As the WHO stated ‘’age is the single most important determinant of NCDs’’. Whereas living in a city contributes to more sedentary lifestyles than life in rural agricultural zones, therefore the incidence of obesity, diabetes type 2 and cardiovascular diseases increases. Rising prevalence of NCDs in India and other developing countries is essentially attributed to urbanization, under the premise that living in a city is conducive to a more sedentary lifestyle (Ramachandran et al., 2008)

8

income levels increase, people’s habits and consumption patterns change. Distinct from Western countries, where people buy more healthy food and spend more time working out as their income levels rise, in developing countries the opposite seems to occur (Eid et al., 2003; Popkin et al., 2001; Andrieu et al., 2006).

An extensive body of biomedical studies links the consumption of unhealthy commodities and an increased risk of chronic diseases. At first, processed foods, containing added sugar and other caloric sweeteners, may artificially stimulate the appetite by creating hormonal imbalances and may cause individuals to eat more food than is naturally needed by their body (Kostoff, 2001; van der Vliet, 2001; Pereira et al., 2005; Li et al., 2003; Berkey et al.,2004). Secondly, refined carbohydrates, readily available in highly processed foods, can be easily and rapidly absorbed by the body (Englyst and Englyst, 2005; Wylie-Rosett et al., 2004). Moreover, a large body of literature demonstrated that processed foods with high glycaemic indexes, concentrated sugars, highly refined starches, and other chemicals such as fructose, can change the hormonal balance and create an addiction to food, leading to obesity, cardiovascular diseases, and diabetes type 2 (Gross et al., 2004; Hu et al., 2001; Kaechele et al., 2006; Pawlak et al., 2002; Liu et al., 2002; MacInnis and Rausser, 2005). Additionally, research shows that obesity is more prevalent in countries that are characterized by less egalitarian distribution of wealth and weaker social welfare systems (Wilkinson and Pickett., 2009; Vogli et al. 2013). The mechanism works twofold, economic insecurity causes stress which can increase obesity directly through adverse metabolic responses or indirectly through behavioural reactions such as diminishing physical activity and surplus energy intake (Brunner et al., 2007).

Previous studies have mainly made use of either cross-country or time-series studies. MacInnis and Rausser (2005) in their research show that energy density is one of the major risk factors to the commonness of childhood obesity in the US. Hence, Basu et al. (2012) research how market sizes of major food products corresponded to diabetes type 2. Since diabetes type 2 develops after cumulative exposure to dietary risk they calculated mean kilojoules per person per day over a 10 year period, with sugar as a separate variable. This was regressed against diabetes type 2 prevalence in 2007. They found that sugar exposure was the major dietary risk related to diabetes type 2 prevalence, so they

Box 2: Food classification Unhealthy commodities in this paper consist of the foods described in group 2 and 3 Group 1: minimally processed foods

Minimal processes are applied to single basic foods with the purpose of preserving them. Not included in the analysis

Group 2: Processed foods Food commodities which are

inedible or unpalatable by

themselves, and have a higher energy and density and lower nutrient density compared with the whole foods from which they were extracted. In this analysis included: confectionary, bakery, dairy.

Group 3: Ultra-processed foods.

Commodities made from

processed substances extracted or refined from whole foods. The

industrial processes are all

designed to create durable, accessible, convenient, attractive ready-to-eat/heat products. They are formulated to have a long shelf-life, to be transportable for long distances, to be extremely palatable and often to be habit forming. In small amounts the products are harmless however, intense palatability, omnipres- ence, and aggressive marketing strategies, all make modest consumption unlikely. In this analysis included:

canned/chilled/dried processed

9

investigated specifically how underlying population level factors – urbanization, population ageing, rising incomes and market integration – may relate to sugar exposure as well as to diabetes type 2 prevalence. Increased income correlated to increased overall joules of food available. However, the level of food importation appears to shift the food supply such that a higher proportion of available joules consist of sugar and related sweeteners. In a case study of Guatemala Asfaw (2001) examines the contribution of processed food consumption to the prevalence of overweight/obesity in the country. The results show that all other things remaining constant, a 10% point increase in the consumption of partially processed food increases the Body Mass Index (BMI) of family members by 3,95%. For highly processed foods the impact was even stronger, a 10% increase in the consumption of highly processed items increases the BMI of individuals by 4,25%.

In the medical field, by making use of clinical trials, consensus is reached on the harmful effect of the consumption of unhealthy commodities and various chronic diseases. For example, recently Dutch and American researchers have come up with the first hard evidence that soft drinks play a role in childhood obesity (de Ruyter et al., 2012). The research involved Dutch children from eight primary schools in Haarlem, Purmerend and the Zaanstreek. The researchers asked 641 normal-weight schoolchildren aged four to eleven to drink eight ounces of a 104 calorie sugar-sweetened or no-calorie sugar-free fruit-flavoured drink every day from identical cans. Over 18 months, the children in the sugar-free group gained 13.9 pounds on average, while those drinking the sugar-added version gained 16.2 pounds. Previous socioeconomic studies have proved their case by making use of one country analysis over time, cross-country regressions, or by just taking one disease of affluence indicator. Alternatively, in this research an extensive group of unhealthy commodity consumption is linked to multiple indicators of diseases of affluence. Based on the above mentioned literature, a positive relation is expected between the consumption of these unhealthy commodities and the prevalence of diseases of affluence leading to the formulation of the following hypothesis.

𝐇𝟏: Unhealthy commodity consumption is positively related to body mass index in low- and middle-income countries

𝐇𝟐: Unhealthy commodity consumption is positively related to total cholesterol in low- and middle-income countries

𝐇𝟑: Unhealthy commodity consumption is positively related to systolic blood pressure in low- and middle-income countries

𝐇𝟒: Unhealthy commodity consumption is positively related to the prevalence diabetes type 2 in low- and middle-income countries

2.2 Trade and investment liberalization and food environments

10

trade and foreign direct investment has profound implications for the structure and nature of food systems. Kenney et al. (2004) stated that globalization is having a major impact on food systems around the world, affecting the availability and accessibility of food. The globalization-related changes in the food system have been influenced by three important features.

At first, the opening of domestic markets towards trade and foreign direct investment. A key mechanism of trade and investment liberalization is the negotiation of international trade and investment agreements and treaties. These bilateral, regional or multilateral trade agreement implicate commitment by member countries to reduce tariffs, export subsidies and domestic supports. Many countries signed multiple trade agreements easing cross-border trade and investment. Mainly because of reduced costs and an increased ease of doing business is food trade a key element of the worldwide food system. Trade in food remained limited since the mid-1990s (Regmi et al., 2009) while FDI, on the other hand, already mushroomed from the beginning of the nineties. FDI can be defined as a long term investment made by a company based in one country, into a company based in another country. The foreign enterprise becomes a foreign affiliate of the parent company, creating or joining a transnational corporation (TNC). FDI is a highly lucrative device for multinational companies to reach foreign markets, allowing them to jump trade barriers and cut costs overcoming tariff and non-tariff barriers to trade. Processed food is of particular interest for multinational companies because they have low production cost, a long-shelf-life, and a high retail value, which is an enormous commercial advantage over fresh whole or minimally processed foods (Stuckler et al., 2012). A number of time series studies confirm this relationship for, amongst others, Central-America and Pacific Island countries (Thow and Hawkes.,2009; Hawkes & Thow., 2008; Clark et al.,2012). Additionally, Stuckler et al. (2012) found a strong correlation between high levels of FDI flows and an exposure to soft drinks. Nonetheless, rising incomes with a limited penetration by transnational companies do not necessarily give rise to higher consumption of processed food. However, this panel data regression does not take into account a loop of causality between FDI and the consumption of unhealthy commodities. Another drawback of the study is the use of FDI flows which might not reveal the long-term effect of FDI on the consumption of unhealthy commodities.

11

position of market leadership and be able to influence the entire production, marketing and distribution process to local tastes and conditions (Hawkes., 2006).

Thirdly, systematic and aggressive mass-marketing campaigns of processed foods and drinks contribute to a rising demand. It works by attracting attention to new products and improving the desirability of these products. TNCs have massive advertisement budgets and make extensively use of various channels to influence consumer behaviour (Popkin., 2002). There is no doubt that access to modern mass media in developing countries has grown very rapidly, particularly with the entrance of the internet. This offers great opportunities for TFCs and leads to an increased exposure to advertising of western convenience and fast food corporations. Currently, six out of the ten most “liked” companies on Facebook are fast food or convenience food corporation with Coca-Cola on top with over 47 million “likes”. However, this advertisement is not only targeted at the main grocery shoppers. Cairns et al. (2009) and MacInnis et al. (2005) show evidence from Western countries targeting a significant proportion of food advertising and promotion at children and youth, and much of it is for nutrient poor, high calorie food. Available studies from developing countries show a similar tendency. In Brazil, close to 60 percent of all food advertisement in 2002 was for foods high in fats and sweeteners (Sawaya et al. 2004). And in Asia food makes up a substantial share of advertising targeted at children, ranging from 25 percent in South Korea, to 40 to 50 percent in India, 50 to 75 percent in Pakistan and the Philippines and 70 percent in Malaysia (Escalante de Cruz et al., 2004). The process of marketing is twofold since it encourages consumers to consume more of the products and producers to produce them, thus advancing the cycle of global market exchange and integration.

Few panel data attempts have been made in empirically linking globalization to changing consumption patterns. However, the above mentioned theoretical frameworks, leads to assume a positive relation between FDI and unhealthy commodity consumption.

𝐇𝟓: Foreign direct investment is positively related with unhealthy commodity consumption in low- and middle-income countries

2.3 The role of the United States of America

12

drinks, refined and processed foods, meat and dairy products as well as significant investment in Mexico’s full spectrum of the food supply chain, from production and processing to distribution and retail. Countries in Central America have more often been a subject for case studies in this area. In Thow et al. (2009) changes in tariff and non-tariff barriers for each country are compared with time-series graphs of food imports and food availability. The study indicates that the policies of trade liberalization in Central-American countries over the past two decades, particularly in relation to the USA, have had severe implications for health in the region, with rising rates of chronic diseases. However, the evidence not only comes from Latin American countries, free trade agreements with the USA are associated with high consumption of soft drinks in several countries (International Centre for Alcohol Policies, 2006; Stuckler et al., 2012). This leads to expect a positive relation between free trade agreements with the USA and unhealthy commodity consumption, leading to the formulation of the following hypothesis.

𝐇𝟔: Free trade agreements with the United States of America are positively associated with unhealthy commodity consumption in low-and middle-income countries.

In the aforementioned review, free trade agreements are hypothesized to be positively associated with unhealthy commodity consumption. Moreover, for a number of Central-American countries a positive relation is found between changing tariff and non-tariff barriers and rising rates of chronic diseases. Additionally, unhealthy commodity consumption seems to play a major role in these rising rates of NCDs (Thow et al., 2009). However, it can also be argued that free trade agreements are associated with increasing rates of diseases of affluence because of the adoption of western lifestyles characterized by physical inactivity encouraged by motorized transport, increasing use of energy-sparing devices, increasingly sedentary employment and the seduction of TV and video games (Prentice, 2005; Carrera-Bastos et al., 2011; Sobngwi et al., 2002). Additionally, changing dietary patterns are increasingly associated with the adoption of western lifestyles (Popkin, 1998), especially when countries are having close ties with the USA (Hawkes, 2006). Taking these associations together, leads to the formulation of the following conjecture.

13

3| Data and methodology

3.1 Study design

The analyses consist of cross-country and panel data analyses of 42 low- and middle-income countries1 over the period 1999 to 2008. In order to adequately analyse the association between FDI, unhealthy commodity consumption and diseases of affluence with a strongly balanced sample, only countries with data on all three variables of interest for the whole period of 1999-2008 are included. A Cook’s distance test (Cook., 1979) is performed to solve the problem of high influential observations, with influence being the result of outliers. An observation is influential if removing the observation substantially changes the estimations of the regression coefficient. The Cook’s distance test identified 31 influential observations and led to the deletion of Saudi Arabia, the United Arab Emirates and Taiwan. Hence, the analysis continues with 39 low- and middle-income countries. The list with countries included in the sample can be found in appendix 1. The interrelation between inward FDI stocks, unhealthy commodity consumption and diseases of affluence is examined in two steps. First, the health model analyses the effect of unhealthy commodity consumption on diseases of affluence indicators. Second, the foreign corporate presence model examines the relation between inward FDI stocks and unhealthy commodity consumption.

Foreign direct investment

Data on foreign direct investment are taken from the United Nations Conference on Trade and Development2_{. Stuckler et al. (2012) use FDI flows as a percentage of Gross Domestic Product (GDP) as} a measurement for greater foreign corporate entrance. However, the effect of FDI on unhealthy commodity consumption is likely to be displayed as a long-term development over years hence, a certain investment flow in a specific year may not reveal the effect of FDI on the consumption of unhealthy commodities. Therefore, the accumulated FDI present in a country is preferred over the year to year flow of investment. To normalize for the size of each country’s GDP3_{, foreign direct investment is reported as} a percentage of GDP (constant 2005 US$, fixed exchange rate 2005).

Packaged food and beverages

Data on unhealthy food and beverages are market data on commodity sales from the Euro monitor Passport Global Information database 2013 edition. The data on ultra-processed and processed foods consumption includes the per capita volume (constant 2005 US$, fixed exchange rate 2013) of bakery, canned/preserved food, chilled/processed food, confectionary, dairy, dried processed food, ice cream, oils

1The classification is adopted from the World Bank. Low-income and middle-income economies are sometimes referred to as developing economies. It is not intended to imply that all economies in the group are experiencing similar development or that other economies have reached a preferred or final stage of development. Classification by income does not necessarily reflect development status. (World Bank, 2013)

14

and fats, ready meals, snack bars and sweet and savoury snacks. The data on soft drinks consumption includes the per capita value (constant 2005 US$, fixed exchange rate 2013) of carbonates, concentrates and sports and energy drinks,

Indicators diseases of affluence

Data on the age standardized mean population levels of BMI kg/m2_{, total cholesterol (mmol/L), systolic} blood pressure (mm/hg), and prevalence of diabetes type 2 (%) are obtained from the Global Burden of Metabolic Risk Factors of Chronic Diseases Collaborating Group4_{. This group produces comparative} estimates of cross-country differences and changes over time in BMI, cholesterol, systolic blood pressure and the prevalence of diabetes type 2 for adults aged 20 years old and older. Data on these measures are reported separately for men and women. Vogli et al. (2013) are followed by estimating an overall indicator, using the proportion of female population from the WDI from 1999 to 20085_.

Free trade agreement dummy

A free trade agreement dummy6 _{is included to examine whether countries which have a free trade} agreement with the USA are associated with either a higher prevalence of diseases of affluence or higher consumption of unhealthy commodities. The regressions are repeated with the inclusion of dummy variable indicating whether countries have a free trade agreement with the USA or not. Countries which have a free trade agreement with the USA receive the value “1”. All other countries receive the value “0”. Referring to the health model, it is argued that free trade agreements with the USA are expected to be positively related with the incidence of diseases of affluence because of the adoption of western lifestyles and unhealthy commodity consumption is hypothesized to be the main influential factor in this association. To formally test for this association, an interaction variable between unhealthy commodity consumption and a free trade agreement dummy is included.

Covariates

More conservative studies have focused on traditional causes of the nutrition transition and changing health patterns. Previous studies are followed in adjusting the model for these socioeconomic risk factors. The proportion of the population living in urban areas7 _{is taken as an indicator of a number of lifestyle} variables, such as physical activity in occupational and transport spheres (Ezatti, 2005). For example, people living in rural areas often have higher levels of physical activity, reflecting the need to walk longer

4_{Source: EuroMonitor International (2013) Passport Global Market Information Database}

5_𝐵𝑀𝐼 𝑟𝑎𝑡𝑖𝑜= ( 𝑓𝑒𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝐵𝑀𝐼𝑓𝑒𝑚𝑎𝑙𝑒) + ( 𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝐵𝑀𝐼𝑚𝑎𝑙𝑒) 𝑇𝐶𝑟𝑎𝑡𝑖𝑜= (𝑓𝑒𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛_{𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛} ∗ 𝑇𝐶𝑓𝑒𝑚𝑎𝑙𝑒) + (𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛_{𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛} ∗ 𝑇𝐶𝑚𝑎𝑙𝑒) 𝑆𝐵𝑃𝑟𝑎𝑡𝑖𝑜= ( 𝑓𝑒𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝑆𝐵𝑃𝑓𝑒𝑚𝑎𝑙𝑒) + ( 𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝑆𝐵𝑃𝑚𝑎𝑙𝑒) 𝐷𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑟𝑎𝑡𝑖𝑜= ( 𝑓𝑒𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝑑𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑓𝑒𝑚𝑎𝑙𝑒) + ( 𝑚𝑎𝑙𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 ∗ 𝑑𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑚𝑎𝑙𝑒)

6_{Data on trade agreements are from sources listed in Kohl, Tristan, 2012, Trade agreements galore: Who, what, where,}

when, why, how and how much? SOM: University of Groningen. All data points were updated by consulting the latest versions of the respective sources in December 2013.

15

distances for day-to-day activities and their agricultural occupations. Moreover, inequality is found to cause stress and excess consumption of unhealthy commodities therefore, the Gini coefficient8 _{is taken to} measure the income distribution of a nation’s residents. A Gini coefficient of zero expresses perfect equality, and a Gini coefficient of one express maximal inequality. Additionally, GDP per capita is the commonly used indicator for a nation’s material well-being. Previous research suggest that as economic development increases, people’s consumption patterns change and an increased consumption of unhealthy commodities seems to be the consequence (Stuckler. 2012). Moreover, life expectancy7 _{is only taken as a} covariate in the health model since NCDs tend to increase with age. As the WHO stated “age is the single most important determinant of NCDs”. However, this promotes the view that chronic diseases are an inevitable consequence of ageing (Stuckler, 2008). The prevalence of NCDs appears to happen differently than in developed countries. As argued by Stuckler (2008) people in developing countries are either moving through the transition from being healthy to dying more rapidly than people in developed countries are or are accumulating behavioural risks at younger ages. Since there is no official threshold found at which NCDs become critical, previous literature is not followed in taking the percentage of the population aged over sixty-five as control variable. Instead, the overall life expectancy from the WDI is taken. Furthermore, the initial level of BMI, total cholesterol, systolic blood pressure and prevalence of diabetes type 2 in 1999 is included in the health model. This is to control for initial level heterogeneity among countries. The initial level of unhealthy commodity consumption is also included to control for country-level heterogeneity in the consumption of unhealthy consumption.

3.2 Trends in the data

As a first step, based on averaged rates of consumption between 1999 and 2008, the trends in the per capita consumption of each major food category are compared. As can be observed in table B.1, the consumption of these unhealthy commodities is growing steadily. However, the data reflect a dip in

8_{Source: UNU-WIDER database: World Institute for Development Economics Research, 2013.}

UPF 1 UPF 2 SD UC 0 200 400 0 200 400 600 0 200 400 600 0 100 200 0 100 200 0 500 1000

16

Figure 3. Trends over time of average diseases of affluence indicators, 39 countries, 1999-2008

24 24,5 25 25,5 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 BMI -r atio (kg/m 2)

Trend over time in BMI-ratio

4,69 4,695 4,7 4,705 4,71 4,715 4,72 4,725 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 TC -r atio (m m o l/L)

Trend over time in TC-ratio

126,2 126,3 126,4 126,5 126,6 126,7 126,8 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 SB P -r atio (m m H g)

Trend over time in SBP-ratio

8,8 9 9,2 9,4 9,6 9,8 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 D iab e te s-ratio (% )

Trend over time in diabetes-ratio

consumption in 2008, this might be associated with the economic recession from 2008 onwards. This is also pointed out by Stuckler et al. (2012). The scatterplots in figure 1 show the unadjusted correlation between unhealthy commodity consumption and inward FDI stocks. The scatterplot disaggregates the trend in developing countries into country-specific patterns. An uneven pattern of unhealthy commodity consumption can be observed. One clue about the underlying causes of this consumption is the observation that the consumption of these commodities are strongly correlated as can be observed in figure 2. In other words, in countries where there are high rates of soft drink consumption, it is quite likely that there is also a high intake of other unhealthy food commodities. Figure 3 shows the trend in changes in diseases of affluence indicators, based on average levels of the indicators over 1999-2008. Results show that in ten years the mean adult BMI increased from 24.5 to 25.4 kg/m2. Which is worrisome since a BMI of 18.5 kg/m2 to 25 kg/m2 indicates optimal weight, a BMI lower than 18.5 kg/m2 suggest underweight and a number above 25 kg/m2 indicates overweight. The figures show that developing countries are already facing overweight. During the same period total cholesterol and systolic blood pressure have not significantly changed. A desirable cholesterol level is lower than 5.2 mmol/L, between 5.2-6.2 mmol/L

17

Figure 4: Trend over time in average FDI/GDP, 39 countries, 1999-2008

figures indicate that developing countries are, although at a steady rate, already at high risk for cardiovascular diseases. Furthermore, diabetes type 2 prevalence is displayed in percentages and increased with 0.6 percentages points over the last decade. Additionally, figure 5 (appendix B), shows the unadjusted correlation between unhealthy commodity consumption and the four diseases of affluence indicators. BMI-ratio and TC-ratio tend to move in the same direction as unhealthy commodity consumption however, SBP-ratio and diabetes-ratio show more ambiguous patterns. To conclude with, figure 4 shows the trend over time of inward FDI stocks as a percentage of GDP. The graph shows a gradual increase in inward FDI stocks in developing countries however, with a sudden dip in 2008, this might be associated with the economic recession from 2008 onwards.

.

3.3 Statistical Analyses

3.3.1 Cross-country approach

To assess the interrelationship between FDI, consumption of unhealthy commodities and diseases of affluence, the analysis proceeded in two steps. First, the effect of the consumption of unhealthy commodities on diseases of affluence indicators is examined. It is examined whether the absolute change in the consumption of unhealthy commodities and the corresponding interaction and control variables over the years 2008 influence the absolute change in diseases of affluence indicators over 1999-2008 The summary statistics are to be found in table C.1, the data sources can be found in table C.2 and the correlation coefficients for all variables are reported in table C.3. A cross-section ordinary least square (OLS) regression model is used leading to the following empirical model.

Health model:

∆𝐵𝑀𝐼𝑖= 𝛼𝑖+ β1∆𝑈𝐶𝑖+ 𝛽2∆𝐺𝐼𝑁𝐼𝑖+ β3∆𝐺𝐷𝑃𝑖+ 𝛽4∆𝑈𝑅𝐵𝑖+ 𝛽5∆𝐴𝐺𝐸𝑖 + 𝛽6𝐵𝑀𝐼1999+ Β7𝑈𝐶1999𝜖𝑖 (1)

∆𝑇𝐶𝑖= 𝛼𝑖+ 𝛽𝑖∆𝑈𝐶𝑖+ 𝛽2∆𝐺𝐼𝑁𝐼𝑖+ β3∆𝐺𝐷𝑃𝑖+ 𝛽4∆𝑈𝑅𝐵𝑖+ 𝛽5∆𝐴𝐺𝐸𝑖 + 𝛽6𝑇𝐶1999+ Β7𝑈𝐶1999+ 𝜖𝑖 (2)

∆𝑆𝑃𝐵𝑖= 𝛼𝑖+ 𝛽𝑖∆𝑈𝐶𝑖+ 𝛽2∆𝐺𝐼𝑁𝐼𝑖+ β3∆𝐺𝐷𝑃𝑖+ 𝛽4∆𝑈𝑅𝐵𝑖+ 𝛽5∆𝐴𝐺𝐸𝑖+ 𝛽6𝑆𝑃𝐵1999+ Β7𝑈𝐶1999+ 𝜖𝑖 (3)

∆𝐷𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑖= 𝛼𝑖+ 𝛽𝑖∆𝑈𝐶𝑖+ 𝛽2∆𝐺𝐼𝑁𝐼𝑖+ β3∆𝐺𝐷𝑃𝑖+ 𝛽4∆𝑈𝑅𝐵𝑖+ 𝛽5∆𝐴𝐺𝐸𝑖+ 𝛽6𝑃𝐷1999+ Β7𝑈𝐶1999+ 𝜖𝑖 (4)

In which i denotes the country, 𝐵𝑀𝐼𝑖, 𝑆𝐵𝑃𝑖, 𝑇𝐶𝑖, and 𝑃𝐷𝑖 denote the absolute change in body mass index, total cholesterol, systolic blood pressure, and diabetes type 2 prevalence, respectively, over 1999-2008. 𝛽1 denotes the absolute change in consumption of unhealthy commodity consumption over 1999-2008,

0 20 40 60 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 p e rc e n tages

18

which is subdivided in: “UPF”, “PF” and “SD”, leading to multiple regressions, 𝛽2 is the regression coefficient for absolute change in the gini-index over 1999-2008, 𝛽3 is the regression coefficient for the absolute change in GDP per capita over 1999-2008, 𝛽4 is the regression coefficient for the absolute change in urbanization over 1999-2008, 𝛽5 is the regression coefficient of the absolute change in life expectancy over 1999-2008, 𝛽6 is the regression coefficient of the initial level of BMI, total cholesterol, systolic blood pressure and prevalence of diabetes type 2 in 1999, respectively, 𝛽7 is the regression coefficient for the initial level of unhealthy commodity consumption in 1999, which is subdivided in: ‘’ultra-processed food consumption’’, ‘’processed food consumption’’ and ‘’soft drink consumption’’, 𝜖_𝑡 denotes the error term and 𝛼 is a constant.

Critical assumptions of OLS regressions require that the error terms are normally distributed and have constant variance. The normality assumption is tested by a Jarque Bera (Jarque et al., 1980) we can reject the hypothesis that the residuals are not normally distributed. Next, the prevalence of heteroskedasticity is tested by performing a Breusch-Pagan test The Breusch-Pagan test does not reject the null-hypothesis of constant variance: (p=0.5923) Therefore, technically spoken, robust standard errors are not necessary. However, robust standard errors will be utilized to be on the safe side. The variance inflation factor is used to test for multicollinearity. As can be observed in table C.4, the VIF-scores of ‘’URB’’, ‘’UC’’, ‘’UPF’’, ‘’PF’’, ‘’SD’’, ‘’AGE’’, ‘’GINI’’ and ‘’GDP’’ are all well below the critical threshold of 10 therefore multicollinearity is not a problem within these models (Belsley 1991).

The role of the United States of America

To examine whether or not free trade agreements with the USA are associated with diseases of affluence, the regressions are repeated with the inclusion of a dummy variable indicating whether countries have a free trade agreement with the USA or not. To formally test whether free trade agreements in the aggregate are associated with a higher occurrence of diseases of affluence or whether free trade agreements are associated with an increase in unhealthy commodity consumption and therefore are associated with the occurrence of diseases of affluence, an interaction variable is included between “FTAUS” and either “UC”, “UPF”, “PF” or “SD” ∆𝐵𝑀𝐼𝑖= 𝛼 + β1∆𝑈𝐶1+ 𝛽2∆𝐹𝑇𝐴𝑈𝑆𝑖+ 𝛽3∆𝐺𝐼𝑁𝐼𝑖+ β4∆𝐺𝐷𝑃𝑖+ 𝛽5∆𝑈𝑅𝐵𝑖+ 𝛽6∆𝐴𝐺𝐸𝑖+ 𝛽7∆𝐹𝑇𝐴𝑈𝑆𝑖∗ ∆𝑈𝐶𝑖+ 𝜖𝑖 (5) ∆𝑇𝐶𝑖= 𝛼 + β1∆𝑈𝐶1+ 𝛽2∆𝐹𝑇𝐴𝑈𝑆𝑖+ 𝛽3∆𝐺𝐼𝑁𝐼𝑖+ β4∆𝐺𝐷𝑃𝑖+ 𝛽5∆𝑈𝑅𝐵𝑖+ 𝛽6∆𝐴𝐺𝐸𝑖+ 𝛽7∆𝐹𝑇𝐴𝑈𝑆𝑖∗ ∆𝑈𝐶𝑖+ 𝜖𝑖 (6) ∆𝑆𝑃𝐵𝑖= 𝛼 + β1∆𝑈𝐶1+ 𝛽2∆𝐹𝑇𝐴𝑈𝑆𝑖+ 𝛽3∆𝐺𝐼𝑁𝐼𝑖+ β4∆𝐺𝐷𝑃𝑖+ 𝛽5∆𝑈𝑅𝐵𝑖+ 𝛽6∆𝐴𝐺𝐸𝑖+ 𝛽7∆𝐹𝑇𝐴𝑈𝑆𝑖∗ ∆𝑈𝐶𝑖+ 𝜖𝑖 (7) ∆𝐷𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑖= 𝛼 + β1∆𝑈𝐶1+ 𝛽2∆𝐹𝑇𝐴𝑈𝑆𝑖+ 𝛽3∆𝐺𝐼𝑁𝐼𝑖+ β4∆𝐺𝐷𝑃𝑖+ 𝛽5∆𝑈𝑅𝐵𝑖+ 𝛽6∆𝐴𝐺𝐸𝑖+ 𝛽7∆𝐹𝑇𝐴𝑈𝑆𝑖∗ ∆𝑈𝐶𝑖+ 𝜖𝑖 (8)

3.3.2 Panel data approach

19

deal with panel data regression are a pooled model, a fixed effects model, and a random effects model. The pooled model pools together the data of different countries however, ignores the relevant different between countries that can be expected. The fixed effects model accounts for the unobserved factors that differ between countries and are constant over time. The constant 𝛼 is different across individuals however, the slope coefficients 𝛽 are assumed to be equal for all individuals. The alternative is the random effects model were these assumptions are made as well, but also takes into account that the countries are randomly chosen (Wooldrige 2009).

The Hausman test (Hausman, 1978) was used to compare random versus fixed-effects models. The Hausman tests indicate that in both cases the random effects model is not appropriate in both cases(p=0.000). Hence, the fixed effects model is the model that fits the data best and is further specified. Panel data models can incur problems of multicollinearity between highly correlated predictors that could result in unstable coefficients and standard errors. As an indicator of multicollinearity the variance inflation factor values greater than or equal to 10 (Belsley 1991) is used. As can be observed in table C.6 and table C.7, the VIF-scores of “URB”, “UC”, “UPF”, “PF”, “SD”, “AGE”, “GINI”, “GDP” and “FDI” are all well below the critical threshold of 10 therefore multicollinearity is not a problem within this model (Belsley 1991). There could also be problems of heteroskedasticity or autocorrelation over time that can further bias estimations. The Breusch-Pagan tests is applied which gives a significant p-value leading to the conclusion that individual heterogeneity is present. Autocorrelation is expected as the variables only change slowly over time. The Wooldrige test for autocorrelation is applied and a significant p-value leads to the conclusion of autocorrelation in both models (Wooldridge, 2002). Cluster-robust standard errors (Baum 2006) help to correct for heteroskedasticity and autocorrelation. Research on international capital flows are followed in including time dummies from 1999 to 2008, to control for all time specific events that vary only across time but not across countries (Kose et al., 2008; IMF April 2011)

20

This two-step estimation involves a two-step regression methodology. In the first step, the first and second lagged instrumental variables for unhealthy commodity consumption are run on the independent unhealthy commodity consumption variable to obtain the residuals9_{. Secondly, the health model is fitted} on the regressors that contain the first step residuals10_{. After conducting a Hausman test for endogeneity,} the residuals show a significant p-value (0.0515), indicating that unhealthy commodity consumption indeed is found endogenous. Secondly, valid and strong instruments are needed for the endogenous variable. A valid instrument is one that influences the endogenous explanatory variable and at the same time is uncorrelated with the error term (Baum et al 2002). It is commonly known that it is hard to find good instruments for macro-economic data therefore, previous macro-economic research is followed and the first and second lagged variables are tested as instruments. To test the strength of the instrumental variables a F-test is performed. Since the F is higher than 10 (p=0.000), we found evidence to reject H0 and the first and second lagged variables are ‘strong’ instruments. To control for the validity of the instruments a Sargan test is performed. The significant p-values indicate that the instrumental variables are uncorrelated to the residuals, and therefore they are acceptable, healthy, instruments. Table 8 shows the Sargan test statistics for all four explanatory variables of the health model.

UC UPF PF SD

Sargan p-value 0.0099 0.0021 0.0094 0.0000 Sargan statistic 6.656 9.441 6.741 20.859

Table 1. Sargan statistics health model

The least square estimator is not found consistent because of endogenous explanatory variables whereas the instrumental variable estimator addresses the simultaneity bias. Therefore. the literature is followed in using an instrumented fixed effects model with first lagged variables used as instruments. Where 𝑇𝑡 denote the time dummies included and 𝜇𝑖 the country fixed effects.

𝐵𝑀𝐼𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (9)

𝑇𝐶𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (10)

𝑆𝑃𝐵𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (11)

𝐷𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (12)

With respect to the foreign corporate presence model, inward FDI stocks is suspect for endogeneity. The same two stage least square instrumenting approach as above is used. At first the first and second lagged instrumental variables for inward FDI stocks are run on the independent inward FDI stocks variable to obtain the residuals11_{. In the second step the foreign corporate presence is fitted on the regressors that}

9_{First stage regression: 𝑈𝐶}

𝑖𝑡= 𝛼𝑖𝑡+ γ1𝑈𝐶𝑖𝑡_1+ γ2𝑈𝐶𝑖𝑡_2+ γ3𝐺𝐷𝑃𝑖𝑡+ γ4𝑈𝑅𝐵𝑖𝑡+ γ5𝐴𝐺𝐸𝑖𝑡+ γ6𝐺𝐼𝑁𝐼𝑖𝑡+ 𝜇𝑖+ 𝑣𝑖𝑡

10_{Second stage regression: 𝐵𝑀𝐼}

𝑖𝑡= 𝛼𝑖𝑡+ β1𝑈𝐶̂ + 𝛽𝑖𝑡−1 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡

11_{First stage regression: 𝐹𝐷𝐼}

21

contain the first step residuals12_{. After conducting a Hausman test for endogeneity, the residuals show a} significant p-value (0.000), indicating inward FDI stocks is found endogeneous. The first and second lagged variables are tested as instruments. To test the strength of the instrumental variables a F-test is performed. Since our F is higher than 10 (p=0.000), we found evidence to reject H0 and the first and second lagged variables are ‘strong’ instruments. To control for the validity of the instruments a Sargan test is performed. The significant p-value (0.0029) indicates that the instrumental variables are uncorrelated to the residuals. Therefore, the instrumented fixed effects model and the first lagged variable is used as an instrument in the foreign corporate presence model. Moreover, this approach can reduce the influence of causality as well. The explanation is simple and intuitive, unhealthy commodity consumption does not influence past inward FDI stocks.

𝑈𝐶𝑖𝑡= 𝛼𝑖𝑡+ β1𝐹𝐷𝐼̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ 𝛽3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (13)

Where i is the country. 𝑈𝐶𝑖𝑡 denotes the consumption of unhealthy commodity consumption, which is subdivided in: “UPF”, “PF” and “SD”, leading to multiple regressions , 𝛽₁ is the regression coefficient for FDI stocks , 𝛽2 is the regression coefficient for the gini-index, 𝛽3 is the regression coefficient for GDP per capita, 𝛽4 is the regression coefficient for urbanization, 𝑇𝑡 denote the time dummies included, 𝜇𝑖 the country fixed effects and 𝜖𝑡 denote the error term and 𝛼 is a constant. The descriptive statistics are to be found in table C.9. Moreover, the correlation coefficients for all variables are reported in table C.10 and table C.11.

The role of the United States of America

Again, it is examined whether free trade agreements with the USA in the aggregate are associated with diseases of affluence or with unhealthy commodity consumption. Countries which have a free trade agreement with the USA receive the value “1”. All other countries receive the value “0”. Concerning the health model, to formally test whether free trade agreements in the aggregate are associated with a higher occurrence of diseases of affluence or whether free trade agreements are associated with an increase in unhealthy commodity consumption and therefore are associated with the occurrence of diseases of affluence, an interaction variable is included between “FTAUS” and either “UC”, “UPF”, “PF” or “SD”. Since unhealthy commodity consumption is instrumented by an one year lag in all previous regressions to overcome endogeneity problems, the free trade agreement dummy is also tested for the presence of endogeneity. First, reverse causality, a higher prevalence of diseases of affluence attracting free trade agreements is very unlikely and is therefore not seen as a problem in this model. Second, simultaneity bias could be a problem, an unobserved variable could influence the free trade agreement as well as the prevalence of diseases of affluence at the same time. The two stage least square instrumenting approach is

12_{Second stage regression: 𝑈𝐶}

22

used to test for the simultaneity bias. The residuals, obtained after the first stage regression13 _{and fitted on} the second stage regression14_{, show an significant p-value (0.090). This indicates that the free trade} agreement is indeed found endogenous. Therefore, the dummy variable as well as the interaction variable are instrumented by an one year lag. Referring to the foreign corporate presence model, the free trade agreement dummy is included to test whether free trade agreements with the USA in the aggregate are associated with unhealthy commodity consumption. The two stage least square instrumenting approach is used to test for the simultaneity bias. The residuals, obtained after the first stage regression15 _{and fitted on} the second stage regression16_{, show an insignificant p-value (0.282). This indicates that the free trade} agreement dummy is not found endogenous. Since reverse causality, unhealthy commodity consumption attracting free trade agreements seems questionable, instrumenting is not found necessary in the foreign corporate presence model.

Health model: 𝐵𝑀𝐼𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡 + 𝛽5𝐴𝐺𝐸𝑖𝑡 +𝛽6𝐹𝑇𝐴𝑈𝑆̂𝑖𝑡+ 𝛽7𝐹𝐷𝐼𝑖𝑡̂∗ 𝐹𝑇𝐴𝑈𝑆𝑖𝑡+ 𝑇_𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (14) 𝑇𝐶𝑖𝑡 = 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝛽6𝐹𝑇𝐴𝑈𝑆̂𝑖𝑡+ 𝛽7𝐹𝐷𝐼𝑖𝑡∗ 𝐹𝑇𝐴𝑈𝑆̂ 𝑖𝑡+ 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (15) 𝑆𝐵𝑃𝑖𝑡= 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝛽6𝐹𝑇𝐴𝑈𝑆̂𝑖𝑡+ 𝛽7𝐹𝐷𝐼𝑖𝑡∗ 𝐹𝑇𝐴𝑈𝑆̂ 𝑖𝑡+ 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (16) 𝐷𝑖𝑎𝑏𝑒𝑡𝑒𝑠𝑖𝑡= 𝛼𝑖𝑡+ 𝛽1𝑈𝐶̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ β3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐴𝐺𝐸𝑖𝑡 + 𝛽6𝐹𝑇𝐴𝑈𝑆̂𝑖𝑡+ 𝛽7𝐹𝐷𝐼𝑖𝑡∗ 𝐹𝑇𝐴𝑈𝑆̂ 𝑖𝑡+ 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (17)

Foreign corporate presence model:

𝑈𝐶𝑖𝑡= 𝛼𝑖𝑡+ β1𝐹𝐷𝐼̂ + 𝛽𝑖𝑡 2𝐺𝐼𝑁𝐼𝑖𝑡+ 𝛽3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝐹𝑇𝐴𝑈𝑆 + 𝑇𝑡+ 𝜇𝑖+ 𝜖𝑖𝑡 (18)

13

First stage regression: 𝐹𝑇𝐴𝑈𝑆𝑖𝑡= 𝛼𝑖𝑡+ 𝛾1𝐹𝑇𝐴𝑈𝑆𝑖𝑡_1+ 𝛾2𝐹𝑇𝐴𝑈𝑆𝑖𝑡_2+ 𝛾3𝐺𝐷𝑃𝑖𝑡+ 𝛾4𝑈𝑅𝐵𝑖𝑡+ 𝛾5𝐺𝐼𝑁𝐼 + 𝛾6𝑈𝐶𝑖𝑡+ 𝜇𝑖+ 𝑣𝑖𝑡

14 _{Second stage regression: 𝐵𝑀𝐼}

𝑖𝑡= 𝛼𝑖𝑡+ 𝛽1𝐹𝑇𝐴𝑈𝑆̂𝑖𝑡+ 𝛽2𝐺𝐼𝑁𝐼𝑖𝑡+ 𝛽3𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑈𝑅𝐵𝑖𝑡+ 𝛽5𝑈𝐶 + 𝜇𝑖+ 𝜖𝑖𝑡

15 _{First stage regression: 𝐹𝑇𝐴𝑈𝑆}

𝑖𝑡= 𝛼𝑖𝑡+ 𝛾1𝐹𝑇𝐴𝑈𝑆𝑖𝑡_1+ 𝛾2𝐹𝑇𝐴𝑈𝑆𝑖𝑡_2+ 𝛾3𝐺𝐷𝑃𝑖𝑡+ 𝛾4𝑈𝑅𝐵𝑖𝑡+ 𝛾5𝐺𝐼𝑁𝐼 + 𝛾6𝐹𝐷𝐼𝑖𝑡+ 𝜇𝑖+ 𝑣𝑖𝑡

16 _{Second stage regression: 𝑈𝐶}

23 OLS regression with robust standard errors, t-statistics between parentheses and asterisks indicate statistical significance at the ***1, ***5, and *10 percent level. 39 developing countries, absolute change between 1999-2008

1_{: Adjusted to explanatory variable included}

a: Base model regression

4| Empirical results

All models are adjusted for more traditional hypothesized causes as urbanization, GDP per capita, inequality and life expectancy. Four different explanatory variables are chosen: “UC”, “UPF”, “PF” and “SD”. Where “UC” is the overall indicator and “UPF”, “PF” and “SD” are subcategories of this overall indicator. Moreover, the disease of affluence indicators chosen consist of “BMI”, “TC”, “SBP” and “Diabetes”. First, the cross-section approach is reported thereafter, the panel data approach for both the health model and the foreign corporate presence model is discussed.

4.1 Cross section approach

4.1.1 Health model

The first disease of affluence indicator discussed is BMI-ratio, the main results are to be found in table 2 The overall indicator “UC” is positive and significantly related with the absolute change in BMI-ratio over the years 1999 to 2008. By unpacking the unhealthy commodity indicator the results show that

“UPF” and “PF” are significantly related with an absolute change in BMI-ratio. “SD” shows a positive coefficient however, this effect is not found significant at conventional significance levels. This indicates that the positive and significant relation between “UC” and BMI-ratio can be clarified by a higher consumption of ultra-processed and processed foods.

1-1a _1-22 _1-3a _1-4a

∆BMIratio ∆_BMIratio ∆_BMIratio ∆_BMIratio

∆UC .0014 (2.92)*** ∆UPF .0048 (3.47)*** ∆PF .0018 (2.58)** ∆SD .0037 (1.36) ∆BMI1999 .1247 (3.27)*** .1066 (3.71)*** .1469 (3.13)*** .0911 (2.72)** ∆ ‘’UC”19991 -.0003 (-1.24) -.0001 (-0.18) -.0006 (-1.63) .0004 (0.47) ∆GDP -2.46e-13 (-3.57)*** -2.66e-13 (-3.87)*** -2.58e-13 (-3.89)*** -2.33e-13 (-3.57)*** ∆GINI -.0091 (-1.95) -.0092 (-1.98) -.0080 (-1.72) -.0079 (-1.32) ∆AGE -.0065 (-0.18) -.0154 (-0.43) .0061 (0.17) .0911 (0.30) ∆URB .0456 (3.39)*** 0.0561 (4.20)*** .0432 (3.00)*** .0497 (3.38)*** R2 _{0.7554 0.7664 0.7387 0.7185}

24

The more traditional hypothesized causes of changing health patterns as “URB” shows a positive and significant relation with BMI-ratio, this effect is visible throughout all regressions. Hence, countries with an increasing percentage of the population living in urban areas tend to experience an increase in BMI-ratio. “GINI” and “AGE” are not found in the aggregate to be associated with BMI-ratios. However, “GDP” shows an unexpected negative and significant relation with BMI-ratio in case of all four unhealthy commodity consumption variables. This indicates that “GDP” in the aggregate is associated with a declining BMI-ratio.

The second health indicator examined is total cholesterol (table D.1). “UC”, “UPF”, “PF” and “SD” all show an unexpected negative relation with total cholesterol. The control variables “GDP”, “GINI” and “AGE” are not found significantly related with TC-ratio. However, after separately including “UPF”, “PF” and “SD” in the regression as explanatory variables, “URB” also has a positive and significant sign. This suggests that “URB” in the aggregate does influence TC-ratios. Hence, countries with an increasing percentage of the population living in urban areas tend to experience an increase in cholesterol levels. As can be observed in table D.2, the regression does not give any notably results when systolic blood pressure is taken as a dependent variable, neither for the explanatory variables nor the control variables. The results of the prevalence of diabetes type 2 are to be found in table D.3. The overall indicator “UC” is positively and significantly related with the absolute change in the prevalence of diabetes type 2. After separately including the unhealthy commodity consumption indicators, the results show that “UPF” is positive and significantly related with diabetes-ratio. “PF” and “SD” show a positive coefficient however, this effect is not found significant at conventional significance levels. This indicates that the positive and significant relation between “UC” and diabetes-ratio can be clarified by a higher consumption of ultra-processed foods. The control variables do not show any significant relation with diabetes-ratio.

25

Table 3. Robustness unhealthy commodity consumption and BMI-ratio (kg/m2)

Table 4. Second robustness test unhealthy commodity consumption and BMI-ratio (kg/m2) 4.1.2. Robustness checks health model

Although the aforementioned results do give an indication of the simultaneous change of unhealthy commodity consumption and changing occurrence of diseases of affluence, it does not give any evidence for their causal relationship. Since the values of the levels of diseases of affluence indicators change slowly, it is expected that the influence of the dependent variable is only noticeable after a period of multiple years. An additional robustness test is performed whereby the absolute change in risk factor indicators over the period 1999-2003 are regressed on the absolute change in diseases of affluence indicators from 2004-2008. The control variables do not give any noteworthy variation compared to the base model therefore only the coefficients of interest are reported. As can be observed in table 3, the regression gives mixed results. Only “SD” is found associated with BMI-ratio whereas ultra-processed food consumption is found positively associated with TC-ratios. SBP-ratio also shows a significant and positive relation with “UC” and “PF”.

However, as argued by Stuckler (2009) the incidence of chronic diseases have a decadal time lag between the exposure to risk factors and the development of chronic diseases. Therefore, an additional robustness test is performed, examining whether the consumption level of processed food in 1999 influences the absolute change in diseases of affluence indicators over 1999-2008. The control variables do not show any significant change and therefore only the variables of interest are reported. The results are to be found in table 4. In line with the previous model, “SD” exhibits a positive and significant effect on BMI-ratio. Furthermore, the signs of “SBP” and “Diabetes” are also found positively associated with soft drink consumption. This confirms the view that soft drink consumption has a decadal time lag between the exposure to the risk factor and the development of the chronic disease.

∆BMIratio ∆TCratio ∆SBPratio ∆Diabetesratio

UC1999 .0004 -.0004 .0004 .0018

UPF1999 .0004 -.0008 .0065 .0002

PF1999 .0005 -.0007 -.0016 .0026

SD1999 .0034* -.0009 .0129** .0167**

∆BMIratio ∆TCratio ∆SBPratio ∆Diabetesratio

∆UC .0004 .0001 .0038* .0024

∆UPF .0007 .0008** .0070 .0066

∆PF .0002 .0002 .0063** .0025

∆SD .0046* -.0010 .0084 .0135

Cross-country regression, independent variables absolute change 1999-2004 and dependent variables absolute change 2004-2008, Model adjusted for urbanization, inequality, life expectancy, GDP per capita, diseases of affluence indicator in 1999 and unhealthy commodity consumption in 1999, 39 countries included, asterisks indicate statistical significance at the ***1, ***5, and *10 percent level,

26

Results instrumented fixed effects regression, Model adjusted for urbanization, inequality, life expectancy, GDP per capita, 39 countries included, asterisks indicate statistical significance at the ***1, ***5, and *10 percent level

1_{: instrumented by one year lag}

Table 5. Unhealthy commodity consumption and diseases of affluence indicators

4.2 Panel data approach

4.2.1 Health model

The coefficients of interest of the instrumented fixed effects model are published in table 5, the overall results can be found in table E.1 and E.2. The effect of “UC”, “UPF”, and “PF” are now found insignificantly associated with BMI-ratio. It seems that over a short period of time, the level of unhealthy commodity consumption does not matter for BMI-ratio. However, “SD” is found positive and significantly associated with BMI-ratio, in consonance with the cross-country model. “GDP” is still found negative and significantly associated, indicating that increasing BMI-ratios are not a consequence of economic development. Moreover, “AGE” is also found negative and significantly associated, this suggests that BMI-ratio has nothing to do with increasing life expectancy in developing countries.

When TC-ratio is taken as dependent variables “UC”, “UPF”, “PF” and “SD” do not show a consistent or significant relation. However, “GDP” is found positive and significantly related with TC-ratios, this indicates that economic development in the aggregate is indeed associated with higher cholesterol rates. “AGE” also shows negative and significant signs, again proving that diseases of affluence are not an inevitable consequence of ageing. In case of SBP-ratio, the explanatory variables as well as the control variables do not show any noteworthy outcomes. Referring to diabetes-ratio, “UPF” and “SD” are found positive and significantly related. Indicating that an increase in “UPF” and “SD” is associated with a higher prevalence of diabetes type 2. The control variables, “GINI” is found positive and significantly related with diabetes-ratio, this indicates that if equality increases diabetes-ratios tend to rise. The other control variables, “URB”, “GDP” and “AGE” do not show any significant association with diabetes-ratio and are therefore not considered central in explaining the hypothesized relation.

Again, a free trade agreement dummy is included to examine whether countries which have a free trade agreement with the USA are associated with a higher prevalence of diseases of affluence. An interaction variable between “UC” and the free trade agreement dummy is included to test whether free trade