Inequality, Globalization and Health - Lunds universitet

nek.lu.se

Inequality, Globalization and Health - Lunds universitet

Inequality, Globalization and Health

Therese Nilsson

Lund Economic Studies Number 157


Distributed by the Department of Economics

Lund University

P.O. Box 7082

S-220 07 Lund

Sweden

Telephone: +46 (0)46 222 00 00

Fax: +46 (0)46 222 46 13

www.nek.lu.se

ISSN 0460-0029

Printed in Sweden by

Wallin & Dahlholm Digital AB, Lund 2009

Cover design: Lund University, 1966

Copyright © Therese Nilsson, 2009


To Henrik


ACKNOWLEDGEMENTS

I remember it as if it were yesterday. It was around Christmas 1982 and I was five years

old. My family and I were preparing the last details for the annual event. Needless to say,

my expectations were high and I could not wait to get started. The kitchen in my parent’s

house was carefully being set according to the ritual, and different pieces of technical

equipment were taken out of the cupboards. It was the yearly date for the baking of

ginger snaps.

The ingredients were mixed together and the dough was ready to be processed. I

rolled out the dough and was eager to punch out the different characters when I realized

that everyone around me carefully used the roller to make an even thinner layer of

dough, enabling the production of a larger quantity of more precise characters. I took my

dough back to the counter-top for some more work. I finally got my thin characters

punched out and they covered half of the baking sheet. At this point I thought it was

baking time, but my parents told me I first had to fill my sheet, otherwise we would be

stuck in the kitchen all night. As I always been eager to see the final result, I somewhat

less enthusiastic started to process the remains of the dough.

Some twenty years later, as I started the work on my PhD-thesis, I got a déjà vu.

In many respects it felt like being back in the kitchen making ginger snaps, and reflecting

upon the past years’ experiences I realize that the evening in December 1982 was my first

encounter with what can be described as the research process: to roll out the dough and

to experience the excitement when only having an idea of what characters to make; to

roll out the dough even thinner and the distress when realizing that you should stay with

one or two characters, since too many types will be difficult to parcel up together; to

bake the ginger-snaps and the frustration when some of them burn in the oven; to put

the icing on the cake and the feeling of satisfaction and accomplishment when someone

taste the ginger snaps and show their appreciation.

There are many people that I owe gratitude for all the guidance and support in

and outside the kitchen – in the process of making and baking this thesis. I would like to

show my largest appreciation to my main supervisor, Carl Hampus Lyttkens, and to my

assistant supervisor, Andreas Bergh. These two excellent and active researchers have with

a seemingly endless enthusiasm and energy taken part in and contributed to this thesis.

Working with these scholars has been incredibly inspiring and enjoyable, partly because

of their specific characteristics and traits: Carl Hampus who combines integrity with an

open mind, the team player who always care about his fellow players, with a true

fascination about economics and ancient Greece, and with a subtle linguistic humor;

Andreas, with his ever so refreshing attitude that a good vacation sometimes can be

defined as getting some inspiring work done, with his fondness of odd food products, an

always broad attitude, an absolute pitch, and a genuine interest in social sciences and

discussions (at Club Cyrus, in Espenäs or in Copenhagen). Carl Hampus and Andreas

have a big part in me gradually growing into the role of a researcher. Without their

support and encouragement much of this thesis never would have been realized.

Motivation and advice have come from many directions during this baking

process. I would therefore also like to thank Göte Hansson and Carl-Johan Belfrage –

for supervision and guidance, and for letting me be a part of the trapca project in

Tanzania, an experience which in many respects has given me that additional perspective

on research and every day life; Sonja Opper – for telling me the truth that baking needs

v


focus and for giving me the opportunity to take part of conference arrangements and to

be a tour guide; and Alia Ahmad and Susanna Thede – for sharing the nice adventure of

teaching development economics.

In the great world of economics outside Lund I would like to thank Martin

Karlsson for being a great economist, co-author and friend. I would also like to express

my thanks to Jesper Roine who gave very valuable comments and suggestions on the

draft version of this thesis. Valuable comments on various versions of the included

chapters have also been given by Peter Lambert, Koen Decancq, Christian Bjørnskov,

Niclas Berggren, Mireia Jofre-Bonet, Björn Ekman, Susanna Thede, Nils Janlöv and

seminar participants in Lund. I would also like to gratefully acknowledge the Central

Statistical Office (CSO) in Lusaka, Zambia, for providing med with data, and all the

financial support that I have received for conferences, field research and research

meetings outside Lund in the last years from: Stiftelsen för främjande av ekonomisk forskning

vid Lunds universitet, Anna Nilssons stipendiefond, Per Westlings Minnesfond, Swedish council for

working life and social research (Health Economics Program, Lund University), Nordiska

Afrikainstitutet, Sida/SAREC, Nordic Network in Economics, British Academy, Gyllenstiernska

Krapperups stiftelsen, Siamon foundation and Willers Stipendiefond,

Baking and similar work requires inspiration, and quite often inspiration requires

perspiration. I would therefore like to thank the brave women of the floorball team

Nutida Ekonomiska Kvinnor: Elvira, Lina Maria, Lu, Maggie, Pernilla, Sofie and Åsa, and all

the participants of the departmental Monday afternoon floorball sessions, for great

spirits, for letting me wildly celebrate when I occasionally manage to put the ball in the

net and for allowing me to be a sore loser. I would also like to take the opportunity to

thank the enthusiasts at the department that have arranged running events, soccer,

badminton, squash and floorball tournaments, where everyone may take part, and the

following highly appreciated ”eftersitsar”.

Inspiration and motivation during the past years have also come from numerous

coffee breaks and fun conversations while having lunch with colleagues, from friendly

“hellos” in the corridor, and from great PhD parties. For the parties the entertaining

group of PhD candidates of 2003/2004 should be particularly accredited. Moreover,

work was often facilitated as I was surrounded by skilled colleagues always willing to help

regardless the nature of the problem: broken copy machines or grumpy computers,

incorrectly filled out travel expense reports or coffee-machines without coffee, instant

need of excerpts from ladok or unwilling stata commands.

Among all the great current and former colleagues at the department, some

require a special thanks for great friendship and for brightening up the times when hands

were tired from working the dough or when the ginger snaps tasted bitter: Maria Persson

– for always having a door open and for a great sense of irony and etiquette; Eric Rehn –

for being the confident moral booster when I was “back in the loop” and for letting me

talk without restraint; Nils Janlöv – for forthright discussions and for being a great

clubber; Erik Jonasson – for precise hilarious analyses on various subjects and behaviors,

and for having a great CD collection; Åsa Ljungvall – for refreshing the departmental

daily grind and for being a “doer”; Per Hjertstrand – for a great sense of humor and for

being generous in sharing his econometric knowledge; Wolfgang Hess – for sharing his

German beer; and Åsa Eriksson – for interesting age-discussions and for helping me feel

at home at the department.

Everybody needs someone to confide in and I am no exception to that rule.

Pernilla “Piff” Johansson, my office roommate that always finds the time for me, is an

important and invaluable piece in this dissertation puzzle. I would like to express my

vi


largest thank to you for being one of my closest friends and for your unselfishness, for all

happy laughter’s in room 265, for almost turning 27 before we met, for being brilliant

and methodical, and for you handling and solving various external problems, such as

heat, empty printers, dirty laundry and annoying red lights. This work would never have

been completed without your support.

I am also grateful to all my friends outside academia for putting things in

perspective and for accepting that I lately all the more often had to prioritize “baking or

icing” when they suggested something fun. Thank you Casper, Claudia, Malin, Morten,

Jouline and Engeli, Peter, Marita, Linnéa, Torbjörn and Astrid, Rebecka, Josahansa, Rofl,

Lias, Elena, Andreas, Josefin, Manfred, the Hanéll-Järvhammar, the Sidenvall-Jegou and

the Sällström families.

Unfortunately it may also seem as if work lately took priority over my family, but

this was just an optical illusion as my thoughts were with them very often. My family has

always been an endless source of encouragement and support and I would therefore like

to thank my mother Erna and my father Björn, my sister Jennie, Martin, my brother

Magnus, Elenor and their children Jacob and Clara, and Henrik’s family; Kerstin, Staffan

and Johan.

Some people say that cookies, bread and pastries taste better if they are baked

with love. If this is a true fact this parcel with ginger snaps will taste delicious as the most

important ingredients were love, never ending support and understanding from Henrik.

Thank you for always believing and showing interest in what I do, for literally scraping

the dough from the counter top, for everything you give me, for your optimism and for

sharing my happiness and sorrow. Thank you for giving me joy in life.

Lund, August 2009

Therese Nilsson

vii


viii


CONTENTS

1 Inequality, Globalization and Health: An Overview 1

1.1 Introduction 1

1.1.1 The inequality, globalization, and health nexus 1

1.1.2 Globalization 7

1.1.3 Inequality 10

1.2 Summaries of the studies 17

References 21

2 Health, Wealth and Wisdom:

Exploring Multidimensional Inequality in a Developing Country 25

2.1 Introduction 25

2.2 Concepts and theoretical framework 28

2.2.1 An item-by-item approach 28

2.2.2 An aggregative approach 30

2.2.3 A non-aggregative approach 33

2.3 Data and variables 38

2.4 Empirical application 40

2.4.1 An item-by-item approach 40

2.4.2 An aggregative approach 44

2.4.3 A non-aggregative approach 51

2.5 Conclusion 56

References 59

3 Income Inequality and Health:

Exploring the Association in a Developing Country 61

3.1 Introduction 61

3.2 Theory and previous empirical findings 65

3.2.1 Absolute income, relative income, income inequality and health 65

3.2.2 What does the empirical evidence tell us? 68

3.3 Data and empirical model 70

3.3.1 Data 70

3.3.2 Variables 71

3.3.3 Empirical model and estimation methods 75

3.4 Empirical analysis 77

3.4.1 Descriptive analysis 77

3.4.2 Regression analysis 80

3.4.3 Sensitivity analysis 88

3.5 Discussion and concluding remarks 93

Appendix 96

References 98

ix


4 Do Liberalization and Globalization Increase Income Inequality? 103

4.1 Introduction 103

4.2 Related literature 105

4.2.1 The different dimensions of economic freedom and globalization 105

4.2.2 Empirical examinations using the EFI and KOF indices 112

4.3 Data and empirical specifications 113

4.3.1 Dependent variables - On the use and misuse of inequality data 114

4.3.2 Independent variables 117

4.3.3 Empirical strategy 119

4.4 Empirical analysis 122

4.4.1 Basic results 122

4.4.2 Sensitivity analysis 125

4.4.3 Distinguishing between development levels 133

4.5 Conclusion 136

Appendix 138

References 144

5 Good for Living?

On the Relationship between Globalization and Life expectancy 149

5.1 Introduction 149

5.2 Background 151

5.2.1 Disentangling the effects of globalization and health 151

5.2.2 Related research 155

5.3 Methods and data 157

5.3.1 Methods 157

5.3.2 Data 158

5.4 Results 161

5.4.1 Baseline estimations 161

5.4.2 Sensitivity analysis 166

5.4.3 Distinguishing between levels of development 171

5.5 Conclusion 175

Appendix 177

References 181

x


Chapter 1

Inequality, Globalization, and Health: An Overview

1.1 INTRODUCTION

This thesis contains four studies dealing with various aspects of the nexus of

inequality, globalization, and health, building our knowledge of some of the empirical

relationships connecting these three concepts. The concept of inequality is central to

the thesis. Regarding this focus, the thesis contributes to our understanding of the

measurement, causes, and consequences of inequality.

This introductory chapter first presents a general discussion of the inequality,

globalization, and health nexus, emphasizing existing linkages and shedding light on

some of the less explored areas in the research field. After this overview, two sections

first discuss globalization and inequality in greater detail, and then define the scope of

the thesis in terms of these concepts. The introduction concludes by summarizing the

four following chapters.

1.1.1 The inequality, globalization, and health nexus

Economic development and welfare are the primary objectives of most of the world’s

nations. Although generally proxied by income, welfare has several obvious

dimensions, such as health, education, power, and assets. This multidimensionality in

one sense explains why the intuitive understanding of development differs between

people. In short, however, many would insist that a minimal requirement for a high


Chapter 1

level of development is that the physical quality of life is good. Moreover, given that

we do not only care about average welfare, it is important that a good quality of life not

be restricted to an affluent minority. Consequently, without its necessarily being a

goal in itself, equal distribution of welfare may be of interest, not least since it can be

a prerequisite for other advances.

To improve welfare, governments create and implement various policies.

Ignoring the details on how the policy process has evolved over the past three

decades, there has clearly been increasing policy support for economic liberalization

in many countries (Stiglitz 2002). As a result, market-oriented policy reforms

concerning, for example, market deregulation, lowering institutional trade barriers,

and protecting property rights, have been instituted. The gradual removal of political

barriers and the market orientation of economic policies are generally viewed as main

drivers of the closer integration of countries and people commonly referred to as

globalization (Dollar 2005). 1

Synthesizing the globalization process there is accordingly on one hand policies

and institutions, such as standards and regulations at the national and international

levels, which support and maintain integration. On the other hand, there are also

factors such as trade in goods and services, investment, and flows of people,

information, and knowledge that bring different societies closer together (ILO 2003).

Hence, the process of increasing globalization consists of economic, social, and

political events and is multidimensional by nature.

Adopting a purely income perspective, there is a vast economic literature on

the consequences of closer integration. Most theoretical arguments stress that

economic liberalization and globalization are positively related to economic growth

(see, e.g., Krugman 1987, Bhagwati and Srinivasan 2002 on trade, De Soto 2000 on

secure property rights). 2 Recent empirical results have confirmed theory in many

respects, leading to the prevalent belief that various aspects of globalization are linked

1 Globalization may also affect economic liberalization. Bjørnskov (2006), for example, discusses how

growth and wealth generated by globalization help maintain institutional quality. Moreover, foreign firms

can apply pressure for liberalization, while the spread of information provides voters with insights into

conditions in other countries, in turn potentially changing voting behavior.

2 Arguments that there could be a negative relationship between free trade and growth suggest, for example,

that trade might reduce growth if countries do not specialize in R&D or that countries that can act as price

makers on a global market may have higher growth rates (Berggren and Jordahl, 2005).

2


3

Inequality, Globalization, and Health: An Overview

to wealth creation (Berggren and Jordahl 2005, Doucouliagos and Ulubasoglu 2006,

Dreher 2006). Since globalization is reshaping several aspects of contemporary life, it

is also worth asking what impact globalization has had on other dimensions of

welfare, a matter that is less considered in economic research. This is the motivation

for examining globalization and population health in Chapter 5.

Beyond its intrinsic value, good health is at the crossroads of many issues

linking the determinants of individual well-being and good physical quality of life,

closely connecting this dimension of welfare to economic outcomes and

development. An intuitive understanding is that unhealthy individuals are less

productive. Empirical evidence indicates that better health is associated with

improved labor market outcomes, particularly in low-income settings (Strauss and

Thomas 1998), and that much of Africa’s growth shortfall versus other developing

regions is explained by a higher disease burden (Bloom and Sachs 1998). 3 Moreover,

mortality stemming from poor health may reduce human capital investment, as

agents potentially have shorter time horizons. In addition, poor health can reduce

human capital investment if children are sick or have less energy to attend school. 4

At the same time, income likely affects health outcomes (Pritchett and

Summers 1996, Smith 1999). 5 Higher individual incomes mean people can invest

more in their health, and higher societal incomes imply higher tax revenues that may

finance public health care. Studying the cross-country association between average

per capita income and life expectancy, Preston (1975) finds a nonlinear relationship

between GDP per capita and population health. Figure 1.1 presents a scatter plot of

life expectancy versus GDP per capita using data on various countries in 2000. While

income is positively related to aggregate health in poorer nations, the relationship

diminishes in high-income settings. 6 Similar results are also found in single countries

3 Acemoglu and Johnson (2007) instead emphasize the indirect effects of good health in determining

institutional choices, which is in turn a first-order determinant of economic development.

4 Research increasingly emphasizes child health as a major factor influencing future economic outcomes

(cf. Currie, 2009). Nutrition seems to be a factor of particular importance; for example, Maluccio et al.

(2006) find that poor nutrition harms cognitive development.

5 Smith (1999) discusses the direction of causality and summarizes evidence for the two-directional

relationship between income and health.

6 However, Deaton (2004) notes that income growth explains little of the variance in improvements in life

expectancy, and states that this is a weak mechanism by which globalization might improve population

health.


Chapter 1

(see e.g. Wilkinson 1992). Higher income elasticity with respect to health among the

poor implies that a progressive redistribution of income, besides reducing income

inequality, will improve population health.

Figure 1.1 The Millennium Preston curve

Life expectancy at birth

35 45 55 65 75 85

0 10000 20000

GDP per capita (PPP adjusted)

30000 40000

Figure created using data from the World Bank (2008) and Heston et al. (2006).

Less income inequality might also affect other features of an economy. A

growing body of microeconomic research finds evidence of economic inefficiency under

conditions of extreme income disparity (cf. Glaeser et al. 2003). 7 An explanatory

factor is that income inequality reduces the opportunities for people to make use of

their full capabilities. For example, at any given average population income, the

higher the income inequality, the smaller the fraction of the population qualifying for

loans or credits. This means that capital might be invested at relatively low marginal

returns and society remains shy of the Pareto frontier.

Extreme income disparities can also undermine social stability. For example,

Fajnzylber et al. (2002) report a robust relationship between income inequality and

violent crime in a sample of rich and poor countries. Moreover, the level of income

inequality might affect the degree of social interaction between people, which is itself

7 Part of this literature also examines the economic inefficiencies deriving from extreme inequalities of

opportunity (see, e.g., Ferreira and Walton, 2006).

4


5

Inequality, Globalization, and Health: An Overview

related to trustworthiness. Several empirical studies establish a negative correlation

between income inequality and the extent to which people trust each other (Knack

and Keefer 1997, Uslaner 2003, Gustavsson and Jordahl 2008). One strand of the

economic literature further suggests that resource distribution affects politics (Alesina

and Rodrik 1994, Persson and Tabellini 1994, Bénabou 2000), since income or wealth

distribution affects political power and hence the economic bargaining position of

interest groups. As a result, policy decisions are not necessarily optimal from a social

point of view.

Alongside the previously discussed traditional view that less income inequality

has an indirect impact on health, as health improves at a decreasing rate with income,

there is ongoing interdisciplinary debate as to whether income inequality has a direct

effect on health outcomes (cf. Wilkinson 1992, Gravelle 1998, Mellor and Milyo

2002, Karlsson et al. 2008). The basic idea emerges from the frequently confirmed

empirical finding of a negative correlation between measures of income inequality

and population health. However, with diminishing returns to income in the

production of individual health, the observed relationship between societal income

inequality and aggregate health measures will, to some extent, be spurious and not

causal. The question is whether the negative relationship remains when testing the

hypothesis using individual-level data. Moreover, although the literature on income

inequality and health is extensive, we possess limited knowledge of the relationship in

a developing context and in the case of extreme income inequality. This is the

rationale for the analysis in Chapter 3, which evaluates the association between income

inequality and health in Zambia.

In the case of a negative relationship between income inequality and

individual health, the relationship is presumably mediated through the economic and

social features arising from economic disparities. Consequently, investment

inefficiency, social instability, and non-optimal policy decisions could all negatively

affect future economic outcomes, while also lowering the average level of welfare in

other dimensions.

Discussion of the potential effects of high societal income inequality naturally

raises the question of what determines wide economic disparities. In recent decades,


Chapter 1

many countries around the world have experienced widening gaps between rich and

poor (Ferreira and Ravallion 2008). Temporally, the trend is concurrent with the

increasing support of economic liberalization and higher globalization levels.

Much research focuses on the consequences of liberalization and globalization

for the distribution of income (cf. Bourguignon and Morrison 1990, Edwards 1997,

Berggren 1999, Carter 2007, Dreher and Gaston 2008). Theoretical predictions are,

however, ambiguous. On the one hand, liberalization that removes legal barriers

protecting affluent groups and provide access to formal property rights create

economic opportunities that are assumed to benefit the less privileged in society (De

Soto 2000). Moreover, under the assumptions of standard trade theories such as the

Hecher–Ohlin model, greater openness will generally reduce income inequality in less

developed countries while widening the gap between the rich and poor in developed

economies. On the other hand, more recent trade models often reveal that the

relationship is more complex, many models featuring multiple equilibriums at certain

openness levels (Krugman and Venables 1995, Das 2005).

Most cross-national empirical research into globalization focuses on foreign

trade and, to some extent, on investment as the primary indicators of closer

integration (Babones 2007). However, overall economic liberalization programs

generally consist of entire reform packages affecting various policy arenas. Hence, it

is reasonable to measure the extent of policy change using a range of indicators.

Moreover, different types of liberalization might have different distributional

consequences. Using the same argument, it is reasonable to examine how the

economic, social, and political events in the globalization process affect income

inequality. Currently, our knowledge of these empirical relationships is limited. This is

the motivation for examining the relationship between globalization, liberalization,

and income inequality in Chapter 4.

Concluding this introductory discussion, it is clearly far from simple to

comprehend the multifaceted interconnections between liberalization, globalization

and inequality. The same complexity suggests that new tools are required to capture

the multidimensionality of changes in individual living conditions due to the

6


7

Inequality, Globalization, and Health: An Overview

globalization process. 8 As stressed by several scholars, economic disparity does not

arise from the distribution of income alone, and analyzing different individual

attributes is crucial to understanding and evaluating inequality among people (cf. Sen

1997, Kolm 1977, Maasoumi 1986). The question of how to measure

multidimensional inequality is the focus of Chapter 2.

1.1.2 Globalization

Although no generally accepted definition exists, globalization typically refers to the

increasing integration of societies and economies. 9 This closer integration concerns,

on one hand, greater openness between countries that speeds transactions, and, on

the other, the development of relationships between individuals at a distance.

Globalization accordingly refers to both the temporal and spatial compression of

interactions. Put differently, globalization implies declining costs of distance and time.

Interconnectedness between economies is not a new phenomenon. Nor is the

level of today’s internationalization unique in history, as the 1870–1914 period was in

some respects characterized by even greater openness. 10 Globalization is thus merely

a new name for a long-standing phenomenon, and the current state falls short of the

textbook benchmark of full integration (Arribas et al. 2009). However, the latest

phase of globalization, taking off in 1945 and accelerating in the 1980s, has some

novel features.

First, the current pace of integration is unprecedented. Second, unlike in

earlier phases of integration, a significant proportion of current capital flows is shortterm.

Third, flows of technology and ideas across borders are larger today than a

century ago, and foreign direct investment more common (Bourguignon et al. 2002).

On top of political support for liberalization, lower transportation and

8 The need to capture the multidimensionality of inequality also relates to the fact that people differ from

each other. As Sen (1997) discusses, the characteristics of inequality in different welfare dimensions tend to

diverge from each other due to the heterogeneity of people.

9 Easterly (2008) lists ten definitions of globalization and discusses how actors in the debate regularly allude

to different aspects of the integration process. While some groups pay particular attention to the power

structures that characterize internationalization, others focus on the liberalization of trade, capital, and

migration flows and the effects thereof. As a result, public debate is often marked by confusion

(Bourguignon et al., 2002).

10 For example, net capital flows are smaller and trade flows in relation to GDP are no higher than they

were before 1914. Migration flows have also been significantly larger at other times in history.


Chapter 1

communication costs and technological change are the main drivers of increasing

integration (Dollar 2005). Increasing political support further relates to the

implementation of the Washington Consensus with its three pillars (i.e., market

liberalization, privatization, and fiscal austerity), which were IMF and World Bank

conditions imposed on many developing countries as an integral part of structural

adjustment plans. Consequently, a fourth characteristic of the current era of

globalization is the increasing integration of developing economies into the world

economy.

An additional distinctive feature is the intensity of the debate concerning the

benefits versus adverse impacts of globalization. In addition to the substantial

empirical literature examining the consequences of increasing integration for

economic growth and income inequality, studies analyze the effect of globalization on

poverty, social expenditure composition, and government size (Nissanke and

Thorbecke 2006, Ravallion 2006, Dreher 2006). As mentioned, however, this

literature is limited in scope as it principally views globalization as a purely economic

phenomenon.

In an attempt to capture how integration is advancing in many directions,

several indices considering different dimensions of liberalization and globalization

have recently been created (Dreher 2006, Kearney 2003, Lockwood and Redoano

2005, Martens and Zywietz 2006). These indices all use statistical criteria to aggregate

partial indicators. Accordingly, they share the general limitations of all aggregation

procedures, namely, information loss (Duclos et al. 2001) and the somewhat arbitrary

weighting of components. 11 Furthermore, Ashby and Sobel (2008) and Heckelman

and Stroup (2005) claim that building on excessively aggregated measures can result

in misspecification bias, since components might suffer from endogeneity problems.

Consequently, beyond the existence of different hypotheses as to how different

dimensions of globalization differently affect welfare and its distribution, there are

11 A relevant reference concerning the pros and cons of using indices is the discussion of the Human

Development Index (HDI), starting in 1990. Besides pointing to the arbitrariness of its weighting structure,

criticism of the HDI emphasizes, for example, that the index permits substitution while neglecting

important dimensions of welfare (see, e.g., Ranis, Stewart, and Samman, 2005).

8


9

Inequality, Globalization, and Health: An Overview

methodological arguments in favor of performing an analysis using the separate

components of an index.

To proxy globalization and liberalization, this thesis uses the KOF Index of

Globalization (Dreher 2006) and the Economic Freedom of the World Index (EFI)

(Gwartney and Lawson 2008). The former index lets us distinguish between

economic, social, and political globalization. Beyond the possibility of separating the

effects of different events, the measure has the advantage over other indices that

comparable figures are available covering a long period, 1970–2009, and many, 122,

countries. 12 Both time and country coverage are relevant, since we assume that the

effects of globalization on welfare level and distribution could appear with a time lag

and since we are interested in examining the relationships in both developed and

developing countries. The EFI captures five types of liberalization: size of

government; legal structure and property rights; access to sound money; freedom to

trade internationally; and regulation of credit, labor, and business. This measure is

available for 141 countries over the 1970–2009 period.

Figure 1.2 displays country-specific changes in the EFI from 1980 to 2005. As

assumed, liberalization has on average clearly increased worldwide in recent decades,

and less than 7% of the countries in the sample experienced a decrease in this

measure. However, there is great variation in change of the EFI across countries.

Moreover, as figure 1.3 shows, there is great variation in terms of levels and changes

of EFI dimensions across countries over time. Notably, this is also true within

countries. 13

12 For example, the Foreign Policy index (Kearney, 2003) covers only 72 countries, while the CSGR index

(Lockwood and Redoano, 2005) covers 106 countries but only from 1982 through 2004.

13 For more information on the development of the dimensions of economic freedom and globalization in

various countries and regions over time, see Gwartney and Lawson (2008) and Dreher (2006).


Chapter 1

Figure 1.2 Percentage change in aggregate EFI, 1980–2005

Percentage change

-50 0 50 100 150

Percentage change in aggregate EFI in 141 countries, 1980–2005.

Calculations made using Gwartney and Lawson (2008)

1.1.3 Inequality

Inequality refers to the distribution of welfare among groups or individuals. Because

welfare includes several dimensions, inequality is multidimensional. As section 1.1

demonstrates, inequality is of interest for functional reasons, as disparities can affect

how an economy works. Furthermore, inequality might also be a concern if equality

per se is regarded as a fundamental value. The intrinsic value of equality relates to the

subject of fairness, and an extensive literature treats the normative issues associated

with distribution. The elementary distributional issue of relevance here relates to the

“cake division problem”: the allocation of a fixed resource among individuals when

allocation is without effect on the total to be allocated (Atkinson and Bourguignon

2000). This problem articulates the normative background for the measurement of

inequality.

Different philosophical perspectives on social justice provide different

foundations for welfare economics and different answers to the question of what type

of equality promotes societal equality with respect to the above distributional

problem (see, e.g., Rawls 1971, Dworkin 1981, Sen 1997, Roemer 1996, Sugden

2004). Should the cake be divided in equal shares to achieve equality of resources, or

10


11

Inequality, Globalization, and Health: An Overview

should equalization focus on making people equal in what is fundamentally valued by

each of them? Should the cake be divided according to the circumstances of a person

that cannot be affected by him- or herself, or should we consider individual effort as

well? In other words, what is the space in which one should measure, understand, and

influence distribution? For example, Roemer (1996) claims that differences in welfare

are acceptable when they stem from characteristics for which individuals can be

deemed responsible. Likewise, Dworkin (1981) argues that society should not aim to

equalize differences resulting from dissimilarities in tastes or preferences.

One particular problem in equalizing welfare relates to the heterogeneity of

individuals. A monetary metric to measure inequality is satisfactory if it can capture

relevant heterogeneity of households or individuals and their different situations. As

stated by Kolm (1977), however, there is no market for variation in characteristics

such as handicaps and talents, which implies that it is impossible to estimate an

indirect utility function allowing the aggregation of these characteristics with income

without adding more assumptions. Moreover, it is unclear what we can do when

adjusting for heterogeneity leads to opposite conclusions when comparing, for

example, income distributions.

Atkinson and Bourguignon (1982, 1987) provide an extension to cases when

there is agreement on ranking households by increasing needs. Needs might refer to,

for example, family size, age, education, or health condition. By extending Kolm’s

(1977) pioneering formal analysis, the framework including needs takes the

measurement literature further in the direction of multidimensional inequality. 14 The

development is also consistent with the work of Sen (1997), emphasizing the need for

the distribution of income to compensate for differences in absolute levels of wellbeing

across heterogeneous individuals. However, the question “Equality of what?”

remains, since it is not evident what needs or welfare dimensions are the most

relevant when measuring inequality in several dimensions.

14 Massoumi (1986), Tsui (1995), and Bourguignon (1999) also head in the direction of multidimensional

inequality by imposing an aggregator function, allowing the problem to be reduced to a single dimension.


Chapter 1

Figure 1.3 Dimensions of EFI, 1975–2005, in various countries

2 4 6 8 10

2 4 6 8 10

2 4 6 8 10

Argentina Nigeria

1975 1985 1995 2005

South Africa Spain

1975 1985 1995 2005

12

2 4 6 8 10

2 4 6 8 10

1975 1985 1995 2005

1975 1985 1995 2005

Tanzania The USA

1975 1985 1995 2005

Calculations made using data from Gwartney and Lawson (2008).

2 4 6 8 10

1975 1985 1995 2005


13

Inequality, Globalization, and Health: An Overview

In measuring inequality, it is also necessary to define our interest in the question

“Between whom?” As stated above, inequality refers to the distribution among

groups or individuals. Consequently, inequality can refer to the distribution of welfare

both among households and among the individuals constituting a particular

household, i.e., intra-household inequality. Moreover, we can measure inequality

using either the full distribution of households/individuals or only the lower or upper

parts of it. Economists further examine between-country and within-country

inequality. In recent years, several studies have also sought to quantify and examine

the dynamics of global inequality by aggregating the disparities among world citizens

into a single inequality measure, which in turn comprises two components, withincountry

and between-country inequality (Bourguignon and Morrisson 2002,

Milanovic 2005).

An increasing amount of data on living conditions in developed and

developing countries has become available in recent decades. The regular

implementation of household budget surveys has contributed, on an ongoing basis, to

increasing this stock of information (Ferreira and Ravallion 2008). Given the practical

problems of observing, identifying, and collecting data on relevant needs or

opportunities, applied economists generally measure monetary indicators and

indirectly address differences in needs using equivalence scales (Deaton 1997).

Regarding potential types of monetary indicators, consumption expenditures

are generally preferred to incomes, at least in a less developed context where people

often rely on seasonal employment. In addition, measurement error is generally larger

in the case of incomes than consumption, and the types of measurement errors differ.

As a result, income inequality generally turns out to be greater than consumption

inequality in a given distribution. Consequently, it is problematic to compare

monetary inequality across or within countries over time when the unit base differs.

Chapter 4 discusses recent developments aiming at overcoming some of the problems

of non-comparable monetary inequality measures across countries, but emphasizes

that measurement errors originating from raw data will remain.

Taking a global perspective on income inequality from the 1950s and onward,

Ferreira and Ravallion (2008) emphasize two emerging tendencies. First, as figure 1.4


Chapter 1

shows, between-country inequality continued to grow, but at a slower pace than

before 1950 when between-country inequality was the main driver of global

inequality. Second, within-country inequality, which decreased between the two world

wars, started to increase in the last decades of the twentieth century. Figure 1.4

moreover indicates the persistence of within-country inequality. Economists usually

explain the slow changes in income distribution by citing the fact that inequalities

reproduce across generations. In other words, inequality begets inequality.

Figure 1.4 Global inequality of individual incomes, 1830–1992

Mean logarithmic devation

0 .2 .4 .6 .8

1820 1840 1860 1880 1900 1920 1940 1960 1980 2000

Global inequality Inequality within countries

Inequality between countries

Figure created using data on 15 single countries and 18 country groups from Bourguignon and

Morrisson (2002).

Information on trends in the within-country inequality of other welfare dimensions is

limited, in particular regarding the longer-term evolution. 15 Chapter 2 of this thesis,

however, discusses how the evolution of inequality might differ between dimensions.

This could reflect that the underlying factors determining inequality in one dimension

differ from those contributing to inequality in others. Moreover, Domeij and Klein

(2002) suggest that public systems offering universal benefits can affect the savings of

low- and high-income earners differently, which in turn might increase wealth

inequality while concurrently reducing income inequality. Another explanation relates

15 One exception is inequality in education. Thomas et al. (2003) have constructed a database including the

education Gini index for 140 countries from 1960 to 2000. In most of these countries, education inequality

declined over the four decades.

14


15

Inequality, Globalization, and Health: An Overview

to time. Measuring inequality in several dimensions over a shorter period naturally

provides a snapshot and does not necessarily capture the inherently dynamic

character of inequality. That inequality begets inequality also implies that decreasing

inequality in one welfare dimension may, over time, reduce inequality in another.

Figure 1.5 Income inequality and development

Net income Gini coefficient

15 25 35 45 55 65

0 10000 20000

GDP per capita (PPP adjusted)

30000 40000

Figure created using data on income inequality from Solt (2008) and Heston et al. (2006).

Figure 1.5 illustrates the existence of a negative correlation between GDP per capita

and within-country inequality, using data on numerous countries in the world from

approximately the year 2000. Moreover, low-income countries display greater

variance, while the Gini coefficients for countries with a GDP per capita above USD

20,000 lie in the relatively narrow 0.20–0.45 interval. This is in line with Ferreira and

Ravallion’s (2008) finding that no country has developed beyond middle-class status

while retaining extreme levels of income inequality.

Economists have long sought some general rule about how development and

income distribution dynamics are related. Kuznets (1955), drawing on Lewis (1954),

assumes the existence of two sectors, traditional and modern. While population is

concentrated in the traditional sector, income is concentrated in the modern sector,

where income per capita is consequently higher. When the modern sector demands

more labor, people move into it from the traditional sector. This increases the income


Chapter 1

per capita of the whole economy while initially increasing overall income inequality.

However, according to the model, income inequality later decreases as people

continue to move from the traditional sector and are absorbed in the workforce of

the modern sector.

Following Kuznets (1955), the level of income inequality was long assumed to

follow an inverted-U relationship with economic development as a country became

more economically developed. As more data have become available, however, it has

been found that this relationship does not generally hold, as the correlation in figure

1.5 also indicates. More recently, economists have developed models with multiple

equilibriums, each characterized by its own income distribution. These indicate that

different combinations of initial conditions, and of the historical processes that

proceed from them, can generate diverse outcomes (see, e.g., Bénabou 2000).

Accordingly, there is no standard theoretical or empirical model of income inequality

and nothing resembling standard control variables for use in econometric analysis

(Carter 2007).

In this thesis, all welfare attributes are taken to correspond to outcomes rather

than opportunities. Consequently, the analyses measure the inequality of where

households or countries end up rather than where and how they begin. The two

studies in the thesis that use data from a single country measure inequality at the

household level. As a result, the calculations do not take account of within-household

inequalities. Concerning the concept of inequality applied in the analyses in the thesis

that use macrodata, the focus is purely on within-country income inequality.

Although global and between-country inequality are equally important from the

normative and functional perspectives, within-country income inequality may be

considered a possible target for national economic policies and to have more

relevance to the survival ability of a globally integrated world.

16


1.2 SUMMARIES OF THE STUDIES

Chapter 2

17

Inequality, Globalization, and Health: An Overview

Health, Wealth, and Wisdom: Exploring Multidimensional Inequality in a Developing Country 16

Chapter 2 relates to the measurement of inequality, given that welfare and its

distribution is multidimensional. In recent decades, economists and policy makers

have increasingly favored considering other dimensions in addition to income when

examining welfare. However, although arguments favor evaluating the social state of

an individual or household using more than one criterion at a time, and although

there is an extensive theoretical literature on multidimensional inequality, the standard

empirical approach to measuring inequality remains that of comparing

unidimensional welfare indicators.

Using data on the expenditures, education levels, health status, and land

holdings of Zambian households in 1998 and 2004, this study explores three

approaches to evaluating inequality changes where there is more than one variable.

Given the lack of consensus on how to measure inequality in several dimensions, this

study assesses the strengths and weaknesses of the applied approaches in an empirical

context. Moreover, it complements the growing theoretical literature on

multidimensional inequality by providing an empirical examination of a specific case.

Applying an item-by-item approach, where changes in the inequality of different

welfare attributes are examined separately, the study finds increasing disparity in three

of four dimensions over time. Correlations between the indicators are, however,

weak. Applying an aggregative approach using the Maasoumi index, which aggregates

various dimensions of welfare into a weighted inequality measure, the study illustrates

how a unidimensional perspective can be misleading, as different dimensions of

household welfare may compensate for each other. We confirm this finding by

examining multidimensional inequality using a non-aggregative approach focusing on

orderings. Using sequential stochastic dominance conditions, it is clear that combined

arrangements of welfare distributions, which independently point toward increased

dispersion, do not always identify a rise in multidimensional inequality.

16 Forthcoming 2009 in Social Indicators Research


Chapter 1

While the first two approaches are useful in empirical research, as they always

reach a conclusion as to the direction of change in inequality, the Massoumi index is

overly sensitive to the degree of substitution. Robustness checks and explicitness

concerning assumptions made are thus crucial when using this approach. Avoiding

the computational complexity of aggregated measures and allowing a less demanding

structure using stochastic dominance, however, come at a cost, as in most cases we

cannot determine whether inequality is lower or higher in one period than in another.

Chapter 3

Income Inequality and Health: Exploring the Association in a Developing Country

Chapter 3 builds on the ongoing interdisciplinary debate as to whether living in an

economically unequal society is harmful to one’s health (cf. Wilkinson 1992, 1996,

Gravelle 1998, Mellor and Milyo 2002, Karlsson et al. 2008). The empirical evidence

in the literature exploring the relationship between income inequality and health using

individual-level data is very mixed. Several studies in a US context find a negative

association; while studies of other developed countries largely reject the so-called

income inequality hypothesis. Yet, we possess very limited knowledge of the

relationship between income inequalities and health in less-developed economies.

Using data from the Zambian Living Conditions Measurement Survey, 2004,

on the anthropometric indicator height-for-age, this chapter examines the

relationship between income, income inequality, and child health in a developing

country by testing three hypotheses: the absolute income hypothesis (AIH), that income

determines individual health and that the positive effect diminishes with higher

income; the relative income hypothesis (RIH), that individual income in relation to average

reference group incomes influences individual health status; and the income inequality

hypothesis (IIH), that income inequality affects individual health, independent from the

effect of income.

Taking account of complex survey design and employing various econometric

techniques we confirm the frequent finding in the literature that economic resources

are related to better health in a concave fashion. However, while we find some weak

evidence supporting the RIH when the reference group corresponds to the local

18


19

Inequality, Globalization, and Health: An Overview

geographical area, the association between average expenditures measured at a

provincial level and child health is positive and significant. Moreover, in contrast to

the IIH and most of the existing literature the empirical analysis shows that there is

robust and significant positive association between income inequality in a previous

time period and child health. The results suggest that the relationship between

income inequality and health in developing contexts might be very different from the

predominant view in the literature. While the findings merits further empirical

research the theoretical aspects on the subject also needs to be revisited as proposed

pathways from inequality to health primarily have been formulated to account for

adversity in health outcomes.

Chapter 4

Do Liberalization and Globalization Increase Income Inequality?

Co-authored with Andreas Bergh

Chapter 4 takes as its starting points (i) the fact that the empirical literature on the

relationship between liberalization, globalization and income inequality has achieved

contradictory empirical results; and (ii) the limited knowledge of how different

dimensions of liberalization and globalization affect income distribution. Using the

KOF and EFI indices we analyze the relationship between dimensions of

liberalization, globalization and income inequality in a panel setting. The chapter

follows, on the one hand, the empirical work of Berggren (1999), Scully (2002), and

Carter (2007) and, on the other, that of Dreher and Gaston (2008). As our main

indicator of inequality, we use the net income Gini coefficient from Solt (2008), who

provides an improved alternative to the commonly used databases containing

inequality measures for cross-country analysis.

Analyzing a sample of 81 developed and developing countries from 1970

through 2005, using different estimation techniques and robustness checks, including

an attempt to control for potential endogeneity, we find clear evidence that policy

reforms promoting trade openness have on average increased within-country income

inequality. Distinguishing between levels of development this distributional impact

appears significant at higher income levels but not in low-income contexts. In most


Chapter 1

cases we also find that social globalization increases within-country income inequality,

suggesting that the globalization process is multifaceted. Lowering our standards

slightly, there is moreover some evidence that credit and labor-market deregulation

might contribute to widening income gaps. This said, it bears emphasizing that most

types of liberalization and globalization studied have no significant impact on income

inequality. Perhaps most interestingly, legal structure and more secure property rights,

has no effect on inequality although often found to be a robust component when if

comes to explaining economic growth.

Chapter 5

Good for Living? On the Relationship between Globalization and Life expectancy

Co-authored with Andreas Bergh

Chapter 5 is motivated by our limited knowledge of the relationship between

globalization and non-income welfare indicators. Using the KOF index we test the

association between the three dimensions of globalization (i.e., economic, social, and

political) and life expectancy at birth using a panel of 92 developed and developing

countries from 1970 through 2005.

Using various estimation techniques, we find that there is a very robust positive

association between economic globalization and life expectancy, even when

controlling for income, education, nutritional intake, literacy, and number of

physicians—four factors repeatedly found to be important for life expectancy in

earlier research. This suggests that some of the effect of globalization on life

expectancy occurs through other mechanisms that might be hard to measure, such as

knowledge transfer or changes in relative prices. The positive impact of economic

globalization on life expectancy also appears when the sample is restricted to lowincome

countries only.

As for the relationships between social and political globalization and life

expectancy, there is a tendency toward a positive relationship for social globalization

and a negative one for political globalization, but these relationships are not robust to

the various sensitivity tests performed. In particular, the effect of political

globalization seems to be very dependent on country-specific circumstances.

20


21

Inequality, Globalization, and Health: An Overview

REFERENCES

Acemoglu, D. and S. Johnson (2007). “Disease and Development: The Effect of Life

Expectancy on Economic Growth.” The Journal of Political Economy, 115 (6), 925

Alesina, A and D. Rodrik (1994). “Distributive Politics and Economic Growth.” Quarterly

Journal of Economics, 109 (2):465–490.

Arribas, I., Pérez, F. and E. Tortosa-Ausina (2009). “Measuring Globalization of

International Trade: Theory and Evidence.” World Development 37: 127–145.

Ashby, N J and R S. Sobel (2008). "Income inequality and economic freedom in the U.S.

states." Public Choice 134:329–346.

Atkinson, A. and Bourguignon, F. (1982). The Comparison of Multi-Dimensioned

Distributions of Economic Status. Review of Economic Studies 49 (2): 183–201.

Atkinson, A. and Bourguignon, F. (1987). Income Distribution and Differences in Needs.

In G. R. Feiwel (ed.), Arrow and the Foundation of the Theory of Economic

Policy. Macmillan, London.

Atkinson, A. and Bourguignon, F. (2000). Income Distribution and Economics. In A. B.

Atkinson and F. Bourguignon (eds.), Handbook of Income Distribution, North-Holland.

Babones, S J. (2007) “Studying Globalization: Methodological Issues,” pp. 144–61 in

George Ritzer (ed.) The Blackwell Companion to Globalization. Oxford: Blackwell.

Bhagwati J and TN Srinivasan (2002). “Trade and poverty in the poor countries” The

American Economic Review, 92(2):180–183

Bénabou, R. (2000). “Unequal Societies: Income Distribution and the Social Contract.”

American Economic Review, 90 (1):96–129.

Berggren, N. (1999). “Economic freedom and equality: Friends or foes?” Public Choice

100:203–223.

Berggren, N and H. Jordahl (2005). “Does free trade really reduce growth? Further testing

using the economic freedom index” Public Choice, 122:99–114.

Bjørnskov, C. (2006). Globalization and Economic Freedom: New Evidence Using the

Dreher Index, Working paper Series of Aarhus School of Business'

Bloom, D.E, and J. D. Sachs (1998). “Geography, Demography and Economic Growth in

Africa.” Brookings Papers on economic Activity, 2, 207–295

Bourguignon, F. (1999). “Comment to: Multidimensional Approaches to Welfare Analysis

by E. Maasoumi” in J. Silber (ed.), Handbook on Income Inequality Measurement, Kluwer,

MA.

Bourguignon et al. (2002). “Making sense of globalization. A guide to the economic issues.”

CEPR Policy Paper No. 8

Bourguignon F. and C. Morrisson (1990). “Income Distribution, Development and Foreign

Trade: A Cross-Sectional Analysis.” European Economic Review, 34(6):1113–1133

Bourguignon F. and C Morrisson (2002) “Inequality among world citizens: 1820–1992.”

The American Economic Review, 92(4):727–744

Carter, J R. (2007). "An empirical note on economic freedom and income inequality." Public

Choice, 130:163–177.

Currie, J. (2009). “Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in

Childhood, and Human Capital Development.” Journal of Economic Literature, 47: 87

De Soto, H. (2000). The Mystery of Capital: Why Capitalism Triumphs in the West and

Fails Everywhere Else. New York: Basic Books.

Deaton, A. (1997). The analysis of household surveys. A microeconometric approach to

development policy. John Hopkins University Press, London

Deaton, A. (2004). "Health in an Age of Globalization” NBER Working Paper No. 10669.


Chapter 1

Dollar, D. (2005). “Globalization, Poverty, and Inequality since 1980.” The World Bank

Research Observer, 20 (2), 145

Domeij, D. and P. Klein (2002). “Public Pensions: To What Extent Do They Account for

Swedish Wealth Inequality?” Review of Economic Dynamics, 5(3):503–534

Doucouliagos, C. and M.A. Ulubasoglu. (2006). "Economic freedom and economic growth:

Does specification make a difference?" European Journal of Political Economy, 22:60–

81.

Dreher, A. (2006). "Does Globalization Affect Growth? Empirical Evidence from a new

Index." Applied Economics, 38:1091–1110.

Dreher, A and N. Gaston (2008). "Has Globalization Increased Inequality?" Review of

International Economics, 16:516–536.

Duclos, J-Y., Sahn, D. and Younger, S. D. (2001). Robust Multidimensional Poverty Comparisons.

Working Paper #0115, Université Laval, Département d’Économie

Dworkin R. (1981). “What is equality? Part 2: Equality of Resources,” Philosophy and Public

Affairs,10, 283–245.

Easterly, W (2008). For the search on the word “Globalization” in The new Palgrave

dictionary of economics, Second Edition

Edwards, S (1997). “Trade policy, growth, and income distribution” The American Economic

Review, 87(2): 205–210

Fajnzylber, P, Lederman D and N. Loayza (2002). “Inequality and Violent Crime” Journal of

Law and Economics, 45(1):1–40

Ferreira F. and M. Walton (2006) “Inequality of Opportunity and Economic

Development,” World Bank Policy Research Working Paper 3816

Ferreira F. and M. Ravallion (2008) “Global poverty and inequality: A review of the

evidence,” World Bank Policy Research Working Paper 4623

Glaeser, E., Scheinkman J. and A. Shleifer (2003). “The Injustice of Inequality.” Journal of

Monetary Economics 50: 199–222

Gustavsson, M. and Jordahl, H (2008), Inequality and trust in Sweden: Some inequalities are more

harmful than others, Journal of Public Economics, Vol. 92(1–2): 348–365

Gravelle, H. (1998). "How much of the relation between population mortality and." BMJ:

British Medical Journal 316:382–386.

Gwartney, J and R. Lawson. (2008). "Economic Freedom of the World Annual Report."

The Fraser Institute, Vancouver BC.

Heckelman, J. C and M. D Stroup. 2005. "A comparison of aggregation methods for

measures of economic freedom." European Journal of Political Economy, 21:953–966

Heston, A., Summers R. and B. Aten (2006), Penn World Table Version 6.2, Center for

International Comparisons of Production, Income and Prices at the University of

Pennsylvania, September 2006.

ILO (2003) World commission on the social dimension of globalization, ILO, Geneva, Switzerland

Karlsson, M., T. Nilsson, C.H Lyttkens and G Leeson (2008), Individual good, public bad, or

societal syndrome? A cross-country study of income, inequality and health, OIA Working Paper

408, Oxford Institute of Ageing

Kearney, A.T. (2003), “Measuring Globalisation: Who’s Up, Who’s down?,” Foreign Policy

Knack, S. and Keefer, P. (1997), Does social capital have an economic payoff? A crosscountry

investigation. Quarterly Journal of Economics, Vol. 112 (4): 1252–1288.

Kolm, S.C. (1977). “Multidimensional Egalitarianism.” Quarterly Journal of Economics, 91:1–

13.

Krugman, P and A Venables, J. (1995). "Globalization and the Inequality of Nations." The

Quarterly Journal of Economics 110:857–880.

22


23

Inequality, Globalization, and Health: An Overview

Krugman, P (1987). “Is Free Trade Passé?.” The Journal of Economic Perspectives, 1(2): 131–144

Kuznets, S. (1955). “Economic growth and income inequality.” American Economic Review.

45:1–28

Lewis, WA (1954). “Economic development with unlimited supplies of labour.” Manchester

School, 22: 139–191

Lockwood, B. and M. Redoano, (2005). “The CSGR Globalisation Index: an Introductory

Guide,” Centre for the Study of Globalisation and Regionalisation Working Paper 155/04.

Maasoumi, E. (1986).”The measurement and decomposition of multidimensional

inequality.” Econometrica, 54, 991–997.

Maluccio, J A., JF. Hoddinott, JR. Behrman, R Martorell, AR. Quisumbing and D. Stein

(2006). “The Impact of Nutrition during Early Childhood on Education among

Guatemalan Adults.” University of Pennsylvania Population Studies Center

Working Paper 06–04.

Martens, P. and D. Zywietz (2006), “Rethinking Globalisation. A Modified Globalisation

Index.” Journal of International Development, 18(3):331–350.

Mellor, J M. and J, Milyo. 2002. “Income inequality and individual health: Evidence from

the Current Population Survey.” Journal of Human Resources 37:510–539

Milanovic, B. (2005). Worlds Apart: Measuring International and Global Inequality. Princeton:

Princeton University Press.

Nissanke, M and E. Thorbecke (2006). “Channels and Policy Debate in the Globalization

Inequality–Poverty Nexus.” World Development, 34:1338–1338.

Persson, T. and G. Tabellini. (1994). “Is Inequality Harmful for Growth?” American

Economic Review 84 (3):600–621

Preston, SH (1975). “The Changing Relation between Mortality and Level of Economic

Development.” Population Studies, 29(2):231–248

Pritchett, L and L. Summers (1996) “Wealthier is healthier.” The Journal of Human Resources,

31(4) 841–868

Ranis, G. Stewart, F. and E. Samman (2005). Development: Beyond the HDI. Yale

University Economic Growth Center Discussion Paper No. 916.

Ravallion, M (2006). "Looking Beyond Averages in the Trade and Poverty Debate." World

Development, 34:1374–1374.

Rawls, J (1971). A Theory of Justice. Cambridge, Mass: Belknap Press.

Roemer, J. (1996). Theories of Distributive Justice, Harvard University Press, Cambridge MA.

Scully, G W. (2002). "Economic Freedom, Government Policy and the Trade-Off Between

Equity and Economic Growth." Public Choice 113:77–77.

Sen, A.K. (1997). On Economic Inequality, (Second eds), Oxford: Clarendon Press.

Smith, J.P. (1999). Healthy bodies and thick wallets: the dual relation between health and

economic status, Journal of Economic Perspectives, Vol. 13: 145–166

Solt, F.(2008). "Standardizing the World Income Inequality Database." Social Science

Quarterly

Stiglitz, J E (2002). Globalization and its Discontents, W. W. Norton and Company,

London

Strauss, J and D, Thomas (1998). “Health, nutrition, and economic development” Journal of

Economic Literature, 36(2):766–818

Sugden, R. (2004). “Living with unfairness: the limits of equal opportunity in a market

economy.” Social Choice and Welfare, 22: 211–236

Thomas, V., Wang Y. and X. Fan (2003). “Measuring Education Inequality: Gini

Coefficients of Education for 140 Countries (1960–2000).” Journal of Educational

Planning and Administration 17(1).


Chapter 1

Trannoy, A (2005). “Multidimensional Egalitarianism and the dominance approach: a lost

paradise.” In Inequality and Economic Integration, ed. F.Farina and E.Savaglio. London:

Routledge. 284–304.

Tsui, K.-Y. (1995). “Multidimensional Generalisations of the Relative and Absolute

Inequality Indices: The Atkinson–Kolm–Sen approach.” Journal of Economic Theory

67: 145–157.

Uslaner, E. (2003), Trust, democracy and governance: can government policies influence

generalized trust? In: Hooghe, M., Stolle, D. (Eds.), Generating Social Capital: Civic

Society and Institutions in Comparative Perspective. Palgrave Macmillan, New York:

p.171–

Wilkinson, R. G. (1992), “National mortality rates: the impact of inequality?,” American

Journal of Public Health, Vol. 82, Issue 8 1082–1084

Wilkinson, R. G. (1996). Unhealthy Societies: The Afflictions of Inequality. London: Routledge.

World Bank. (2008). World Development Indicators. Washington DC: World Bank.

24


Chapter 2

Health, Wealth and Wisdom: Exploring Multidimensional

Inequality in a Developing Country

Forthcoming 2009 in Social Indicators Research

2.1 INTRODUCTION

During the past decades there has been a growing opinion in favor of including other

dimensions than a monetary perspective in analyzing inequality. The justification is

based on the idea that no matter how good an income indicator is - it is incomplete,

as individuals and households have different characteristics and needs, and such

shortfalls leads to inaccurate estimations of disparity. These concerns have received

increasing attention from economists resulting in a broad theoretical literature on the

subject of multidimensional inequality (c.f. Sen 1973, Kolm 1977, Atkinson and

Bourguignon 1982, 1987, Maasoumi 1986 and Tsui 1995, 1999). However, less

attention has been paid to empirical research on the subject and most work on

inequality applies a one-dimension monetary perspective. 1 Welfare examinations on

the situation in developing countries do not differ on this point, although the nature

of development per se is not unidimensional. Clearly, the operationalization of the

1 The multidimensional approach to poverty has also been addressed theoretically by several researchers;

see Duclos et al. (2001), Deutch and Silber (2005). In contrast to the study of inequality, there are currently

more empirical applications in this field, see e.g. Chambaz and Maurin (1998) and Maltzahn and Durrheim

(2008).


Chapter 2

extended inequality concept has not followed the theoretical developments. There is

often an assumption in the empirical literature that income inequality is closely related

to other forms of inequality and thus can be used as a single proxy for the level of

and changes in overall inequality (Maasoumi 1999).

The purpose of this study is to empirically explore techniques on how to

evaluate multidimensional inequality. Taking as a starting point the fact that

multidimensional inequality comparisons are ethically and theoretically attractive and

that there hitherto is no pronounced agreement on how to assess inequality of several

dimensions, there is an evident rationale for examining existing methods on how to

measure changes in multidimensional inequality in order to assess their strengths and

weaknesses in an empirical context. This is further justified by the limited number of

empirical applications at present.

For our empirical applications we employ three existing theoretical

approaches to the measurement of multidimensional inequality using Zambian

household data. First we examine inequality in the different attributes separately,

applying inequality indices customarily used in the univariate study of inequality.

Concerning the second method we make use of a multidimensional index developed

by Maasoumi (1986). In implementing the third technique, we employ sequential

stochastic dominance conditions, derived by Muller and Trannoy (2003) and Trannoy

(2005), which follow the work of Atkinson and Bourguignon (1982, 1987).

This analysis takes four welfare dimensions into consideration; consumption,

education, health, and land. These are four out of several dimensions of interest when

studying inequality, all of them well known from the poverty literature. As there are

reasons to believe that economic conditions in monetary terms drive other aspects of

living standards, there are arguments for including a consumption variable in the

study of inequality. On the other hand, non-monetary attributes such as education

and health may capture dimensions of a household’s welfare that are non-tradable

and thus not well proxied by consumption. Moreover, it seems motivated to include

these variables when studying multidimensional inequality in less developed countries

as they, in contrast to monetary information, are less likely to be influenced by

seasonal fluctuations or measurement problems. In a development context it also

26


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

seems relevant to include an asset perspective such as land, not least because land

might contribute to household food security.

The contribution of this paper is threefold. Firstly, it complements the

growing theoretical literature on multidimensional inequality with an empirical

application. In particular, we use a range of measurement techniques. To our

knowledge there are no empirical examinations of multidimensional inequality fully

exploring the multivariate stochastic dominance approach here considered also

analyzing aggregated indices of multidimensional inequality. Secondly, we apply a

multidimensional perspective to inequality using household level data. Existing

empirical examinations of multidimensional inequality are mainly introduced in a

cross-country or regional-based setting with population data (Hirschberg et al. 1991,

Lugo 2004 and Quadrado et al. 2001). An exception is Justino (2004), who analyze

multidimensional inequality among Brazilian households. 2 Clearly, less aggregated

statistical units imply more information and variation. Thirdly, the paper provides

indications of the development of inequality over the past six years in Zambia with

respect to several dimensions of welfare. With respect to our purpose, we believe it is

relevant and interesting to study the Zambian case as it is well known that monetary

inequality is very high, allowing for variation in at least one welfare dimension, and as

existing policy conclusions are mainly based on a monetary perspective. Furthermore,

in many respects the country is representative for developing economies.

Our examination points to the fact that inequality comparisons taking

interrelations between attributes into account are repeatedly at odds with

comparisons of independent distributions. The assessment of the multidimensional

index shows evidence of that dimensions of wellbeing compensate and reinforce each

other with respect to inequality in this empirical context. However, a majority of the

results using this technique are very sensitive to the degree of substitution between

attributes. Sensitivity analyses and explicitness should thus accompany examinations

of this kind. In applying a stochastic dominance method few combinations fulfill the

required dominance conditions. Accordingly, generality and less imposed structure

2 Although being a thorough review on different techniques, Justino(2004) does not include a test of

sequential dominance as developed by Trannoy(2005).

27


Chapter 2

come at a cost. Although the overall findings does not give a clear cut picture, we

conclude that the results are very informative in some selected dimensions and

combinations of dimensions and that empirical usefulness of these existing

techniques is reasonable as long as we stay aware of intrinsic weaknesses. Clearly,

careful interpretations and analyses involving more than one technique are

constructive in portraying multidimensional inequality.

This paper takes off with a brief presentation on the three theoretical

frameworks for comparisons of inequality in more than one dimension. Secondly we

present the nature of the data. With this foundation an empirical analysis on changes

in multidimensional inequality is performed. The paper ends with some concluding

remarks.

2.2 CONCEPTS AND THEORETICAL FRAMEWORK

Generally the theoretical literature on multidimensional inequality can be divided into

three different parts. A first element of the literature applies an item-by-item approach,

where comparisons over time or space are made independently for each dimension of

interest. A second part applies an aggregative approach, where conclusions on inequality

are established by the magnitude of different indices based on an aggregation of

multiple indicators. The third method concerns a non-aggregative approach. This part of

the literature focuses on orderings rather than levels and use stochastic dominance

techniques for analyzing changes in inequality.

2.2.1 An item-by-item approach

In applying an item-by-item approach to examining multidimensional inequality, each

attribute of concern is regarded separately and no structure on the relations between

different dimensions is introduced. Comparisons of inequality between units, groups

and over time are performed by using measures frequently applied in unidimensional

income inequality analysis (e.g. the Gini coefficient, General Entropy measures, the

Atkinson index), or comparisons based on orderings (e.g. first or higher order

stochastic dominance conditions) using one variable at a time (Lugo 2004, Justino

2004).

28


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

It is of importance to note that the item-by-item approach says nothing about

the degree of household substitution of attributes or about the relative weight society

puts on different attributes of welfare. As a result, by independent one-at-a-time

analysis of welfare attributes, it is not possible to conclude whether there is joint

incidence of inequality, along different dimensions of interest. An attempt to capture

what interdependencies exist between different distributions of welfare when

applying an item-by-item approach is generally to perform a cross-correlation

analysis.

Our examination includes four inequality indices: the Gini coefficient and

three indices belonging to the Generalized Entropy class. The Gini coefficient is a

statistical measure of inequality that takes values between zero (0) and one (1), where

0 implies complete equality and 1 complete inequality. If there are n households in the

population, μ is the mean household value of the attribute x studied and xi and

x j denote the allocation of this attribute in household i and j, respectively, the Gini

coefficient (G) can be written as follows

1

G = 2

2n

μ

n

n

∑∑

i=

1 j=

1

x

i

− x

j

Consequently, we here measure an average of pair-wise differences between the

individual observations in a population, weighted by the overall population mean

(Cowell 2000). Blackorby and Donaldson (1978) derive the particular form that a

social welfare function consistent with the Gini coefficient must take. This

unidimensonal inequality measure satisfies the Pigou-Dalton principle and is most

sensitive to differences in allocations about the middle of a distribution.

Measures belonging to the Generalized Entropy class of inequality indices

have their foundation in information theory. These indices allow us to examine the

stability of welfare rankings for different weightings by selecting different choices of a

parameterα . By following the above notation, members of the GE class are derived

by

29

(2.1)

α

⎡ n 1 1 ⎛ x ⎤

i ⎞

GE(

α ) = ⎢ ∑ ⎜ ⎟ −1⎥

α ≠ 0,1 (2.2)

2

α − α ⎢ i=


n 1 ⎝ μ ⎠ ⎥⎦


Chapter 2

For the special cases when α → 0and

1the above general form becomes

=

n

1

GE(

0)

1

GE(

1)

=

n

n

∑ ⎜

i= i x

log

1

n


i=

1

⎛ μ ⎞

and (2.3)

⎝ ⎠

⎛ xi

⎞ ⎛ xi


⎜ ⎟log⎜

⎟ respectively. (2.4)

⎝ μ ⎠ ⎝ μ ⎠

GE(0) corresponds to the mean-log deviation and is particularly sensitive to low

values in the distribution. GE(1) corresponds to Theil’s inequality index which allots

equal weight to all observations in the distribution, and the GE(2) places greater

weight on differences in the upper tail of a distribution (Myles 2002).

The above inequality measures range from zero (0) to infinity (∞) and higher

values indicate higher levels of inequality. Although the indices of the Generalized

Entropy class are not based on a welfare theoretic approach, we note that these tools

for distributional analysis are ordinally equivalent to the Atkinson measure, in turn

directly derived from a social welfare function, when α = 1 − ε . The

parameterε defines inequality aversion (Cowell 2000). As the Gini coefficient,

measures belonging to the Generalized Entropy class follow the Pigou-Dalton

principle. In addition, both the Gini and the GE measures are symmetric and obey

the axioms of continuity and invariance to scalar multiplication.

2.2.2 An aggregative approach

A second approach to analyzing inequality with respect to different dimensions

includes a more explicit multidimensional framework. In this case direct aggregated

composite indices of multidimensional inequality, which synthesize information on

distributions of interest into a single real-valued measure, are derived. The principal

critique regarding multivariate indices is that an aggregation procedure leads to loss of

information, and that it is difficult to develop consensus axioms (Duclos et al. 2001).

In addition, all aggregated indices have in common the fact that it is impossible to

reach any conclusion regarding the value of a particular measure if we do not take a

stand in the aggregation phase regarding (1) weighting structure, (2) degree of

substitution between attributes and (3) degree of inequality aversion (Bourguignon

30


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

1999). On the other hand, we here get a complete ordering of distributions since a

scalar measure is received.

The literature on the aggregative approach includes indices developed by

means of an axiomatic approach (Tsui 1995, 1999) as well as those derived ad-hoc

(Maasoumi 1999, Bourguignon 1999). To examine changes in multidimensional

inequality when applying an aggregative approach, we make use of the Maasoumi

index. This is one of the first indices proposed in the literature.

Consider j=1,2,….m dimensions of wellbeing represented by attributes and

i=1,2,…n statistical units representing individuals, households etc. For each statistical

unit there is a non-negative value X ij for every m dimension and thus a welfare

matrix X = X ij . The first step in calculating the Maasoumi index of multivariate

inequality is to aggregate the chosen welfare attributes into a summary wellbeing

function for the i-th statistical unit, S = S X , X ,..., X ) . Assume further the

i

31

( i1

i2

im

existence of a scalar function (e.g. a social welfare function) of the matrix X.

Maasoumi (1986) here defines a multivariate generalization of the GE measures of

divergence or closeness between the m densities corresponding to

β

m ⎧ n ⎡


⎪ ⎛ ⎞ ⎤


∑ ⎨∑

⎢⎜

S i

D

⎟ − ⎥

β ( S,

X , w)

= w j S i 1 / β ( β + 1)

⎬ β ≠ 0, −1

(2.5)

⎪ ⎢⎜


j i= 1


⎩ ⎣⎝

X ij ⎠ ⎦ ⎪


When β → 0 or -1, one obtains the following indicators

m ⎡ n ⎛ ⎞⎤

∑ ⎢∑


S i

D


0 ( S,

X , w)

= w j S i log ⎥


⎜ ⎟

j i= 1 ⎣ ⎝ X ij ⎠⎥⎦

(2.6)

m ⎡ n ⎛ X ij ⎞⎤

D ∑ ⎢∑



−1

( S,

X , w)

= w j S i log ⎥

j ⎢⎣

i= 1 ⎝ S i ⎠⎥⎦

(2.7)

where w j are the weights allotted to each welfare attribute and where the parameter

β is the coefficient of substitution between the different welfare dimensions. This

coefficient guarantees that changes in inequality not only take place due to changes in

rankings, but also to changes in the dependence between various welfare attributes.


Chapter 2

Minimizing Dβ with respect to i

S , such that S = 1,

generates ‘optimal’

aggregation functions interpreted by Maasoumi (1986, 1999) as the wellbeing for the

i-th unit

S

i


⎪⎛


⎪⎜

⎪⎝


∝ ⎨







j=


m


j

m

−1


⎞ β

−β

w ⎟ ⎪

j X ij ⎟

, β ≠ 0,

−1


1 ⎠ ⎪

w j


X →

j ij , β 0 ⎬


w → −


j X ij , β 1




32

n


i=

1

i

(2.8)

The composite welfare indicator S i , that for strictly positive X ij is defined for all

different degrees of substitution, can be interpreted as a utility function of the CES

type with an elasticity of substitution defined by σ = 1/( 1 + β ) when β ≠ 0, −1

(Deutch and Silbert 2005). Different values for the coefficient of substitution give

different degrees of curvature to the social indifference curves with respect to

household attributes. Consequently, the value of β depends on the degrees to which

equality with respect to attributes is valued (Jehle and Reny 2001). As β increases,

there is less and less substitution between attributes. At the limit, when β → ∞ and

σ → 0 , the included dimensions of welfare are assumed to be perfect complements

and accordingly there is no substitution between attributes. During these

circumstances S i will mirror the attribute with the lowest value for every statistical

unit, i.e., the worst performer of the selected welfare dimensions for each household

or individual. As a result the composite indicator here approaches a Rawlsian form,

where social bias in favor of equality between attributes is absolute. The aggregate

function here is of Leontief type with L-shaped contour curves.

With β → 0 the welfare indicator corresponds to a Cobb-Douglas utility with

unit elasticity with respect to different dimensions. On the other hand, when

β → −1

andσ → ∞ S i can be interpreted as a linear utility function of the m

attributes. In this case, low levels in one dimension can be fully compensated by high

levels in another and attributes are assumed to be perfect substitutes. Accordingly,

this corresponds to social indifference to how household or individual welfare is


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

distributed among attributes, and the composite indicator of every statistical unit

corresponds to the weighted arithmetic mean of the different dimensions included. In

the literature β > −1

is a common restriction which implies a non-negative elasticity

of substitution.

After obtaining aggregation functions over desired household welfare

attributes, where multivariate welfare is composed of a weighted sum of attribute

inequalities and an adjustment due to the covariation between the attributes,

Maasoumi (1986) makes use of the above presented General Entropy measures, here

applied to the obtained S distribution. Defining d i as the population share

corresponding to the i-th unit in the distribution, in general equal to 1/n. Moreover

*

defining S i as S i divided by the total sum of the welfare function S i over all units.

By also introducing a parameter α , representing inequality aversion, with more

sensitivity to dispersion in the lower part of a distribution the lower the α ,

Maasoumi presents the following two multidimensional inequality measures,

corresponding to Theil’s two inequality indices:

n

*

* ⎡ S ⎤ i

M 0 ( S)

= ∑ Si

log⎢


(2.9)

i= 1 ⎣ d i ⎦

n ⎡d

⎤ i

M1

( S)

= ∑ d i log⎢


(2.10)

*

i= 1 ⎣ Si


2.2.3 A non-aggregative approach

Turning to the non-aggregative approach, the goal is to order different states and the

method does not generate a specific real valued measure of the degree of inequality of

a distribution. This approach allows for joint distributional analysis and uses

stochastic dominance analysis to make judgments about which distribution is the

more equal (Savaglio 2005, Tsui 1999). Although the inequality measures discussed

above generally meet a set of desirable axioms, it might be that these indices rank the

same set of distributions in different ways, since they attribute different sensitivity to

allocations in different parts of the distributions. The general idea of using this

method in comparisons of inequality is to verify whether an ordering of distributions

can be considered to remain the same under a wide spectrum of indices.

33


Chapter 2

With a stochastic dominance approach there is a possibility of agreement over

classes of welfare functions of different dimensions of welfare, without having to

specify the precise form of a social welfare function (Maasoumi 1999). Assumptions

are made for classes of utility functions from which conditions of dominance of

different orders are derived. There exist several orders of stochastic dominance and

all have an ethical interpretation in the context of social welfare (Trannoy 2005). If

dominance is achieved, one distribution is unambiguously preferred to another.

Our exploration makes use of a recently developed framework by Muller and

Trannoy (2003) and Trannoy (2005). Their work belongs to the strand of the

theoretical literature on multidimensional stochastic dominance where particular

attention is directed towards one distribution within a joint-distribution, i.e. attributes

of interest are assigned an asymmetric role. The attribute given a particular position

plays the role of a compensating variable since it is assumed to be able to compensate

for deficiencies in other characteristics. The pioneering work in an asymmetric setting

is by Atkinson and Bourguignon (1987), where focus is on the measurement of

income inequality while at the same time accounting for households’ different needs.

One attribute, representing needs, is used to partition the population into

homogenous groups, while social welfare defined from the attribute of particular

concern, income, is considered within the need groups and in the whole society.

Muller and Trannoy begin with an arrangement of ALEP substitutability

utility functions where the partial cross derivative is assumed to be non-increasing

U ALEP

{ u u ≥ 0,

u ≤ 0,

u ≤ 0,

≤ 0}

= u

1,

2 11 22 12

This implies that the marginal utility of one attribute (the compensating) decreases

with the level of another attribute (the compensated). For example, a household’s

marginal utility of income is lower if a household is well educated compared to if the

household’s educational level is low. This can be interpreted as a Pigou-Dalton

principle in a multidimensional context, since a monetary transfer from a richer

household to a poorer one with the same educational level should not result in an

increase in multidimensional inequality (Justino 2004). The framework developed by

Muller and Trannoy (2003) consequently secures the idea that compensation is good

for welfare. In addition to requiring a non-increasing partial derivative, it is demanded

34


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

here that the marginal utility of income decreases as a statistical unit gets richer in

terms of money, and that the marginal utility of education decreases as a statistical

unit gets more educated.

On top of securing the idea of substitutability Muller and Trannoy (2003)

capture the intuition that compensation seems even more appropriate if people in the

lower tail of one distribution in a population also have a poor situation in terms of

another attribute. Let x1 represent the income and x2 the educational level of a

statistical unit. It is now assumed that the third cross partial derivative is nondecreasing

which means the difference in marginal utility of income among well

educated households is lower compared to among households with low education.

U MTx 1

{ u u ≥ 0,

u ≤ 0,

u ≤ 0,

u ≤ 0,

≥ 0}

= u

1,

2 11 22 12 112

In this subset income is the compensating variable of particular interest and

education is the compensated variable. Specifically, this class is designed to capture

the view that we are primarily interested in the distribution of income among the less

educated.

The framework also includes a set of conditions for the family of utility

functions satisfying

U MTx 2

{ u u ≥ 0,

u ≤ 0,

u ≤ 0,

u ≤ 0,

≥ 0}

= u

1,

2 11 22 12 221

where education now is the compensating attribute and income the compensated. In

this case there is agreement that the poor in terms of money must have priority when

constructing educational policies. In line with the asymmetric role, the two families of

utility functions U MTx and U 1

MTx are not anonymous with respect to the set of

2

attributes.

When the compensated attribute is discrete some of the assumptions in the

work by Muller and Trannoy (2003) and Trannoy (2005) are akin to what is

introduced in Atkinson and Bourguignon (1987). In this situation it is assumed that

the marginal valuation of the compensating attribute is different between diverse

groups, and that it is possible to identify and rank households according to this

valuation. This particular separation allows for diverse judgments of welfare for

different partitions identified by given characteristics different from income

35


Chapter 2

(Maasoumi 1999). As above, welfare cannot decrease as a result of increasing

incomes. Moreover, it is possible to identify what households with the same income

level would benefit the most from a monetary increase as the effect on social welfare

of a given increase in income is larger the needier the group that receives this money.

The assumption on the third cross partial derivative now implies that differences in

the marginal valuation of income between groups decrease as we move toward higher

monetary levels (Chambaz and Maurin 1998). This reflects less concern with

differences in needs for higher income groups.

The obtained criterion for one multivariate distribution (A) to second order

dominate another (B), both containing a marginal distribution of income (x1) and

education (x2), in this setting consists of two conditions. One condition for the

compensating attribute, and one for the compensated. When the distribution of the

latter variable is discrete, the distribution of the compensating variable of particular

concern has to meet a sequential generalized Lorenz condition. A sequential test is

implemented by initially examining whether there is dominance in e.g. income for the

neediest group. If one distribution dominates the other, the exercise is continued by

adding in the second neediest group and now testing for dominance of the income

variable for these two groups combined and so forth until the total population is

included (Bazen and Moyes 2003). We should note that a sequential dominance

approach demands weaker conditions compared to examining income distributions

of different groups, characterized by their different needs, separately. This is the case

since a negative distributional effect in one group can be offset by a favorable

distributional effect in another as the groups are gradually cumulated. Concerning the

second dominance criterion derived by Muller and Trannoy (2003), the distribution

of the compensated variable, in addition, has to satisfy a generalized Lorenz (GL)

condition.

The statistical tool to examine of the distribution of the compensated attribute

is to check the GL curves of the compensated variable in the two multivariate

distributions, (A) and (B). When primary concern is the monetary distribution of the

least educated, the GL curve of x2 in distribution (A) must not be below the

corresponding curve in distribution (B). To enable a test of second order dominance

36


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

of the chosen compensating variable income, a definition of the absolute poverty gap

is defined in a multivariate setting

1

PG ( x1

| x2

) = ∑ ( x1

− xi1

)

(2.11)

n { i|

( xi1

, xi

2 ) ≤(

x1

, x2

) }

This is the cumulated poverty gap for all statistical units with income xi1 below or

equal to an income limit x1 and with an educational level below or equal to x2. If this

gap is not larger in (A) compared to (B) for all choices of x1 and x2, distribution (A)

income poverty gap dominates (B), i.e.

PGA(x1|x2) ≤ PGB(x1|x2) ∀ x1 ∈ X 1 , ∀x

2 ∈ X 2 ↔ A ≥ PGX1 B (2.12)

In a corresponding way the educational poverty gap is defined

1

PG ( x2

| x1

) = ∑ ( x2

− xi

2 )

(2.13)

n { i|

( xi1

, xi

2 ) ≤(

x1

, x2

) }

where xi2 is the educational level of an individual or a household and x2 is equivalent

to an educational poverty line. If the main concern is the educational distribution

among the poor, the educational poverty gap in distribution (A) must be no higher

than in distribution (B), i.e.

PGA(x2|x1) ≤ PGB(x2|x1) ∀ x1 ∈ X 1 , ∀x

2 ∈ X 2 ↔ A ≥ PGX 2 B (2.14)

The poverty gap dominance condition is implemented sequentially. Achieved poverty

gap dominance for the compensating attribute implies that generalized Lorenz

dominance is fulfilled for this variable as well. Consequently, Trannoy (2005) states

the following conditions as sufficient to realize dominance for the multivariate

distribution (A) over the corresponding distribution (B) according to the asymmetric

classes U MTx and U

1 MTx2

≥ B and ≥ B � ≥ B

A GL

A PGX1

A U MTx1

A ≥

U MTx 2

≥ B and ≥ B � B

A GL

A PGX 2

37


Chapter 2

2.3 DATA AND VARIABLES

The Zambian Living Conditions Monitoring Survey 1998 (LCMS II) and the

Zambian Living Conditions Monitoring Survey 2004 (LCMS IV) are used for this

empirical application. These household surveys were conducted in November and

December 1998 and in October 2004 to January 2005, respectively, by the Zambian

Central Statistical Office (CSO 1999, 2005). In addition to being conducted during

the same period of the year the two surveys have similar questionnaires which make

them comparable. The CSO has been conducting household-based LCMS in the

country since 1996, all being independent surveys interviewing different households

in each year. Consequently, the datasets are not a panel of households, but two

representative cross-sections.

The two surveys were employed using similar survey design with a multiplestage

sample selection process. The LCMS II and the LCMS IV have nationwide

reporting and cover 16710 and 19340 households respectively and 93 471 and

103 242 individuals respectively. Sample weights are applied in all calculations to

correct for differential representation of the sample at national and sub-national

levels. Our analysis is based on 16445 and 19179 households. 3

The unit of analysis in this paper is the household. A household is defined as

the head, the spouse, children, relatives and other dependents living in the household

but also usual members who at the moment of the conduction of the survey were

away visiting, hospitalized etc. Visitors are generally excluded, unless they have lived

with the household for six months or more. Although the standard apparatus of

welfare economics and measurement concerns the wellbeing of individuals, in a

development context an individual’s wellbeing often depends on the resources

available to the household, the size and the structure and in what manner resources

are shared within the household (Deaton 1997).

Four dimensions of household welfare are examined: expenditures, education,

health and land holdings. In both surveys there is information on household monthly

3 We drop 265 households in 1998 and 161 households in 2004 from the original samples since these units

do not have complete information concerning one or more of the variables used in the analysis. This

represents a drop of 1.6% and 0.8%, respectively, of the original samples.

38


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

expenditures and on total land area under crop on household level. Data on years in

education and on health status are reported on an individual level. Representing a

monetary distribution, household monthly per adult equivalent expenditures are used

in the empirical application. To adjust for inflation over the time period we use

information on the Consumer Price Index in Zambia (CSO 2005).

To represent a first human capital distribution, we use the maximum

educational grade obtained by the head of the household ranging between 0 and 17,

with 17 corresponding to doctoral level. To examine inequalities in health we estimate

a health distribution by using an ordered probit regression using a categorical

subjective health variable as the dependent variable. Each observation in the resulting

variable corresponds to household average health status. The estimation follows a

procedure in the World Bank’s technical note #3 (2005) on quantitative techniques

for health equity analysis, further described in van Doorslaer and Jones (2003). 4

Naturally, the results including the health dimension of welfare should be interpreted

with particular care. To represent an asset distribution, we use information on

household land area under crop in the last agricultural season measured in hectares.

In applying the non-aggregative approach, the four welfare variables in turn

are used as discrete partitions. The aim is to generate a division where groups are

relatively homogenous so that it is reasonable to consider the household marginal

valuation of a second welfare dimension, corresponding to the compensating

attribute, to be similar within partitions. When employing the monetary distribution

as the compensated attribute, households are divided into the groups extremely poor

(households with a monthly per adult equivalent real expenditure of less than

K32,681), moderately poor (households with a monthly per adult equivalent real

expenditure higher than K32,681 but less than K47,187) and non-poor (monthly per

adult equivalent real expenditure higher than K47,187) following the poverty lines

derived by the Zambian CSO (CSO 2005). The distribution of education is split up

into the groups no or primary education (0-7 years of schooling), secondary

education (8-12 years of schooling) and higher education (more than 12 years of

schooling) corresponding to levels of educational grade in the Zambian education

4 Detailed information of this procedure and estimation results are available from author upon request.

39


Chapter 2

system. The land variable is divided into three sub groupings according to the

definitions landless, small scale (households cultivating land more than or equal to 0.1

hectares but less than or equal to 2 hectares of land) and medium and large scale

cultivators (households cultivating more than 2 hectares of land). This grouping

relates to the criteria for rural stratification of households in Zambia used in the

previous National Census of Agriculture (CSO 1994). The estimated health

distribution is divided into population quintiles each representing 20% of the

population. 5

2.4 EMPIRICAL APPLICATION

2.4.1 An item-by-item approach

Monetary inequality in Zambia is generally stated to be high and the country is often

referred to as one of the most unequal societies in the world. Table 2.1 confirms that

the level of monetary inequality in the country is high in an international context. As

measured by Gini coefficients of consumption distributions the World Bank only

identifies 5 countries out of 83 that are less equal (World Bank 2006). 6 Although

McCullough et. al (2001) establish that the Zambian Gini coefficient somewhat

decreased during the 1990s, table 2.1 indicates that consumption inequality at the

national level slightly increased between the two survey periods studied here, with the

Gini coefficient rising from 0.533 in 1998 to 0.544 in 2004.

The three GE inequality measures parallel the pattern revealed by the Gini

coefficient as they point to an increase in expenditure inequality over the time period.

This indicates that the point estimates of inequality portrayed by the Gini coefficients

are robust with respect to different weighting. Concerning specific changes in the

expenditure distribution between 1998 and 2004 we note a sharp increase in the

consumption growth of the very richest households relative to the sample as a whole.

5 To ensure that empirical results are not driven by how we create the different partitions we will also use

discrete partitions where the dataset is divided into quintiles corresponding to 20% of the population for

all distributions.

6 These countries are the Central African Republic (0.61), Lesotho (0.63), Panama (0.55), South Africa

(0.58) and Zimbabwe (0.57).

40


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

The three additional welfare dimensions, education, health and land, have not

been examined as much as income or consumption from an inequality perspective.

Cross country studies on Sub-Saharan Africa state regional average education Gini

coefficients to be 0.66 (World Bank 2006). In relation to this, the magnitude of

inequality in education in Zambia is low and education inequality decreased over the

six-year period. Measured in terms of the Gini coefficient, inequality in this

dimension was reduced by 9%. This fall in education inequality is further confirmed

by additional inequality indices, and seems to be a decreasing function of education as

the more weight given to distances between educational outcomes in the lower end of

the distribution, the higher the fall in inequality. In particular, the decrease in

education inequality between 1998 and 2004 is large when studying GE(0).

Table 2.1 Unidimensional inequality measures

Gini GE(0) GE(1) GE(2)

1998

Expenditures 0.533 0.533 0.596 1.661

Education 0.354 2.728 0.262 0.194

Health 0.127 0.030 0.027 0.025

Land 0.675 5.654 0.901 1.615

2004

Expenditures 0.544 0.551 0.738 19.281

Education 0.322 2.079 0.215 0.162

Health 0.138 0.037 0.032 0.030

Land 0.698 5.981 1.174 15.491

Change Gini (1998-2004) %

Expenditures 2.167

Education -9.098

Health 9.470

Land 3.362

Author's calculations from LCMS II and LCMS IV

Concerning the level of inequalities, land seems to be the more unequally distributed

welfare attribute in Zambia and inequality in this dimension is also high by

international standards. Cross-country studies on Sub-Saharan Africa state regional

average land Gini coefficient of 0.5 (World Bank 2006). Large numbers in the GE(0)

and the GE(2) column reflect a situation with a small share of the total population

being involved in large-scale farming that, although not many in terms of numbers,

41


Chapter 2

cultivates very large areas of land, and major disparities at the bottom of the

distribution, mirroring that several households do not cultivate any land. Irrespective

of what indicator is examined, inequality in landholdings seems to have increased

over time. Above all, the GE(2) indices point to a major rise in inequality between

1998 and 2004, indicating that the increase in land inequality has mainly been located

in the higher percentiles.

Health status appears to be relatively homogenous among Zambian

households resulting in a relatively low level of health inequality. On the other hand,

the health Gini coefficient increased by more than 9% between 1998 and 2004. This

increase in inequality over time is also confirmed by changes in the class of GE(α )

measures. The augmented health inequality does not seem to be an outcome of larger

disparities in certain percentiles of the distribution.

A general conclusion from the above exercise on monetary and non-monetary

inequalities, concerning the magnitude of inequality rather than changes over time, is

that there are differences in the pattern of inequality across the various welfare

distributions. The monetary distribution is the least unequal when putting extra

weight on the observations in the lower part of the distribution, and increases with

the choice ofα . Conversely, the GE inequality measures for the non-monetary

distributions of education and health decrease with the choice of α , implying large

differences in the lower end of these distributions. Given the impact of human capital

on economic development, and assuming abilities to be normally distributed across

the population, this result represents a loss in aggregate welfare.

To get a first indication of multivariate inequality, Spearman's rank correlation

coefficients are calculated. A coefficient value of 1 indicates that the rankings of the

two distributions are perfect and -1 that rankings are reversed. If rankings are

completely independent the correlation coefficient is 0.

From a general point of view, the absolute values of the different correlation

coefficients in table 2.2 do not indicate very strong relationships among the different

attributes. Although the null hypothesis of independence of the distributions of

expenditures, education, health and land is rejected in all cases, the extent of overlap

42


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

among the distributions generally seems to be rather low. 7 The strongest relationship

appears between expenditures and education in 1998 with a correlation coefficient of

0.457. In the 2004 sample the rankings of the educational and health distribution are

somewhat more similar than the rankings in the expenditure and the educational

distribution.

Table 2.2 Spearman rank correlation

1998

Expenditures Education Health Land

Expenditures 1

Education 0.457* 1

Health 0.100* 0.245* 1

Land

2004

-0.267* -0.290* -0.085* 1

Expenditures Education Health Land

Expenditures 1

Education 0.349* 1

Health 0.118* 0.398* 1

Land -0.200* -0.329* -0.253* 1

Author's calculations from LCMS II and LCMS IV.

H0, that the variables are independent, is rejected for all cases. * Significance at 5% level.

Despite the fact that the distribution of expenditure or income is often used as a

proxy for the distributions of other welfare attributes in empirical work, we find the

relationship between expenditures and the other two indicators, health and land, to

be weak both in 1998 and in 2004. In examining the relation involving expenditures

and health, the Spearman rank correlation coefficient is positive and about 10-12%

correlated in ranking relationship. Concerning expenditures and land the relationship

is negative. Moreover, the welfare indicator land is negatively correlated to all other

variables, indicating that one of the variables decreases as the other increases. Weak

relationships between the distribution of expenditure and other welfare distributions

are recurrent in the empirical literature (Lovell et al. 1994, Ramos and Silber 2005). In

addition, static comparisons of monetary and non-monetary inequality from African

developing countries suggest that a lack of overlap between the different indicators is

common (Sahn and Stifel 2003).

7 The Pearson’s test of correlation generates the same general result as above with the difference that the

correlations are overall lower and that correlation between expenditures and land is positive, but

insignificant.

43


Chapter 2

The results from the item-by-item analysis give support to the plan of

continuing our exploration on multidimensional inequality to techniques combining

welfare dimensions. Not only do the magnitudes of monetary and non-monetary

inequalities seem to be dissimilar, there are also different trends in changes over time.

In addition, theoretical motives point to the relevance of examining how potential

complementarities between different welfare dimensions affect conclusions on

inequality. This is further supported empirically by the results from the rank

correlation examination.

2.4.2 An aggregative approach

Before aggregating the different attributes of concern into a multivariate welfare

function, it is necessary to transform the chosen attributes into the same unit of

measurement. The technique used in this application is based on the HDI approach

(UNDP, 1995) and the following formula generates distributions with observations

varying between 0 and 1.

S

ij

xij

− min xij

= with i=1,2…n and j=1,2,3,4 (2.15)

max x − min x

ij

ij

As recommended in the literature, the degree of substitution between the

attributes, β , is set to be larger than -1 in the calculations of the distribution

functions, which implies a positive elasticity of substitution. To include the case when

attributes are seen as complements we also let β take positive values. Concerning the

possible choice of relative inequality aversion, we operate with α = 0 and α = 1

which corresponds to Theil’s first and second measures of inequality. In all cases we

weight the four attributes equally (w1=w2=w3=w4).

Starting from a traditional viewpoint, the first examination of the Maasoumi

inequality index includes expenditures and one of the three other attributes in turn.

The results in table 2.3 clearly point out that we gain additional insights by applying a

multivariate technique compared to when examining inequality with a monetary

perspective exclusively. For example, inequality with respect to household

expenditures is only lower than the composite indices when β > 0 .These results

44


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

demonstrate that ignoring important ranges of economic conditions such as

education and health here seems to result in underestimation of inequality if attributes

are assumed to be complements, but overestimation if the elasticity of substitution is

larger or equal to one. One realizes also that these inequality indices of several

dimensions take an increasing magnitude across higher values of β . When β ≤ 0 we

assume that more of one attribute can compensate for a poor situation in terms of

another, and the level of multidimensional inequality should reasonably be relatively

low. As we impose less substitution, the indices capture the social bias in favor of

equality between different attributes and we have to accept larger magnitudes of

inequity.

Examining changes in multidimensional inequality over time, all but one

combination of attributes in table 2.3 show that inequality increased between 1998

and 2004. 8 The assessment including expenditures and land points to a significant

increase in multidimensional inequality across all degrees of substitution. This is also

true when a health dimension is included in the composite welfare function. The

exception is found in the second column combining a monetary and an educational

dimension of household welfare, as we find evidence of decreasing multidimensional

inequality over time. With this arrangement M(0) in 2004 is significantly lower

compared to in 1998 when we assume attributes to be perfect substitutes.

Consequently the results using the aggregative approach to some extent mirrors the

decrease in educational inequality observed when examining attributes separately.

The decreased dispersion of the educational distribution between 1998 and

2004 and the compensational effect from education on both levels and changes in

multidimensional inequality revealed in table 2.3, motivate an examination of another

set of combinations of the four attributes with this particular variable as a baseline.

Table 2.4 includes inequality indices of three multivariate S functions, together with

the GE(0) measure of the educational distribution S1.

The inclusion of the health variable generates a situation where inequality

seems to have decreased over the time period, although health inequality increased

8 For a test of statistical inference, a bootstrap procedure is used to generate estimates of the standard

errors of M(0) and M(1). The bootstrapped samples mimic the empirical distributions of the LCMS II and

LCMS IV survey samples.

45


Chapter 2

between 1998 and 2004 according to the examination of independent attributes.

When operating with high elasticities of substitution, this robustness is somewhat

unexpected as the rank correlation between education and health increased over time.

As the preceding conclusion might be a result of our choice of inequality aversion, we

also operate withα = 1,

which allots equal weight to all observations. However, in

this setting as well the method generates the same outcome with significantly lower

magnitude of inequality in 2004 compared to in 1998. Consequently, the effect of less

dispersion of the observations in the educational distribution in the latter period can

counterweigh both the increased spread in the health distribution as well as the

increase in the correlation of the attributes over moments in time.

When examining multidimensional inequality including education, health and

expenditures, S3, we find increasing inequality across a majority of choices of β .

Consequently, the effect generated by declining inequality in household educational

status between 1998 and 2004 is no longer dominant once the monetary outcome of

households is taken into account. Interestingly there is no monotonous development

with respect to changes in inequality across degrees of substitution in this setting.

When β = −1and

β = ½ or greater, multidimensional inequality significantly

increases over time. If one lets the degree of substitution take an intermediate value

instead, the conclusion is that inequality significantly decreases. Similar results appear

when household welfare is assumed to depend on all four attributes. This is not in

line with earlier findings using this aggregative approach, for example Maasoumi and

Nickelsburg (1988) and Lugo (2004), who conclude that changes in multidimensional

inequality using this method are robust with respect to choice of β .

Although not shown in the table, changing the parameter of inequality

aversion, so thatα = 1,and

thereby allotting equal weight to all households in the

aggregated distributions, do not change the results for S3 or S4. Also, in this setting

changes with respect to equality are generally negative over time, when monetary and

a land perspective are assumed to contribute to household welfare together with

education. Moreover, despite the fact that the dispersion among households increases

over time when examining three out of the four distributions independently,

multivariate inequality is sensitive to the choice of β here as well.

46


Table 2.3 Maasoumi multidimensional inequality index – 1998 and 2004

S1:Expenditures

S2 Expenditures

& Education

S2: Expenditures

& Health

S2: Expenditures

& Land

Zambia 1998

M(0) β = -0.99999 0.260 (0.006) 0.026 (0.000) 0.544 (0.023)

M(0) β = 0.000001 0.372 (0.007) 0.152 (0.003) 0.694 (0.022)

M(0) β =1/2 0.696 (0.022) 0.549 (0.023) 0.735 (0.018)

M(0) β =1 0.780 (0.031) 0.643 (0.039) 0.742 (0.017)

M(0) β=20 0.533 0.800 (0.034) 0.663 (0.047) 0.775 (0.018)

Zambia 2004

M(0) β = -0.99999 0.214 (0.004) *** 0.032 (0.001) *** 0.748 (0.138) **

M(0) β = 0.000001 0.366 (0.009) 0.198 (0.006) *** 0.796 (0.018) ***

M(0) β =1/2 0.811 (0.048) ** 0.698 (0.039) *** 0.801 (0.015) ***

M(0) β =1 0.938 (0.080) ** 0.821 (0.074) ** 0.820 (0.016) ***

M(0) β=20 0.551 1.012 (0.098) ** 0.881 (0.105) ** 0.877 (0.017) ***

Author's calculations from LCMS II and LCMS IV

Standard errors in parentheses, estimated by bootstrapping with 500 replications.

*** Difference between the two periods significant at 1% level

**Difference between the two periods significant at 5% level * Difference between the two periods significant at 10% level


Table 2.4 Maasoumi multidimensional inequality index – 1998 and 2004

S1: Education

S2: Education

& Health

S3: Education,

Health &

Expenditures

S4: Education,

Health,

Expenditures &

Land

Zambia 1998

M(0) β = -0.99999 0.047 (0.001) 0.047 (0.001) 0.047 (0.001)

M(0) β=-1/2 0.090 (0.002) 0.090 (0.001) 0.086 (0.002)

M(0) β =0.000001 0.207 (0.006) 0.260 (0.006) 0.460 (0.007)

M(0) β=1/2 0.216 (0.006) 0.624 (0.015) 0.923 (0.018)

M(0) β=1 0.222 (0.006) 0.762 (0.026) 0.962 (0.021)

M(0) β=20 2.728 0.253 (0.006) 0.799 (0.034) 0.998 (0.020)

Zambia 2004

M(0) β = -0.99999 0.032 (0.001) *** 0.050 (0.001) *** 0.049 (0.001) **

M(0) β=-1/2 0.083 (0.001) *** 0.083 (0.001) *** 0.077 (0.001) ***

M(0) β =0.000001 0.168 (0.004) *** 0.238 (0.004) *** 0.625 (0.008) ***

M(0) β=1/2 0.175 (0.003) *** 0.708 (0.018) *** 0.919 (0.014)

M(0) β=1 0.181 (0.004) *** 0.880 (0.045) ** 0.975 (0.017)

M(0) β=20 2.079 0.209 (0.004) *** 0.990 (0.117) ** 1.032 (0.019) *

Author's calculations from LCMS II and LCMS IV

Standard errors in parentheses, estimated by bootstrapping with 500 replications.

*** Difference between the two periods significant at 1% level

**Difference between the two periods significant at 5% level * Difference between the two periods significant at 10% level


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

The non-monotonous development with respect to changes in multivariate inequality

across β motivates a more detailed examination of the choice of degree of

substitution. We here focus on the range where the difference in inequality over time

hovers between positive and negative outcomes when operating the S4 aggregation.

Interestingly, the spans of β equaling (-0.7 - -0.2) and (0.2- 0.6) all point to the fact

that inequality decreased between 1998 and 2004 in table 2.5. Thus, in this setting the

method seems to capture exceptions in changes in multidimensional inequality

when β is close to -1 and 0. This is particularly noteworthy since these exact values of

degrees of substitution have been examined in the existing empirical literature on

multidimensional poverty and inequality using the same kind of aggregation

procedure (c.f. Maasoumi and Nicklesburg 1988, Lugo 2004, Deutsch and Silber

2005). Our results suggest further examinations of the consequences of different

choices of β for the inference on changes in inequality of several dimensions in the

above literature.

As emphasized by Anand and Sen (2003) the degree of substitution among

various welfare attributes, the choice of inequality aversion and the different weights

attributed to different variables do not take a predominant role in the current

theoretical literature. However, the choice of β appears to be of major importance

when examining multidimensional inequality empirically in the above context. Three

combinations of attributes generate the same conclusion on whether inequality

including several dimensions increased or decreased over time across all ranges of

degree of substitution. All other aggregations of welfare dimensions, (also

combinations not presented) point to the same conclusion, namely that inequality is

sensitive to the particular assumption on the degree of substitution among

dimensions of welfare. This emphasizes the importance of not drawing major

conclusions from one or two point estimates but rather testing a range of values of

β when examining multidimensional inequality by this approach.

Turning to the choice of inequality aversion, inference on changes in welfare

does not seem to be sensitive to whether we employ M(0) or M(1). The general

conclusion on inequality changes over time outlined previously is invariant with

respect to the choice inequality measure using this aggregative technique.

49


Table 2.5 Maasoumi multidimensional index across ranges of different degrees of substitution – 1998 and 2004

S4: Expenditures,

Education

Health & Land β = -0.99999 β=-0.9 β=-0.8 β=-0.7 β=-0.6 β=-0.5 β=-0.4 β=-0.3 β=-0.2 β=-0.1

** *** *** *** *** *** *** *** ***

Zambia 1998 M(0) 0.047 (0.001) 0.052 (0.003) 0.057 (0.001) 0.065 (0.001) 0.074 (0.001) 0.086 (0.002) 0.102 (0.002) 0.122 (0.003) 0.158 (0.004) 0.237 (0.005)

Zambia 2004 M(0) 0.049 (0.001) 0.055 (0.001) 0.057 (0.001) 0.062 (0.001) 0.069 (0.001) 0.077 (0.001) 0.087 (0.001) 0.102 (0.002) 0.138 (0.002) 0.296 (0.005)

β =0.000001 β=0.1 β=0.2 β=0.3 β=0.4 β=0.5 β=0.6 β=0.7 β=0.8 β=0.9

*** ***

Zambia 1998 M(0) 0.460 (0.007) 0.694 (0.010) 0.806 (0.013) 0.864 (0.015) 0.899 (0.017) 0.923 (0.018) 0.938 (0.018) 0.947 (0.016) 0.954 (0.019) 0.959 (0.017)

Zambia 2004 M(0) 0.625 (0.008) 0.740 (0.010) 0.800 (0.011) 0.851 (0.013) 0.891 (0.014) 0.919 (0.014) 0.936 (0.016) 0.952 (0.016) 0.962 (0.016) 0.969 (0.018)

Source: Author's calculations from LCMS II and LCMS IV

Standard errors in parentheses, estimated by bootstrapping with 500 replications.

*** Difference between the two periods significant at 1% level **Difference between the two periods significant at 5% level * Difference between the two periods significant at 10% level


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

2.4.3 A non-aggregative approach

Turning to the examination of changes in multidimensional inequality by applying an

asymmetric non-aggregative approach, figure 2.1 illustrates how the distribution of

expenditures of 1998 dominates the corresponding distribution of 2004 according to

the poverty gap condition when partitioned by the three groupings representing

different levels of households’ land holdings. As the cumulated curves all take nonpositive

values, this indicates that the expenditure poverty gap is lower in the first

time period compared to the latter.

Figure 2.1 Difference between the poverty gap curves – Expenditures across land groups

From this examination it is also apparent that the generalized Lorenz dominance

criterion is fulfilled for the expenditure distribution, as this is an implication when

dominance according to the poverty gap condition is satisfied. This result is moreover

shown in figure 2.2 where it is clear that the difference between the generalized

expenditure Lorenz curves in 2004 and 1998 is negative. In other words, the

expenditure distribution of 1998 dominates the corresponding distribution in 2004

according to the generalized Lorenz requirement.

As multidimensional inequality can only be stated to be lower in one year

compared to another if the Generalized Lorenz criteria for both marginal

distributions of concern are fulfilled, we also implement this test for the distributions

51


Chapter 2

of land in 1998 and 2004. In studying figure 2.3 it is apparent that the generalized

Lorenz condition of land holds since the difference between the two generalized

Lorenz curves is equal to or less than zero.

Figure 2.2 Difference between the generalized Lorenz curves – Expenditures 2004 & 1998

Figure 2.3 Difference between the generalized Lorenz curves – Land 2004 & 1998

52


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

To ensure that the above results on expenditures partitioned by land are robust, we

perform a conditional poverty gap dominance test, using a poverty line

corresponding to the maximum value within the particular distribution. The results

confirm that the expenditure poverty gap is larger in 2004 than in 1998 for all poverty

lines, for all land partitions. As the differences are statistically significant for all test

points we declare dominance. Consequently we here come to an unambiguous

conclusion concerning changes in inequality including a monetary and a land

perspective. Multidimensional inequality in this setting, with expenditures as a

compensating attribute, increased over the time period.

Turning to the analogous asymmetric class U MTx where the distribution of

2

land among the poor is of primary interest, we test for land poverty gap dominance

across expenditure groups over time. As can be understood from the graphs in figure

2.4, this dominance condition is fulfilled for the group consisting of extremely poor

households.

Figure 2.4 Difference between the poverty gap curves – Land across expenditure groups

Furthermore the land poverty gap does not seem to be larger in 1998 compared to

2004 for the two following cumulated expenditure groups, including moderately poor

households and finally the total population. These results are confirmed when

53


Chapter 2

running the poverty gap dominance test. Together with the above result on

generalized expenditure Lorenz dominance, this implies that multidimensional

inequality increased during the six year period also when land is set as an attribute

compensating for low monetary levels.

The examination of changes in multidimensional inequality when including

expenditures and education once again shows that the expenditure distribution in

1998 dominates the corresponding distribution in 2004 according to the poverty gap

condition. As a matter of fact we find that the income poverty gap condition is

fulfilled for all educational groups separately, i.e. not cumulated. This is a stronger

condition that corresponds to a situation where there is no agreement on how the

marginal utility of income varies with the need of a household in terms of the second

attribute education, i.e. no conformity on the second cross partial derivative

(Atkinson and Bourguignon 1987). This implies that the mean expenditure of all

education groups is not lower in 1998 than in 2004. On the other hand, in

interpreting figure 2.5, we find that the generalized educational Lorenz curve in 2004

dominates the corresponding curve in 1998. Consequently, we can not come to an

unambiguous conclusion about whether multidimensional inequality including a

monetary and an educational viewpoint increased or decreased over the time period.

This outcome is moreover confirmed when studying the educational poverty gap

across expenditure groups where the dominance condition does not hold in the first

partition.

Turning to the combination of expenditures and health, the results are a

resemblance of appears in the expenditure and education setting. The income poverty

gap is lower in 1998 than in 2004 for the cumulated health quintiles but the

generalized Lorenz condition is not met for the attribute health as the two

generalized health Lorenz curves cross at their lower ends. In addition, the health

poverty gap condition, when the distribution of health is partitioned by poverty

groups, is not fulfilled. Accordingly, this signals a conflict among the points of views

captured by the dominance test.

54


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

Figure 2.5 Difference between the generalized Lorenz curves – Education 2004 & 1998

The results of the different tests for the combinations land - education, land -

health and education - health are not included. Since the generalized Lorenz

condition does not hold or point in a different direction for one or more of the

attributes in these combinations, we conclude that there is no possibility of finding an

unambiguous conclusion concerning changes in multidimensional inequality including

these attributes. 9

The generality of the conclusions that can be drawn by using the stochastic

dominance technique developed by Muller and Trannoy (2003) and Trannoy (2005) is

attractive. However, as can be seen in this empirical application, such generality

comes at a cost. When the cumulative density functions of a particular compensating

distribution cross one or more times or when the generalized Lorenz condition does

not hold for the attribute being compensated for, there is no clear ordering.

Consequently, by imposing less structure on how to exactly combine distributions

compared to the procedure followed when applying an aggregative approach and the

striving for ordinality to cardinality, we cannot tell whether inequality is lower in one

year or the other in a majority of cases.

9 All results in this section are confirmed using a different discrete partition of the variables. In this case the

dataset is divided into quintiles corresponding to 20% of the population.

55


Chapter 2

2.5 CONCLUSION

Although there is no complete agreement on how to measure inequality of several

dimensions of welfare, there are numerous reasons to consider inequality as a

multidimensional phenomenon. Moreover, with a lack of consensus on how to

proceed there is rationale for applying available methods on real data to gain

knowledge of their strengths, weaknesses and usefulness in an empirical context and

to contrast differing results. We have applied three techniques representing the main

theoretical strands of the literature on multidimensional inequality, using two

Zambian sets of household data.

Cardinal measures are clearly practical when examining changes in welfare as

these exercises always generate a yes or a no to whether equality increased or

decreased across time, regions or socio-demographic groups. The outcome of a scalar

is accordingly an advantage of the item-by-item as well as the aggregative approach.

When examining the attributes expenditures, education, health and land holdings

separately, all distributions but education prove to have less dispersion at the first

point in time compared to the second, for all inequality measures examined.

However, this approach also discloses the fact that correlations of different

dimensions of welfare are not very strong, and somewhat changed over time.

Accordingly, we get a first indication here that it is not necessarily the same

households that experience a poor situation in one welfare dimension that also face

an underprivileged position in another. This information has implications for the

results generated in employing an aggregative and a non-aggregative approach,

explicitly taking the interrelation of attributes into account.

In applying Maasoumi’s multidimensional index we find evidence of how

inequality measurement in any single attribute might be misleading as an overall

measure. Noticeably, the increase in health inequality between 1998 and 2004 is no

longer evident when we combine the health and the education distribution. Clearly,

different dimensions of household welfare can compensate for each other. Moreover,

it is by no means certain that combinations of distributions that separately indicate

increasing inequality over time or space necessarily point to the fact that

multidimensional inequality increases. To say the least, this implies that the use of one

56


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

indicator of inequality alone may generate a rather incomplete picture and

multidimensional measures have value in illuminating subtle differences.

Concerning the empirical usefulness the aggregative technique, it is of

importance to note that a number of results are very sensitive to the assumed degree

of substitution between attributes. In a majority of cases the choice of β matters for

what conclusions on changes in inequality can be drawn. Although our conclusions

on changes in inequality are robust to the choice of inequality aversion, the sensitivity

of outcomes to a particular aggregation function chosen, points to the importance of

sensitivity analysis and explicitness regarding what particular assumptions are made

when applying an aggregative approach.

With reference to the application of the non-aggregative approach, only the

combination of land and expenditures fulfill the required dominance conditions both

when expenditure is treated as a compensating attribute and land is compensated for

and vice versa. These results indicate that multidimensional inequality was

unambiguously higher in 2004. Consequently, an agreement among the outcomes of

the different techniques concerns the setting with land and expenditure for which

there are straightforward results from all three approaches. Regarding all other

arrangements of attributes we cannot come to a clear-cut conclusion when using the

non-aggregative approach. Thus, also in this application it is evident that combined

arrangements of distributions, which independently points toward increased

dispersion, do not identify a rise in multidimensional inequality. Avoiding the

computational complexity of aggregated measures of multidimensional inequality and

allowing a less demanding structure come at a cost, as we in a majority of cases

cannot tell whether inequality is lower or higher in one time period or the other.

In reviewing three theoretical methods for measurement of changes in

multidimensional inequality we find additional support for the claim in existing

welfare literature that the analysis of inequality ought to be multidimensional. Bearing

in mind that weak relationships between welfare distributions are recurrent, it is not

unlikely that examinations, combining and taking into account the interrelations of

different household attributes, may lead to unexpected multivariate comparisons. The

usefulness of the different techniques for measurement and policy analysis is

57


Chapter 2

reasonable given that we are aware of their intrinsic weaknesses. Careful

interpretations and analyses involving more than one technique are constructive

when portraying multidimensional inequality.

Our examination does not give a clear cut picture to whether

multidimensional inequality, using four attributes of household welfare, increased or

decreased in Zambia between 1998 and 2004. However, as declared by Sen (1997),

the concept of inequality is ambiguous and rather than leaving analyzes of inequality

of several dimensions undone, we should continue the research in this field.

Moreover, in some selected dimensions and combinations of dimensions the results

are very informative, representing contributions to knowledge. Clearly, more

empirical applications employing existing techniques are needed as the gap between

theoretical and empirical research in this field is substantial. There is also a call for

deeper research on ways to proceed when analyzing more than two dimensions

simultaneously employing the non-aggregative approach.

Acknowledgements

The author which to thank Peter Lambert, Koen Decancq and participants at the 2 nd

Meeting of the Society for the Study of Economic Inequality (ECINEQ), Berlin,

2007, Carl Hampus Lyttkens, Andreas Bergh, Göte Hansson, Carl-Johan Belfrage,

Nils Janlöv, Pernilla Johansson and seminal participants in Lund for useful comments

and suggestions. Financial support from Anna Nilssons stipendiefond and Per Westlings

fond is gratefully acknowledged.

58


Health, Wealth and Wisdom: Exploring Multidimensional Inequality in a Developing Country

REFERENCES

Anand, S and A K Sen. (2003). “Concepts of human development and poverty: a

multidimensional perspective”, in Readings in human development: concepts, measures and

policies for a development paradigm, ed. Fukuda-Parr and A.K. Shiva Kumar. New Delhi:

Oxford University Press.

Atkinson, A and F Bourguignon. (1982). “The Comparison of Multi-Dimensioned

Distributions of Economic Status”. Review of Economic Studies 49, no. 2: 183-201.

Atkinson, A and F Bourguignon. 1987. “Income Distribution and Differences in Needs”.

In Arrow and the Foundation of the Theory of Economic Policy, ed. G. R. Feiwel. London:

Macmillan.

Bazen, S and P Moyes. (2003), “Comparisons of Income Distributions across

Heterogeneous Populations.” In Research in Economic Inequality, Volume 9, ed. Bishop,

J.A.;Amiel, Y.

Blackbory, C and D Donaldson. (1978). “Measures of Relative Inequality and their

Meaning in terms of Social Welfare”. Journal of Economic Theory 18, no.1: 58-80.

Bourguignon, F. 1999. “Comment to: Multidimensional Approaches to Welfare Analysis

by E. Maasoumi.” In Handbook on Income Inequality Measurement, ed. J. Silber, Kluwer,

MA.

Chambaz, C. and E Maurin. (1998). “Atkinson’s and Bourguignon dominance criteria:

extended and applied to the measurement of poverty in France”. Review of Income and

Wealth 44, no.4: 497-513.

Cowell, F. (2000). “Measurement of Inequality”. In Handbook of Income distribution, ed.

A.B. Atkinson and F. Bourguignon. Amsterdam: Elsevier

CSO – Central Statistical Office of Zambia. (1994). “National Census of Agriculture

1990-92”

CSO – Central Statistical Office of Zambia. (1999, 2005). “Living Conditions Monitoring

Survey II/Living Conditions Monitoring Survey IV”

Deaton, A. (1997). The Analysis of Household Surveys: A Microeconomic Approach to Development

Policy. Johns Hopkins University Press.

Deutsch, J and J Silber. (2005). “Measuring Multidimensional Poverty: An Empirical

Comparison of Various Approaches”. The Review of Income and Wealth 51, no.1: 145-

174.

Duclos, J-Y,D Sahn and S D. Younger. (2001). “Robust Multidimensional Poverty

Comparisons”. Working Paper no. 0115, Université Laval, Département

d’Économie, Québec.

Hirschberg, J, E, Maasoumi and D.J Slottje. (1991). “Cluster Analysis for Measuring

Welfare and Quality of Life Across Countries”, Journal of Econometrics 50, no.1-2: 131-

150.

Jehle, G A and P J. Reny. (2001). Advanced Microeconomic Theory. Boston, MA: Addison-

Wesley.

Justino, P. (2004). “Multidimensional Inequality: An Empirical Framework”. PRUS

Working Paper no.23, Poverty Research Unit at Sussex, Department of Economics,

University of Sussex.

Kolm, S.C. (1977). “Multidimensional Egalitarianisms”, Quarterly Journal of Economics 91,

no.1: 1-13.

Lovell, K, S Richardson, P. Travers, and Wood, L. (2004). “Resources and Functionings:

A New View of Inequality in Australia”. In Models and Measurement of Welfare and

Inequality, ed. W. Eichhorn. Heidelberg: Springer-Verlag.

59


Chapter 2

Lugo, A.M. (2004). “On Multivariate Distributions of Well-Being: The case of Argentine

Provinces in the 1990s”, paper presented at the general conference of IARIW, Cork ,

Ireland , August 2004.

Maasoumi, E. (1986). ‘The Measurement and Decomposition of Multi-Dimensional

Inequality’. Econometrica 54, no. 4: 991-997.

Maasoumi, E. (1999). “Multidimensioned Approaches to Welfare Analysis”. In Handbook

of Income Inequality Measurement, ed. Silber, J. Boston: Kluwer Academic.

Maasoumi, E and G. Nickelsburg. (1988). “Multivariate Measures of Well-Being and an

analysis of inequality in the Michigan Data”. Journal of Business & Economic Statistics 6,

no. 3: 327-335

Maltzahn M. and Durrheim. (2008). “Is poverty multidimensional? A comparison of

income and asset based measures in five Southern African countries”. Social Indicators

Research, no 86:149-162

McCulloch, N, B. Baulch and M. Cherel-Robson. (2001). “Poverty, Inequality and

Growth in Zambia during the 1990s” WIDER Discussion Paper no. 2001/123,

Helsinki.

Muller, C. and A Trannoy. (2003). “Multidimensional inequality comparisons: a

compensation perspective”. Working Paper THEMA, University of Cergy-Pontoise,

France.

Myles, G D. (2002). Public Economics, London: Cambridge University Press

Quadrado, L, W Heijman and H Folmer. (2001). “Multidimensional Analysis of Regional

Inequality: The case of Hungary”, Social Indicators Research 56, no.1: 21-42

Ramos, X. and J Silber. (2005). “On the Application of Efficiency Analysis to the Study

of the Dimensions of Human Development”, Review of Income and Wealth 51, no.2:

285-310.

Sahn, D E. and D C. Stifel. (2000). “Poverty Comparisons Over Time and Across

Countries in Africa” World Development 28, no.12: 2123-2155.

Savaglio, E. (2005). “Three Approaches to the Analysis of Multidimensional Inequality

In Inequality and Economic Integration, ed. F.Farina and E.Savaglio. London: Routledge.

Sen, A K.(1973). On Economic Inequality, Clarendon Press, Oxford.

Tsui, K-Y. (1995). “Multidimensional Generalisations of the Relative and Absolute

Inequality Indices:The Atkinson-Kolm-Sen approach”. Journal of Economic Theory 67,

no.1: 145-157.

Tsui, K-Y. (1999). “Multidimensional Inequality and Multidimensional Generalised

Entropy Measures: An Axiomatic Derivation”. Social Choice and Welfare 16, no.1: 145-

157.

Trannoy, A. (2005). “Multidimensional Egalitarianism and the dominance approach: a

lost paradise”. In Inequality and Economic Integration, ed. F.Farina and E.Savaglio.

London: Routledge. 284-304.

UNDP. (1995). Human Development Report, New York and Oxford: United Nations

Development Programme

van Doorslaer, E and A M. Jones. (2003). “Inequalities in self-reported health: validation

of a new approach to measurement” Journal of Health Economics 22, no 1: 61-87.

World Bank. (2005). Technical Note # 3 on quantitative techniques for health equity analysis.

Washington DC: World Bank.

World Bank (2006), World Development Report: Equity and Development. New York:

Oxford

60


Chapter 3

Income Inequality and Health:

Exploring the Association in a Developing Country

3.1 INTRODUCTION

There is an on-going debate as to whether health is negatively affected by greater

levels of economic inequality within a society. This issue has received abundant

research interest in economics and other disciplines (cf. Rodgers 1979, Wilkinson

1992, Kawachi and Kennedy 1997, Mellor and Milyo 2001, Deaton 2003, Leigh and

Jencks 2007).

The idea that economic inequality causes poor health originates from an often

noted negative correlation between various income inequality measures and the

average health status of a population (e.g. Babones 2008, Ram 2006, Waldman 1992).

These aggregate associations, however, might stem from a non-linear relationship

between income and individual health, making it vital to use data indicative of the

situation of individuals (Gravelle 1998). Individual data, moreover, enables the

distinction between different theoretical hypotheses consistent with a negative

association between inequality and population health (Wagstaff and van Doorslaer

2000).

Empirical studies using individual level data have produced largely

contradictory conclusions. Some of the most consistent evidence of a negative

association between income inequality and individual health is found in analyses of

the situation in the United States (e.g. Kennedy et al. 1998, Subramanian et al. 2001,


Chapter 3

Lopez 2004). In contrast, studies focusing on other developed countries often reject

the hypothesis that income inequality has a direct detrimental effect on individual

health and mortality (e.g. Shibya et al. 2002 on Japan, Gerdtham and Johannesson

2004 on Sweden, Jones et al. 2004, Gravelle and Sutton 2009 on the UK), suggesting

that a general association between inequality and health does not exist. As stated by

Deaton (2003), judgments whether there is no such general relationship will, however,

depend a good deal on findings in different contexts than the ones hitherto studied.

Yet, we possess very limited knowledge of the relationship between inequality and

individual health in less-developed economies.

Figure 3.1 The cross-country correlation between income inequality and prevalence of stunting, 2000.

0 20 40 60

Prevalence of stunting, %

20 30 40 50 60

Net income Gini coefficient

Source: World Bank(2008) and Solt(2008)

Figure 3.2 Distinguishing between development levels.

0 20 40 60

Prevalence of stunting

20

Source: World Bank(2008) and Solt(2008)

30 40

Net income Gini coefficient

50 60

62


Income Inequality and Health: Exploring the Association in a Developing Country

Figure 3.1 displays the relationship between stunting and income inequality using

national Gini coefficients from various countries in 2000. 1,2 Although there is a good

deal of variation, it suggests that malnourishment in children is more prevalent in

more unequal societies, supporting findings from earlier research using aggregate

data. When the relationship is examined using different levels of development,

however, the relationship appears to stem from the association between inequality

and average child health status in developed countries. In fact, while figure 3.2 reveals

a positive correlation between income inequality and poor health among middle- and

high-income countries there is a negative correlation among low-income countries.

Several caveats apply if we would like to interpret these correlations, above

and beyond the use of aggregate data. For example, correlations do not take into

account income levels and other potentially important health determinants. In

particular we assume absolute income to be crucial in low-income contexts.

Moreover, if there are omitted factors that are of importance to health that in turn are

affected by inequality, the noted associations would be capturing the effect of these

mediators. There is also the issue of causality. If health status improves

disproportionately across a population, then average health may improve

concurrently with changes in income distributions as, healthier individuals are more

productive. 3 Nevertheless, the different relationships appearing across development

levels are interesting and deserving of further exploration. Whereas previous

individual level studies have analyzed the subject within middle- or high-income

countries, this chapter seeks to extend our knowledge of the relationship between

income inequality and individual health by extending the analysis to a less developed

context.

Zambia is classified as one of the poorest and most unequal countries in the

world with poor key indicators on health (WHO 2007). Using data from Zambia we

1 The prevalence of stunting refers to the share of children under 5 years of age with a height-for-age zscore

lower than -2. Height-for-age is a long-term chronic health indicator, generally acknowledged as a

good, objective indicator or children’s health status (Mosley and Chen 1984, WHO 1995).

2 N.B. Few OECD member countries (3 out of 30) report prevalence of stunting.

3 Another caveat is measurement errors. Inequality measures in Figure 3.1 and 3.2 come from Solt(2008)

who provides an alternative to commonly used cross-country inequality databases (for a discussion see

Chapter 4). However, Deaton (2003) emphasizes that conceptual and practical issues affect the

comparability of inequality measures across countries.

63


Chapter 3

test three hypotheses: the absolute income hypothesis (AIH) – stating that individual health

is determined by individual income and that the positive effect from higher income is

subject to diminishing return; the relative income hypothesis (RIH) – assuming that health

is influenced by the relation of individual income to the average income in a reference

group; and the income inequality hypothesis (IIH) - emphasizing that individual health

status is impacted by inequality in the distribution of income.

The dependent variable to be used in this analysis is height-for-age for

children. Research increasingly emphasizes the important role of child health as a

major factor influencing future economic outcomes (cf. Currie, 2009). As nutritional

status seems to be of particular importance in determining these outcomes it is

chosen to serve as our indicator. For example, stunted children suffer from elevated

risk of morbidity and mortality (Bengtsson and Lindström, 2000) and poor nutrition

seems to harm cognitive development (Maluccio et al. 2006). Moreover,

anthropometrical indicators are objective, relatively precise, and consistent across

subgroups (Heltberg 2009). To test the AIH we include household expenditures. At

the contextual level we test the RIH and the IIH by using average expenditures and

expenditure Gini coefficients calculated at three geographical levels; provincial,

district and constituency.

Taking account of complex survey design into account and using different

econometric techniques we confirm that more household economic resources

produce better individual health in a non-linear fashion. Moreover, although we find

some evidence in line with the RIH at the constituency level, different mechanisms

seem to be at work at different contexts as the association between average

expenditures and health is positive at the provincial level. Moreover, in contrast to

the IIH, but analogous to the above correlation between income inequality and

stunting among low-income countries, the analysis provides evidence of a positive

association between inequality and healthier children. This outcome is robust to

measuring inequality at different geographical levels, and to alternative inequality

measures and alternative specifications.

The next section of the paper reviews theoretical relationships between

income, income inequality and individual health and shortly reviews existing empirical

64


Income Inequality and Health: Exploring the Association in a Developing Country

results. Section 3.3 describes the data and discusses methodological choices. Section

3.4 presents the empirical results, while section 3.5 interprets the findings and

concludes the chapter.

3.2 THEORY AND PREVIOUS EMPIRICAL FINDINGS

Researchers within different disciplines have for long reported a negative correlation

between income inequality and indicators of population health (Rodgers 1979,

Wilkinson 1992, 1996, Babones 2008, Ram 2006). As revealed by Wagstaff and Van

Doorslaer (2000) various theoretical hypotheses concerning the relationship between

income, income inequalities and health at the individual level are consistent with this

observed association. 4 This section reviews the suggested theoretical relationships and

suggestions on how they may appear in developing contexts. As most of the literature

focus on adults we briefly discuss the proposed mechanisms through a child health

perspective.

3.2.1 Absolute income, relative income, income inequality and health

The absolute income hypothesis (AIH) states that additional monetary means improves

health but that the impact dimishies as an individual obtain greater levels of wealth.

As poor health will restrict the individual’s ability to earn income or accumulate

wealth, the causal link from income to health is debatable (c.f. Deaton 2003, Smith

1999). However, it is generally assumed that higher income improves health as more

resources can be devoted to health service or goods that are beneficial to one’s wellbeing.

Correspondingly, although higher incomes also might correlate with health

production “bads”, rich households are generally thought to be more efficient at

producing child health because of their ability to purchase greater quantities and

quality of health inputs and provide healthier environments, (Khanam et al. 2009). 5

4 There are hypotheses in addition to the three presented in this paper, e.g. the relative deprivation hypothesis

and the relative position hypothesis, both relating to the relationship between relative income and individual

health. As a result, the RIH has been defined differently in parts of the literature and is sometimes used as

a heading for all hypotheses. We follow the definitions in Wagstaff and van Doorslaer (2000).

5 Moreover education may mediate the household income-child health relationship. Education might also

be an unobserved factor that makes people both healthier and wealthier. If education and not income

matters to health, correlation between income and health is induced by effects of education on income

(Grossman, 1975, 2000).

65


Chapter 3

According to the AIH we therefore assume absolute income to play a significant role

in improving health in countries with a high incidence of poverty.

The relative income hypothesis (RIH) states that individual health is affected by the

individual’s economic situation relative to others in some reference group.

Consequently, holding individual income constant, average income in the reference

group will be negatively related to good health. The mechanism through which

relative income is assumed to harm health is psychosocial. Wilkinson (1996) emphasizes

that health status is determined by perceptions of place in the social hierarchy, i.e. the

relative income situation. Poorer individuals might for example feel stress, loss of

respect, distrust and shame when comparing themselves to their richer counterparts,

in turn affecting health and well-being. Perceptions could either translate directly into

physical afflictions through biochemical responses to stress and anxiety, increasing

the probability of disease (Brunner and Marmot 1999), or into unhealthy behavior

such as smoking (Lynch et al. 2000).

Psychosocial stress induced by the relative income situation can deteriorate

parental health and may also indirectly influence child health negatively (Olivius et al.

2004). Parental economic stress might transfer over to their children as parental

health problems may affect household income. Alternatively, or simultaneously,

parental stress may also cause deficient child care through such forms as lower levels

of stimulation to the child and decreased capacity for supportive parenting (Wachs,

2000). 6

The income inequality hypothesis (IIH) states that there is a direct negative effect

on individual health from income inequality, independent from the effect of income.

The first potential mechanism underlying the link between income inequality and

health relates to trust and social capital Inequality may create distrust at the individual

level, which translates into antisocial behavior and reduced civic participation at the

societal level. 7 Low social capital or lack of social connectedness may in turn have

6 These relations support the RIH and the AIH as parental economic stress does not necessarily have to be

the result of welfare comparisons, but rather a consequence of scarcity of economic resources per se.

7 Several empirical studies find income inequality to display a strong, negative correlation with the extent to

which people trust each other (see e.g. Knack and Keefer 1997, Gustavsson and Jordahl 2008).

66


Income Inequality and Health: Exploring the Association in a Developing Country

health consequences (Durkheim 1897, Putnam 2000, Kawachi et al. 2007). 8 For

example, socially integrated people have been shown to display greater

immunological resistance to certain diseases while social isolation correlates with

unhappiness (Grant 2000). Social capital may also promote child health (Morrow

2002, Berkman and Kawachi 2000). In particular, higher levels of maternal social

capital may improve child nutritional status by potentially permitting mothers greater

access to more services and assets, improving maternal health and promoting health

awareness (Baum 1999, De Silva et al. 2007) or by providing protection in times of

crisis (Harpham et al 2006, Cuny 1994). 9

A second mechanism potentially mediating the relationship between inequality

and health is political. Greater differences between rich and poor are assumed to

coexist with less common resources (e.g. public health care) in turn affecting

individual health (Kaplan et al. 1996, Lynch et al. 2000). Inequality may translate into

less public spending as large income differences often reflect heterogeneity in

interests between the rich and the poor (Krugman 1996, Alesina et al. 1999). In

developing contexts, public hospitals and a public infrastructure are generally found

to be important determinants of child health (c.f. Rajkumar and Swaarop 2008).

Underinvestment in common resources, in particular, can affect child nutrition as it

may affect maternity care, the number of child medical check-ups performed and

immunization rates. 10

As empirical evidence shows that income inequality and violence are

positively correlated (Demombynes and Özler 2005) violent crime is a third factor

potentially mediating the relationship between income inequality and health status.

Obviously, violence, by nature, can directly affect health, but it could also increase

stress among those worrying that they or someone they care about could become a

8 Social capital is defined differently in various paradigms. In general social capital can however be defined

as the sum of trust and respect of individuals in a society and by the features of social organization that

facilitate cooperation of mutual benefit. This characterization incorporates Putnam’s (2000) perspective, in

which social capital mainly relates to social trust, and Bourdieu’s (2007) view were it is primarily defined as

networks.

9 Crisis protection is likely of great importance to the physical condition of young children and infants who

rely on household’s strategies to ensure basic health needs.

10 The WHO (2007) states that poor feeding practices and infections often undermine child nutritional

status. These outcomes likely correlate with the access to and quality of maternity care and to whether

children are immunized.

67


Chapter 3

victim. Reasonably this mechanism will only be relevant to child health if there are

second-order effects such as higher parental stress.

Some scholars have discussed the relationship between income, inequality and

health in relation to development levels. Relating to the epidemiological transition,

Wilkinson (1996) states that income should have a stronger association with health in

less developed countries, while economic inequalities should be relatively more

important in developed ones. However, exploring the theoretical basis, Deaton

(2003) concludes that many of the arguments that income inequality is a health risk

are as plausible for poor as for rich countries and that there is no need to assume that

the relationship changes with economic development. Yet, Deaton (2003) emphasizes

that there is little to suggest that income inequality is important to individual health,

but that it is something that correlates with inequality. Building on this argument he

suggest that income inequality, conditional on average income, only is related to poor

health outcomes because inequality is effectively a measure of poverty.

3.2.2 What does the empirical evidence tell us?

Numerous empirical studies have related individual income or socioeconomic status

to individual health (cf. Ettner 1996). Generally this research demonstrates a positive

and non-linear association between income and health at the individual level, in line

with the AIH. A positive relationship between household income and child health is

also well documented in the child health literature, particularly in poor countries

where malnutrition is a phenomenon, although the precise mechanisms by which

income transmits to better health remain unresolved (Khanam et al. 2009). Moreover,

the literature examining potential health hazards from economic inequalities generally

confirm the AIH. For example, Mellor and Milyo (2002) and Karlsson et al. (2008)

find a concave relationship with self-assessed health (SAH) when controlling for

individual income and income squared in estimations.

The RIH has mainly been tested by examining the individual health impact of

average incomes of the geographical area where the individual reside. As noted by

scholars, the reference group to which the individual compare could evidently be

different (Deaton 2003, Miller and Paxon 2006). Concerning the RIH evidence is

68


Income Inequality and Health: Exploring the Association in a Developing Country

relatively weak, with several studies finding virtually no such effects (Lorgelly and

Lindley 2008, Li and Zhu 2006). Moreover, among the articles finding a significant

association, there is both evidence of a negative and a positive RIH. Using data on

the United States, Luttmer (2005) finds evidence of a negative RIH when employing

individual happiness as the dependent variable. Similarly, Karlsson et al. (2008)

cannot reject the RIH across development levels, although the reference group seems

to differ across economic contexts. While average incomes within the individual’s

own age-group negatively affect SAH in OECD countries, regional average incomes

leads to deteriorating health in less developed contexts. Gerdtham and Johanesson

(2004) and Miller and Paxson (2006), however, find evidence of average incomes to

reduce individual mortality in Sweden and the United States, respectively.

Most empirical studies testing the IIH with data at the individual level reflects

findings on the United States. Several of these studies find that income inequality

correlates with adverse health impacts. 11 For example, Kennedy et al. (1998), Lopez

(2004) and Subramanian and Kawachi (2001) conclude that income inequality is

detrimental to health when regressing SAH on confounding individual variables and

state level inequality. Moreover, there is evidence that individuals living in states with

higher income inequality are at increased risk of mortality, hypertension and having

harmful levels of BMI (Lochner et al. 2001, Diaz-Roux et al. 2000). However, Chang

and Christiakis (2005) find an inverse association between inequality and weight

status, and controlling for state-specific effects Mellor and Milyo (2002) do not find a

significant association between income inequality and SAH in the USA.

Moreover, studies within other developed countries generally reject the IIH.

For example, Shibya et al. (2002) conclude that income inequality does not have a

detrimental effect on self-rated health in Japan. Blakely et al (2006), Gerdtham and

Johannesson (2004) and Jones et al. (2004) come to similar in studies of mortality

within New Zealand, Sweden and the UK, respectively. Moreover, Lorgelly and

Lindeley (2008) and Gravelle and Sutton (2009) reject the IIH using panel data on

11 Several articles review the existing literature in this research field e.g. Wagstaff and van Doorslaer (2000),

Mellor and Milyo (2001), Deaton (2003), Submaranian and Kawachi (2004) and Wilkinson and Picket

(2007). For a review specifically of the literature using individual level data see Subramanian and Kawachi

(2004).

69


Chapter 3

SAH in Britain. 12 Using a cross-national individual level panel, however, Hildebrand

and van Kerm (2009) find consistent evidence of a negative effect from income

inequality on good SAH in Europe. Similarly, testing the IIH using individual level

data from 21 countries, Karlsson et al. (2008) locate a negative association between

within-country inequality and good SAH in OECD countries. When examining the

relationship among respondents in countries at lower development levels, however,

the negative health impact from inequality disappears. 13

Four studies test the IIH using data within middle-income countries and all

come to similar conclusions. In Chile, community inequality is found to have an

independent impact on the probability of reporting poor health (Subramanian et al.

2003). Work on child health in Ecuador, moreover, supports the IIH when income

inequality is measured at the provincial level, but not at smaller geographical areas

(Larrea and Kawachi 2005). Testing the IIH within China, Li and Zhu (2006) find

that the negative relationship between income inequality and SAH exists in

communities with a relatively high degree of inequality. Similar results are also

confirmed among men in Russia (Carlson 2005). To our knowledge, the relationship

between inequality and individual health has not been tested within a low-income

country.

3.3 DATA AND EMPIRICAL MODEL

3.3.1 Data

To test the three hypotheses we use the 2004 Zambian Living Condition Monitoring

Survey (LCMS IV). This is the fourth such survey conducted by the Central Statistical

Office (CSO) in Zambia since 1996. Carried out from October 2004 to January 2005,

the survey provides nationwide reporting on indicators such as health, education,

consumption and expenditures and food production for more than 103 200

individuals in 19340 households. The number of observations represents a sampling

12 The contradictory evidence on the IIH has also generated a discussion whether the lack of association

between inequality and individual health reflects the existence of an inequality threshold, i.e. that the

phenomenon should be restricted to relatively unequal societies (Subramanian and Kawachi 2004).

13 Except for respondents living in India, individuals in this sub-group reside in middle-income countries.

70


Income Inequality and Health: Exploring the Association in a Developing Country

fraction of about one household for every 110 household in the population. The

survey design employed by the CSO is a multiple stage sample selection process,

where clusters and households within clusters are randomly selected. Moreover the

survey sample is stratified. The sampling frame of the survey was developed from the

Census of Copulation and Housing carried out in 2000.

3.3.2 Variables

Health Status

The LCMS IV contains health and nutritional data for all children aged 0-59 months

in surveyed households, including information on physical body measurements.

Using this information we derive our dependent variable, the anthropometric

measure height-for-age. Height-for-age, usually referred to as stunting, is a long-term

health indicator reflecting the accumulation of health and nutrition over a child’s

entire lifetime (Falkner and Tanner, 1986). Unlike other anthropometric measures,

height-for-age is generally not affected by acute episodes of poor nutritional intake or

sickness that might have occurred around the time of measurement. The measure is

acknowledged as a good, objective indicator of children’s general health status

(Mosley and Chen 1984, WHO 1995), and of a household’s well-being (Thomas et al.

1991, Zere and McIntyre, 2003). 14

We derive the individual height-for-age measures expressed in the form of zscores,

which compares the height of a child with that of a child of the same age and

gender from a healthy reference population consisting of healthy individuals (WHO,

1995). 15 As research indicates that the height distributions of healthy individuals,

regardless or racial or ethnic makeup, are comparable around the world until the age

of five (Habicht et al. 1974, Graitcher and Gentry 1981), we derive the variable for

children younger than 60 months. The height-for-age z-score (HAZ) for individual i

is formulated as:

14 As height, length and weight are measured by survey enumerators, anthropometric indices are not

susceptible to self-reported bias. Errors in measurement, therefore, are unlikely to be correlated with socioeconomic

characteristics of the household where children reside.

15 The dependent variable was derived by using the anthropometric statistical software Epi-Info. We use

the sex specific 2000 CDC normalized version of the NCHS reference.

71


Chapter 3

HAZi

xi

=

− xmedian

σ x

where i x is the height for child i; xmedian is the median height for a healthy and well-

nourished child from a reference population of the same age and gender, and σ x is

the standard deviation from the mean of the reference population. Children with a

HAZ more than 2 and 3 standard deviations below the international referenced

median are conventionally defined as stunted and severely stunted, respectively

(WHO, 1995).

Following WHO (1995) recommendations we exclude observations with a zscore

lower than -6 or larger than +3. 16 Children younger than three months are also

excluded as their health status might possibly be explained by the weight of the

mother (Skoufias, 1998). Following these guidelines we generate a sample of 10,316

children with valid information on health status. 17

Income, Mean Income and Income Inequality

As income in developing countries often is subject to seasonal fluctuations,

household expenditure is generally viewed as a better measure of long-run resource

availability in the context we study. To test the AIH we include the log of household

monthly per capita expenditures in our estimations. In testing the RIH we include a log of

average expenditures per capita in the geographical area where respondents reside.

Moreover, we use three measures of expenditure inequality to examine the IIH. Our

primary inequality measure is the Gini coefficient which takes a value between 0 and

1, where 0 implies complete equality and 1 complete inequality respectively. As a test

of sensitivity we also use Theil’s two inequality indices, GE(0) and GE(1). In line with

the IIH we expect the regression coefficient to be negative in our estimations using

these measures, as more inequality should be accompanied by a lower value of the

dependent variable (i.e. poorer health).

16 As the HAZ mean in our sample is below -1.5 we follow recommendations on a flexible exclusion range.

The lower limit of -6 is based on the recommended formula 4*s.d of the sample.

17 We drop 1797 observations from the original sample due to either (a) incomplete information

concerning one or more of the variables needed in the calculations of the standardized height-for-age

health indicator (gender, age in months, height), (b) too high or low z-scores or (c) child being younger

than three months old. This represents a drop of 14.8 per cent of the original sample.

72


Income Inequality and Health: Exploring the Association in a Developing Country

As it is not clear at what aggregation level economic inequalities matter to

health we test the RIH and the IIH at three different aggregation levels – province,

district and constituency. Zambia is demarcated into nine provinces, which are

further divided into 72 districts and 155 constituencies. Moreover, as changes in

income distributions are unlikely to have an instantaneous health impact (Mellor and

Milyo 2002) we examine the association between inequality and health under different

assumptions about lag periods, using data on current as well as past inequality levels.

Inequality measures in a previous time-period are derived from the LCMS II survey. 18

The variables average per capita expenditures and expenditure inequality are calculated from

the full data set rather than just the sample on households with children. To correct

for differential representation sample weights are applied in these calculations. As

Gini, GE(0), GE(1) and average expenditures are contextual, values differ across

geographical areas, but not across individuals within an area.

Additional Control Variables

To assess biological characteristics of importance to health outcomes we use the

variables gender and age. Research on gender bias and nutrition in developing countries

has come to mixed conclusions. While girls seem to be disadvantaged in Asia, the

opposite seems to hold in African contexts with worse nutrition for boys than for

girls (Svedberg 1990, Madise et al. 1999). Gender is specified as a binary variable

where 1 indicates female and 0 male. Age is measured in terms of months and ranges

from 3 to 59. As evidence points to increasing prevalence of stunting during the first

two years of life before stabilizing, we also include age square. Child health research

has established a relationship between birth order or birth intervals and nutritional

status. Although the underlying mechanisms influencing nutritional status remain

unsolved, children with a higher birth order are generally at higher risk of being

malnourished (Behrman 1988). As a final individual element we derive the birth order

of the children in our sample, based on the sequence of all siblings within a

household, excluding those who had already died.

18 The LCMS II was carried out in 1998. The survey has nationwide reporting, covering 93417 individuals

in 16710 households. Using this data set we also derive the income inequality measures of particular

interest employing sample weights.

73


Chapter 3

A mother’s level of education is a social characteristic that we assume

correlates positively with child health (Kahn et al., 2002). Moreover, as education is

related to family decision-making processes and child feeding practices (Mosley,

1984), the educational level of other household members can be a strong determinant

of nutritional status. We use two variables related to the stock of human capital

within a household in our models: the educational level of the mother to the child, and the

maximum level of education of any person in the household older than 12. Education is

categorized into one of four levels of schooling corresponding to the educational

system in Zambia: none (no years of schooling), primary (1-7 years of schooling),

secondary (8-12 years of schooling) and higher education (more than 12 years of

schooling).

To assess family structure and social support, estimations include information

on marital status of the mother. As women living with the father of the child can

receive more support and are less likely to engage in risky behaviors (Albrecht et al.,

1997) we assume this indicator to correlate positively with child health. We also

control for the age of the mother in trying to capture her experience. On the household

level we moreover take account of the household size and whether the household

resides in a rural or an urban area. The literature suggests that children living in urban

centers in developing countries have better growth and nutritional status than their

rural counterparts (Skoufias 1998).

Theories on household economics suggest that a household’s composition

and characteristics affect intra-household allocations, which may have nutritional

consequences (see e.g. Becker 1964, 1965, Behrman et. al. 1986). Empirical studies

contrasting child malnutrition in male- and female-headed households provide

contradictory evidence. On one hand, as the earnings in female-headed households

are generally lower in those headed by men, lower levels of child health may result

within the former group. The same conclusion could potentially also be drawn if a

majority of the household is composed of women. On the other hand it is possible

that that a mother’s control of resources is associated with a more equitable

distribution of available food within a household, implying better nutritional status of

children (Thomas, 1994). In order to take account for this, we use two binary

74


Income Inequality and Health: Exploring the Association in a Developing Country

variables to proxy household composition and characteristics: household head, where 0

refers to male-headed and 1 to female-headed households, and household gender share,

corresponding to the percentage of female household members above 12 years of

age.

Malnutrition is commonly caused by an inadequate availability of energy and

proteins. Two dummy variables, meals and animal products, are derived with the former

taking on a value of 1 if the household normally have more than two meals a day

(excluding snacks); and the latter equaling 1 if the household eats fish, poultry or

animal products more than once a week. 19 The frequency with which a household

eats animal products, moreover, can be a proxy of dietary diversity which has been

associated with improved nutritional status for young children (Arimond and Ruel,

2004). 20 In addition, as we believe sanitary standards might affect nutritional status we

derive two binary household variables to proxy the standard of water and toilet facilities.

Finally, empirical studies indicate that the estimated impact of income and

education on health are biased if the role of community factors, for example access to

health facilities, is not taken into account (e.g. Attanasio et al. 2004). As understood

from the discussion in section 3.2.1, however, the degree of inequality this might

affect the level of this accessibility. If the effect of inequalities on health is mediated

through a political mechanism it seems questionable to include proxies for the access

to health care along with an inequality variable in a first step. For the same reason we

do not include variables on stress, social capital or violence.

3.3.3 Empirical model and estimation methods

The empirical model, where individuals are indexed by i, the households where

individuals i live by j, and geographical level where household j reside by g, is

formulated as:

HAZ α + v 'β

+ x 'β

+ z 'β

+ ε

i = i i j j g g

19 Fish, meat and soya beans are provisions generally referred to as protein-rich (FAO, 2006). The choice

of delimitations to more than two meals a day and the consumption of animal products more than once a

week is arbitrary, but intend to capture regularity. According to World Bank (2005) estimates, a typical

poor person in Zambia consumes a small serving of chicken, beef or fish every 3-4 days. Importantly, our

results are not sensitive to this cut-off point.

20 As these binary variables are measured at the household level, we cannot control for intra-household

differences with respect to food intake. Consequently the proxies relate to how often and what the children

(and their parents) may have eaten.

75


Chapter 3

The dependent variable HAZ is increasing with better child health. The vector v

corresponds to individual characteristics of children, the vector x relates to features

of households where children live, including household expenditures, while the

vector z corresponds to the contextual factors of interest. Although there are reasons

to distinguish between the AIH, RIH and IIH in empirical analysis, the hypotheses

are not necessarily mutually exclusive. Each of them could be a partial explanation to

the aggregate relationship between inequality and health suggesting the health impact

of individual income, average income and income inequality can be tested in the same

specification. The residual termε includes unobserved child, household and

community characteristics.

First we estimate the hypothesized relationships using OLS. 21 As clustering

and stratification will typically skew standard errors, complex survey design features

are taken into account. 22 It also seems important, however, to seek to control for

potential endogeneity between household expenditures and the dependent variable.

In terms of child health, parental or household behavioral responses may cause a

negative correlation between consumption and the regression error term resulting

from families spending more money on health care, medicines or food consumption

when children are ill, or a positive correlation resulting from bad child health

translating into low parental incomes (Attanasio et al. 2004). Consumption

expenditures may also be subject to measurement errors correlated to the error term.

In an attempt to solve the potential endogeneity problem we use instrumental

variable techniques. A common identification strategy is to consider household assets as

valid instruments. This approach assumes that decisions over household assets do not

depend on or respond to shocks in child health. As a result we need to avoid using

assets related to the dietary quality, assets that may correlate to the access of health

related information, or assets influencing household access to health care (Thomas et

21 Although the HAZ variable is amenable to linear regression analysis (c.f. Thomas et al. 1991, Alderman

et al. 2000) some empirical work on nutritional status use non-linear estimation techniques, such as logit or

probit, using a dependent variable refering to whether or not a child is malnourished. We prefer estimating

the subject relationship using linear regression analysis also using information on the depth of malnutrition.

22 We make use of the Stata command svyset, which allows for specification of stratification scheme and

primary sampling unit (the LCMS IV has eight strata and sample size of 1048 PSUs). While clustering likely

reduces the precision of sampling estimates, as households living in the same cluster usually are more

similar to another in behavior and characteristics, stratification will likely enhance it (Deaton, 2000).

76


Income Inequality and Health: Exploring the Association in a Developing Country

al., 1991). Following Skoufias (1998) and Lawson (2004) we use a set of identifying

instruments consisting of the value of electrical goods (excluding radio and TV) and the

type of energy used for lighting in the household.

As the causal processes affecting our dependent variable are assumed to

operate at more than one level simultaneously we test the robustness of our results

using a two-level random intercept model. 23 The method allows estimation of the fraction

of variance of the dependent variable occurring at each level of the analysis and

adjusts for spatial correlation and heteroscedasticity. This might be important as the

nutritional status of children and infants living in the same families and

neighborhoods can be correlated due to common household and neighborhood

influences (Hox 2002). The empirical model is specified as above with the difference

being that the residual term consists of components referring to the individual and

the household, and the contextual level. The residuals are assumed to be uncorrelated

and have normal distributions with zero mean and variances.

3.4 EMPIRICAL ANALYSIS

3.4.1 Descriptive analysis

Despite considerable efforts to improve the health situation over the past decade, key

health indicators in Zambia are very poor. Life expectancy at birth decreased from

45.8 years in 1990 to 38.4 years in 2005 and the mortality rate of children under five

increased from 180 to 182 (per 1000 individuals) in the same time period (WDI,

2008). In terms of malnutrition the Zambian situation is also unfortunate. Table 3.2

indicates that almost every second child in the sample suffers from stunted growth,

and every fourth child is severely stunted, corresponding to a very high degree of

malnutrition across the population when using the WHO classification (WHO, 2005).

Figure 3.3 illustrates a histogram of the current HAZ distribution and the

corresponding distribution for the healthy reference group. Clearly, the average value

23 In single level multiple regression analysis unexplained variability is only the variance of the residual

term. Consequently, data sets with a nesting structure that includes unexplained variability at each level

might be less adequately represented by e.g. OLS (Hox, 2002).

77


Chapter 3

of the health variable (-1.88) is significantly lower than the mean value of the healthy

population (0). Moreover, examining the mean z-score by age-groups in table 3.4 it

appears that stunting displays prevalence in children as young as ten months old. This

is true irrespective of whether the exercise is performed on the full sample or

separately for boys and girls, although average height-for-age z-scores are somewhat

lower among the former group, implying a higher prevalence of stunting. At the age

of 25 months the health variable stabilizes around - 2. 24

Table 3.2 Prevalence of stunting

Variable Mean S.D % below -2 S.D % below -3 S.D n

HAZ -1.88 1.81 47.30 25.97 10316

HAZ, boys -1.95 1.79 48.78 26.92 5128

HAZ, girls -1.80 1.82 45.83 25.02 5188

Author's calculations using LCMS IV. Calculations are made using sample weights.

Figure 3.3 The HAZ distribution

Density

0 .1 .2 .3 .4

-6 -4 -2 0 2 4

z-score

24 Clearly the height-for-age indicator can be affected by long periods of droughts or other natural

catastrophes which could decrease the dispersion in the health distribution as the most undernourished

children may pass away. We have found no reports of preceding periods of common severe hardship in the

country along these lines.

78


Figure 3.4 Malnutrition and age

-2.5 -2 -1.5 -1 -.5

Mean z-score (height-for-age)

Income Inequality and Health: Exploring the Association in a Developing Country

5 10 15 20 25

Age in months

As in many developing countries, household per capita expenditures are on average

low and monetary poverty levels high. In 2004, 68 per cent of the Zambian

population had a consumption level below the national poverty line. In particular the

poverty-situation was more severe in rural than urban areas and examining regional

average expenditure levels it appears there is a large variation across geographical

areas. With regard to the distribution of monetary means, Zambia is often referred to

as one of the most unequal societies in the world. Table 3.3 confirms that

expenditure inequality at the national level is high and indicates that inequality slightly

increased in the late 1990’s.

Table 3.3 Expenditure inequality

1996 1998 2004

Gini 0.518 0.533 0.544

GE(0) 0.485 0.533 0.551

GE(1) 0.587 0.596 0.738

Source: Author's calculations using LCMSIII, LCMS IV and McCulloch et al. (2000)

Table 3A in the Appendix provides summary statistics for variables in our analysis.

Most individuals and households in the sample reside in rural areas, although the

relatively high degree of urbanization in Zambia is mirrored. Naturally the mean

household size of the sample is larger than the national average as we are solely

studying households with children and as expected the maximum level of education

in the household exceeds the maternal education level. Some of the children in the

79


Chapter 3

sample do not live with their biological mother, which potentially indicates that the

biological mother is not alive. As a result the number of observations decreases from

10,316 to 9,752 when including explanatory variables referring to the biological

mother in the regression analysis.

3.4.2 Regression analysis

Previous individual level studies come to different conclusions with respect to the

RIH and the IIH depending on the geographical scale at which contextual variables

are assessed. Table 3.3, 3.4 and 3.5 present baseline regression estimates using OLS

when inequality and average expenditures are aggregated at the provincial, the district

and the constituency level, respectively.

The empirical analysis shows that household expenditure is an important

determinant of children’s nutritional status. In line with the AIH the logged

expenditure coefficient is positive across all specifications, suggesting that more

economic resources correlate with better child health in a non-linear fashion. Without

confounding with other household characteristics, a 10 percent increase in household

per capita expenditures would increase HAZ by more than 1, corresponding to a

substantial nutritional improvement for most children in the sample. Controlling for

additional household variables somewhat decreases the magnitude of the association,

suggesting that the relationship between household expenditures and child health is

partially mediated through for example food consumption. The same relationship

holds for specifications including information on the biological mother, somewhat

decreasing the sample.

The baseline analysis does not provide any evidence of the RIH. In fact,

holding household and individual characteristics constant, children in households

residing in richer provinces are less malnourished. Following the discussion in

Gerdtham and Johannesson (2004) and Miller and Paxson (2006) the protective

health effect from provincial mean expenditures could indicate a potential positive

spill-over effect on disadvantaged children from living among richer households,

from better provision of public goods or better environmental quality. An indication

of a positive relationship also appears when evaluating average expenditures

80


Income Inequality and Health: Exploring the Association in a Developing Country

measured at the district and constituency level, but these results are not robust to the

inclusion of household characteristics.

Examining the association between expenditure inequality and child health we

arrive at an interesting finding. Testing the health impact from present levels of

inequality at the district and constituency level, the regression coefficients are

negative, in line with the IIH, but far from being significant. However, replacing the

inequality variable with information from a previous time period the estimation

results provides evidence that expenditure inequality correlates positively with child

nutrition. Moreover, evaluating the relationship at the provincial level both present

and lagged inequality levels associate with better, rather than worse, health outcomes.

Comparing the size of the regression coefficients of present and lagged inequality

levels in table 3.3 it moreover appears that the positive health impact is larger when

allowing for some time of exposure. This indication supports the view that the main

health impact from inequality is not instantaneous and suggests that mediating factors

requires a certain time period to affect health status (Subramanian and Kawachi 2004,

Mellor and Milyo 2002, 2001).

In sum, counter to the IIH, we find a statistically inverse association between

inequality and individual health when allowing for some time of exposure, despite

adjustment for individual and household level covariates. Moreover, in contrast to

previous empirical work reporting income inequality to have a health impact, there is

evidence of a significant association also at lower geographical levels.

81


Chapter 3

Table 3.3 Regression estimates - provincial level – OLS

(1) (2) (3) (4) (5) (6) (7)

Gender 0.204*** 0.203*** 0.200*** 0.200*** 0.201*** 0.189*** 0.190***

[0.035] [0.035] [0.035] [0.034] [0.034] [0.035] [0.035]

Age -0.050*** -0.051*** -0.051*** -0.053*** -0.053*** -0.058*** -0.058***

[0.005] [0.005] [0.005] [0.005] [0.005] [0.005] [0.005]

Age^2 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

Birth order 0.010 0.006 0.006 -0.069*** -0.068*** -0.080*** -0.080***

[0.014] [0.014] [0.014] [0.018] [0.018] [0.019] [0.019]

Household head 0.010 0.005 0.032 0.025

[0.065] [0.065] [0.082] [0.082]

HH gender share (adult) 0.227* 0.230* 0.198 0.207

[0.120] [0.120] [0.129] [0.128]

HH size 0.044*** 0.043*** 0.045*** 0.043***

[0.009] [0.009] [0.009] [0.009]

Rural -0.197*** -0.177*** -0.183*** -0.161***

[0.059] [0.060] [0.058] [0.059]

HH edu1 0.269** 0.249**

[0.123] [0.124]

HH edu2 0.340*** 0.305**

[0.127] [0.128]

HH edu3 0.516*** 0.504***

[0.145] [0.147]

Water -0.012 -0.027 -0.032 -0.050

[0.046] [0.045] [0.047] [0.046]

Toilet facility -0.004 0.000 -0.009 -0.006

[0.065] [0.064] [0.066] [0.065]

Meals 0.182*** 0.171*** 0.163*** 0.150***

[0.044] [0.043] [0.044] [0.044]

Animal products 0.118*** 0.110** 0.120*** 0.114**

[0.045] [0.045] [0.046] [0.046]

Mother's age 0.015*** 0.016***

[0.003] [0.003]

Marital status 0.09 0.089

[0.067] [0.066]

Mother's edu1 0.091 0.071

[0.063] [0.063]

Mother's edu2 0.306*** 0.285***

[0.074] [0.074]

Mother's edu3 0.551*** 0.545***

[0.131] [0.132]

HH expenditures per capita 0.168*** 0.124*** 0.130*** 0.045* 0.051** 0.065** 0.049*

[0.024] [0.025] [0.025] [0.026] [0.025] [0.026] [0.026]

Average province expenditures 0.650*** 0.636*** 0.494*** 0.583*** 0.505*** 0.598***

[0.094] [0.094] [0.101] [0.102] [0.101] [0.102]

Gini province 1.784** 1.947** 2.016***

[0.744] [0.775] [0.773]

Gini province t-1 3.263*** 3.430***

[0.708] [0.730]

Constant -1.854*** -9.228*** -10.020*** -8.592*** -10.273*** -8.963*** -10.755***

[0.142] [1.083] [1.163] [1.256] [1.290] [1.265] [1.297]

Observations 10316 10316 10316 10316 10316 9752 9752

R-squared 0.03 0.04 0.04 0.06 0.06 0.06 0.07

Complex survey design accounted for. Standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%

82


Table 3.4 Regression estimates – district level – OLS

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

Gender 0.205*** 0.205*** 0.202*** 0.203*** 0.191***

[0.035] [0.035] [0.035] [0.035] [0.035]

Age -0.050*** -0.052*** -0.050*** -0.052*** -0.057***

[0.005] [0.005] [0.005] [0.005] [0.005]

Age^2 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000]

Birth order 0.007 -0.071*** 0.008 -0.070*** -0.081***

[0.014] [0.018] [0.014] [0.018] [0.019]

Household head 0.01 0.004 0.026

[0.065] [0.065] [0.082]

HH gender share (adult) 0.215* 0.221* 0.195

[0.120] [0.120] [0.129]

HH size 0.046*** 0.046*** 0.047***

[0.009] [0.009] [0.009]

Rural -0.222*** -0.220*** -0.209***

[0.063] [0.063] [0.061]

HH edu1 0.297** 0.292**

[0.124] [0.125]

HH edu2 0.363*** 0.358***

[0.129] [0.129]

HH edu3 0.540*** 0.526***

[0.147] [0.147]

Water 0.003 0.005 -0.017

[0.046] [0.046] [0.046]

Toilet facility -0.075 -0.055 -0.06

[0.063] [0.063] [0.065]

Meals 0.248*** 0.233*** 0.213***

[0.043] [0.043] [0.043]

Animal products 0.064 0.074 0.076

[0.045] [0.045] [0.047]

Mother's age 0.015***

[0.003]

Marital status 0.094

[0.067]

Mother's edu1 0.089

[0.064]

Mother's edu2 0.295***

[0.076]

Mother's edu3 0.539***

[0.136]

HH expenditure per capita 0.136*** 0.059** 0.144*** 0.065** 0.064**

[0.025] [0.026] [0.025] [0.026] [0.027]

Average district expenditures 0.254*** 0.071 0.263*** 0.079 0.073

[0.079] [0.085] [0.081] [0.087] [0.088]

Gini district -0.618 -0.584

[0.520] [0.507]

Gini district t-1 0.943*** 0.680** 0.867***

[0.319] [0.310] [0.320]

Constant -4.345*** -2.369** -5.261*** -3.131*** -3.343***

[0.917] [1.007] [0.956] [1.043] [1.060]

Observations 10316 10316 10316 10316 9752

R-squared 0.03 0.05 0.03 0.05 0.06

Complex survey design accounted for. Standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%


Chapter 3

Table 3.5 Regression estimates – constituency level – OLS

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

Gender 0.204*** 0.205*** 0.201*** 0.203*** 0.192***

[0.035] [0.035] [0.035] [0.035] [0.035]

Age -0.050*** -0.052*** -0.050*** -0.052*** -0.057***

[0.005] [0.005] [0.005] [0.005] [0.005]

Age^2 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000]

Birth order 0.006 -0.071*** 0.007 -0.070*** -0.081***

[0.014] [0.018] [0.014] [0.018] [0.019]

Household head 0.008 0.004 0.027

[0.065] [0.065] [0.082]

HH gender share (adult) 0.217* 0.222* 0.194

[0.121] [0.120] [0.129]

HH size 0.047*** 0.046*** 0.047***

[0.009] [0.009] [0.009]

Rural -0.228*** -0.232*** -0.221***

[0.062] [0.062] [0.061]

HH edu1 0.302** 0.303**

[0.124] [0.124]

HH edu2 0.368*** 0.370***

[0.128] [0.129]

HH edu3 0.541*** 0.535***

[0.147] [0.147]

Water 0.009 0.007 -0.015

[0.046] [0.046] [0.046]

Toilet facility -0.071 -0.048 -0.055

[0.063] [0.064] [0.065]

Meals 0.248*** 0.239*** 0.222***

[0.043] [0.043] [0.043]

Animal products 0.06 0.062 0.063

[0.045] [0.045] [0.046]

Mother's age 0.015***

[0.003]

Marital status 0.095

[0.067]

Mother's edu1 0.096

[0.064]

Mother's edu2 0.301***

[0.076]

Mother's edu3 0.548***

[0.136]

HH expenditure per capita 0.133*** 0.063** 0.141*** 0.067** 0.065**

[0.025] [0.026] [0.026] [0.026] [0.027]

Average constituency expenditures 0.194*** 0.018 0.202*** 0.025 0.019

[0.070] [0.073] [0.071] [0.073] [0.074]

Gini constituency -0.649 -0.388

[0.435] [0.426]

Gini constituency t-1 0.756*** 0.620** 0.677***

[0.247] [0.243] [0.248]

Constant -3.631*** -1.881** -4.426*** -2.478*** -2.603***

[0.794] [0.845] [0.809] [0.865] [0.875]

Observations 10316 10316 10316 10316 9752

R-squared 0.03 0.05 0.03 0.05 0.06

Complex survey design accounted for. Standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%

84


2SLS

Income Inequality and Health: Exploring the Association in a Developing Country

Performing Durbin-Wu-Hausman tests we find evidence of endogeneity of

household per capita expenditures implying that initial results could be biased. As a

result we re-estimate the full model employing 2SLS, where instruments refer to the

value of electrical goods and type of energy used for lighting. Before examining the

outcomes from this exercise, we test the relevance and validity of instruments. The

Anderson canonical correlation LR test is rejected in all specifications. Moreover,

instruments seem to be uncorrelated to the error term and excluded instruments are

correctly excluded as p-values of Sargan’s test of overidentifying restrictions are large.

The regression coefficient of household expenditures per capita remains

significantly positive and similar across specifications. IV estimates in table 3.6 are in

general larger than OLS estimates. This suggests that the latter is biased downward,

potentially following from households spending resources for children who are ill.

Reassuringly, the baseline indication on the association between expenditure

inequality and health is also robust to this exercise, as higher inequality in an earlier

time period correlate to better child health outcomes at all aggregation levels. With

respect to current inequality levels the relationship is only significant at the provincial

level.

The previous findings regarding the RIH change slightly when we employ the

IV-technique. On one hand the positive relationship between provincial level average

expenditures and better health in baseline estimations remains robust (column 1 and

2). On the other hand, there is some evidence in line with the RIH when the

constituency level constitutes our reference group, although the significant

association is not robust to the inclusion of lagged inequality levels. Also the

coefficient of average district expenditures is negative, but the correlation is

insignificant. That the reference group matters for the sign and significance of relative

expenditures on health is in line with previous work (Osler et al. 2003, Gravelle and

Sutton 2009). Interestingly the negative relationship between average expenditures

and child health merely appears at the lowest aggregation level, suggesting this is

where a psychosocial mechanism potentially is at work.

85


Chapter 3

Control variables

Examining the role of the various control variables, we confirm many of the findings

in the child-health literature. The relation between the age of children and height-forage

seems to be convex, in line with the previously noted indication that average

health status decreases sharply with age until children turn two years old. Moreover,

the dummy on gender is positive and significant in all estimations, supporting the

results in Madise et al. (1999). That female children generally appear to be better

nourished than their male counterparts may relate to differential feeding practices and

gender discrimination in favor of female infants in the allocation of food, but it is also

consistent with recent findings on that boys are more vulnerable to health shocks

early in life. In support of existing evidence, the empirical analysis also show that

higher birth order is significantly associated with poorer nutritional status. Following

Behrman (1988) this could either be explained by behavioral factors, with younger

children in the company of older siblings receiving benign neglect from parents,

potentially related to an adjustment between quantity and quality, or by competition

between children within the same household.

Regarding household variables the coefficient of the rural dummy is negative

and generally significant through OLS and IV estimations, suggesting that children

residing in the countryside have poorer nutritional status on average. Moreover,

household size is positive and significant, possibly capturing an effect of child care

within the extended family. However, neither the gender of the household head nor

the marital status of the mother has a significant impact on child nutritional status. As

expected children in households having more meals per day are less malnourished,

but the intake of animal product does not seem to significantly matter to child health.

86


Table 3.6 Regression estimates – 2SLS

Income Inequality and Health: Exploring the Association in a Developing Country

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

Gender 0.205*** 0.206*** 0.214*** 0.211*** 0.216*** 0.214***

[0.035] [0.035] [0.035] [0.035] [0.036] [0.036]

Age -0.053*** -0.053*** -0.052*** -0.052*** -0.052*** -0.052***

[0.005] [0.005] [0.005] [0.005] [0.005] [0.005]

Age^2 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

Birth order -0.058*** -0.059*** -0.054** -0.054** -0.049** -0.052**

[0.021] [0.021] [0.021] [0.021] [0.021] [0.021]

Household head 0.048 0.036 0.063 0.053 0.061 0.059

[0.073] [0.072] [0.074] [0.073] [0.075] [0.074]

HH gender share (adult) 0.192 0.2 0.165 0.172 0.181 0.17

[0.125] [0.124] [0.127] [0.127] [0.129] [0.127]

HH size 0.058*** 0.054*** 0.067*** 0.066*** 0.075*** 0.069***

[0.015] [0.015] [0.016] [0.015] [0.015] [0.016]

Rural -0.138* -0.127 -0.162** -0.163** -0.187** -0.173**

[0.077] [0.078] [0.073] [0.074] [0.073] [0.073]

HH edu1 0.260** 0.239* 0.290** 0.280** 0.301** 0.298**

[0.126] [0.127] [0.129] [0.130] [0.131] [0.129]

HH edu2 0.293** 0.264* 0.299** 0.293** 0.316** 0.305**

[0.136] [0.137] [0.138] [0.140] [0.140] [0.139]

HH edu3 0.380** 0.388** 0.341* 0.339* 0.358* 0.340*

[0.190] [0.192] [0.195] [0.196] [0.196] [0.193]

Water -0.05 -0.06 -0.048 -0.046 -0.047 -0.049

[0.057] [0.056] [0.056] [0.056] [0.057] [0.057]

Toilet facility -0.025 -0.015 -0.102 -0.084 -0.113* -0.079

[0.070] [0.067] [0.068] [0.069] [0.068] [0.069]

Meals 0.140** 0.133** 0.183*** 0.173*** 0.178*** 0.174***

[0.058] [0.058] [0.060] [0.058] [0.060] [0.059]

Animal products 0.059 0.061 -0.029 -0.012 -0.037 -0.035

[0.068] [0.068] [0.073] [0.072] [0.074] [0.074]

HH expenditures per capita 0.317* 0.240* 0.403* 0.382* 0.513** 0.430**

[0.181] [0.127] [0.207] [0.203] [0.200] [0.215]

Average province expenditures 0.401*** 0.509***

[0.129] [0.127]

Gini province 1.960**

[0.784]

Gini province t-1 3.440***

[0.737]

Average district expenditures -0.136 -0.101

[0.149] [0.142]

Gini district -0.189

[0.579]

Gini district t-1 0.838**

[0.335]

Average constituency expenditures -0.247* -0.196

[0.148] [0.146]

Gini constituency 0.245

[0.503]

Gini constituency t-1 0.678***

[0.249]

Constant -8.484*** -10.318*** -1.639 -2.467** -0.998 -1.496

[1.276] [1.305] [1.107] [1.133] [1.003] [1.037]

Observations 10302 10302 10302 10302 10302 10302

Complex survey design accounted for. Standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%

87


Chapter 3

As anticipated, households with more educated members have significantly

better nourished children when compared to households with no formal education.

This is also true for the level of schooling of the biological mother. A mother’s

education level has shown to be important to child health as educated mothers often

are more likely to follow instructions about feeding practices and child care. Madise et

al. (1999) also suggest that educated mothers use curative and preventive health

services more often than women with little of no education. Finally, with reference to

the influence of maternal experience, the coefficient on our proxy, age, is positive and

significant as expected.

Alternative inequality indices

3.4.3 Sensitivity analysis

All inequality measures implicitly make value judgments by their way of weighting

differences between observations in a distribution. Table 3.7 presents estimation

results when we replace the Gini coefficient with alternative inequality indices, GE(0)

and GE(1). Reassuringly this robustness test does not alter baseline findings. Once

again the empirical results are not in line with the suggested IIH. Using contemporary

inequality levels, a positive significant association between inequality and health is

established for provincial level inequality. Moreover, when using lagged information

instead, all specifications report that inequality is positively and significantly related to

greater height-for-age z-scores, regardless of aggregation level or inequality measure.

Inconsistent with the RIH, also the analysis using alternative inequality

measures suggest that the average expenditures of the province have a protective

health impact. However, this exercise confirms the negative correlation between

average constituency expenditures and child health found in table 3.6, when

confounding by measures on current inequality. This analysis also supports the AIH.

Outliers

To test whether results are driven by outliers, we re-estimate our results first

excluding the lowest deciles in the expenditure distribution. If the baseline AIH

results are sensitive to this robustness test it would suggest that income only improve

health for them living in absolute or relative poverty (see discussion Wagstaff and van

Doorslaer 2000). We also re-estimate our results excluding the highest decile in the

88


Income Inequality and Health: Exploring the Association in a Developing Country

expenditure distribution. These exercises do not change our conclusions with respect

to any of the hypotheses. In particular, the protective effect of household income on

child health remains also when the poorest are excluded.

With reference to inequality and average expenditures a corresponding

dilemma would be a measurement error problem following from that derivations of

some of the contextual variables were made using a small number of observations.

Relatively few observations could bias the coefficients for income inequality and

average expenditures upwards. Following Gerdtham and Johannesson (2004) we reestimate

the models excluding constituencies with fewer than 50 and 100

observations. In line with our previous findings the association between present

inequality and child health remains insignificant in this exercise, while the relationship

is positive and significant using inequality measures from a previous time period.

Likewise the deteriorating impact on child health from average expenditures reappears

in specifications using current inequality levels. The protective child health

impact from household expenditure is also stable towards this sensitivity test.

Alternative explanatory variables and instruments

As several studies on child nutrition find evidence of a positive effect from a higher

intake of energy and proteins we replace the variable animal products, with a proxy for

the proportion of high protein foods in diet. As there is no quantitative information on

households’ intake of different categories of food in the dataset, the new variable

refers to expenditures spent on fish, meat, eggs and milk products by the household

divided by total household expenditures on food. 25 Moreover, as sanitary standards

are assumed matter to nutritional status, we replace the indicator on the standard of a

household’s toilet facility with information on its method of garbage disposal. These

variables, however, also seem to be poor predictors of child health in the current

setting.

Furthermore we replace the set of instruments used in 2SLS estimations with

information on housing conditions (type of floor and roof) and type of energy used in

cooking. While results with respect to the AIH, RIH and IIH are re-confirmed, some

of the additional covariates at the household level are now insignificant. For example

household education is not associated with health status, nor is the rural dummy. We

25 Important to note this ratio is likely biased by geographical price differences.

89


Chapter 3

therefore suspect some of these alternative instruments to be correlated with factors

such as household location.

Weights included in regressions

Following Deaton’s (2000) and Korn and Graubard’s (2003) recommendations on

regression analysis of survey data, we initially do not apply weighting procedures in

the econometric modelling. As a sensitivity test we do, however, take them into

account. Baseline results are robust to including weights in estimations.

Correlation

Table 3B in the appendix presents a correlation matrix of the various variables. The

overall pair wise correlation between our indicators is low and a majority of the

coefficients do not exceed an absolute value of 0.3. However, we notify a negative

relationship between higher education levels and the dummy variable indicating if

households reside in rural areas. Moreover there is a fairly strong correlation between

gender share of adult household members and the variable head of household.

Excluding the gender share variable generates a significantly positive correlation

between female headed households and better child health. Notably, this alternative

specification does not change our baseline results with respect to the AIH, RIH or

IIH.

Alternative estimation techniques

Following the estimation approach in some of the empirical studies on the

determinants of child nutritional status we create two binary variables, referring to

whether or not children are stunted, respectively severely stunted, and re-estimate the

results using a probit model. In terms of the three hypotheses all baseline results are

confirmed. The results, however, suggest some differences regarding some of the

covariates compared to when using OLS. In particular, households with primary

education do not exhibit significantly better child health compared to households

with no education.

90


Table 3.7 Sensitivity analysis – alternative inequality measures – 2SLS

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

HH expenditure per capita 0.320* 0.332* 0.284* 0.288* 0.478** 0.478** 0.451** 0.453** 0.517** 0.515** 0.504** 0.505**

[0.182] [0.183] [0.156] [0.158] [0.194] [0.194] [0.192] [0.191] [0.202] [0.202] [0.202] [0.202]

Average province expenditures 0.431*** 0.307* 0.587*** 0.559***

[0.135] [0.164] [0.135] [0.133]

GE(0) province 1.047***

[0.342]

GE(1) province 0.246

[0.163]

GE(0) province t-1 1.800***

[0.321]

GE(1) province t-1 1.056***

[0.201]

Average district expenditures -0.161 -0.140 -0.096 -0.093

[0.148] [0.160] [0.144] [0.142]

GE(0) district 0.067

[0.242]

GE(1) district -0.078

[0.139]

GE(0) district t-1 0.427***

[0.122]

GE(1) district t-1 0.319***

[0.109]

Average constituency expenditures -0.257* -0.287* -0.219 -0.220

[0.150] [0.158] [0.146] [0.146]

GE(0) constituency 0.123

[0.225]

GE(1) constituency 0.156

[0.104]

GE(0) constituency t-1 0.348***

[0.102]

GE(1) constituency t-1 0.278***

[0.097]

Household characteristics included yes yes yes yes yes yes yes yes yes yes yes yes

Constant -8.517*** -6.692*** -10.504*** -9.882*** -1.744 -1.900 -2.553** -2.564** -0.829 -0.503 -1.324 -1.292

[1.250] [1.451] [1.289] [1.246] [1.151] [1.249] [1.167] [1.160] [1.049] [1.136] [1.067] [1.060]

Observations 10316 10316 10316 10316 10316 10316 10316 10316 10316 10316 10316 10316


Chapter 3

If outcomes of observations are not independent, standard errors will be

underestimated. Accordingly it is important to take complex survey design into

account of as done in the previous analysis. However, there is concern about

outcomes also being related, as some children live in the same household. This might

be problematic if nutritional status depends on characteristics common to a family, or

as an effect of common influences related to the geographical area where children

live. To overcome these potential problems we run a two-level random intercept

model including individual characteristics, household expenditures as well as the

contextual variables. 26

Along the lines of the AIH, the association between household expenditures

and health remains positive and significant. Moreover the relationship between

greater inequalities in a previous time period is associated with less malnutrition also

in this setting. Neither of the conclusions on the RIH differs from baseline

regressions. The household level variance is significant for all models, suggesting that

significant unobserved heterogeneity in nutritional status between families exist. In

other words, some households have higher risk of malnutrition among their children

and the determinants of this risk are not explained by any of the included control

variables. The significant household effects are not surprising given that the

determinants of child health operate at many levels. This could be related to genetic

frailty, variation in parental competence and that some households, given the same

economic constraints, are better than others in using economic resources to

maximize child health. The contextual level variance is smaller but significant in those

cases where the geographic level refers to the province or the district. Potential

explanations of homogeneity within a geographic area might relate to cultural

practices, similar environmental conditions and the same level of infrastructure.

26 Multilevel models are also referred to as variance-component models. We use the Stata command

xtmixed and let the random part at level 2 households and at level 1 the geographical area of interest.

92


Income Inequality and Health: Exploring the Association in a Developing Country

3.5 DISCUSSION AND CONCLUDING REMARKS

The relationship between income inequality and individual health is well tested in

high-income contexts and to some extent within middle-income countries. However,

knowledge is limited on how inequality relates to health in less developed contexts.

Testing three suggested hypotheses using Zambian data our analysis arrives at several

interesting findings.

First, as expected in a setting with high poverty levels, household monetary

means is an important determinant of child nutritional status. Furthermore,

consistent with the AIH, the protective health effect from more household economic

resources appears non-linear. Although few studies in the inequality-health literature

undertake preventive measures for potential reverse causality, our findings are clearly

in line with previous findings (c.f Subramanian and Kawachi 2004, Wagstaff and van

Doorslaer 2000, Deaton 2003, Wilkinson and Picket 2006).

Second, we only find weak evidence of the RIH. In line with previous

findings, the association between relative expenditures and health appears different

depending on the choice of reference group, suggesting a single explanation cannot

be identified and reasonably that different mechanisms mediate the relationship in

various settings. While there is some evidence that relative expenditures correlates

with poorer child health when the reference group corresponds to households in a

local geographic area, the relationship is reverse and positive when testing the

association between average provincial economic expenditures and health status. The

former result is reasonable since day-to-day comparisons are likely to exist in

neighboring contexts. The protective health impact of living in a richer province

suggests there might be spill-over effects. For example, higher levels of income per

capita may expand the provision of public goods which may improve parental and

child health status. A positive RIH also corresponds with Senik (2004) who suggest

that individuals form their expectations based on reference group average incomes

and that average income could exert a positive influence on individual satisfaction if

we see that people around us receive higher incomes

Third, in contrast to the traditional view in the inequality – health literature

and to most empirical findings, the analysis suggests that greater expenditure

93


Chapter 3

inequality correlates with better nutritional status. Allowing for some time of

inequality exposure this finding appears regardless of whether inequality is measured

at the provincial, the district or the constituency level, suggesting that mechanisms

mediating the relationship between inequality and child health are similar at all levels.

The positive outcomes also appear when we use alternative inequality indicators,

alternative specifications, and different estimations techniques, implying that the

results are robust. Interestingly our result using individual level data is consistent with

the correlation between income inequality and less prevalence of stunting at the

population level found among various low-income countries. This suggests that the

relation found may not be atypical for Zambia.

As proposed pathways from inequality to health predominantly have been

formulated to account for adversity in health outcomes in the related literature, our

findings require a different framework for interpretation. First, maintaining that the

proposed mediators are relevant and that higher levels of inequality correlates, as

expected, with for example less social capital, one possible explanation to our result

could be that lower social interaction is associated to better health status. Although in

contrast with the general findings on the relationship between social capital and

health, less connected households may for example have children with better health

than their more socially integrated equals as they have lower risk of being

contaminated by infections.

We should also note that although overall inequality is high in a society,

within-group inequality may still be low. For example table 3.3 indicates that

differences at the bottom of the expenditure distribution in Zambia are smaller than

overall expenditure differences, when every observation in the distribution is given an

equal weight. As evidence suggests that lower trust in particular associates with

income differences among people in the bottom half of the distribution (Gustavsson

and Jordahl 2008), trust and social capital may still be important factors to better

nutrition in unequal contexts if income differences among the poor are relatively

small. Our results could also follow from that there exists an inverse relationship

between inequality and any of the suggested mediators in less developed contexts.

For example, along the lines of the Meltzer-Richard theorem, greater inequality

94


Income Inequality and Health: Exploring the Association in a Developing Country

among voters may increase government spending as the median voter is inclined to

support large public expenditures when a majority of the population is poor.

An alternative interpretation to our findings would be that higher levels of

inequality correlate with other important health determinants. Given that most

households in developing contexts engage in agricultural activities, it seems

reasonable to assume that income differences to some extent may appear as a result

of idiosyncratic income shocks following from crop failures. Crop failures will turn

affect the possibility for households to smooth consumption and calorie intake which

generally is seen as an important factor to good long term nutritional status.

However, if households that are not hit by a shock view this fact as brute luck it is

rational to have a food sharing among households (cf. Gurven 2004, Kaplan and Hill

1985) that in turn can neutralize negative health impacts. As discussed by Bergh

(2008) sharing norms are efficient if surplus is generated by brute luck as such sharing

will reduce consumption variance. Various agricultural production shocks, in turn

generating higher income inequality, could also stimulate more formal solutions to

the consumption smoothing problem. If formal insurance and credit markets are

imperfect or non-existent, households may instead protect their consumption and

health by relying on friends and village networks. As discussed by Attanasio and Rìos-

Rull (2000), enforceability of contracts may in this situation relate to the cost for

households of being left alone to deal with shocks in the future. 27

Altogether, our findings merit further research on the relationship between

inequality and health in developing contexts. A possible direction for further studies

is to include proxies for various mediators in order to determine through what

linkages the positive income inequality-health relationship mediates. Moreover it is

desirable to move beyond the correlation strategy used in this analysis to enable an

identification of causal effects. Ideally, the relationship between income inequality

and individual health would be analyzed in a panel data setting. In addition, the

theoretical aspects on the relationship between inequality and health need to be

revisited with the above results in mind.

27 Evidence in developing contexts indicate partial consumption sharing among households (Udry 1995)

and that households have a better ability to insure caloric consumption than spending of total consumption

against idiosyncratic shocks to income (Deininger et al. 2007).

95


Chapter 3

Table 3A Summary statistics

APPENDIX

Variable Mean sd Min Max Obs

Individual level

HAZ -1.79 1.82 -5.99 3 10316

Gender 0.50 0.50 0 1 10316

Age 28.80 15.81 3 59 10316

Age^2 1079.19 963.63 9 3600 10316

Birth order 2.84 1.51 1 18 10316

Household level

HH expenditure per capita 94468 10299 286 1967600 10316

Rural 0.60 0.49 0 1 10316

HH head 0.15 0.35 0 1 10316

HH gender share 0.54 0.18 0 1 10316

HH size 6.81 3.12 2 33 10316

HH edu 0 0.03 0.16 0 1 10316

HH edu 1 0.36 0.48 0 1 10316

HH edu 2 0.52 0.50 0 1 10316

HH edu 3 0.10 0.29 0 1 10316

Meals 0.47 0.50 0 1 10316

Animal products 0.49 0.50 0 1 10316

Water 0.38 0.49 0 1 10316

Toilet facility 0.84 0.36 0 1 10316

Mother's age 28.57 7.20 15 61 9752

Mother edu 0 0.12 0.32 0 1 9752

Mother edu 1 0.56 0.50 0 1 9752

Mother edu 2 0.29 0.45 0 1 9752

Mother edu 3 0.03 0.18 0 1 9752

Marital status 0.81 0.39 0 1 9752

Contextual level

Gini province 0.51 0.04 0.47 0.59 10316

Gini district 0.51 0.05 0.43 0.68 10316

Gini constituency 0.49 0.06 0.31 0.70 10316

Average expenditures province 121708 35709 78300 204000 10316

Average expenditures district 117806 41118 51700 227000 10316

Average expenditures constituency 118818 52423 37700 439000 10316

96


Table 3B Correlation matrix

Income Inequality and Health: Exploring the Association in a Developing Country

HAZ Gender Age Age^2 Birth

order

97

HH exp

per capita

Rural HH head HH share

(adult)

Heigh-for-age 1

Gender 0.053 1

Age -0.112 0.017 1

Age^2 -0.090 0.016 0.971 1

Birth order 0.001 -0.016 -0.127 -0.122 1

HH expenditure per capita 0.084 -0.012 0.001 0.000 -0.243 1

Rural -0.109 0.008 -0.029 -0.024 0.080 -0.345 1

HH head -0.007 0.006 0.017 0.014 -0.026 -0.080 0.004 1

HH share (adult) 0.014 0.007 -0.010 -0.013 -0.027 -0.004 -0.019 0.545 1

HH size 0.035 -0.012 0.045 0.044 0.704 -0.223 0.011 -0.071 -0.094

Water 0.055 -0.004 0.013 0.012 -0.011 0.252 -0.299 -0.061 -0.017

Toilet 0.033 -0.012 0.006 0.005 0.005 0.182 -0.320 -0.054 -0.042

HH edu 1 -0.076 0.006 -0.028 -0.027 -0.018 -0.225 0.383 0.069 0.065

HH edu 2 0.040 -0.009 0.010 0.009 0.045 0.086 -0.244 -0.061 -0.090

HH edu 3 0.081 0.007 0.034 0.033 -0.027 0.254 -0.263 -0.043 0.002

Animal products 0.057 0.001 0.011 0.014 -0.023 0.237 -0.171 -0.084 -0.056

Meals 0.111 -0.006 0.019 0.019 0.003 0.258 -0.272 -0.063 0.002

Gini province 0.033 0.028 0.005 0.002 0.012 -0.076 0.161 0.035 0.012

Gini district -0.020 0.024 0.015 0.014 -0.013 -0.078 0.063 0.036 0.008

Gini constituency -0.029 0.022 0.012 0.009 0.000 -0.115 0.139 0.030 0.005

Average province expenditures 0.107 0.000 0.007 0.000 -0.028 0.231 -0.275 -0.044 -0.008

Average district expenditures 0.065 -0.003 0.011 0.007 -0.041 0.328 -0.397 -0.047 -0.009

Average constituency expenditures 0.064 0.006 0.006 0.003 -0.038 0.374 -0.394 -0.050 -0.005

HH size Water Toilet HH edu 1 HH edu 2 HH edu 3 Animal

products

Meals

HH size 1

Water 0.037 1

Toilet 0.038 0.210 1

HH edu 1 -0.161 -0.227 -0.213 1

HH edu 2 0.137 0.133 0.155 -0.780 1

HH edu 3 0.084 0.184 0.130 -0.245 -0.337 1

Animal products 0.007 0.149 0.139 -0.142 0.047 0.188 1

Meals 0.064 0.182 0.084 -0.246 0.121 0.230 0.194 1

Gini province 0.035 -0.160 -0.355 0.100 -0.090 -0.025 -0.096 0.131

Gini district -0.020 -0.094 -0.164 0.060 -0.080 0.013 -0.085 0.055

Gini constituency -0.010 -0.089 -0.148 0.083 -0.086 -0.015 -0.075 0.007

Average province expenditures 0.004 0.225 0.098 -0.143 0.109 0.089 -0.073 0.210

Average district expenditures -0.012 0.243 0.142 -0.174 0.116 0.129 -0.032 0.207

Average constituency expenditures 0.000 0.241 0.157 -0.178 0.098 0.170 0.013 0.207

Gini

province

Gini

district

Gini

constituency

Average

province

expenditures

Average

district

expenditures

Average

constituency

expenditures

Gini province 1

Gini district 0.556 1

Gini constituency 0.394 0.703 1

Average province expenditures 0.040 -0.024 -0.075 1

Average district expenditures -0.039 0.064 -0.055 0.709 1

Average constituency expenditures -0.068 0.007 0.034 0.583 0.841 1


Chapter 3

REFERENCES

Alesina, A. et al. (1999), “Public goods and ethnic divisions.” Quarterly Journal of Economics.

Vol. 114,1243-1284

Arimond M, and M.T Ruel (2004), “Dietary Diversity Is Associated with Child

Nutritional Status: Evidence from 11 Demographic and Health Surveys”. Journal

of nutrition, Vol. 134(10): 2579-2585

Attanasio, O.; Gomez, L.C.; Rojas, A.G.; Vera-Hernandez, M. (2004). “Child health in

rural Colombia: determinants and policy interventions”. Economics and Human

Biology, Vol. 2(3): 411-438

Attanasio O, J.V Rios-Rull (2000). ”Consumption smoothing in island economies: Can

public insurance reduce welfare?” European Economic Review Vol 44(7): 1225-1258

Babones, S.J. (2008).”Income inequality and population health: Correlation and

causality”. Social Science & Medicine Vol 66(7): 1614-1626

Baum, F.(1999), “Social Capital: Is it Good for your Health? Issues for a Public Health

Agenda.”, Journal of Epidemiology and Community Health, 53:195-196.

Becker G (1964). Human capital, New York: Columbia University Press.

Becker G (1965). “The theory of the allocation of time, Economic Journal, Vol. 75, 493-

517.

Behrman J R, Pollak R A, Taubman P (1986) “Do parents favor boys?”. International

Economic Review, Vol. 17(1), 33-54.

Behrman, J. R. (1988) “Nutrition, health, birth order and seasonality: intra-household

allocation among children in rural India”. Journal of Developing Economics. 28(1), 43–

62.

Bengtsson T., M. Lindström (2000). “Childhood misery and disease in later life: The

effects on mortality in old age of hazard experienced in early life, sothern

Sweden, 1760-1894”. Population Studies, Vol 54: 263-277

Bergh, A. (2008). “A critical note on the theory of inequity aversion” Journal of Socio –

Economics, Vol 37(5):1789

Berkman, LF, Kawachi I (Eds) (2000), Social Epidemiology, Oxford University Press, New

York

Blakely, T, Atkinson, J, Ivory, V, Collings, S., Wilton, J, Howden-Chapman, P. (2006).

“No association of neighbourhood volunteerism with mortality in New Zealand:

a national multilevel cohort study” International journal of epidemiology Vol 35 (4):

981-989

Bourdieu, P. (2007). “The Forms of Capital”. i A. R. Sadovnik, Sociology of Education. New

York: Routledge.

Brunner,E. and M Marmot(1999). “Social organization, stress and health” Chapter 2 in

M. Marmot and R.G. Wilkinson Social determinants of health Oxford

Carlson, P (2005). “Relatively poor, absolutely ill? A study of regional income inequality

in Russia and its possible health consequences”. Journal of epidemiology and

community health Vol 59(5): 389-394

Chang VW, Christakis NA. (2005). “Income inequality and weight status in US

metropolitan areas”. Social Science & Medicine Vol 61:83–96.

CSO (1999, 2005), Living Conditions Monitoring Survey II , IV

Cuny, F C. (1994), Disasters and Development. Dallas: Intertect Press.

Currie, J. (2009). “Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in

Childhood, and Human Capital Development.” Journal of Economic Literature, 47: 87

98


Income Inequality and Health: Exploring the Association in a Developing Country

Deaton, A (2000), The analysis of household surveys – A microeconomteric approach to development

policy, World Bank, John Hopkins UP, Baltimore and London

Deaton, A (2003). “Health, Inequality and Economic Development”. Journal of Economic

Literature, Vol. 41

Deininger K, J Songqing and Y Xinhua. (2007). “Risk coping and starvation in rural

China” Applied Economics Vol 39(11): 1341-1352

De Silva, M.J and S.R.; Harpham (2007). ”Maternal social capital and child nutritional

status in four developing countries”. Health and Place, Vol. 13(2): 341-355

Demombynes, G and O. Berk (2005), “Crime and local inequality in South Africa”,

Journal of Development Economics Vol 76(2): 265-292

Diaz-Roux AV, Link BG, Northridge ME. (2000). “A multilevel analysis of income

inequality and cardiovascular disease”. Social Science & Medicine Vol 50:673–687.

Durkheim, E. (1897) Suicide. London: Routledge 1992

Ettner, S. L (1996), “New evidence on the relationship between income and health”.

Journal of Health Economics, Vol. 15(1): 67-86

Falkner F, and J. M. Tanner (1986) Human growth, 2nd edition, Plenum Press

Gerdtham, U and M. Johannesson (2004). “Absolute Income, Relative Income, Income

Inequality, and Mortality”, The Journal of Human Resources, Vol 39, 228-247

Graitcer, P.; Gentry, E. (1981), “Measuring Children: One Reference for All”, The Lancet,

Vol. 318(8241): 297-299

Gravelle, H. (1998). “How much of the relation between population mortality and

unequal distribution of income is a statistical artefact?”, British Medical Journal,

316, 382–385.

Gravelle Hugh and Sutton Matt (2009). “Income, relative income, and self-reported

health in Britain 1979-2000”, Health Economics, Vol 18(2): 125

Grossman, M. (1975). The correlation between health and schooling, in Household

Production and Consumption, N. Terleckyj ed. NY: Columbia U. Press for NBER,

pp. 147-211

Grossman, M. (2000) The human capital model, in Handbook of Health Economics, A.J.

Culyer and J.P Newhouse, eds. Vol 1A, Amsterdam, Elsevier, pp. 347-408

Gurven, M, (2004). “Reciprocal altruism and food sharing decisions among Hiwi and

Ache hunter-gatherers”. Behavioral Ecology and Sociobiology 56, 366–380.

Gustavsson, M. and Jordahl, H (2008). “Inequality and trust in Sweden: Some inequalities

are more harmful than others”, Journal of Public Economics, Vol. 92(1-2): 348-365

Habicht, J.-P., Yarbrough, C., Martorell, R., Malina, R and Klein, R. (1974), “Height and

weight standards for preschool children - how relevant are ethnic differences in

growth potential?”, The Lancet, Vol. 303(7858): 611-615

Harpham, T., MJ. De Silva, and T. Tran (2006). ”Maternal social capital and child health

in Vietnam” Journal of epidemiology and community health, Vol. 60( 10): 865-871

Heltberg, R.(2009). “Malnutrition, poverty, and economic growth” Health Economics Vol

18: S77

Hildebrand,V and P. Van Kerm (2005). “Income Inequality and Self-Rated Health Status:

Evidence from the European Community Household Panel”, McMaster

University, Social and Economic Dimensions of an Aging Population Research

Papers, 2005

Hox, J. (2002), Multilevel analysis: techniques and applications, Lawrence Erlbaum Ass.

Publisher, London

Jones K, Duncan C, Twigg L. (2004), “Evaluating the absolute and relative income

hypothesis in an exploratory analysis of deaths in the Health and Lifestyle

99


Chapter 3

Survey”. In The Geography of Health Inequalities in the Developed World, Boyle P, Curtis

S, Graham E (eds). Ashgate Press: London, UK.

Kaplan, G.A, E.R. Pamuk, J.W Lynch, R..D. Cohen and J.L. Balfour (1996). “Inequality

in income and mortality in the United States: analysis of mortality and potential

pathways”. British Medicine Journal, Vol. 312:999-1003

Kaplan, H., Hill, K., (1985). “Food sharing among Ache foragers: tests of explanatory

hypotheses”. Current Anthropology 26, 223–245

Kawachi, I. and B.P. Kennedy (1997). “The relationship of income inequality to

mortality: does the choice of indicator matter?” Social Science Medicine, Vol. 45:

1121-1127

Kawachi, I. and B.P. Kennedy (1999). “Income inequality and health: Pathways and

mechanisms”. Health Services Research, Vol. 34(1):215-227

Kahn, B. Zuckerman, H. Bauchner, C.J. Homer and P.H. Wise (2002). “Women's health

after pregnancy and child outcomes at age 3 years: a prospective cohort study”.

American Journal of Public Health, Vol 92:1312–1318

Karlsson, M., T. Nilsson, C.H Lyttkens and G Leeson (2008). “Individual good, public

bad, or societal syndrome? A cross-country study of income, inequality and

health”, OIA Working Paper 408, Oxford Institute of Ageing

Khanam, R. H S Nghiem and LB., Connelly (2009). “Child Health and the Income

Gradient: Evidence from Australia”, Forthcoming in Journal of Health Economics

Kawachi I., S.V. Subramanian, D. Kim (2007), Social Capital and Health, Springer, New

York

Kennedy, Kawachi, Glass and Prothrow-Stist (1998). “Income distribution, socioeconomic

status, and self-related health in the US: Multilevel analysis.” British

Medical Journal, Vol 317

Knack, S. and Keefer, P. (1997) “Does social capital have an economic payoff? A cross-

country investigation”. Quarterly Journal of Economics, Vol. 112 (4): 1252–1288.

Korn, E.L. and B.I. Graubard (2003). “Estimating variance components by using survey

data”. Journal of the Royal Statistical Society. Series B (Methodological), Vol. 65(1):

175-190

Krugman, P. (1996). “The Spiral of Inequality”. Mother Jones (November/December): 44-

49

Larrea, C and Kawachi, I. (2005). “Does economic inequality affect child malnutrition?

The case of Ecuador”. Social Science and Medicine, Vol. 60: 165-178

Lawson, D. (2004) “Determinants of Health Seeking Behaviour in Uganda –Is it Just

Income and User Fees That Are Important?”, Working Paper University of

Manchester, Manchester, United Kingdom

Leigh, A and Jencks C (2007). “Inequality and mortality: Long-run evidence from a panel

of countries”. Journal of Health Economics, Vol 26 (1)

Li, H. & Zhu Y. (2006). Income, Income Inequality and Health – Evidence from China.

Journal of Comparative Economics, Vol 34, 668-693

Lochner, K, E. Pamuk, and D. Makuc (2001). “State-Level Income Inequality and

Individual Mortality Risk: A Prospective, Multilevel Study”. American Journal of

Public Health, Vol. 91(3): 385-392

Lopez, R. (2004). “Income inequality and self-rated health in US metropolitan areas: a

multi-level analysis” Social Science & Medicine, Vol 59: 2409-2419

Lorgelly, P.K., J.K. Lindley (2008). “What is the relationship between income inequality

and health? Evidence from the BHPS”, Health Economics Vol 17(2):249-265

100


Income Inequality and Health: Exploring the Association in a Developing Country

Luttmer, EFP (2005). “Neighbors as negatives: Relative earnings and well-being”.

Quarterly Journal of Economics, 120 (3): 963-1002

Lynch, J.W, G.D Smith, G.A Kaplan, J.S House (2000). “Income inequality and

mortality: importance to health of individual income, psychosocial environment,

or material conditions”, British Medical Journal, Vol. 320:1200-1204

Madise, N.J, Z. Matthews, B. Margetts (1999). “Heterogeneity of Child Nutritional Status

between Households: A Comparison of Six Sub-Saharan African Countries”.

Population Studies Vol.53(3): 331-343

Maluccio, J A., JF. Hoddinott, J R. Behrman,R Martorell, A R. Quisumbing, and AD.

Stein. (2006). “The Impact of Nutrition during Early Childhood on Education

among Guatemalan Adults.” University of Pennsylvania Population Studies

Center Working Paper 06-04

Miller, D. L. and C. Paxson (2006). “Relative income, race, and mortality”, Journal of

Health Economics, 25 (5): 979-1003

Mellor, J.M and J.D. Milyo (2001). “Income Inequality and Health.”, Journal of Policy-

Analysis and Management. Winter 20(1): 151-55

Mellor, J.M. and J.D. Milyo (2002). “Income inequality and individual health: Evidence

from the Current Population Survey”, Journal of Human Resources, Vol 37

Morrow, V. (2004). “Children's "social capital": implications for health and well-being”.

Health Education, Vol. 104(4): 211-225

Mosley, W.H and L.C. Chen (1984). “An Analytical Framework for the Study of Child

Survival in Developing Countries”. Population and Development Review, Vol. 10: 25-

45

Olivius, G., P.O Ostergren, B.S Hanson and C.H. Lyttkens (2004). “Parental economic

stress”, European journal of public health, Vol. 14(4): 354-360

Osler M, Christensen U, Due P, Lund R, Andersen I, Diderichsen F, Prescott E. (2003).

Income inequality and ischaemic heart disease in Danish men and women.

International Journal of Epidemiology 32:375–380.

Preston, S.H. (1975). “The changing relation between mortality and level of economic

development”. Populations Studies, Vol. 29(231-48)

Putnam, R. D. (2000). Bowling Alone. The Collapse and Revival of American Community. New

York: Touchstone.

Rajkumar, A. and S. Vinaya. (2008). Public spending and outcomes: Does governance matter?

Journal of Development Economics. Vol. 86(1): 96

Ram, R. (2006). “Further examination of the cross-country association between income

inequality and population health”. Social Science and Medicine Vol 62(3): 779-791

Rodgers, G.B. (1979). Income and inequality as determinants of mortality: an international crosssection

analysis, Population Studies, Vol. 33: 343-351

Senik, C. (2004). “When information dominates comparison: Learning from Russian

subjective panel data”. Journal of Public Economics, 88, 2099-2123

Shibuya, K., H. Hashimoto, and E. Yano (2002), Individual income, income distribution, and self

rated health in Japan: cross sectional analysis of nationally representative sample, BMJ

2002;324:16

Skoufias, E. (1998). Determinants of child health during the economic transition in Romania, World

Development, Vol. 26(11): 2045-2056

Smith, J.P. (1999). Healthy bodies and thick wallets: the dual relation between health and

economic status, Journal of Economic Perspectives, Vol. 13: 145-166

Solt, F.(2008). "Standardizing the World Income Inequality Database." Social Science

Quarterly

101


Chapter 3

Submramanian S.V, Kawachi and Kennedy (2001). Does the state you live in make a difference?

Multilevel analysis of self related health in the US, Soc. Sci. Med., Vol 53

Subramanian S.V, I Delgado, L Jadue, J Vega, I Kawachi, (2003), Income inequality and

health: multilevel analysis of Chilean communities, J Epidemiol Comm Health

2003;57:844–8

Subramanian, S.V and Kwachi I (2004). Income Inequality and Health: What Have We Learned

So Far?, Epidemiologic Reviews, Vol. 26

Svedberg, P. (1990), Undernutrition in Sub-Saharan Africa: Is there a gender bias?, Journal of

Development Studies, Vol. 26(3): 469-486

Thomas D, Strauss, J. and M-H. Henriques (1991). “How Does Mother's Education

Affect Child Height?” The Journal of Human Resources Vol 26(2): 183

Thomas, D (1994), Like father like son: like mother, like daughter: Parental resource and child

height, Journal of Human Resources, Vol. 29(4), 950-988.

Udry, C. (1995). “Risk and Saving in Northern Nigeria”. American Economic Review Vol

85(5): 1287-1301

Waldman, R. (1992). “Income distribution and infant mortality”. In The society and

population health reader, edited by I. Kawachi, B. Kennedy, and R. G. Wilkinson, 14-

27. New York: New Press.

Wagstaff and van Doorslaer (2000). Income Inequality and Health: What Does the Literature

Tell Us?, Annual Review of Public Health, Vol. 21

WHO (1995), Physical status: the use and integration of anthropometry. Report of a WHO

Expert committee, World Health Organ Tech Rep Ser 854:1-452

WHO (2007), http://www.who.int/nutgrowthdb/database/countries/zmb/en/

Wilkinson, R. G. (1992), “National mortality rates: the impact of inequality?”, American

Journal of Public Health, Vol. 82, Issue 8 1082-1084

Wilkinson, R.G. (1996). Unhealthy Societies: The Afflictions of Inequality, Routledge, London

Wilkinson R, Pickett K. (2006). “Income inequality and population health: a review and

explanation of the evidence”. Social Science & Medicine Vol 62:1768–1784.

World Bank (2008), World Development Indicators

Zere, E. and D. McIntyre (2003), Inequities in under-five child malnutrition in South Africa,

International Journal for Equity in Health, Vol. 2(1): 7

Acknowledgements

The author is thankful for comments and suggestions from Carl Hampus Lyttkens,

Andreas Bergh, Jesper Roine, Mireia Jofre-Bonet, Pernilla Johansson and participants at

the Joint Meeting of the UK Health Economists’ Study Group & the Nordic Health

Economists’, Aberdeen, 2008, Björn Ekman and seminal participants in Lund for useful

comments and suggestions. Financial support from the Swedish Council for Working life and

Social research (Health Economics Program, Lund University) is gratefully acknowledged.

102


Chapter 4

Do Liberalization and Globalization Increase Income

Inequality?

Co-author: Andreas Bergh

4.1 INTRODUCTION

Over the past 30 years, most countries around the world have experienced substantial

increases in economic freedom and globalization. There is a prevalent belief that such

changes may benefit economic growth, but at the expense of increased income

inequality within countries. Regarding the first issue, the current consensus among

researchers seems to be that economic freedom (often referred to as liberalization)

and globalization are indeed linked to economic growth (see, e.g., Berggren and

Jordahl 2005, Doucouliagos and Ulubasoglu 2006, and Dreher 2006). 1

This paper examines the second question: Are increases in economic freedom

and globalization associated with increasing income inequality within countries?

Although participants in public debate on this topic generally have a clear opinion on

the relationships, empirical evidence is surprisingly contradictory (Berggren 1999,

Scully 2002, Carter 2007, Dreher and Gaston 2008). Moreover, knowledge is limited

as to whether all the dimensions of liberalization and globalization have similar impacts

on income distributions. While the focus is often on the overall economic

relationship with income inequality, past decades of closer integration among

1 It should be noted, however, that the proper measurement of globalization or liberalization and the

direction of the causality in question is still subject to intense debate; see, for example, Rodriguez and

Rodrik (2000) and the response of Lee Ha et al. (2004).


Chapter 4

countries, and of interrelated policy processes, have led to development in many

directions, justifying the examination of individual dimensions. Due to previous data

limitations, empirical studies have also neglected the issue of how different

dimensions of economic freedom and globalization influence income inequality at

different development levels. Beyond the intrinsic value of such an analysis, economic

theory often provides ambiguous or no predictions of these relationships in different

economic contexts, a matter that calls for empirical analysis.

Using Gini coefficients of household net income from Solt's (2008) recently

developed Standardized World Income Inequality Database (SWIID) as our preferred

inequality measure, we can construct a panel from 1970 through 2005 with more

observations on within-country income inequality than do other studies in this area.

This setup also allows for rigorous analysis of the differential impact across rich and

poor contexts and for the use of sophisticated techniques to handle possible

endogeneity problems. To quantify globalization and economic freedom, we use the

KOF Index of Globalization (KOF), developed and first used by Dreher (2006), and

the well-known Economic Freedom of the World Index (EFI) of Gwartney et al.

(2008). Beyond the general empirical approach of using the composite indices these

sources provide in order to measure the overall level of economic freedom and

globalization, we exploit the fact that both indices consist of several dimensions,

allowing for an analysis of the impact of different types of liberalization and

globalization on income inequality.

By estimating a fixed-effect model of country-level income inequality as a

function of liberalization or globalization, and employing a battery of robustness

tests, our analysis arrives at several findings. In particular, while there is evidence that

some dimensions of liberalization and globalization are indeed linked to greater

income inequality, several dimensions display no robust, significant effect.

First, the analysis supports the notion that policy reforms favoring trade

openness have on average increased income inequality in recent decades. Exploring

the relationship at different levels of development, however, indicates that, in line

with theoretical predictions, this significant relationship only appears in middle- and

high-income contexts. Second, findings repeatedly also indicate that policy reforms

104


Do Liberalization and Globalization Increase Income Inequality?

promoting deregulation and social globalization on average have a non-equalizing

distributional impact. Moreover, the coefficient of economic globalization is positive,

but is sensitive to the exclusion of certain countries from the sample. Third, in

estimating a dynamic model, consistent in the case of endogenous variables (Arellano

and Bover 1995, Blundell and Bond 1998), we again confirm that trade liberalization

and social globalization increase income inequality.

The chapter proceeds as follows. The next section describes the related

theoretical and empirical literature. Section 4.3 describes our data, with a special focus

on the measurement of income inequality, and introduces our empirical strategy.

Section 4.4 presents our main results and various robustness checks, while section 4.5

concludes the chapter.

4.2 RELATED LITERATURE

In this section, we first describe the different dimensions of economic freedom and

globalization measured by the indices used in the empirical analysis. In relation to

each dimension, we discuss the expected relationship with income inequality by

referring to existing theoretical and empirical research. When possible, we also

discuss the predicted effects of certain types of liberalization or globalization on

income inequality at different levels of development. As will be apparent from the

presentation below, there are sometimes theoretical arguments suggesting that

various dimensions may both increase and decrease income inequality, wheras some

of these dimensions lack sound theoretical underpinnings. We then summarize some

earlier studies that share our approach of analyzing the effect of liberalization or

globalization on income inequality using the EFI and KOF indices.

4.2.1 The different dimensions of economic freedom and globalization

The Economic Freedom of the World Index (EFI) is a composite index that weighs

together five dimensions of economic freedom, EFI1–EFI5, which are in turn based

on several indicators.

105


Chapter 4

Size of government (EFI1)

The first dimension of the EFI measures government size using indicators such as

public consumption and transfers relative to GDP. It also includes top marginal tax

rates and state-owned enterprises. The index is coded so that bigger government

means a lower economic freedom value in this dimension.

Theoretically, there are several reasons to expect states with larger welfare

systems to have lower income inequality; for example, public sector transfers are

assumed to have net equalizing effects on income distribution (see, e.g., Rothstein

1998, Åberg 1989). The welfare state may also stimulate risky but profitable incomeequalizing

activities such as education, as people are more likely to engage in such

activities when enjoying some protection provided by the welfare state (Sinn 1995).

Evidence also suggests that the welfare state is particularly beneficial to the middle

class (see, e.g., Bergh 2007 and Le Grand and Winter 1986), again suggesting a more

compressed income distribution.

Importantly, bigger government, as measured by the index, does not

necessarily imply a larger welfare state. In poor countries, where government may be

corrupt or even predatory, smaller government may not increase income inequality at

all. A study by Odedokun and Round (2004) examining the relationship between

government size and income inequality in 35 African countries supports this view.

Following the above, we hypothesize that an increase in EFI1 will have a nonequalizing

effect on the within-country income distribution and that this effect will be

larger in richer than poorer contexts.

Legal structure and security of property rights (EFI2)

The second dimension of the EFI quantifies the quality and integrity of the legal

system and the protection of property rights. This dimension can be thought of as an

attempt to quantify rule of law.

How do we expect it to influence inequality? It seems intuitive that better

protection of property rights should mainly benefit those with more property, as this

protection increases tenure security for the owner, which in turn is expected to

increase the value of the property itself. However, several scholars suggest the

106


Do Liberalization and Globalization Increase Income Inequality?

opposite. Inspired by the Russian oligarchs of the 1990s, Sonin (2003) notes that

poor protection of property rights may actually be relatively more beneficial to those

already rich, resulting in greater inequality. In many developing countries, elites are

rich because of corruption and inefficient property rights, and improvements in the

legal system may actually be relatively more important for less privileged groups,

thereby reducing inequality, as described by, for example, De Soto (2000). However,

the policy process may also let actors in the political bodies involved choose the

format of the legal framework that is most favorable for them (see, e.g., Lund 2001

on land titling in a developing context).

Accordingly, there is no clear theoretical prediction of the overall relationship

between legal structure, property rights security, and income inequality, or of how the

relationship changes with level of development.

Access to sound money (EFI3)

The sound money dimension of the EFI captures the effect of large and

unpredictable changes in inflation and money supply. This component is coded so

that the greater the unpredicted inflation, the lower the value.

The literature on the cost of inflation presents various theoretical mechanisms

by which inflation could affect income distribution, in particular through returns to

capital. High inflation is primarily expected to be relatively more harmful to lowincome

earners, whose assets are often less protected against inflation, increasing

income inequality. Moreover, unanticipated inflation may also lead to resource

misallocation and to the absorption of considerable resources in information

gathering, in an attempt to mitigate the uncertainty of future price levels (Fischer and

Modigliani 1978). This will have negative welfare effects that will reduce the

possibility of progressive redistribution.

Confirming theory, most empirical studies of the subject indicate a positive

relationship between inflation and inequality. For example, Albanesi (2007) presents

cross-country evidence that inflation and income inequality are positively correlated.

A link between pro-poor growth and low inflation is also found by Son and Kakwani

(2008). Hence, we hypothesize that an increase in the EFI3 index will be associated

107


Chapter 4

with a narrower spread in the income distribution, irrespective of the level of

economic development at which such an increase takes place.

Freedom to trade internationally (EFI4)

This component of the EFI combines measures of trade taxes, tariff rates and trade

barriers, and capital market controls to create a composite measure of freedom to

trade.

The effect of trade openness on inequality is highly debatable, both

theoretically and empirically. Kanbur (2000) describes a widespread and simple

intuition into the theoretical relationship between openness and inequality based on

Heckscher–Ohlin (H–O) theory in a model including both skilled and unskilled

workers, the former being more abundant in rich countries. In this case, trade

openness will exert downward pressure on the wages of unskilled workers in rich

countries while increasing income from capital, raising inequality within these

economies. Versions of this theoretical model are at the core of the debate in many

developed countries, where increased trade and outsourcing are assumed to be

harmful to unskilled workers. The same theoretical model, however, predicts that the

wages of unskilled workers in less developed countries will increase, lowering withincountry

inequality there.

The above reasoning assumes, however, that factor supply is constant. This is

a workable approximation only in the short term, before general equilibrium effects

kick in. Although wages decline and jobs are lost in some sectors in rich countries,

other sectors will benefit from trade and demand more labor, as emphasized by

Richardson (1995). Furthermore, more sophisticated theoretical models often feature

multiple equilibria at certain openness levels, which complicates the issue substantially

(see, e.g., Krugman and Venables 1995 and Das 2005).

Empirical evidence is also inconclusive. Sebastian (1997) finds that openness

to trade leads to increased income inequality in more developed economies, but not

in less developed countries. Savvides (1998), however, concludes that more open, less

developed economies experienced increased income inequality in the late 1980s.

Gourdon et al. (2008) find that the effects of lower tariffs on income inequality

108


Do Liberalization and Globalization Increase Income Inequality?

depend on relative factor endowments: In poor countries with a high share of lesseducated

labor, lower tariffs will raise inequality. Moreover, while both Lindert and

Williamson (2001) and O'Rourke (2001) support the position that economic

globalization is a force for income convergence between countries, they state that the

effect on inequality within countries is less clear. 2

Contradictory theoretical and empirical results leave us in a vacuum when it

comes to predicting the direction of any distributional effect of an increase in EFI4.

However, given that our use of data based on five-year averages is unlikely to capture

long-term general equilibrium effects, we rely on traditional trade theories and

hypothesize that higher EFI4 levels will increase income inequality in more developed

contexts but not in less developed ones. Overall, the relationship between EFI4 and

income inequality is ambiguous.

Regulation of credit, labor, and business (EFI5)

In this dimension, greater economic freedom means less regulation of credit markets,

labor markets, and business in general. Regarding the income distribution impact, this

issue is theoretically ambiguous. On one hand, increasing the availability of credit, for

example, will likely reduce income inequality, as a larger fraction of people will be

able to realize their potential (see, e.g., Galor and Zeira 1993 on economic growth).

On the other hand, such reform might also increase income inequality in cases where

the political elite can influence the format of a deregulation policy (see, e.g., Claessens

and Perotti 2007). If economic regulation creates monopoly rents, the impact of

deregulation on income inequality depends on how these rents are redistributed when

deregulation increases competition.

Existing empirical studies suggest that we should expect deregulation to

increase income inequality. For example, Calderón and Chong (2009) find that labor

market regulations reduce income inequality. Similarly, Fortin and Lemieux (1997)

argue that the declining real value of the minimum wage, declining unionization, and

2 Several relevant country studies also exist. According to Kumar and Mishra (2008), trade liberalization has

reduced wage inequality between skilled and unskilled workers in India. Acosta and Montes-Rojas (2008),

however, present more mixed evidence of the effect of trade liberalization on skill premiums in Mexico

and Argentina.

109


Chapter 4

general deregulation together explain a third of the increase in wage inequality in the

United States in the 1980s. Roine et al. (2009) moreover find financial development

to be particularly pro-rich in a context of relatively low incomes.

Since there are no clear theoretical underpinnings for the relationship between

deregulation and income inequality, we have no hypothesis regarding the sign of the

EFI5 component.

Now we focus on our measure of globalization, the KOF Index of Globalization

(KOF). This is a composite index weighting together three dimensions of

globalization, KOF1–KOF3.

Economic globalization (KOF1)

According to the KOF index, economic globalization is closely related to the fourth

dimension of the economic freedom index (EFI4). Several components are identical,

so our theoretical expectations regarding the relationship with income inequality are

more or less the same.

Two important aspects, however, are noteworthy. First, in contrast to the

EFI4, KOF1 includes information on foreign direct investment (FDI). As Feenstra

and Hanson (1997) argue that FDI increases the relative demand for skilled labor in

developed and developing economies, higher economic globalization levels should

relate to increasing income inequality in the latter context as well. Second, comparing

the individual components of KOF1 and EFI4, we see that the latter is slightly more

institutional, whereas the former relies more on actual flows of trade. This distinction

could be important. Rodriguez and Rodrik (2000) note that the significant link

between openness and growth, when openness is measured using flow variables, is

not robust when instead using institutional measures such as mean tariff rates. 3

Based on the theoretical underpinnings, we hypothesize that increases in

KOF1 will be associated with increasing income inequality in high-income settings,

while the same theoretical underpinnings suggest that such a change will have

ambiguous effects at lower development levels.

3 Their criticism is directed toward often-cited papers such as that of Sachs and Warner (1995).

110


Do Liberalization and Globalization Increase Income Inequality?

Social globalization (KOF2)

The social component of the KOF index captures factors such as outgoing telephone

traffic, number of Internet users, and number of IKEA and McDonalds outlets per

capita. No formal theory forecasts any specific effect of social globalization on

income distribution. Nevertheless, Atkinson (1997) notes that changing social norms

(which may follow from increased interaction and more integration among countries),

can affect economic inequality, for example, by influencing the behavior of unions,

resulting in larger wage differentials becoming more socially acceptable.

Following this argument, we predict that an increase in KOF2 will be

associated with higher income inequality, and do not assume that the impact will

differ between low- and high-income settings.

Political globalization (KOF3)

The third dimension of the KOF index measures the number of embassies,

membership in international organization, and participation in UN Security Council

Missions.

There is no obvious reason to expect such political cooperation to influence

income inequality. Studying the effect of globalization on human well-being, Tsai

(2007) notes that the international political system can bring supra-territorial interests

into domestic policy arenas, such as epidemic management, human rights issues, and

global environmental concerns, contributing to the advance of human well-being.

Using the KOF index, he finds political globalization to positively associate with the

Human Development Index (HDI). Furthermore, findings here indicate that

globalization increases the state’s revenue-extracting capacity, rejecting the idea that

closer integration constrains state capacity.

Since neither theory nor empiricism provides any indication that KOF3 is

associated with income inequality, we hypothesize that this component will have no

distributional effect.

From the above we are uncertain about the inequality impact in several cases, but to

summaraize: in this setting we hypothesize that the coefficients of EF1, EFI4 and

111


Chapter 4

KOF1 will be positive in richer contexts, while inequality chould decrease and

increase, respectively, with EFI3 and KOF2 focusing on the full sample.

The relationship between different types of liberalization and globalization

As table 4.1 illustrates, several dimensions of economic freedom and globalization are

highly correlated. In particular, EFI4 is, as expected, highly correlated with KOF1.

Thus, the matrix somewhat confirms the common view that nations with liberal

policies in some areas also tend have them in other areas. However, the cross-country

correlations also reveal that EFI1 is negatively correlated with both EFI2 and all

other measures of globalization. In other words, countries with big governments are

on average more globalized and have more legal system integrity. This indication

corroborates the view that reform programs may affect diverse policy arenas

differently.

Table 4.1 Correlations between components of the economic freedom index and the KOF

globalization index

EFI EFI1 EFI2 EFI3 EFI4 EFI5 KOF KOF1 KOF2 KOF3

EFI 1

EFI1 0.23 1

EFI2 0.71 -0.33 1

EFI3 0.80 0.10 0.42 1

EFI4 0.85 0.05 0.62 0.58 1

EFI5 0.79 0.09 0.62 0.49 0.63 1

KOF 0.76 -0.22 0.75 0.57 0.77 0.65 1

KOF1 0.76 -0.13 0.67 0.53 0.82 0.66 0.91 1

KOF2 0.78 -0.17 0.73 0.58 0.75 0.68 0.96 0.86 1

KOF3 0.40 -0.32 0.53 0.35 0.40 0.31 0.75 0.46 0.59 1

4.2.2 Empirical examinations using the EFI and KOF indices

Three studies, i.e. Berggren (1999), Scully (2002), and Carter (2007), examine the

economic freedom index in relation to income inequality in a cross-country setting.

Only Dreher and Gaston (2008) have so far analyzed the relationship between the

KOF index and inequality. 4

4 A paper that also deserves mention is that of Ashby and Sobel (2008), who use data on US states. They

find that increased economic freedom between 1980 and 2003 have reduced income inequality by

increasing incomes relatively more for low-income groups.

112


Do Liberalization and Globalization Increase Income Inequality?

The results regarding economic freedom are surprisingly contradictory. Scully

(2002) and, to a lesser extent, Berggren (1999) claim to find evidence that economic

freedom reduces income inequality by examining the issue using data on 26 and 66

countries, respectively, over the 1975–1990 period. These early studies, however,

suffer from several problems, including the problem of non-comparable Gini

coefficients and limited data availability. 5 Improving on several weaknesses, Carter

(2007) is the first to analyze the question in a panel setting. In contrast to Berggren

and Scully, he finds a positive but relatively inelastic relationship, where an increase in

economic freedom of two standard deviations leads to an increase in the Gini

coefficient of 0.33 standard deviations. 6

Carter’s unbalanced panel runs from 1980 to 2000 but contains data for only

seven countries in 1980 and 15 countries in 1985. The efficient sample in the

empirical analysis refers to 104 observations from 39 countries, most of which are

OECD members, which limits the possibility of analyzing the relationship between

economic freedom and income inequality at different development levels. As Carter’s

study focuses on the income distribution effect within countries of overall economic

freedom, the examination focuses entirely on the composite index.

Using the KOF index and income and wage inequality data, Dreher and

Gaston (2008) find evidence that globalization on average has increased income

inequality in OECD countries from 1970 through 2000. However, their findings do

not identify any robust impact in less-developed nations. Not discriminating between

economic, social, and political globalization, their baseline examination refers to a

sample of approximately 400 observations (varying somewhat in size depending on

whether focus is on wage or income inequality).

4.3 DATA AND EMPIRICAL SPECIFICATIONS

The data comprise a panel of observation from 81 countries covering the 1970–2005

period (table 4A in the Appendix provides information on the country coverage of

5 Carter (2007) provides a comprehensive review of the problems in earlier studies.

6 In fact, Carter estimates a quadratic relationship between economic freedom and inequality. For all but

three observations, however, the index value is high enough that an increase in freedom is estimated to

raise inequality. We estimate a linear relationship, but estimate a quadratic model as one of our robustness

checks.

113


Chapter 4

the data). To reduce the possibility that short-term movements and measurement

errors may affect the results, the data are averaged over five-year periods resulting in

eight distinct periods. With regard to income inequality, the initial measure is the

average for the 1965–1970 period. 7 This results in the first period of the panel

containing information on 51 and 41 countries, depending on whether we are

examining the distributional effects of the KOF or EFI, respectively.

Since country data are not available for all periods, the panel is unbalanced.

Still, the efficient sample consists of more than 500 observations meeting baseline

specifications. Roughly 40% of these observations refer to conditions in countries

classified as low- or low-middle-income countries with a 2007 GNI per capita of

USD 3705 or less.

4.3.1 Dependent variables - On the use and misuse of inequality data

Among the most commonly used measures of inequality are the Gini coefficients.

For completely egalitarian income distributions in which the whole population has

the same income, the Gini coefficient takes a value of 0. A value of 1 indicates that all

incomes are concentrated in one person.

Gini coefficients can be calculated in several ways: for gross income (before

taxes and transfers), net income (after taxes and transfers), and consumption

expenditure. Furthermore, the unit of analysis can be individuals or households. The

lack of comparable Gini coefficients both between countries and over time has long

been a major obstacle in inequality research. Many consider the Luxembourg Income

Study (LIS) to be the best option, as it is based on reliable microdata from national

household income surveys. Unfortunately, LIS data are available for only thirty

countries, almost exclusively rich ones, and contain few observations from before

1990.

As a second best solution, many scholars resort to the World Income

Inequality Database (WIID), created by the World Institute for Development

Economics Research of the United Nations University (UNU-WIDER). This is an

updated and expanded version of the Deininger and Squire (1996) dataset, used by,

7 The eight periods are thus 1965–1970, 1971–1975, 1976–1980, 1981–1985, 1986–1990, 1991–1995,

1996–2000, and 2001–2005. The first period includes 1965 because of data availability.

114


Do Liberalization and Globalization Increase Income Inequality?

for example, Berggren (1999). The WIID contains a large set of inequality statistics

from several sources, totaling over 5000 observations from 160 countries. However,

as Deininger and Squire have themselves pointed out, the observations are rarely

comparable across countries or over time within a single country. 8

Two recent papers have attempted to handle the problem of few and noncomparable

Gini measures: the Standardized Income Distribution Database (SIDD)

created by Babones and Alvarez-Rivadulla (2007), and the Standardized World

Income Inequality Database (SWIID) created by Solt (2008). Both the SWIID and

the SIDD aim to improve data availability and comparability for cross-national research

by exploiting the fact that different types of Gini coefficients display systematic

relationships. The Gini coefficient of gross income is typically larger than the

coefficient of net income, which in turn is larger than the Gini coefficient of

expenditure. Similarly, Gini coefficients for households are typically lower than

coefficients calculated on an individual basis. 9 For example, Deininger and Squire

(1996) recommend adding three points to net-income-based inequality observations

to make them comparable with the gross-income-based observations.

There are problems, however, with such a constant adjustment procedure.

For reasons explained in Bergh (2005) and Uusitalo (1985), the difference between

gross and net income Gini coefficients depends on the degree to which taxes and

transfers are progressive and redistribute income from rich to poor. As a result, the

difference varies across countries and within countries over time, so constant

adjustment will introduce systematic errors into the data. The same line of reasoning

applies to the empirical strategy of including dummy variables to correct for different

types of Gini coefficients being used in the same regression, as this also assumes that

differences between different types of Gini coefficients remain constant over time.

The adjustment procedure is the major reason for preferring the SWIID to

the SIDD: Babones and Alvarez-Rivadulla (2007) use a constant adjustment

8 Nevertheless, hundreds of cross-country studies use the Deininger and Squire dataset. It is often hard to

tell how or even whether authors have dealt with the problem of non-comparable Gini coefficients. Solt

(2008) notes that Deininger and Squire’s recommendations on how to use their data are often entirely

overlooked by researchers.

9 Gross Gini coefficients are larger than net Gini coefficients because taxes and transfers typically equalize

the income distribution. The Gini coefficient of consumption is smaller because people use savings and

loans to smooth consumption.

115


Chapter 4

procedure to compensate for missing data. Solt (2008), however, uses various

techniques to estimate the ratios between different types of Gini coefficients, relying

more on information about the ratio in the same country nearby in time, to increase

the number of comparable observations. 10 An additional advantage of Solt (2008) is

the provision of estimates of uncertainty for observations, which implies that users

can easily do robustness tests with respect to their chosen inequality measure.

Our preferred distributional measure and dependent variable is the net income

Gini coefficient from Solt (2008). Despite the improvements in data availability and

comparability in this source versus other available inequality databases, it is important

to emphasize that measurement errors originating from the raw data will remain

through the adjustment procedure, which might affect trends and levels of withincountry

inequality. Such errors can be traced to differences or errors in the collection

of underlying data, and to dissimilarities in methodology across countries or over

time. 11 Moreover, in cases when underlying datapoints are few, SWIID estimates are

close to a linear interpolation between available observations (Solt 2008). This might

affect the variation in within-country income inequality measures, which could bias

estimation results.

As a test of sensitivity, we use the gross income Gini coefficient, also from Solt

(2008), and the Kuznets ratio as dependent variables. The latter measure is calculated as

the ratio of income shares of high- to low-income earners (80th to the 40th percentile).

In contrast to the Gini coefficient, which is most sensitive to changes at the mode of

the income distribution, the Kuznets ratio is sensitive to changes in the upper and

lower parts of the distribution. The effects of liberalization or globalization gauged

using this measure might thus be different from those gauged using the Gini

coefficient if, for example, the very top income earners are the ones who benefit the

most from a reform (see, e.g., Roine et al. 2009 on the effect of trade openness and

financial development on top incomes). Information on income shares comes from

WIID 2.0c (Wider 2008). To maximize comparability, the country datapoints are well

matched. In other words, information from time t in country i is determined as in t-1,

10 Another advantage of the SWIID over the SIDD is that the former is based on version 2.0c of the

WIID data, released in May 2008, whereas the latter relies on the older WIID version 1.0.

11 For a general discussion of the use of secondary inequality datasets, see Atkinson and Brandolini (2001).

116


Do Liberalization and Globalization Increase Income Inequality?

with respect to income measure, unit of analysis, etc.; however, the above caveats still

apply.

4.3.2 Independent variables

To measure economic freedom and globalization, we use the EFI and KOF indices.

The former was developed by Gwartney and Lawson (2003) and covers a large

number of countries every fifth year since 1970, and yearly since 2000; we use the

2008 dataset. The composite index and its subcomponents range from 0 to 10, 0

indicating the lowest and 10 the greatest economic freedom. 12 The composite EFI

exists in a chain-linked version, suitable for analysis over time, which we use in our

analysis. As discussed previously, we also examine the association with inequality and

the five subdimensions. Since the subcomponents are not completely comparable

over time, these results should be interpreted with care. 13

The KOF index was developed by Dreher (2006) and covers more than 120

countries on a yearly basis from 1970 through 2008. The composite index and its

subcomponents take values between 0 and 100, higher values representing more

globalization. Table 4B and 4C in the Appendix provides the details of the areas and

components of the EFI and KOF indices.

We include a number of control variables in specifications to correct for the

influence that factors other than economic freedom and globalization may have on

income inequality. Adopting measures similar to those used in previous studies, we

add three control variables to our baseline regression. First, the model includes log of

real GDP per capita to correct for any distributional effects driven by income levels.

Following Kuznets (1955), we expect inequality to follow an inverted U-curve over

levels of development and this control variable to be positive.

Second, we include a variable for the share of population above 25 years old with

higher education to correct for human capital effects. Theoretically the impact of higher

education on inequality is ambiguous. More people with higher education implies that

12 As discussed by De Haan et al. (2006), the EFI has been criticized for being ideologically biased, but it

has nevertheless often been used in research as a descriptive device.

13 Numerous authors, however, have successfully obtained results using the subcomponents of the index;

see Carlsson and Lundstrom (2002) and Berggren and Jordahl (2005) on the relationship between types of

economic freedom and growth, and Berggren (1999) on the relationship between types of economic

freedom and inequality.

117


Chapter 4

a larger share of the population will enjoy the wage premium. Such a development,

however, may also serve to reduce the premium associated with higher education. 14

Moreover, Krusell et al. (2000) present a model emphasizing the complementarity of

capital and skills as drivers of increasing income inequality in high-income contexts.

Lindquist (2005) moreover demonstrates that such complementarities raised the

premium to higher education in Sweden (and thus income inequality), even during a

period of increasing relative supply of well-educated workers.

Third, the baseline model includes a dependency ratio, corresponding to the

share of population younger than 15 years and older than 64 years. The primary

effect of demographic change is the modification of the population age distribution.

Following Higgins and Williamson (1999), and assuming relatively large cohorts to

obtain low earning rewards, income inequality will decrease when relatively large

cohorts are mature and are situated at the top of the age–earnings curve. When these

cohorts are young adults or old, inequality increases. 15 Although an ideal indicator

would measure the size of the mature cohort in relation to the number of adults in

the population, we predict that a higher dependency ratio will be associated with

higher income inequality.

To examine the robustness of our results, we modify the baseline model in

several ways. One set of sensitivity tests involves adding further covariates. Following

Kuznets (1955), we include a variable for the share of labor force employed in the industrial

sector and its square to control for the structure of the economy. A small but

increasing share of people employed in the modern sector will widen the gap between

rich and poor. When the manufacturing sector provides a larger share of less-skilled

workers an opportunity to earn higher wages, income inequality will eventually

decrease.

We also test the robustness of our results by including the population share living

in urban areas. Following traditional development theories, urbanization mirrors

14 Education data exist on a five-year basis from 1960 through 2000. To fully explore existing information

on inequality, economic freedom, and globalization, data on human capital in 2005 are estimated as the

year 2000 value plus the country change over the 1995–2000 period. Our results do not depend on this

operation. Results are robust to excluding the final period, to assuming human capital to be constant

between 2000 and 2005, and to the use of a lagged human capital variable.

15 This framework also assumes poor people to be evenly distributed in the population.

118


Do Liberalization and Globalization Increase Income Inequality?

economic development, so we also expect this variable to be nonlinear to inequality.

Nevertheless, there are also arguments in the inequality literature that relate larger

shares of urban populations to higher degrees of population heterogeneity (Wirth

1938), arguing for the existence of only a positive association. Finally, we perform a

sensitivity test in which we replace our measure of human capital with the average years

of education in the population above 15 years old.

Except for the data corresponding to human capital and wealth that come

from Barro and Lee (2000) and Heston (2006), respectively, independent variables

come from the World Development Indicators (World Bank 2008). Tables 4C and

4D in the Appendix provide descriptive statistics, information on exact definitions,

and the sources of all variables and cross correlations.

4.3.3 Empirical strategy

To analyze the effect of economic freedom and globalization on inequality, we

formulate the following empirical model, where countries are represented by i and

time by t :

y it = α + libit β + xit


+ δ i + ρ t + ε it

' (4.1)

Here, it y is the dependent variable of interest, lib it is a vector of indices of

liberalization, and it x includes the additional covariates presented above. δ i

corresponds to a country fixed effect that captures stable differences in economic

inequality between countries, while ρ t is a period fixed effect, capturing the influence

of shocks that affect economic inequality in multiple countries at the same time. ε it is

a normally distributed error term. As a baseline, the EFI1–EFI5 and KOF1–KOF3

subindices are included separately, and we estimate the relationships of interest by

least squares and country fixed-effects. 16

16 Views differ as to whether to include the various dimensions of liberalization and globalization

simultaneously or not. Heckelman and Stroup (2005) argue that any summary index may result in a

misspecification bias, and suggest also performing the analysis using the actual individual components.

Dreher and Gaston (2008), however, argue that components of globalization should be regressed in the

same specification, as the different components are highly correlated, to control for other globalization

dimensions. To avoid problems caused by multicollinearity, our preferred approach is to include the

indices separately.

119


Chapter 4

The above specification has a potential endogeneity problem: The levels of

economic freedom or globalization might be influenced by the changes in income

inequality, and not just the other way around. Gradstein (2007), for example, states

that the more equal the income distribution in a society, the greater the support for

property rights protection. Politicians may respond to increases in income inequality

by implementing certain policies, favoring either more or less economic freedom or

globalization depending on their preferences and beliefs about the causes of

inequality. If an increase in inequality reduces liberalization and globalization, we

belive our analysis may underestimate the inequality impact.

The endogeneity problem has been handled in different ways in the related

literature. Berggren (1999), using cross-country data, let the independent variables

(policy reforms) predate the dependent one (inequality); this is also the strategy used

by Ashby and Sobel (2008). In our setting, this would correspond to regressing

inequality on the lagged indices, lib it−1

, and we estimate this specification as one of

several robustness tests. The panel data structure, however, also lets us handle

potential endogeneity by estimating our model using a system GMM estimator

(Arellano and Bover 1995, Blundell and Bond 1998). 17 In a single system, this

estimator combines the regression equations of both differences and in levels, each

having a particular set of instrumental variables. Specifically, the system is jointly

estimated using first-difference equations instrumented by lagged levels and using

level equations instrumented by the first differences of the regressors. If these

variables are appropriate instruments, the estimator should be consistent in the

presence of endogenous variables. The system GMM estimator is also consistent in

the presence of country-specific effects and the estimation method works for

unbalanced panels and situations with few periods and many countries. 18

GMM specifications are preferred by Dreher and Gaston (2008), whereas

Carter (2007) does not discuss the potential endogeneity problem. We regress our

17 We use the Stata command xtabond2 to estimate the system GMM. See Bond et al. (2001) and Roodman

(2006) for a rigorous outline of the method and the syntax.

18 As demonstrated by Arellano and Bond (1991), the GMM difference estimator could also be used in this

context. In empirical examinations, however, the difference estimator often performs poorly when the

number of periods, as in our case, is limited (Bond et al., 2001). Moreover, the difference estimator does

not allow for country-specific effects.

120


Do Liberalization and Globalization Increase Income Inequality?

model using the system GMM estimator as another robustness test of our baseline

results. This also lets us test the sensitivity of our findings when including a lagged

dependent variable.

While the benefits of a panel dataset are evident, the choice to construct a

panel based on five-year averages is not an obvious one as within-country Gini

coefficients remain relatively stable over time. As an alternative to the panel

specification, we study the development of globalization and inequality by

considering the difference over a longer period, a method used by, for example,

Sylwester (2002), in analyzing the effects of education policy on income inequality. 19

In this exercise, we examine whether the changes in the dimensions of liberalization

and globalization between 1980 and 2000 are associated with increasing income

inequality over the 1985–2005 period. 20 In this case, our empirical specification is

Δ yi

= α + β ( Δlib)

i + γ ( xi

1980 ) + ε i

, (4.2)

where i y Δ and Δ libi

correspond to the differences in income inequality and

liberalization in country i, respectively, computed as the level of inequality or index of

interest in the last period minus the level in the first one. This method also lets us

examine whether the results are robust to using an alternative measure of income

inequality. Thus, equation 2 is estimated using both the Gini coefficient and Kuznets

ratio as the dependent variable.

19 The same method is also used by Bergh and Fink (2008), Savvides (1998), and Sebastian (1997).

20 The periods differ to minimize the risk of reverse causality between policy affect and inequality.

121


Chapter 4

4.4 EMPIRICAL ANALYSIS

Before estimating our model, we study the pairwise correlations among independent

variables. As table 4E in the Appendix illustrates, some of the indicators are closely

related. Examining the variance inflation factor (VIF), however, suggests that there is

no incidence of multicollinearity. 21

4.4.1 Basic results

We begin the empirical analysis by estimating baseline specifications by least squares

and country fixed effects. The dependent variable refers to the country Gini

coefficients of net incomes. All regressions include period dummies and we employ

robust standard errors throughout the empirical examination to account for

heteroscedasticity. To maximize comparability, in all estimations focusing on the

distributional impact of dimensions of the EFI, the sample contains the same

countries. The equivalent approach is applied in the regressions including KOF

indices. The number of observations might, however, vary across index-specific

estimations.

Some interesting evidence is presented in tables 4.2 and 4.3. First, overall

economic freedom is positively associated with income inequality. Testing the

components of the EFI separately, this result appears to be driven by the

disequalizing impact of three subcomponents: EFI1, EFI4, and EFI5. The first

finding supports the hypothesis that reduced government size on average increases

income inequality. Moreover, the evidence that a greater amount of international

trade is positively associated with inequality is also theoretically reasonable, given that

most sample observations correspond to middle- or high-income countries.

Interestingly, market deregulation seems to widen the gap between rich and poor,

which supports the political economy argument that an elite gains most of the

benefits of such liberalization initiatives while the risks are shared by a larger group.

Among the EFI components, deregulation in fact has the quantitatively greatest

impact on inequality.

21 The VIF test can only be calculated for pooled regressions. Numbers for individual variables range from

1.9 (EFW) to 6.2 (GDP per capita), which is below the critical value of 7. In most cases, the average VIF is

well below 3.

122


Do Liberalization and Globalization Increase Income Inequality?

Table 4.2 Net income inequality and the dimensions of economic freedom, OLS fixed effects

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

GDP per capita 3.861** 4.212*** 3.606** 3.641** 4.149*** 4.338**

[1.476] [1.305] [1.424] [1.424] [1.369] [1.647]

Human capital 0.345** 0.347** 0.316** 0.310** 0.385** 0.328*

[0.161] [0.145] [0.139] [0.154] [0.147] [0.172]

Dependency ratio 4.665 6.545** 2.972 5.426 4.008 4.916

[3.432] [3.200] [3.425] [3.301] [3.011] [3.614]

EFI 0.701**

[0.334]

EFI1 0.453*

[0.257]

EFI2 0.142

[0.202]

EFI3 0.009

[0.151]

EFI4 0.604**

[0.249]

EFI5 1.091***

[0.379]

Constant 4.001 -9.631 -0.116 -0.838 3.309 -1.810

[10.413] [12.365] [13.560] [12.643] [9.772] [11.884]

Number of countries 81 81 81 81 81 81

Observations 505 524 491 518 508 496

R-squared (within) 0.146 0.163 0.132 0.147 0.159 0.153

Period dummies (Prob > F) 0.01 0.00 0.01 0.00 0.02 0.01

Country dummies (Prob > F) 0.00 0.00 0.00 0.00 0.00 0.00

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.

Second, neither more secure property rights nor better access to sound money has a

significant impact on income inequality. The latter finding contradicts the theoretical

prediction and much of the existing literature regarding the distributional costs of

unexpected inflation. Potentially, the result follows from our using a measure of

income inequality excluding capital returns.

Third, repeating the above exercise using KOF indices, it is clear that the

positive link between overall globalization and income inequality comes from a

significant relationship between the economic and social dimensions of globalization

and inequality, respectively. Political globalization, however, has no significant effect

on income inequality. These results support the hypotheses in section 4.2.1 and

evidently suggest that a purely economic perspective on globalization might be too

narrow in analyzing distributional effects across countries.

123


Chapter 4

Table 4.3 Net income inequality and the dimensions of globalization, OLS fixed effects

(1) (2) (3) (4)

GDP per capita 3.490** 3.899*** 3.590** 4.127***

[1.327] [1.441] [1.384] [1.464]

Human capital 0.304** 0.296* 0.270* 0.338**

[0.150] [0.151] [0.153] [0.162]

Dependency ratio 5.949* 5.722 5.103 5.459

[3.543] [3.498] [3.465] [3.419]

KOF 0.145*

[0.077]

KOF1 0.090*

[0.052]

KOF2 0.094**

[0.039]

KOF3 0.006

[0.029]

Constant -8.724 -8.988 -4.975 -5.783

[13.256] [13.402] [12.736] [13.864]

Number of countries 80 80 80 80

Observations 523 523 523 523

R-squared (within) 0.161 0.154 0.158 0.138

Period dummies (Prob > F) 0.01 0.00 0.01 0.01

Country dummies (Prob > F) 0.00 0.00 0.00 0.00

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.

The influence of GDP per capita on income inequality is in line with the theoretical

expectation and consistently positive across estimations. Results moreover suggest

that a larger share of the population having a higher education increases inequality,

which might mirror an average rise in returns on human capital investments. The

positive association between education and income inequality corroborates the

findings of Carter (2007) and Berggren (1999), who use measures of the average

education of the population and illiteracy levels respectively. The coefficient of the

demographic indicator is positive, consistent with the view that smaller cohorts

benefit from greater income rewards; however, this control variable is significant in

only two of ten cases.

The null hypothesis of no country effects is rejected in all estimations,

implying that a pooled regression model is inappropriate. Moreover, the randomeffect

model is rejected by a standard Hausman test against the fixed-effect model,

which supports our methodological choice. Time dummies are jointly significant in all

specifications, implying they should be included in the model.

124


Do Liberalization and Globalization Increase Income Inequality?

4.4.2 Sensitivity analysis

(a) Variations on baseline specification

To examine the robustness of the baseline findings, we carry out several sensitivity

checks; table 4.4 summarizes these results. All regressions include all control

variables, but we only present the variables of main interest, i.e., the composite

indices and the significant subcomponents.

The first sensitivity assessment addresses the EFI and KOF indices. It is

reasonable to assume that the income distribution impact of liberalization could

appear with a time lag. However, using lagged index values does not support this

notion, and the regression results turn out more or less the same as in our baseline

scenario. In fact, in contrast to initial findings, this exercise provides no evidence that

smaller governments augment income inequality. The same result appears when we

examine whether the results are driven by outliers. Excluding countries with extreme

values on the KOF and EFI indices from the sample also reveals that the initially

positive association between economic globalization and income inequality is not

robust.

The results of simultaneously including all dimensions of the EFI (or of the

KOF index) in one specification provide additional support for the notion that

increasing trade liberalization and social globalization raise income inequality. In this

test, the significant association between the deregulation component and inequality

disappears; moreover, government size and economic globalization again lose

significance. Following Carter (2007), we also include quadratic terms for the indices.

This entirely eliminates the significance of the economic freedom index, resulting in

only the social globalization component being robust to this specification. The

quadratic term is small and insignificant, giving little support to the quadratic

specification. To sum up, it is evident that the baseline finding regarding EFI1 is not

robust. In fact, all of the remaining sensitivity assessments indicate a non-significant

relationship between government size and income inequality.

The second type of robustness test involves the dependent variable.

Excluding ten countries with extreme Gini values does not change our findings thus far.

Similarly, excluding relatively uncertain Gini estimates from the sample changes the

125


Chapter 4

results only marginally. We moreover replace Gini coefficients of net income with

their gross income equivalents. If anything, the significant coefficients (EFI4, KOF1,

and KOF2) are now bigger, in line with the idea that the difference between gross

and net income reflects behavioral adjustment to policy changes (see Bergh 2005).

The next assessment addresses how the initial results are affected by including

additional covariates or by replacing indicators in our baseline model. When replacing

the human capital variable with an alternative measure, the initial findings change with

respect to only one component. As previously, EFI1 is no longer significantly related

to income inequality. Furthermore, when adding information on the degree of

urbanization or on the share of the labor force employed in the industrial sector to the

specifications, EFI4, EFI5, KOF1, and KOF2 remain positive and significant. The

urbanization coefficient is never significantly different from zero, irrespective of

whether we include only the level covariate or also a squared version. Moreover, in

contrast to theoretical predictions, the employment share variable and its square are

never significant when included together. Specifying a model including only the

employment share, however, provides evidence of a negative relationship with

inequality in the models focusing on economic freedom. The same exercise generally

removes significance with respect to the GDP per capita and human capital variables,

which highlights the structure of the economy as a potentially more important

determinant of income inequality. However, the result seems to be driven by the

unavailability of data on this indicator for certain countries. 22

Finally, we examine the robustness of our findings by excluding countries in

the geographical regions sub-Saharan Africa, Latin America, and East Asia from the

sample. With respect to economic freedom, this changes the results only marginally.

The distributional impact of EFI4 and EFI5 is somewhat stronger and of greater

magnitude when excluding sub-Saharan African and Latin American countries,

suggesting that the positive effect of trade liberalization and market deregulation on

inequality is more perceptible in the remaining sample, which includes relatively

22 The inclusion of share of employment in industry reduces the sample to 325 observations from 72

countries. Running baseline regressions on the same sample, GDP per capita and human capital are also

insignificant, while the relationship between EFI components and inequality is robust and similar to the

findings in section 4.4.

126


Do Liberalization and Globalization Increase Income Inequality?

developed European and North American countries. However, the baseline

indication of a positive association between economic globalization and income

inequality completely disappears in all three settings. Excluding countries from sub-

Saharan Africa also removes the significant effect of KOF2. Since several of the

excluded countries are developing countries, this result could indicate that

globalization mainly has a distributional impact in low-income contexts, a

consideration we examine in section 4.4.2.

(b) Longer time period

In addition to the sensitivity tests reported in table 4.4, we analyze whether changes

in liberalization and globalization over a longer period are associated with higher

income inequality. Table 4.5 presents the results when using Gini coefficients of net

income (column 1) or the Kuznets ratio (column 2) as the dependent variable. Due to

data limitations, the sample is smaller than that used in baseline estimations. In

particular, the sample is small and mainly consists of high- and upper-middle-income

countries when using the Kuznets ratio (table 4A in the Appendix provides

information on the country coverage). Still, the results in general confirm our

conclusions so far.

Column 1 shows that market deregulation stands out, in terms of both size

and significance, in Gini estimations. Furthermore, social globalization has an

unequalizing impact when using this dependent variable, while estimations using the

Kuznets ratio provide evidence that freedom to trade internationally has a significant

positive influence. This empirical estimate possibly confirms theoretical predictions.

According to H–O theory, capital owners benefit more than do low skilled workers

from trade in high-income countries. Assuming that capital owners appear in the

upper tail of the income distribution, the positive effect of EFI4 should particularly

appear when using the Kuznets ratio as our dependent variable.

127


Chapter 4

Table 4.4 Summary of sensitivity tests

Variation Composite index Significant components

Baseline EFI 0.701** [0.334] EFI1 0.453* [0.257]

EFI4 0.604** [0.249]

EFI5 1.091*** [0.379]

KOF 0.145* [0.077] KOF1 0.090* [0.052]

KOF2 0.094** [0.039]

Lagged EFI and KOF indices (t-1) EFI 0.609 [0.403] EFI4 0.425* [0.228]

EFI5 1.120*** [0.384]

KOF 0.177** [0.081] KOF1 0.141** [0.061]

KOF2 0.084** [0.037]

Excluding countries with extreme EFI indices EFI 0.670* [0.343] EFI4 0.634** [0.253]

(4 countries) EFI5 1.108*** [0.391]

Excluding countries with extreme KOF indices KOF 0.142* [0.082] KOF2 0.106** [0.040]

(4 countries)

All EFI sub-indices together EFI4 0.706*** [0.244]

All KOF sub-indices together KOF2 0.073* [0.041]

Quadratic specification EFI EFI -1.958 [1.761]

EFI^2 0.243 [0.164]

Quadratic specification KOF KOF 0.211* [0.119] KOF2 0.190** [0.076]

KOF^2 -0.001 [0.001] KOF2^2 -0.001 [0.001]

Excluding countries with extreme Gini coefficients EFI 0.587* [0.324] EFI4 0.653*** [0.239]

(10 countries) EFI5 0.863** [0.347]

KOF 0.096 [0.063] KOF1 0.065* [0.038]

KOF2 0.078** [0.037]

Excluding Gini coefficients with extreme standard errors EFI 0.388 [0.380] EFI4 0.557** [0.262]

(19 observations) EFI5 1.033*** [0.377]

KOF 0.124 [0.075] KOF1 0.088* [0.052]

KOF2 0.079** [0.037]

Replacing net income Gini by gross income Gini EFI 0.742 [0.480] EFI4 0.731** [0.357]

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.

KOF 0.243** [0.095] KOF1 0.145** [0.067]

KOF2 0.173*** [0.049]

128


Table 4.4 continued

Do Liberalization and Globalization Increase Income Inequality?

Variation Composite index Significant components

Alternative measure of education EFI 0.735* [0.370] EFI4 0.449* [0.270]

EFI5 1.556*** [0.440]

KOF 0.173* [0.090] KOF1 0.102* [0.060]

KOF2 0.122*** [0.041]

Including urban population EFI 0.663* [0.334] EFI4 0.598** [0.246]

EFI5 1.066*** [0.382]

KOF 0.144* [0.077] KOF1 0.090* [0.052]

KOF2 0.099** [0.039]

Including share of employment in industry EFI 0.480 [0.343] EFI4 0.668** [0.262]

EFI5 0.604* [0.322]

KOF 0.102* [0.052] KOF1 0.086** [0.035]

KOF2 0.090*** [0.032]

Excluding sub-Saharan countries EFI 0.888*** [0.313] EFI4 0.752*** [0.238]

( 14 countries) EFI5 1.245*** [0.392]

KOF 0.049 [0.064]

Excluding Latin American countries EFI 0.703 [0.540] EFI4 0.599** [0.270]

(22 countries) EFI5 1.350** [0.521]

KOF 0.224*** [0.080] KOF2 0.103** [0.040]

Excluding East Asian countries EFI 0.572 [0.370] EFI4 0.473* [0.245]

(11 countries) EFI5 0.905** [0.366]

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.

KOF 0.117 [0.076] KOF2 0.068* [0.040]

129


Chapter 4

Table 4.5 Liberalization, globalization, and inequality increase between 1985 and 2005 23

Gini Kuznets ratio

EFI 2000-1980 0.625 0.156

[0.506] [0.118]

EFI1 2000-1980 0.195 0.041

[0.322] [0.065]

EFI2 2000-1980 -0.243 -0.021

[0.333] [0.061]

EFI3 2000-1980 0.273 0.021

[0.226] [0.043]

EFI4 2000-1980 0.122 0.127*

[0.386] [0.072]

EFI5 2000-1980 1.340* 0.238

[0.691] [0.170]

KOF 2000-1980 0.129* 0.011

[0.0757] [0.013]

KOF1 2000-1980 0.059 0.004

[0.0620] [0.011]

KOF2 2000-1980 0.098* 0.006

[0.054] [0.008]

KOF3 2000-1980 0.023 0.004

[0.045] [0.010]

Observations 59 or 60 39 or 41

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets.

(c) GMM estimation

As a final robustness test, we apply the system GMM estimator developed by

Arellano and Bover (1995) and Blundell and Bond (1998). In addition to providing

consistent results in the presence of endogenous variables, this estimator allows the

inclusion of a lagged value of inequality. While fixed-effect estimations control for

country characteristics that are constant over time, including the lag of inequality will

control for the longer-term impacts of our existing independent variables and for

omitted variables that change over time in a way that could drive the results. As noted

by Owen and Wu (2007), the inclusion of the lagged independent variable also

changes the model slightly, from examining how liberalization and globalization

relates to the level of income inequality, to examining the growth of inequality.

23 Regressions control for the 1980 values of GDP per capita, human capital, and the dependency ratio.

Moreover, the initial inequality value and dummies for Latin America and East Asia are included. As

above, indices are included separately in the estimations.

130


Do Liberalization and Globalization Increase Income Inequality?

Following Roodman’s (2006) recommendations, we use a technique to reduce

the number of instruments and test the sensitivity of our results with respect to lag

lengths, since the GMM estimator easily becomes biased due to over-identification of

instruments. The estimations treat the lagged information on income inequality and

the different indices as endogenous and all other variables as exogenous. Moreover,

we use a two-step estimator, including Windmeijer’s (2005) finite sample correction.

Before interpreting the results, we note that the Hansen J-test suggests that

the instruments are valid. We also conduct the Arellano–Bond test for second-order

autocorrelation. As table 4.6 shows, there is no significant serial correlation in our

specifications, so the estimator should be consistent. As previously, EFI4, KOF,

KOF1, and KOF2 all come out significant and with the same positive sign, so these

results are not driven by an endogenous relationship. However, the positive effect of

market deregulation on income inequality disappears when using the system GMM

estimator. This result suggests that this liberalization reform might be affected by

changes in the income distribution. Surprisingly, the GDP per capita coefficient is

now negative and significant. Moreover the relationship between the human capital

variable and income inequality no longer appears. However, in contrast to what was

previously the case, the effect of the dependency ratio is now significant and large.

To summarize our sensitivity analysis, the positive effect of EFI4 is very robust. We

are confident in this result, as economic freedom remains significant across all

sensitivity tests. Moreover, KOF2 is repeatedly found to have a positive impact on

income inequality. This is also often true for KOF1 when using a full sample, but

these results are sensitive to the exclusion of particular countries. Regarding EFI5,

the component is generally found to be significant in FE estimations, though the

effect could be driven by reverse causality. The baseline finding with respect to EF1

never reappears in any of the robustness assessments, indicating that government size

has no clear effect on income inequality. Moreover, in line with our previous findings,

EFI2, EFI3, and KOF3 never have a significant distributional impact. As a result, the

composite indices are often insignificant in our specifications, though essentially

masking the inequality impact of certain subcomponents.

131


Table 4.6 System GMM estimation with lagged inequality and liberalization and globalization index values as endogenous

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Inequality (t-1) 0.500** 0.290 0.636*** 0.446* 0.724*** 0.404 0.278 0.259 0.220 0.461**

[0.233] [0.307] [0.199] [0.265] [0.148] [0.296] [0.253] [0.253] [0.309] [0.207]

GDP per capita -1.434* -1.292 -1.853** -0.799 -1.400* -1.436 -2.542** -3.006** -3.003** -0.238

[0.850] [1.022] [0.930] [0.743] [0.811] [1.310] [1.214] [1.263] [1.527] [0.648]

Human capital 0.021 0.040 0.073 0.026 0.077** 0.014 -0.005 0.058 -0.015 -0.066

[0.086] [0.091] [0.065] [0.085] [0.036] [0.102] [0.098] [0.104] [0.112] [0.108]

Dependency ratio 14.744** 22.009* 9.230* 16.118* 6.920** 15.747* 23.755*** 23.219** 23.240** 14.166*

[7.456] [11.514] [5.037] [8.982] [3.337] [9.244] [9.045] [9.206] [10.195] [8.089]

EFI 0.895

[0.691]

EFI1 -0.383

[0.911]

EFI2 0.558

[0.496]

EFI3 0.001

[0.293]

EFI4 0.638*

[0.376]

EFI5 0.434

[0.942]

KOF 0.120*

[0.066]

KOF1 0.131*

[0.075]

KOF2 0.114*

[0.067]

KOF3 -0.037

[0.061]

Constant 17.468 28.139 21.486* 19.026** 14.055 24.163 29.221*** 32.814*** 37.287** 17.973**

[11.009] [17.285] [11.056] [9.467] [9.971] [15.654] [10.619] [11.313] [15.599] [7.235]

Number of i 81 81 81 81 81 81 80 80 80 80

Observations 473 485 461 480 483 465 483 483 483 483

Number of instruments 25 25 25 25 25 25 25 25 25 25

Period dummies 0.00 0.37 0.00 0.00 0.00 0.04 0.00 0.00 0.00 0.00

Serial correlation [p-value] 0.620 0.627 0.0340 0.382 0.527 0.743 0.995 0.971 0.879 0.380

Hansen J-test [p-value] 0.892 0.519 0.528 0.476 0.351 0.266 0.698 0.763 0.362 0.164

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.


Do Liberalization and Globalization Increase Income Inequality?

4.4.3 Distinguishing between development levels

As noted in section 4.2.1, theory suggests that certain types of liberalization and

globalization may have different inequality consequences at different development

levels. However, for some dimensions, theory provides ambiguous or no predictions

regarding such relationships, a matter that merits examination. In this section, we

split the sample into three equally sized groups based on GDP per capita and interact

these groups with the independent variables.

Tables 4.7 and 4.8 report how our variables of interest fare in this exercise.

Confirming the predictions of traditional trade theories, the positive effect of

freedom to trade on inequality appears in middle- and high-income settings. The

EFI4 coefficient is significantly different from zero at higher development levels, and

tests of equal coefficients across income groups (p-values in the bottom of table 4.7

and 4.8) indicate that the point estimates for the middle- and the high-income groups

are significantly different from the one corresponding to the low-income context.

The analysis also indicates that market deregulation has a distributional impact at

higher development levels. The size of the coefficients suggests that the inequality

effect is double the size in the high- compared with the low-income group. However,

we cannot rule out that coefficients may be equal across development groups.

Regarding the relationship between overall globalization and inequality, there

is little evidence that the distributional effect depends on the development level.

However, we note some vague indications of such differences when examining the

individual dimensions of globalization. On one hand, greater economic and social

globalization significantly increases inequality in the middle- and high-income parts of

our sample, while the coefficient is not significantly different from zero in the lowincome

group. On the other hand, the coefficient of political globalization is positive

and significant at lower development levels. However, in line with the relationship

found between overall globalization and inequality, we cannot rule out that

coefficients may be equal across development groups.

While the demographic measure remains insignificant throughout this

exercise, our results provide evidence that the distributional effect of education only

appears in high-income contexts. This finding is in line with the view that the general

133


Chapter 4

shift in economic activity in developed countries in recent decades has generated a

growing demand for educated people, in turn widening the earning gap.

Table 4.7 Analyzing the effects of economic freedom at different development levels

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

EFI x Low Development 0.392

[0.368]

EFI x Middle Development 1.046***

[0.342]

EFI x High Development 0.977**

[0.405]

EFI1 x Low Development 0.155

[0.397]

EFI1 x Middle Development 0.393

[0.295]

EFI1 x High Development 0.614*

[0.359]

EFI2 x Low Development 0.112

[0.295]

EFI2 x Middle Development 0.264

[0.201]

EFI2 x High Development 0.122

[0.199]

EFI3 x Low Development 0.030

[0.251]

EFI3 x Middle Development 0.342*

[0.178]

EFI3 x High Development 0.159

[0.187]

EFI4 x Low Development 0.216

[0.298]

EFI4 x Middle Development 0.826***

[0.277]

EFI4 x High Development 0.799***

[0.299]

EFI5 x Low Development 0.681

[0.480]

EFI5 x Middle Development 1.101***

[0.376]

EFI5 x High Development 1.252***

[0.380]

Human capital x Low Development 0.118 0.148 -0.117 0.100 0.068 0.038

[0.280] [0.283] [0.257] [0.239] [0.277] [0.277]

Human capital x Middle Development 0.070 0.183 0.104 0.154 0.055 0.104

[0.194] [0.173] [0.183] [0.176] [0.186] [0.184]

Human capital x High Development 0.385** 0.408** 0.316** 0.425*** 0.387** 0.341*

[0.168] [0.156] [0.143] [0.150] [0.148] [0.175]

Dependency x Low Development 1.922 0.340 1.669 1.647 0.782 2.190

[2.689] [2.141] [2.197] [2.018] [3.625] [3.439]

Dependency x Middle Development -1.940 -1.184 -0.044 -0.440 -3.196 -0.495

[2.615] [2.244] [1.925] [1.911] [3.999] [2.235]

Dependency x High Development -4.665 -5.129 0.767 -1.920 -6.676 -4.387

[3.640] [3.278] [3.486] [3.016] [4.826] [3.285]

Constant 35.251*** 38.487*** 34.515*** 33.770*** 32.684*** 33.794***

[1.835] [1.389] [2.394] [2.084] [3.356] [1.390]

Number of countries 81 83 85 87 89 91

Observations 505 524 491 518 508 496

R-squared (within) 0.142 0.135 0.120 0.135 0.139 0.143

Period dummies (Prob > F) 0.01 0.00 0.01 0.01 0.00 0.00

Country dummies (Prob > F) 0.00 0.00 0.00 0.00 0.00 0.00

Low=Middle 0.061 0.415 0.517 0.247 0.032 0.333

Low=High 0.145 0.204 0.974 0.631 0.064 0.251

High=Middle 0.814 0.375 0.527 0.325 0.916 0.632

*** Denotes significant at 1% level, ** significant at 5 % level, * significant at 10 % level.

Robust standard errors in brackets. All estimations include period dummies.

134


Do Liberalization and Globalization Increase Income Inequality?

Table 4.8 Analyzing the effects of globalization at different development levels

(1) (2) (3) (4)

KOF x Low Development 0.129*

[0.072]

KOF x Middle Development 0.174***

[0.064]

KOF x High Development 0.142**

[0.058]

KOF1 x Low Development 0.052

[0.050]

KOF1 x Middle Development 0.099**

[0.049]

KOF1 x High Development 0.088*

[0.048]

KOF2 x Low Development 0.077

[0.062]

KOF2 x Middle Development 0.156***

[0.052]

KOF2 x High Development 0.095**

[0.037]

KOF3 x Low Development 0.054*

[0.031]

KOF3 x Middle Development 0.058*

[0.033]

KOF3 x High Development 0.050

[0.035]

Human capital x Low Development -0.101 0.050 -0.050 0.003

[0.261] [0.268] [0.265] [0.229]

Human capital x Middle Development -0.023 0.088 -0.011 0.126

[0.196] [0.194] [0.194] [0.189]

Human capital x High Development 0.372** 0.383*** 0.377** 0.450***

[0.145] [0.143] [0.144] [0.154]

Dependency x Low Development -1.611 0.536 0.433 0.246

[2.465] [2.053] [1.994] [1.771]

Dependency x Middle Development -3.836 -1.567 -2.422 0.054

[2.600] [2.262] [1.939] [2.046]

Dependency x High Development -5.710 -4.191 -3.150 -1.951

[4.161] [4.126] [3.353] [3.529]

Constant 36.207*** 31.203*** 36.557*** 31.860***

[1.263] [2.668] [1.257] [2.271]

Number of countries 80 80 80 80

Observations 523 523 523 523

R-squared (within) 0.164 0.135 0.160 0.131

Period dummies (Prob > F) 0.00 0.00 0.01 0.00

Country dummies (Prob > F) 0.00 0.00 0.00 0.00

Low=Middle 0.387 0.194 0.156 0.898

Low=High 0.800 0.351 0.744 0.884

High=Middle 0.385 0.758 0.097 0.708

*** Denotes significant at 1% level, ** significant

Robust standard errors in brackets. All estimations include period dummies.

135


Chapter 4

4.5 CONCLUSION

In short, our analysis establishes that freedom to trade, as measured by the economic

freedom index, has a positive robust effect on within-country inequality. In line with

traditional trade theories, this distributional impact appears significant at higher

income levels, which also confirms the commonly held view of many developed

countries. In contrast, freedom to trade does not increase income inequality in lowincome

contexts.

The estimates of the EFI4 coefficient are stable at approximately 0.7–0.8. To

put the size of the effect into perspective, the standard deviation of EFI4 in our

sample is 1.6, corresponding to a one-point increase in the Gini coefficient. As a

concrete example, we note that Sweden increased its EFI4 score by 1.4 points

between 1980 and 2000. According to our estimates, this explains a quarter of

Sweden’s inequality increase from 21 to 25 points in the SWIID data. Since our

lowest estimated coefficient of EFI4 is approximately 0.5, and the highest is

approximately 1, our guess is that increasing freedom to trade explains between a

fifth and a third of rising inequality in Sweden from 1980 to 2000.

In most cases, we also find that social globalization, as measured by the KOF

index, has a significant positive effect on income inequality. This finding indicates

that the globalization process is multifaceted and that the lack of a conceptual

framework and of knowledge of the mechanisms by which this dimension affects

income inequality merit further theoretical research. Moreover, policy reforms

favoring less market regulation seem to increase inequality in our baseline model.

GMM estimations, however, suggest that these effects may be caused by reverse

causality in the short term.

This said, it bears emphasizing that most types of liberalization studied here

have no significant effect on income inequality, which supports our empirical

approach to distinguishing between dimensions of economic freedom and

globalization. Smaller government size has no robust distributional impact, and

improvements in the monetary system can probably also be made without increasing

inequality. Perhaps most interestingly, legal structure as measured by EFI2 has no

effect on income inequality. At the same time, Berggren and Jordahl (2005)

136


Do Liberalization and Globalization Increase Income Inequality?

demonstrate that EFI2 is in fact the most robust component of the economic

freedom index when it comes to explaining economic growth. This finding suggests

that building a well-functioning legal system may offer a way to promote growth

without inducing negative distributional consequences.

137


Chapter 4

Table 4A Country list

APPENDIX

Algeria* El Salvador* Jordan*¤ Portugal

Argentina Fiji* Kenya* Senegal*

Australia¤ Finland¤ Korea, Rep.¤ Sierra Leone*

Austria¤ France¤ Malawi* Singapore

Bangladesh*¤ Germany Malaysia South Africa

Barbados Ghana*¤ Mali*¤ Spain¤

Belgium¤ Greece Malta Sri Lanka*

Bolivia* Guatemala*¤ Mauritius Sweden¤

Botswana Guyana* Mexico¤ Switzerland

Brazil¤ Haiti* Mozambique* Thailand

Cameroon* Honduras*¤ Nepal* Trinidad and Tobago

Canada¤ Hong Kong, China¤ Netherlands¤ Tunisia*¤

Chile¤ Hungary¤ New Zealand¤ Turkey

China*¤ India*¤ Nicaragua* Uganda*¤

Colombia* Indonesia*¤ Norway¤ United Kingdom¤

Costa Rica¤ Iran, Islamic Rep.* Pakistan*¤ United States¤

Cyprus Ireland¤ Panama¤ Uruguay

Denmark Israel¤ Paraguay* Venezuela, RB¤

Dominican Republic*¤ Italy¤ Peru* Zambia*

Ecuador* Jamaica*¤ Philippines*¤ Zimbabwe*

Egypt, Arab Rep.* Japan Poland¤

* Low or lower middle income countries

¤ Countries included in estimations employing Kuznets ratio as the dependent variable

Table 4B

The Economic Freedom of the World Index

1: Size of Government: Expenditures, Taxes, and Enterprises

A. General government consumption spending as a percentage of total consumption

B. Transfers and subsidies as a percentage of GDP

C. Government enterprises and investment as a percentage of GDP

D. Top marginal tax rate (and income threshold at which it applies)

i. Top marginal income tax rate (and income threshold at which it applies)

ii. Top marginal income and payroll tax rate (and income threshold at which it

applies)

2: Legal Structure and Security of Property Rights

A. Judicial independence: the judiciary is independent and not subject to interference

from the government or parties in disputes

B. Impartial courts: A trusted legal framework exists for private businesses to challenge

the legality of government actions or regulation

C. Protection of intellectual property

D. Military interference in rule of law and the political process

E. Integrity of the legal system

138


Do Liberalization and Globalization Increase Income Inequality?

3: Access to Sound Money

A. Average annual growth of the money supply in the last five years minus average

annual growth of real GDP in the last ten years

B. Standard inflation variability in the last five years

C. Recent inflation rate

D. Freedom to own foreign currency bank accounts domestically and abroad

4: Freedom to Trade Internationally

A. Taxes on international trade

i. Revenue from taxes on international trade as a percentage of exports plus

imports

ii. Mean tariff rate

iii. Standard deviation of tariff rates

B. Regulatory trade barriers

i. Hidden import barriers: no barriers other than published tariffs and quotas

ii. Costs of importing: the combined effect of import tariffs, license fees, bank

fees, and the time required for administrative red tape raises costs of importing

equipment: by 10% or less = 10, by more than 50% = 0

C. Actual size of trade sector compared with expected size

D. Difference between official exchange rate and black market rate

E. International capital market controls

i. Access of citizens to foreign capital markets and foreign access to domestic

capital markets

ii. Restrictions on the freedom of citizens to engage in capital market exchange

with foreigners—index of capital controls among 13 IMF categories

5: Regulation of Credit, Labor, and Business

A. Credit market regulations

i. Ownership of banks: percentage of deposits held in privately owned banks

ii. Competition: domestic banks face competition from foreign banks

iii. Extension of credit: percentage of credit extended to private sector

iv. Avoidance of interest rate controls and regulations that lead to negative real

interest rates

v. Interest rate controls: interest rate controls on bank deposits and/or loans are

freely determined by the market

B. Labor market regulations

i. Impact of minimum wage: the minimum wage, set by law, has little impact on

wages because it is too low or not obeyed

ii. Hiring and firing practices: hiring and firing practices of companies are

determined by private contract

iii. Share of labor force whose wages are set by centralized collective bargaining

iv. Unemployment benefits: the unemployment benefits system preserves the

incentive to work

v. Use of conscripts to obtain military personnel

C. Business regulations

i. Price controls: extent to which businesses are free to set their own prices

139


Chapter 4

ii. Administrative conditions and new businesses: administrative procedures are

an important obstacle to starting a new business

iii. Time spent dealing with government bureaucracy: senior management spends

a substantial amount of time dealing with government bureaucracy

iv. Starting a new business: starting a new business is generally easy

v. Irregular payments: irregular, additional payments connected with import and

export permits, business licenses, exchange controls, tax assessments, police

protection, or loan applications are very rare

Table 4C

The KOF Index of Globalization

A. Economic Globalization

i) Actual flows

Trade (percent of GDP)

Foreign direct investment, flows (percent of GDP)

Foreign direct investment, stocks (percent of GDP)

Portfolio investment (percent of GDP)

Income payments to foreign nationals (percent of GDP)

ii) Restrictions

Hidden import barriers

Mean tariff rate

Taxes on international trade (percent of current revenue)

Capital account restrictions

B. Social Globalization

i) Data on personal contacts

Outgoing telephone traffic

Transfers (percent of GDP)

International tourism

Foreign population (percent of total population)

International letters (per capita)

ii) Data on information flows

Internet hosts (per 1000 people)

Internet users (per 1000 people)

Cable television (per 1000 people)

Trade in newspapers (percent of GDP)

Radios (per 1000 people)

iii) Data on cultural proximity

Number of McDonald’s restaurants (per capita)

Number of IKEA outlets (per capita)

Trade in books (percent of GDP)

C. Political Globalization

Embassies in country

Membership in international organizations

Participation in U.N. Security Council missions

140


Table 4D Summary statistics for different samples

Variable Explanation Countries Obs. Mean Std. Dev. Min Max Source

Baseline

Gini Gini coefficient of net incomes 81 524 38.27 9.54 20.86 63.11 Solt (2008)

Gini gross Gini coefficient of gross incomes 81 499 51.60 9.80 28.37 82.54 Solt (2008)

GDP per capita Natural logarithm of real GDP per capita (PPP adjusted) 81 524 8.33 1.12 5.23 10.52 Heston (2006)

Human capital Share of total population over 25 years with higher education 81 524 6.11 5.07 0.10 32.50 Barro and Lee (2000)

Dependency Dependency ratio (dependents to working-age population) 81 524 0.69 0.18 0.39 1.12 WDI (2008)

Employment industry Share of total employment in industry 73 327 25.14 7.42 2.30 50.20 WDI (2008)

Urban Share of population living in urban areas 81 524 56.56 23.37 5.58 100 WDI (2008)

Unemployment Share of labor force unemployed 76 323 7.99 4.77 0.60 35.50 WDI (2008)

Human capital II Average years of schooling in population over 15 years 79 436 6.11 2.68 0.61 12.05 Barro and Lee (2000)

EFI Aggregated chain-linked economic freedom of the world index 81 505 6.14 1.22 2.80 9.23 Gwartney and Lawson (2008)

EFI1 Size of government 81 524 5.60 1.66 1.68 9.96 Gwartney and Lawson (2008)

EFI2 Legal structure and secure property rights 81 491 6.01 2.30 1.21 9.89 Gwartney and Lawson (2008)

EFI3 Access to sound money 81 518 7.05 2.19 0.00 9.84 Gwartney and Lawson (2008)

EFI4 Freedom to exchange with foreigners 81 508 6.40 1.59 1.53 9.77 Gwartney and Lawson (2008)

EFI5 Regulation of credit, labor and business 81 496 5.84 1.09 2.73 8.85 Gwartney and Lawson (2008)

KOF Aggregated globalization index 80 523 48.56 18.98 8.51 92.61 Dreher (2008)

KOF1 Economic globalization 80 523 50.56 19.80 7.80 94.76 Dreher (2008)

KOF2 Social globalization 80 523 42.46 21.94 5.52 95.19 Dreher (2008)

KOF3 Political globalization 80 523 54.97 24.35 4.27 98.45 Dreher (2008)

Longer time period - Gini

Gini 2005-1985 Change in Gini coefficient of net income 60 60 1.37 5.32 -20.90 12.90 Solt (2008)

Gini 1985 Initial value of Gini coefficient of net income 60 60 37.32 10.11 20.95 61.46 Solt (2008)

EFI 2000-1980 Change in aggregated chain-linked index 59 59 1.19 0.92 -0.80 3.32 Gwartney and Lawson (2008)

EFI1 2000-1980 Change in EFI1 59 59 1.27 1.57 -3.56 5.52 Gwartney and Lawson (2008)

EFI2 2000-1980 Change in EFI2 59 59 0.39 1.41 -2.47 4.68 Gwartney and Lawson (2008)

EFI3 2000-1980 Change in EFI3 59 59 1.84 2.18 -3.22 7.21 Gwartney and Lawson (2008)

EFI4 2000-1980 Change in EFI4 59 59 1.52 1.14 -0.45 4.68 Gwartney and Lawson (2008)

EFI5 2000-1980 Change in EFI5 59 59 0.61 0.67 -1.29 2.70 Gwartney and Lawson (2008)

KOF 2000-1980 Change in aggregated globalization index 60 60 16.98 6.90 1.32 30.08 Dreher (2008)

KOF1 2000-1980 Change in KOF1 60 60 19.69 8.79 -4.67 39.86 Dreher (2008)

KOF2 2000-1980 Change in KOF2 60 60 20.14 8.60 4.46 39.13 Dreher (2008)

KOF3 2000-1980 Change in KOF3 60 60 8.09 13.75 -19.40 35.51 Dreher (2008)

GDP per capita 1980 Initial value log real GDP per capital 60 60 7.81 1.01 5.82 9.25 Heston (2006)

Human capital 1980 Initial value share of population with higher education 60 60 5.56 2.76 0.54 11.87 Barro and Lee (2000)

Dependency 1980 Initial value of dependency ratio 60 60 0.74 0.16 0.48 1.07 WDI(2008)

Latin America Dummy for countries in Latin America 60 60 0.29 0.46 0 1 WDI(2008)

East Asia Dummy for countries in East Asia 60 60 0.14 0.35 0 1 WDI(2008)


Table 4D continued

Longer time period - Kuznet Explanation Countries Obs. Mean Std. Dev. Min Max Source

Kuznets ration 2005-1985 Change in Kuznets ratio 41 41 0.06 0.65 -2.22 1.23 WIID2.c(2008)

Kuznets ratio 1985 Initial value of Kuznets ratio 41 41 3.12 1.74 1.19 7.43 WIID2.c(2008)

EFI 2000-1980 Change in aggregated chain-linked index 39 39 1.24 0.87 -0.80 3.32 Gwartney and Lawson (2008)

EFI1 2000-1980 Change in EFI1 39 39 1.42 1.32 -0.91 5.52 Gwartney and Lawson (2008)

EFI2 2000-1980 Change in EFI2 39 39 0.55 1.48 -2.47 4.68 Gwartney and Lawson (2008)

EFI3 2000-1980 Change in EFI3 39 39 1.85 2.02 -1.72 6.99 Gwartney and Lawson (2008)

EFI4 2000-1980 Change in EFI4 39 39 1.41 1.27 -0.12 5.17 Gwartney and Lawson (2008)

EFI5 2000-1980 Change in EFI5 39 39 0.58 0.64 -1.29 2.00 Gwartney and Lawson (2008)

KOF 2000-1980 Change in aggregated globalization index 41 41 17.17 5.93 3.13 30.08 Dreher (2008)

KOF1 2000-1980 Change in KOF1 41 41 20.27 7.74 9.76 39.86 Dreher (2008)

KOF2 2000-1980 Change in KOF2 41 41 20.14 8.19 4.40 39.13 Dreher (2008)

KOF3 2000-1980 Change in KOF3 41 41 7.90 13.35 -19.40 35.51 Dreher (2008)

GDP per capita 1980 Initial value log real GDP per capital 41 41 7.98 0.98 5.91 9.22 Heston (2006)

Human capital 1980 Initial value share of population with higher education 41 41 5.95 2.88 0.54 11.87 Barro and Lee (2000)

Dependency 1980 Initial value of dependency ratio 41 41 0.74 0.17 0.48 1.07 WDI(2008)

Latin America Dummy for countries in Latin america 41 41 0.26 0.44 0 1 WDI(2008)

East Asia Dummy for countries in East Asia 41 41 0.18 0.39 0 1 WDI(2008)


Table 4E Correlation matrix

GDP per capita Human capital Dependency Employment industry Urban Unemployment Human capital II Gini Gini gross

GDP per capita 1

Human capital 0.70 1

Dependency -0.81 -0.57 1

Employment industry 0.58 0.14 -0.57 1

Urban 0.77 0.62 -0.63 0.54 1

Unemployment -0.04 -0.04 0.14 -0.02 0.03 1

Human capital II 0.82 0.78 -0.77 0.46 0.71 -0.06 1

Gini -0.48 -0.30 0.61 -0.42 -0.34 0.21 -0.56 1

Gini gross -0.40 -0.26 0.57 -0.41 -0.30 0.15 -0.49 0.92 1

EFI 0.72 0.58 -0.63 0.36 0.55 -0.14 0.68 -0.39 -0.34

EFI1 -0.06 0.09 0.06 -0.22 -0.04 -0.08 -0.14 0.41 0.31

EFI2 0.70 0.45 -0.67 0.54 0.55 -0.15 0.73 -0.62 -0.53

EFI3 0.53 0.45 -0.42 0.14 0.33 -0.13 0.44 -0.32 -0.26

EFI4 0.67 0.49 -0.57 0.40 0.58 -0.08 0.62 -0.34 -0.28

EFI5 0.61 0.51 -0.52 0.29 0.54 0.00 0.63 -0.32 -0.29

KOF 0.87 0.65 -0.74 0.44 0.72 0.00 0.80 -0.55 -0.45

KOF1 0.79 0.56 -0.64 0.39 0.65 0.08 0.70 -0.36 -0.28

KOF2 0.85 0.64 -0.70 0.44 0.69 0.02 0.80 -0.52 -0.42

KOF3 0.63 0.49 -0.62 0.29 0.55 -0.13 0.58 -0.60 -0.51


Chapter 4

REFERENCES

Acosta, Pablo and Gabriel V. Montes-Rojas. 2008. “Trade Reform and Inequality: The

Case of Mexico and Argentina in the 1990s.” The World Economy 31:763–763.

Albanesi, Stefania. 2007. “Inflation and inequality.” Journal of Monetary Economics 54:1088–

1088.

Arellano, M and O Bover. 1995. “Another look at the instrumental variable estimation of

error-components models.” Journal of Econometrics 68:29–51

Arellano, Manuel and Stephen Bond. 1991. “Some Tests of Specification for Panel Data:

Monte Carlo Evidence and an Application to Employment Equations.” Review of

Economic Studies 58:277–97.

Ashby, Nathan J. and Russell S. Sobel. 2008. “Income inequality and economic freedom

in the U.S. states.” Public Choice 134:329–346.

Atkinson, A. B. 1997. “Bringing Income Distribution in From the Cold.” The Economic

Journal 107:297–321.

Atkinson, A.B. and A, Brandolini. 2001. "Promise and Pitfalls in the Use of "Secondary"

Data-Sets: Income Inequality in OECD Countries as a Case Study" Journal of

Economic Literature 39(3):771-799

Babones, Salvatore J and Maria Jose Alvarez-Rivadulla. 2007. “Standardized Income

Inequality Data for Use in Cross-National Research.” Sociological Inquiry 77:3–22.

Barro, Robert J. and Jong-Wha Lee. 2000. "International Data on Educational

Attainment: Updates and Implications." CID Working Paper No. 42, April 2000.

Berggren, Niclas. 1999. “Economic freedom and equality: Friends or foes?” Public Choice

100:203–223.

Berggren, Niclas and Henrik Jordahl. 2005. “Does free trade really reduce growth?

Further testing using the economic freedom index.” Public Choice 122:99–114.

Bergh, Andreas. 2005. “On the counterfactual problem of welfare state research: How

can we measure redistribution?” European Sociological Review 21:345–357.

—. 2007. “The Middle Class and the Swedish Welfare State.” The Independent Review.

Bergh, Andreas and Günther Fink. 2008. “Higher Education Policy, Enrollment, and

Income Inequality.” Social Science Quarterly 89:217–235.

Blundell, R. and S. Bond. 1998. “Initial conditions and moment restrictions in dynamic

panel data models “ Journal of Econometrics 87:115–143.

Bond, S. R, A Hoeffler, and J Temple. 2001. “GMM Estimation of Empirical Growth

Models.” CEPR Discussion Papers 3048.

Calderón, César and Alberto Chong. 2009. “Labor market institutions and income

inequality: An empirical exploration.” Public choice: Center for study of public choice,

Virginia polytechnic institute and State university 138:65–81.

Carlsson, Fredrik and Susanna Lundstrom. 2002. “Economic Freedom and Growth:

Decomposing the Effects.” Public Choice 112:335–344.

Carter, John R. 2007. “An Empirical Note on Economic Freedom and Income

Inequality.” Public Choice 130:163–177.

Claessens, S. and E, Perotti (2007). “Finance and inequality: Channels and evidence”

Journal of Comparative Economics: 35(4 ):748–773

Das, Satya P. 2005. “Gradual Globalization and Inequality Between and Within

Countries.” Canadian Journal of Economics 38:852–869.

De Haan, J, S Lundstrom, and J Egbert Sturm. 2006. “Market-oriented institutions and

policies and economic Growth: A critical survey.” Journal of Economic Surveys

20:157–324.

144


Do Liberalization and Globalization Increase Income Inequality?

De Soto, Hernando. 2000. The Mystery of Capital: Why Capitalism Triumphs in the West and

Fails Everywhere Else. New York: Basic Books.

Deininger, Klaus and Lyn Squire. 1996. “A New Data Set Measuring Income Inequality.”

World Bank Economic Review 10:565–591.

Doucouliagos, C. and M.A. Ulubasoglu. 2006. “Economic freedom and economic

growth: Does specification make a difference?” European Journal of Political

Economy 22:60–81.

Dreher, Axel. 2006. “Does Globalization Affect Growth? Empirical Evidence from a

new Index.” Applied Economics 38:1091–1110.

Dreher, Axel and Noel Gaston. 2008. “Has Globalization Increased Inequality?” Review of

International Economics 16:516–536.

Feenstra, R.C and G, Hanson. 1997. "Foreign direct investment and relative wages:

Evidence from Mexico's maquiladoras." Journal of International Economics 42 (3-4):

371-393

Fischer, S. and F, Modigliani (1978) “Towards An Understanding of the Real Effects and

Costs of Inflation” Review of World Economics (Weltwirtschaftliches Archiv),

Springer, vol. 114(4):810–833

Fortin, Nicole, M and Thomas Lemieux. 1997. “Institutional Changes and Rising Wage

Inequality: Is there a Linkage?” The Journal of Economic Perspectives 11:75–96.

Galor, O and J. Zeira 1993. "Income Distribution and Macroeconomics" The Review of

Economic Studies 60(1):35-52

Gourdon, Julien, Nicolas Maystre, and Jaime de Melo. 2008. “Openness, inequality and

poverty: Endowments matter.” The Journal of International Trade & Economic

Development 17:343–343.

Gradstein, M. 2007. “Inequality, democracy and the protection of property rights.” The

Economic Journal 117(516): 252

Gwartney, J and R. Lawson. 2003. “Economic Freedom of the World Annual Report.”

The Fraser Institute, Vancouver BC.

Gwartney, James, Robert Lawson, and Seth Norton. 2008. “Economic Freedom of the

World: 2008 Annual Report.” The Fraser Institute. Data retrieved from

www.freetheworld.com.

Heckelman, J. C and M. D Stroup. 2005. “A comparison of aggregation methods for

measures of economic freedom.” European Journal of Political Economy 21:953–966.

Heston, A., Summers R. and B. Aten. 2006. Penn World Table Version 6.2, Center for

International Comparisons of Production, Income and Prices at the University of

Pennsylvania, September 2006.

Higgins, M. and J.G. Williamson. 1999. “Explaining Inequality the World Round: Cohort

Size, Kuznets Curves, and Openness.” NBER Working Paper 7224. Cambridge,

MA: National Bureau of Economic Research

Kanbur, R. 2000. “Income Distribution and Development.” in Handbook of Income

Distribution, edited by A. Atkinson and F. Bourguignon: Elsevier Publisher.

Krugman, Paul and Anthony Venables, J. 1995. “Globalization and the Inequality of

Nations.” The Quarterly Journal of Economics 110:857–880.

Krusell, P., Ohanian, L. Giovanni L., Rios-Rull, J-V. and G. Violante. 2000. "Capital-Skill

Complementarity and Inequality: A Macroeconomic Analysis" Econometrica

68(5):1029-1053

Kumar, Utsav and Prachi Mishra. 2008. “Trade Liberalization and Wage Inequality:

Evidence from India.” Review of Development Economics 12:291–311.

145


Chapter 4

Le Grand, Julian and David Winter. 1986. “The Middle Classes and the Welfare State

under Conservative and Labour Governments.” Journal of Public Policy 6:399–430.

Lee Ha, Yan, Antonio Ricci Luca, and Roberto Rigobon. 2004. “Once again, is openness

good for growth?” Journal of Development Economics 75:451–472.

Lindert, Peter H and Jeffrey G Williamson. 2001. “Does Globalization Make the World

More Unequal?” NBER Working Paper No. 8228.

Lindquist, M.J. 2005. "Capital-Skill Complementarity and Inequality in Sweden." The

Scandinavian Journal of Economics 107(4): 711

Odedokun M O and J. Round (2004). "Determinants of Income Inequality and its

Effects on Economic Growth: Evidence from African Countries." African

Development Review 16(2): 287-327

O'Rourke, Kevin H. 2001. “Globalization and Inequality: Historical Trends.” NBER

Working Paper No. 8339.

Owen, Ann L. and Stephen Wu. 2007. "Is Trade Good for Your Health?" Review of

International Economics 15:660-682.

Richardson, J. David. 1995. “Income Inequality and Trade: How to Think, What to

Conclude.” The Journal of Economic Perspectives 9:33–55.

Rodriguez, F and R Rodrik. 2000. “Trade policy and economic growth: a skeptic’s guide

to the cross-national evidence.” NBER Macroeconomics Annual:261–325.

Roine, J., Vlachos, J. and D. Waldenström. 2009. "The long-run determinants of

inequality: What can we learn from top income data?", Forthcoming in Journal of

Public Economics

Roodman, D. 2006. “How to Do xtabond2: An Introduction to “Difference” and

“System” GMM in Stata “ Working Paper Number 103, Center for Global Development.

Rothstein, Bo. 1998. Just Institutions Matter: The Moral and Political Logic of the Universal

Welfare State. Cambridge: Cambridge University Press.

Sachs, Jeffrey D. and Andrew Warner. 1995. “Economic reform and the process of

global integration.” Brookings Papers on Economic Activity:1–119.

Savvides, A. 1998. “Trade policy and income inequality: new evidence.” Economics Letters

61:365–372.

Scully, Gerald, W. 2002. “Economic Freedom, Government Policy and the Trade-Off

between Equity and Economic Growth.” Public Choice 113:77–96.

Sebastian, Edwards. 1997. “Trade Policy, Growth, and Income Distribution.” The

American Economic Review 87:205–210.

Sinn, H-W 1995.”A theory of the welfare state,” Scandinavian Journal of Economics 97:495–

526

Solt, Frederick. 2008. “Standardizing the World Income Inequality Database.” Social

Science Quarterly Forthcoming.

Son, Hyun, H. and Nanak Kakwani. 2008. “Global Estimates of Pro-Poor Growth.”

World Development 36:1048–1048.

Sonin, K. 2003. “Why the rich may favor poor protection of property rights.” Journal of

Comparative Economics 31:715–731.

Sylwester, Kevin. 2002. “Can Education Expenditures Reduce Income Inequality?”

Economics of Education Review 21:43–52.

Tsai, Ming-Chang. 2007. “Does globalization affect human well-being?” Social Indicators

Research 81:103–126.

Uusitalo, Hannu. 1985. “Redistribution and equality in the welfare state: An effort to

interpret the major findings of research on the redistributive effects of the

welfare state.” European Sociological Review 1:163–176.

146


Do Liberalization and Globalization Increase Income Inequality?

WIID2c, 2008. World Income Inequality Data base. UNU-Wider, Helsinki

Windmeijer, F. 2005. "A finite sample correction for the variance of linear efficient twostep

GMM estimators" Journal of Econometrics 126(1): 25-51

Wirth, L. 1938. “Urbanism as a way of life.” American Journal of Sociology 44: 3–24.

World Bank. 2008. World Development Indicators. Washington DC: World Bank.

Åberg, R. 1989. “Distributive Mechanisms of the Welfare State: a Formal Analysis and an

Empirical Application.” European Sociological Review 5:167–182.

Acknowledgements

The authors wish to thank Christian Bjørnskov, participants in the 2008 Public

Choice Meeting (St. Antonio, Texas), Carl Hampus Lyttkens, Jesper Roine, Susanna

Thede, Pernilla Johansson, and seminar participants in Lund for their helpful

comments and suggestions. Therese Nilsson gratefully acknowledges financial

support from Per Westlings minnesfond.

147


148


Chapter 5

Good for Living? On the Relationship between Globalization

and Life expectancy

Co-author: Andreas Bergh

5.1 INTRODUCTION

Concurrent with increasing worldwide globalization, there has been much research

into its consequences. A recent volume by Dreher et al. (2008) comprehensively

summarizes the empirical findings on the effects of closer integration between

economies, on growth, taxation and government spending, within-country inequality,

de-unionization, and the environment. Additional studies include Nissanke and

Thorbecke (2006) and Ravallion (2006), which treat the relationship between

globalization and poverty reduction, and Tsai (2007), which focuses on the Human

Development Index (HDI). However, little is known about the effects of

globalization on physical health.

Studies of the determinants of population health suggest there are several

channels by which globalization may affect health. Many relate to the movement of

goods and services, such as the availability of imported pharmaceuticals and changes

in relative prices. As a result, the limited literature on the relationship between

globalization and health typically adopts an economic perspective and focuses on the

health effects of increased trade openness or economic freedom (Bussmann, 2009;

Owen and Wu, 2007; Stroup, 2007). Globalization, however, could also affect health

through, for example, life style change, faster spread of contagious diseases, and


Chapter 5

altered international relations. Analyzing the health effects of increasing

internationalization therefore requires distinguishing between different dimensions of

globalization. Moreover, given the numerous potential channels at work, it is essential

to control for possible mediating factors in the globalization–health relationship.

This paper analyzes the relationship between globalization and an objective

and easily quantifiable measure of health: life expectancy at birth. Unlike the few

existing studies in this field, we examine the effects of economic, social, and political

globalization, not only the impact of trade and investments, by using the KOF Index

developed by Dreher (2006). It is hoped that a more comprehensive perspective on

globalization will provide a better understanding of health in relation to increasing

internationalization. Figure 5.1 plots the cross-correlation between the composite

KOF Index (which assigns a value of 0–100, indicating the level of globalization of

each country) and life expectancy at birth in 2000. The scatterplot depicts a positive

but non-linear relationship. However, whether the relationship holds in a more

formal analysis and whether it is robust across different types of economies remains

to be explored. Our analysis particularly emphasizes how the relationship between

types of globalization and life expectancy varies between levels of development.

Figure 5.1. The cross-country correlation between life expectancy and globalization, 2000.

40 50 60 70 80

Life expectancy at birth, years

Australia

Austria Belgium

Albania

Barbados

Bahrain

Argentina

Algeria Bahamas, The Brazil

Bulgaria

Japan

Canada

Cyprus

Israel Iceland Italy France

Costa Malta Rica Greece New Germany

Kuwait Chile

Luxembourg

Zealand Spain

Sweden Switzerland

United Arab EmiratesUnited Singapore

Norway

Finland Netherlands

Ireland Denmark

Korea, Rep. Portugal

United States

Kingdom

Slovenia

Czech Republic

Ecuador Oman Mexico

Uruguay Panama

Sri Lanka Croatia Poland

Syrian Arab Republic Tunisia Venezuela, Malaysia RB Slovak Republic

Mauritius

Dominican Colombia

El Salvador Republic Jamaica Lithuania

China Jordan Estonia

Hungary

Paraguay Nicaragua Latvia

Romania

Iran, Islamic Rep.

Philippines

Morocco Peru Egypt, Trinidad

Turkey

Arab and Rep. Tobago

Guatemala HondurasThailand

Fiji

Ukraine

Indonesia Russian Federation

Bangladesh

Myanmar Nepal

Pakistan IndiaGuyana

Bolivia

Senegal

Haiti

TogoGabonGhana

Madagascar Papua New Guinea

Benin

Niger Congo, Kenya Rep.

Chad Mali Cameroon

Namibia

Tanzania Botswana South Africa

Cote d'Ivoire

Burundi Uganda

Nigeria

Malawi

Guinea-Bissau

Central African Congo, Republic Dem. Rep.

Zimbabwe

Rwanda Sierra Leone

Zambia

20 40 60

Aggregated KOF index

80 100

Data source: Dreher(2008) and World Bank(2008)

150


Good for Living? On the Relationship between Globalization and Life Expectancy

We construct a 92-country panel covering the 1970–2005 period and control for

demographic structure and four factors repeatedly found to influence life expectancy,

i.e., public health measures (e.g. health care availability or sanitation), education,

nutrition, and GDP per capita. Using different estimation techniques, we find that

economic globalization has a strong and robust positive effect on life expectancy.

Moreover, using a procedure by which we gradually exclude high-income-country

observations from our sample and re-run the estimation, we examine how the

globalization effect varies with income level in a way that interaction terms do not

capture. We find evidence that the positive effect of economic globalization is

present also in a low-income context.

The paper continues as follows. The next section reviews recent research into

the determinants of life expectancy and discusses how these might be influenced by

globalization. Section 5.3 discusses the methodological choices and describes the

data, while section 5.4 presents the empirical analysis, including several robustness

checks. Section 5.5 summarizes our results and presents our conclusions.

5.2 BACKGROUND

5.2.1 Disentangling the effects of globalization and health

Globalization typically refers to the process by which different economies and

societies become more closely integrated. This closer integration entails, on one hand,

more open borders, which speed up transactions, and, on the other, the development

of relationships between individuals at a distance. Globalization accordingly refers to

both the temporal and spatial compression of interaction. Moreover, as discussed by

Arribas et al. (2009), the progress of globalization has many facets, because of the

range of interactions it involves. In other words, globalization is multidimensional.

We can roughly distinguish three different dimensions of globalization.

Economic globalization refers to the increased exchange of goods and services and

larger investment flows across countries and regions of the world. Political

globalization refers to the trend for economies to become more integrated at a

political level. Finally, social globalization refers to how the closer interaction between

countries can influence norms and cultural values.

151


Chapter 5

Several studies attempt to explain variations in life expectancy across

countries; recent such studies include Kabir (2008), Cutler et al. (2006), Fayissa and

Gutema (2005), and Husain (2002), while an older study is Grosse and Perry (1982).

Four broad factors repeatedly found to be significantly and positively related to life

expectancy are nutritional status, education, public health measures, and income.

Most studies focus on less-developed countries where factors such as water sanitation

and literacy are crucial determinants (as demonstrated by Grosse and Perry 1982). In

contrast, dietary and nutritional factors often explain variations within developed

countries. For example, Shaw et al. (2005) examine 29 OECD countries, 1960–1999,

and find positive effects for the per capita consumption of pharmaceuticals, fruits

and vegetables, and butter; moreover, consumption of alcohol and tobacco generally

has the expected negative sign.

Figure 5.2. Important determinants of life expectancy according to the existing literature.

Economic Social Political

Globalization

Income

Income Education Nutrition Public Health

Life Expectancy

A major point of disagreement in the literature is the relative importance of income

in determining life expectancy, some studies finding no effect and other studies

finding small or large positive effects. 1 There are several possible explanations for this

discrepancy. According to standard economic theory, income should have no direct

1 For example, Soares (2007) argues that increases in life expectancy between 1960 and 2000 were largely

independent of improvements in income.

152


Good for Living? On the Relationship between Globalization and Life Expectancy

effect on health: Income is only instrumentally important by enabling purchasing

power that can be used to consume, for example, food, safety measures, health care,

and vaccinations, which in turn affect health. When more control variables are added

to a regression on life expectancy, the coefficient for income will decrease.

Furthermore, the degree to which countries spend their income on health-improving

consumption is likely to differ, and, to some degree, income can be spent on areas

likely having negative health effects, such as the military or fast food.

Globalization can affect life expectancy through the four factors presented in

figure 5.2 and through other mechanisms. However, theory makes ambiguous

predictions as to whether the health impact of globalization is positive or negative on

average. First, if globalization is positively related to GDP per capita, it will be

beneficial for life expectancy. Such an effect may occur through the static effects of

trade liberalization or because globalization is good for economic growth, as found by

Dreher (2006). 2 Second, globalization may positively affect education levels, including

literacy. For example, the possibility of working abroad may increase the education

premium and thus strengthen education incentives, as suggested by Stark (2004). In

addition, social globalization via tourism and information flows may increase literacy

levels.

Third, globalization can affect public health by improving access to new

technologies for water sanitation, medical treatment, and pharmaceuticals. For

example, Papageorgiou et al. (2007) argue and find empirical support for the view

that pharmaceutical R&D is highly concentrated in a small group of ten countries that

export these goods to the rest of the world. Using a cross-section of 63 technologyimporting

countries, they demonstrate that technology diffusion through medical

exports is an important contributor to improved life expectancy. 3

Fourth, globalization may affect nutritional intakes both directly, through

increased availability of imports, and indirectly, because relative prices change when

2 Note, however, that the empirical link between globalization and growth is debatable and depends on

how globalization is measured (cf. Rodriguez and Rodrik, 2000; Lee Ha et al., 2004).

3 The relevance of medical technologies, specifically new drugs, is supported by Lichtenberg (2003). In a

sample of 50 upper-middle-income developing and developed countries, he demonstrates that the launch

of new drugs between 1986 and 2000 had a strong positive impact on the probability of survival. He claims

that these new drugs were responsible for 40% of the gains in life expectancy observed in the sample

during the period.

153


Chapter 5

an economy becomes more open. Furthermore, social globalization may lead to

changes in lifestyle and dietary habits that have health consequences. Medez and

Popkin (2004) note that the structure of dietary intakes is rapidly changing in lessdeveloped

countries around the world, converging towards a “Western diet” high in

saturated fats and sugar, which might affect health negatively. Yach et al. (2007) note

that waves of cultural interaction have also extended the mass consumption of

“bads,” such as tobacco, in turn increasing the spread of non-infectious diseases. On

the other hand, Deaton (2004) emphasizes the counter-effect of globalization, since

closer integration facilitates the transmission of health-related knowledge.

While most of the mechanisms discussed above suggest that globalization

positively affects life expectancy, there are several complicating factors. One

important possible negative link between globalization and health is the faster and

geographically broader spread of infectious diseases such as HIV and the H5N1 avian

influenza virus (Kawachi and Wamala, 2007). However, political globalization may

allow governments to react faster and with greater coordination to counter emerging

heath threats. Another potentially negative health effect of globalization is the stress

effect of having more choices and more available information. While economists

typically expect more choices to be welfare enhancing, Schwartz (2004), for example,

has argued that an excessive range of choices causes stress and regret, making us less

happy. 4 Cutler et al. (2006) note that cumulative distress leads to increased probability

of disease, particularly cardiovascular disease.

A third reason why globalization and health may be negatively related is the

effect of globalization on income distribution. An emerging consensus in empirical

studies is that, while many aspects of globalization have no significant effect on

income inequality, trade liberalization and economic openness probably increase

within-country income inequality, especially in developed countries (see recent studies

by Dreher and Gaston, 2008, Bergh and Nilsson, 2008). If there is also a link between

income inequality and health, as suggested by, for example, Wilkinson (1996) and

Babones (2008), but disputed by, for example, Gravelle (1998) and Mellor and Milyo

4 Reviewing Schwartz’s book, Veenhoven (2005) claims it to be “persuasive at first sight,” but adds that “a

closer look shows the evidence to be flimsy” (p. 94).

154


Good for Living? On the Relationship between Globalization and Life Expectancy

(2002), this is a mechanism by which globalization can negatively affect life

expectancy. 5 Another important factor may be that even though globalization may

increase GDP per capita, it does so by changing the structure of the economy;

structural adjustment can be painful for those in the labor force who must switch

jobs, which in turn affects health. Furthermore, some aspects of globalization, such as

trade, may also affect the environment and thereby health (Owen and Wu 2007).

Finally, globalization might affect government size and, for example, social

spending, in turn affecting population health. Economic theory suggests two opposite

scenarios: the race to the bottom hypothesis (Sinn 1997), according to which

economies compete, for example, by lowering taxes, and the compensation

hypothesis (Rodrik 1998, Katzenstein 1985, Lindbeck 1975), according to which

open economies develop larger welfare states. 6

To summarize, few of the possible links between globalization and health are

theoretically unambiguous, a situation that calls for empirical examination.

5.2.2 Related research

There is limited empirical literature on the relationship between economic

globalization and objective or subjective health. The study closest to ours is that of

Owen and Wu (2007), who analyze a panel of 219 countries using observations in

five-year intervals from 1960 to 1995. They find that increased economic openness,

i.e., (exports + imports)/GDP, is associated with lower rates of infant mortality and

higher life expectancies, especially in developing countries. Their findings also

indicate that some of the positive correlation between trade and health can be

attributed to knowledge spillovers. 7 In contrast, using a panel of 134 countries

containing annual data from 1970 to 2000, Bussmann (2009) fails to find evidence

5 As pointed out by, for example, Gravelle (1998), the often-noted correlation between income inequality

and population health indicators (such as life expectancy) can be flawed by the non-linear relationship

between income and individual health.

6 It should be noted that empirical evidence on the relationship between government size and population

health (Björnskov et al., 2007, Tsai, 2007) is inconclusive.

7 Their results imply that a one-standard-deviation increase in the log of openness for a country in the

lowest quintile of real GDP is associated with a drop of approximately seven infant deaths per 1000 (a

reduction in average infant mortality of approximately 8%). The increase in female (male) life expectancy

associated with a one-standard-deviation increase in log openness is 1.39 (0.84) years.

155


Chapter 5

that economic integration improves health care provision, proxied by female life

expectancy, in her study of the effect of trade openness on women’s welfare and

work life. 8

Stroup (2007) uses panel data and finds evidence that the economic freedom

index (Gwartney et al., 2008) is positively linked to life expectancy and other welfare

outcomes. As well, Ovaska and Takashima (2006) examine the effects of economic

freedom and trade on self-reported levels of happiness and life satisfaction, using a

cross-country sample of 68 countries in the 1990s. They found that economic growth

had robust positive effects on life expectancy, and that in many cases economic

freedom also had considerable positive impact.

Three of the four related studies include controls for income and education in

their estimations. The exception is Stroup (2007), where the only explanatory variable

competing with economic freedom is an index of political rights, which to some

extent makes it problematic to evaluate the effect of globalization. None of the

studies controls for nutritional intake or public health availability (indicated, for

example, by physicians per capita).

Another closely related study is that of Tsai (2007), who finds a positive

relationship between the KOF Index of Globalization and the HDI, but more so in

developed than developing countries. The data cover 112 countries in three waves

(i.e., 1980, 1990, and 2000) and exclude developing countries with populations less

than one million. The interpretation of Tsai’s results is complicated by the fact that

the HDI is a composite measure, aggregating life expectancy, adult literacy, combined

primary, secondary, and tertiary school enrolment, and GDP per capita (PPP US$). 9

8 This result might be explained by Bussmann’s use of annual data for trade/GDP, the dependent variable

being interpolated for missing years. While economic openness (i.e., trade/GDP) fluctuates from year to

year, changes in health outcomes likely evolve over a number of years.

9 An obvious problem in Tsai’s (2007) study is that per capita income is used as both an explanatory

variable and as part of the HDI. This is addressed by the author in a footnote, where it is also reported that

“economic globalization generated significantly favorable impacts on life expectancy, and all but political

globalization measures produced positive impacts on infant mortality” (p. 124).

156


Good for Living? On the Relationship between Globalization and Life Expectancy

5.3 METHODS AND DATA

5.3.1 Methods

To examine the relationships of interest we specify an equation that relates

globalization to life expectancy and to a set of control variables:

health ε

it = α + β1

X it−1

+ β 2Vit−1

+ β 3Z

it + it

(5.1)

where X is a vector for the types of globalization believed to affect health. Since the

impact of closer integration on health is unlikely to be instant, these variables are

lagged: average globalization in 1970–1973 is consequently used to explain average

life expectancy in 1974–1977. This specification also reduces the bias following from

potentially reverse causality between globalization and health. V and Z are vectors for

additional covariates that can be classified as potential mediators through which

globalization influences population health, and as exogenous factors affecting population

health but not themselves influenced by globalization. Importantly, the inclusion of a

mediator as a regressor should reduce the estimated effect of globalization on health.

In equation (5.1), ε is an error term. Ordinary least squares (OLS) regression

assumes error processes to have the same variance and to be independent of each

other. In the presence of non-spherical errors, the estimated coefficients are still

consistent, but standard errors are not efficient and are likely biased, in turn affecting

statistical inference. By means of correction, robust standard errors of the fixed-effect

(FE) OLS estimator can be estimated in case of heteroscedasticity and

autocorrelation within panels. 10 However, because globalization means greater

integration between economies, increasing inter-country linkages imply that errors

may be contemporaneously correlated across countries. We therefore estimate the

relationship using a panel-corrected standard errors (PCSE) procedure, allowing for

disturbances that are heteroscedastic and contemporaneously correlated across

countries (Beck and Katz, 1995). 11 Estimations correct for first-order autocorrelation,

by treating the AR(1) process as specific to each country. From Monte Carlo

10 Using the Stata command “xtreg, fe,” SE estimates are robust to disturbances being heteroscedastic if

using the robust option. In the case of heteroscedasticity and autocorrelation within panels, one should use

the “cluster( )” option (Wiggins, 2001; Hoechle, 2007).

11 We use the Stata command “xtpcse.”

157


Chapter 5

experimentation, Reed and Ye (2009) recommend using this estimator when the

discussed non-spherical errors are present, the number of units is greater than the

number of periods, and the primary concern is accurate inference. To control for

potential unobserved heterogeneity, the specifications include country dummies,

capturing stable differences between countries in population health status, and period

dummies, capturing the influence of health shocks in multiple countries at the same

time.

Following Wiggins (2001), we also estimate the relationship by OLS fixed

effects regression, using a variant of the White estimator of robust standard errors

adjusting for clustering over country. This estimator yields consistent estimates of the

covariance matrix under general conditions of heteroscedasticity and autocorrelation

within panels. 12 All FE estimates include period dummies.

5.3.2 Data

Using several data sources, we create a panel dataset for the 1970–2005 period. The

dependent variable and indicator of population health refer to Life expectancy at birth.

This is the average number of years newborns would live, assuming that current

levels and patterns of mortality remain constant over their lifetimes. The measure

refers to the whole population in each country and comes from the World

Development Indicators (World Bank, 2008). Information on life expectancy at birth

is also available for men and women separately, and we use this information in the

sensitivity analysis.

Our globalization indicator is the KOF Index (Dreher et al., 2008), which

measures economic globalization (e.g., using trade flows and trade restrictions), social

globalization (e.g., using tourism and outgoing telephone calls), and political globalization

(e.g., using number of embassies and membership in international organizations). 13

We use the index both as a composite measure, in which the three dimensions of

globalization are equally weighted together, and in a disaggregated format. In either

12 The FE estimator cannot correct for contemporaneous correlation. Moreover, the FE and PCSE

estimators differ in that the former is asymptotic in the number of panels while the latter is asymptotic in

the number of periods.

13 For detailed information on the KOF Index and its various dimensions, see Table 5A in the Appendix.

158


Good for Living? On the Relationship between Globalization and Life Expectancy

case, the index takes values between 0 and 100, higher values indicating more

globalization. To capture the non-linearity between globalization and life expectancy,

we log these indices.

The selection of additional control variables is mainly informed by the

discussion in section 5.2. To indicate the level of economic development, the

specifications include country real GDP per capita (PPP adjusted) from the Heston

(2006). Although the data sample is large, implying that skewness is less of a concern,

we log GDP per capita; a histogram indicates that the empirical variation still is large

after this operation. Furthermore, we use data on the log average years of education in

the population over 15 years old (Barro and Lee, 2000), nutritional status, measured by

log average national calorie intake per day per capita (FAO, 2009), and the log

number of physicians per 1000 people (World Bank, 2008). These controls are all

conservatively assumed to relate positively to life expectancy. To capture economic

and demographic structure, we correct for the urban share of the population and national

dependency ratio in our specifications (World Bank, 2008). The latter refers to the share

of young (age 64) people relative to the working-age population.

To test the robustness of the results, we include several control variables.

Government consumption as a share of GDP (World Bank, 2008) is included to check

whether globalization affects government size in a way that changes its effect on life

expectancy. We also test whether the results are sensitive to the inclusion of Gini

coefficients for net income (taken from Solt, 2008), and to alternative data on the

level of human capital, namely, the log average educational level in the population over

15 years old and in the population over 25 years old. The last two variables come

from Lutz et al. (2007), who derive them by backward simulation using detailed

recent sources on education levels and demographic information. Finally, as a proxy

for instability and rapid change, we include the growth of the urban population.

159


Chapter 5

Table 5.1. Summary statistics

Variable Mean Std. Dev. Min Max n N Source

Life expectancy at birth (years) 65.84 10.47 27.72 81.86 92 608 World Bank, 2008

Life expectancy at birth (years, female) 68.23 11.13 29.63 85.44 92 608 World Bank, 2008

Life expectancy at birth (years, male) 63.55 9.91 25.89 78.45 92 608 World Bank, 2009

Globalization - Kof* 3.77 0.43 2.54 4.53 92 608 Dreher, 2008

Economic globalization - Kof1* 3.81 0.46 2.05 4.56 88 583 Dreher, 2008

Social globalization - Kof2* 3.57 0.58 1.90 4.56 91 604 Dreher, 2008

Political globalization - Kof3* 3.85 0.54 0.76 4.59 92 608 Dreher, 2008

GDP per capita (PPP)* 8.28 1.19 5.46 10.53 92 608 Heston, 2006

Years of education (population 15+)* 1.59 0.65 -1.34 2.49 92 608 Barro and Lee, 2000

Years of education (population 15+, female)* 1.44 0.81 -2.32 2.49 92 608 Barro and Lee, 2000

Years of education (population 15+, male)* 1.71 0.56 -1.34 2.50 92 608 Barro and Lee, 2000

Years of education (population 15+, simulated)* 1.72 0.61 -1.61 2.55 75 445 Lutz et al., 2007

Years of education (population 25+, simulated)* 1.59 0.75 -2.30 2.56 75 445 Lutz et al., 2007

Number of physicians (per 1000 people)* -0.55 1.43 -4.17 1.61 92 608 World Bank, 2008

Nutritional status (average calorie intake per capita)* 7.88 0.19 7.38 8.23 92 608 FAO, 2009

Dependency ratio 0.71 0.19 0.35 1.14 92 608 World Bank, 2008

Urban population 52.40 23.82 4.07 98.27 92 608 World Bank, 2008

Government consumption 20.18 8.06 2.47 67.54 92 608 Heston, 2008

Net income Gini coefficient 37.80 9.59 20.95 63.11 79 448 Solt, 2008

Urban population growth 0.05 0.06 -0.08 0.45 92 608 World Bank, 2008

Low-income country 0.23 0.42 0 1 92 608 World Bank, 2008

Middle-income country 0.46 0.50 0 1 92 608 World Bank, 2008

High-income country

* indicates that the variable is logged

0.31 0.46 0 1 92 608 World Bank, 2008

The initial sample is an unbalanced panel consisting of 121 countries for which the

composite KOF Index is available together with nine periods: 1970–1973, 1974–1977,

1978–1981, 1982–1985, 1986–1989, 1990–1993, 1994–1997, 1998–2001, and 2002–

2005. The observations are period averages, except for the average years of

education, which is only available for particular years. 14 Due to missing data, the

effective sample is smaller than the population of possible observations. Moreover, to

ease interpretation of how additional covariates affect the results, we do not allow the

sample size to vary across tested specifications. The final sample refers to 92

countries (28 high-income, 41 middle-income, and 23 low-income countries) and

more than 600 observations. Table 5B in the Appendix presents a complete list of

countries included in the panel.

Table 5.1 presents summary statistics on the variables of interest. The standard

deviation of life expectancy at birth (female, male, and total) indicates great variation

14 Data on the average number of years of schooling are reported on a five-year basis from 1960 to 2000.

In this study, we linearly interpolate for intervening years. In the final period, “average years of education”

refers to the average number of years of schooling in period t-1. Regression results are robust to the

exclusion of the final period.

160


Good for Living? On the Relationship between Globalization and Life Expectancy

between the population health outcomes of the various countries. The same is true

with respect to globalization.

5.4 RESULTS

Before running estimations, we perform various diagnostic tests. First, using the Hadi

method, we do not detect any outliers. Second, examining pairwise correlations

between variables reveals a close relationship between some of the indicators (see

table 5C in the Appendix), which might inflate standard errors. However, examining

the variance inflation factor (VIF) suggests no incidence of multicollinearity.

Individual figures range from 3.6 (urban) to 6.5 (GDP per capita), which is below the

critical value of 7.

5.4.1 Baseline estimations

Table 4.2 presents estimation results for the relationship between globalization and

life expectancy, controlling for development level and demographic structure.

Regressions using panel-corrected standard errors (PCSE) suggest that the composite

KOF Index is positively related to life expectancy. From testing the components of the

index separately (columns 2–4), it appears that this result is driven by economic

globalization. In baseline estimations, we find no significant relationship between

social or political globalization and life expectancy. As expected, the effect of GDP

per capita is positive while a high dependency ratio is negatively related to life

expectancy. R-squared statistics are omitted for the PCSE regressions, as they include

the influences of country dummies that serve only to control for the influences of

unobserved variables.

Fixed-effect (FE) estimations support the finding that economic globalization

has a positive health effect. However, there is also evidence that political globalization

has a negative health effect, indicating that countries with a greater number of

diplomatic contacts and more involved in the international community have lower

average life expectancies. We will return to this result in the sensitivity analysis.

161


Chapter 5

Table 5.2. Globalization and life expectancy

PCSE PCSE PCSE PCSE FE FE FE FE

KOF (t-1) 1.661** 3.266

[0.732] [3.475]

KOF1 (t-1) 2.702*** 4.473**

[0.756] [2.098]

KOF2 (t-1) 0.572 1.804

[0.300] [1.968]

KOF3 (t-1) -1.181 -2.094*

[0.800] [1.119]

GDP per capita (t-1) 0.867** 0.834 0.832* 1.248** 0.884 0.196 0.753 1.082

[0.449] [0.616] [0.445] [0.622] [1.623] [1.723] [1.737] [1.465]

Dependency -4.388** -2.944 -5.102** -4.809* -2.332 -1.874 -2.884 -5.365

[2.189] [2.474] [2.344] [2.483] [5.117] [5.168] [5.593] [4.562]

Observations 608 583 604 608 608 583 604 608

Number of countries 92 88 91 92 92 88 91 92

R-squared (within) 0.448 0.452 0.433 0.448

*, **, and *** denote statistical significance at the 10 %, 5%, and 1% levels, respectively.

PCSE: Estimations include country dummies and period dummies; panel-corrected standard errors in brackets.

FE: Estimations with country and period fixed effects; robust standard errors in brackets.

A Hausman specification test suggests that an FE model matches the data better than

does a random-effects model. Moreover, period dummies are jointly significant in the

specifications and thus should be included. In this stage, we also assess the presence

of serial correlation. Using a test derived by Wooldridge (2002), the null hypothesis of

no serial correlation is strongly rejected, which supports the clustering at the panel

level and the AR correction.

Tables 5.3a and 5.3b show how the results change when including additional

control variables. The positive association between economic globalization and life

expectancy remains significant across all specifications. The magnitude of the effect is

rather stable, with a coefficient estimate of approximately 3 in PCSE estimations,

suggesting that a 10% increase in economic globalization increases life expectancy by

0.3 years. This result confirms the findings of Owen and Wu (2007) and Stroup

(2007) that more openness and economic freedom are associated with higher life

expectancies. Regarding the social dimension of globalization, none of the models

indicates that such integration is a significant determinant of life expectancy. The

negative relationship between political integration and population health remains in

the FE estimations and also appears in the PCSE estimations that include additional

control variables.

162


Good for Living? On the Relationship between Globalization and Life Expectancy

Regardless of estimation technique, we find that a greater number of

physicians per capita and a larger per capita calorie intake have strong and robust

positive effects on life expectancy, confirming previous findings in the literature. On

the other hand, neither the average education level of the population nor the share of

people living in urban areas is significantly associated with longevity. The former

result is unexpected, as we believe that schooling provides people with new skills

relevant to health outcomes (the same outcome appears in a related study by Ovaska

and Takashima, 2006). One interpretation is that it is the quality of learning that is

important to better health, not how many people from the same cohort graduate at a

particular level.

Relating to the discussion of the relative importance of income to population

health, it appears that the coefficient estimates of GDP per capita become

insignificant when adding more covariates to the model. The non-positive effect of

average income on life expectancy corresponds to the findings of some related

studies (Owen and Wu 2007, Ovaska and Takashima 2006) and confirms the results

of Pritchett and Summers (1996), who estimate the effect of income on life

expectancy in a panel of countries. Moreover, the demographic structure indicator

loses significance when including additional control variables.

163


Table 5.3a. Including additional control variables: PCSE

PCSE PCSE PCSE PCSE PCSE PCSE PCSE PCSE

KOF (t-1) 1.406 1.950**

[0.923] [0.902]

KOF1 (t-1) 2.771*** 3.372***

[0.774] [0.752]

KOF2 (t-1) -0.240 0.685

[0.569] [0.621]

KOF3 (t-1) -1.454* -1.543*

[0.776] [0.804]

GDP per capita (t-1) 1.073** -0.182 0.805 -0.705 0.931* -0.098 1.289** -0.143

[0.518] [0.635] [0.589] [0.555] [0.521] [0.636] [0.609] [0.610]

Dependency -4.469** -1.561 -2.821 -0.626 -5.156** -2.136 -4.569** -3.116

[2.070] [2.534] [2.486] [2.736] [2.281] [2.659] [2.325] [2.286]

Urban share of population 0.008 0.042 0.024 0.050 0.046* 0.048 0.019 0.050

[0.025] [0.035] [0.029] [0.033] [0.026] [0.035] [0.028] [0.034]

Average year of education -0.206 -0.701 -0.477 -0.855 1.257 -0.470 1.292 0.615

[1.091] [1.164] [1.142] [1.184] [0.803] [1.225] [0.808] [0.897]

Physicians 1.000** 0.978** 0.983** 0.803**

[0.395] [0.408] [0.382] [0.341]

Nutrition 11.34*** 11.02*** 11.43*** 11.25***

[3.192] [3.121] [3.279] [3.161]

Observations 608 608 583 583 604 604 608 608

Number of countries 92 92 88 88 91 91 92 92

R-squared (within)

*, **, and *** denotes statistical significance at the 10 %, 5%, and 1% levels respectively.

Estimations include country dummies and period dummies; panel-corrected standard errors in brackets.


Table 5.3b. Including additional control variables: Fixed effects

FE FE FE FE FE FE FE FE

KOF (t-1) 3.290 3.577

[3.618] [3.003]

KOF1 (t-1) 4.448** 4.126**

[1.963] [1.731]

KOF2 (t-1) 1.825 1.584

[1.931] [1.675]

KOF3 (t-1) -2.253* -1.612*

[1.210] [0.878]

GDP per capita (t-1) 0.895 -1.146 0.214 -1.591 0.852 -1.095 1.139 -0.714

[1.528] [1.401] [1.535] [1.434] [1.521] [1.413] [1.276] [1.288]

Dependency -2.749 0.586 -1.792 1.212 -3.178 -0.259 -4.924 -1.764

[4.525] [4.288] [4.763] [4.500] [4.861] [4.467] [3.992] [3.806]

Urban share of population -0.017 -0.025 0.0006 -0.009 -0.028 -0.037 -0.022 -0.031

[0.103] [0.087] [0.104] [0.089] [0.110] [0.094] [0.120] [0.104]

Average year of education 0.656 -0.564 0.751 -0.308 1.078 -0.170 2.647 1.165

[2.188] [1.724] [2.124] [1.631] [2.203] [1.692] [1.878] [1.563]

Physicians 1.711*** 1.658** 1.736*** 1.389***

[0.632] [0.648] [0.630] [0.527]

Nutrition 15.93*** 15.29*** 15.83*** 15.52***

[4.198] [4.380] [4.163] [4.063]

Observations 608 608 583 583 604 604 608 608

Number of countries 92 92 88 88 91 91 92 92

R-squared (within) 0.438 0.521 0.453 0.529 0.434 0.516 0.454 0.524

*, **, and *** denotes statistical significance at the 10 %, 5%, and 1% levels respectively.

Estimations with country- and period fixed effects; robust standard errors in brackets.


Chapter 5

5.4.2 Sensitivity analysis

Tables 5.4a and 5.4b list the PCSE regression coefficient estimates of the

globalization indices for variations of sensitivity analyses using the preferred

specification with the complete set of control variables. Although our FE

specifications limit the potential for omitted variables to drive the results, the first

type of robustness assessment involves the adding of covariates. Baseline findings could

be spurious if, on one hand, lack of globalization is correlated with instability and if

instability, on the other hand, is a cause of poor health outcomes. Following Tsai

(2007), we therefore control for the influence of instability and rapid change on

health by including urban population growth. The urbanization rate is not

significantly associated with life expectancy and the inclusion of the covariate does

not alter our previous findings regarding economic and political globalization. This is

also true when controlling for within-country net income Gini coefficients, an

exercise that significantly reduces the number of observations examined. With this

specification and sample, moreover, there is evidence that the social dimension of

globalization has a positive effect on life expectancy. Unlike several studies of the

relationship between income inequality and population health (e.g., Babones, 2008),

we find that higher income inequality correlates with good health status. 15

Testing the robustness of our results by including information on government

consumption leaves the relationship between dimensions of globalization on life

expectancy unchanged compared to baseline estimations. The effect of government

consumption is negative but statistically insignificant suggesting that this is not an

omitted variable influencing our conclusions. In addition, when including all subcomponents

of the globalization index simultaneously in one specification, economic

globalization remains positive and significant while greater political integration has a

negative effect. The same is true when testing baseline results with respect to sample

coverage, allowing the sample size to vary across specifications. Including a maximum

of 117 countries and 725 observations in the analysis does not alter the baseline

findings.

15 The reason why we test the effect of income inequality in the sensitivity analysis, rather than in our main

scenario, is that we lose many observations when including standardized Gini coefficients.

166


Good for Living? On the Relationship between Globalization and Life Expectancy

A second type of robustness test addresses the timing of effects. Using current

rather than lagged GDP per capita does not change our initial conclusions, with

respect to either the effect of globalization or the role of income. More income does

not directly contribute to better health. Furthermore, we also test the assumption that

the impact of globalization on health is contemporaneous. Interestingly, unlike the

baseline findings, increasing political collaboration between economies has no

immediate negative effect on health status. However, there is a significant

simultaneous relationship between economic globalization and life expectancy.

Notably, the magnitude of the coefficient indicates that the health benefit of

economic globalization is greater when the process is allowed to work for some years.

A third set of sensitivity tests involves replacement of variables. Replacing

information on average education level with corresponding information from an

alternate data source (Lutz et al., 2007) generates a smaller sample to analyze. In this

setting, the negative effect of political globalization on life expectancy again

disappears, while economic integration remains beneficial. We also replace the

dependent variable and run separate regressions focusing separately on female and

male life expectancy. Unlike Bussmann’s (2009) results, which indicate no significant

relationship between economic openness and female life expectancy, our baseline

results remain unaltered when modeling female or male longevity. In fact, our

findings indicate that globalization is more beneficial to women than to men: the

positive association with economic globalization is stronger, whereas the negative

impact of political globalization is weaker. We have also verified that our results hold

when not using any logged variable values.

A fourth type of sensitivity assessment examines whether baseline outcomes

change when excluding various groups of countries. Excluding East Asian countries in

the sample has little effect, keeping the effect of economic globalization significant

and positive and that of political globalization significant and negative. Excluding

Latin American countries, however, renders a situation in which political

globalization does not reduce life expectancy. The negative influence of the political

dimension of globalization also disappears when excluding the Sub-Saharan African

countries from the analysis. Excluding these countries further reveals a positive

167


Chapter 5

relationship between social globalization and population health. Apparently, closer

social integration and more personal cross-border contacts generally improve

population health, but not in Sub-Saharan Africa. The non-positive effect of social

integration in this region likely relates to the negative health effects of HIV/AIDS, as

mobility generally is found to be a key factor promoting HIV diffusion.

Finally, we exclude the five countries in our sample with the highest

prevalence of HIV, where life expectancy has decreased over the 1990–2005 period.

In addition, this exercise renders the effect of political globalization insignificant

while the effect of economic globalization remains positive and significant.

To summarize, the positive effect of economic globalization on life

expectancy is very robust. Increasing economic interaction with other countries is

important in improving average health outcomes. Conversely, the initially stated

negative relationship between political globalization and health is sensitive to the

selection of countries. Closer examination of the data reveals that many countries in

Latin America have experienced decreasing political globalization, increasing

economic globalization, and increasing life expectancy since the 1970s. This opposite

trend in types of globalization is difficult to interpret but the positive trend in

economic globalization is possibly an effect of what Biglaiser (2002) calls “the

internationalization of Chicago’s economics in Latin America.” 16 In any case, the

negative effect of political globalization found in our main specification is not robust

and seems to be driven by specific circumstances in certain countries, not picked up

by the country fixed effects.

16 Biglaiser (2002) analyzes how US government-supported training of Latin American economists at the

University of Chicago translated into general support for economically liberal reforms in many Latin

American countries.

168


Table 5.4a. Sensitivity analysis

Variation Composite KOF index Significant components Comments

Baseline model 1.950** [0.902] KOF1 (t-1) 3.372*** [0.752] Baseline estimates

KOF2 (t-1) 0.685 [0.621] Corresponds to the results in Table 3

KOF3 (t-1) -1.543* [0.804]

Controling for income inequality 2.628** [1.028] KOF1 (t-1) 3.195*** [0.909] Income inequality positive and significant except

KOF2 (t-1) 1.178** [0.514] when controlling for social globalization

KOF3 (t-1) -1.278** [0.544] Reduced sample: 79 countries, 448 observations

Controlling for government consumption 2.075** [1.050] KOF1 (t-1) 3.377*** [0.715] Government consumption negative and insignificant

KOF3 (t-1) -1.466** [0.744]

Controlling for urban population growth 2.321** [1.054] KOF1 (t-1) 3.279*** [0.713] Urbanization rate positive and insignificant

KOF3 (t-1) -1.518** [0.725]

All sub-indices of globalization KOF1 (t-1) 3.141*** [0.946]

together in the same specification KOF3 (t-1) -1.119** [0.542]

Controling for non-lagged GDP per capita 1.895* [1.072] KOF1 (t-1) 3.539*** [0.718] GDP per capita insignificant

KOF3 (t-1) -1.530** [0.724]

Controling for non-lagged globalization 0.765 [1.164] KOF1 2.757*** [0.696] GDP per capita insignificant

and non-lagged GDP per capita

Alternative data on education (+15 years) 1.113 [1.069] KOF1 (t-1) 3.756*** [0.732] Education insignificant

Reduced sample: 75 countries. 445 observations.

Alternative data on education (+25 years) 0.996 [1.085] KOF1 (t-1) 3.694*** [0.732] Education negative and significant

*, **, and *** denote statistical significance at the 10 %, 5%, and 1% levels, respectively.

Estimations include country dummies and period dummies; panel-corrected standard errors in brackets.


Table 5.4b. Sensitivity analysis (cont.)

Variation Composite KOF index Significant components Comments

Replacing life expectancy with female life expectancy 2.497** [1.136] KOF1 (t-1) 3.473*** [0.769] Average years of education refers in this case to

KOF3 (t-1) -1.472* [0.771] average years of education in female population

Replacing life expectancy with male life expectancy 1.321 [0.987] KOF1 (t-1) 3.203*** [0.689] Average years of education refers in this case to

KOF3 (t-1) -1.586** [0.675] average years of education in male population

Excluding countries with high prevalence of HIV 2.266*** [0.651] KOF1 (t-1) 1.718*** [0.625] Botswana, Namibia, South Africa, Zambia and Zimbabwe

(5 countries) all have an estimated prevalence of HIV of +15 per cent in the adult population

Dependency negative and significant. Education positive and significant.

Excluding sub-Saharan African coutries 1.713*** [0.532] KOF1 (t-1) 1.568*** [0.479] Estimations exclude countries with very high

(23 countries) KOF2 (t-1) 0.639** [0.325] and high adult prevalence of HIV

Excluding Latin American countries 2.380* [1.296] KOF1 (t-1) 2.727*** [0.913]

(23 countries)

Excluding East Asian countries 1.999 [1.221] KOF1 (t-1) 3.535*** [0.761]

(10 countries) KOF3 (t-1) -1.903** [0.839]

*, **, and *** denote statistical significance at the 10 %, 5%, and 1% levels, respectively.

Estimations include country dummies and period dummies; panel-corrected standard errors in brackets.


Good for Living? On the Relationship between Globalization and Life expectancy

5.4.3 Distinguishing between levels of development

The relationship between globalization and life expectancy may well differ between

rich and poor countries. For one thing, Cutler et al. (2006) note that the mortality

pattern is very different: in low-income countries, 30% of all deaths occur before age

4, while the corresponding proportion in high-income countries is 0.9%. For another,

high-income countries have more deaths caused by cancer and cardiovascular disease,

while low-income countries have more deaths from respiratory infections and

HIV/AIDS. This suggests that even small improvements in knowledge, nutrition,

and access to pharmaceuticals may have large positive health effects in low-income

countries. Finally, the sensitivity analysis indicated a negative relationship between

political integration and health in (some) low- and middle-income countries.

We first examine the relationship between globalization and population health

for countries with low GDP per capita in 1970. These 47 countries are kept in the

sample regardless of whether they remained poor throughout the period or whether

they moved up the income per capita ladder. 17 As shown in Table 5.5, both economic

and social globalization seem to increase life expectancy under these circumstances.

The size of the effect of economic globalization is about the same as in the full

sample. Notably, there is in this case no negative relationship between political

globalization and life expectancy. The estimation results moreover confirm the initial

finding that public health and nutrition matter to longevity. In fact, the magnitude of

the positive impact of higher calorie intake is greater in this setting than when using

the full sample.

A standard approach when examining whether coefficients vary with income

level is to include interaction terms. For example, Owen and Wu (2007) find a

negative multiplicative effect, suggesting that trade openness has a greater effect in

low-income countries, using this technique.

As noted by Braumoeller (2004), multiplicative interaction terms, however,

make it harder to interpret other coefficients in the model, and the use of interaction

17 Tables 5D and 5E in the Appendix present pairwise correlations between variables and summary

statistics for this sub-sample. Summary statistics indicate that 50% of the observations in this sample

belong to countries currently classified as middle-income countries.

171


Chapter 5

terms assumes a simple linear relationship between (in our case) the effect of

globalization and income. When we include interaction terms between dimensions of

globalization and income, both globalization coefficients and interactions terms are

insignificant, suggesting that there is no simple linear relationship between size of

globalization coefficient and income. 18

Table 5.5. Globalization and life expectancy: Low-income countries in 1970

PCSE PCSE PCSE PCSE

KOF (t-1) 2.621*

[1.467]

KOF1 (t-1) 2.601***

[0.852]

KOF2 (t-1) 1.525**

[0.763]

KOF3 (t-1) -0.948

[0.851]

GDP per capita (t-1) 1.211 0.851 1.066 1.131

[0.825] [0.875] [0.697] [0.813]

Dependency 0.813 3.657 0.599 1.519

[4.280] [4.949] [4.242] [4.095]

Urban share of population 0.0188 0.0408 0.0239 -0.0643

[0.0620] [0.0564] [0.0644] [0.0659]

Average years of education -0.785 -1.394 -0.614 -0.783

[1.287] [1.328] [1.405] [0.942]

Physicians 1.500*** 1.544*** 1.460*** 1.149**

[0.501] [0.551] [0.487] [0.536]

Nutrition 15.54*** 15.07*** 15.90*** 16.70***

[4.222] [4.324] [4.192] [4.655]

Observations 307 282 303 307

Number of countries 47 43 46 47

*, **, and *** denotes statistical significance at the 10 %, 5%, and 1% levels respectively.

Estimations include country dummies and period dummies; panel-correscted standard errors in brackets.

To get a more thorough and meaningful understanding of how the globalization–

health relationship varies with income level, we estimate the globalization coefficients

repeatedly while we exclude, one-by-one, observations from the highest-income

countries and re-estimate the equation.

Figures 5.3-5.5 demonstrate how the coefficient estimates and panel-corrected

standard errors (for a 95% confidence interval) of economic, social, and political

18 In our case, adding an interaction term turns the coefficients of the lower-order terms into conditional

effects, measuring the effect of types of globalization when GDP per capita equals zero.

172


Good for Living? On the Relationship between Globalization and Life expectancy

globalization vary as we gradually move from full sample to focusing only on the

observations from the lowest-income countries. 19 The graph shows that little happens

to the different globalization estimates as we gradually restrict the full sample by

excluding all observations from countries with incomes higher than approximately

4000 PPP dollars. The relationship is insignificant at lower GDP levels, but when we

focus only on the lowest-income countries in our sample, the effect is actually

positive and significant. A similar pattern holds for social globalization, except that

the effect in most regressions is not significantly different from zero.

Political globalization, however, is negative and often borderline significant

until we exclude countries with incomes higher than approximately 3000 PPP dollars.

Below this level, the effect is actually sometimes positive and significant. However,

we know from the sensitivity analysis that the effects of political globalization are

likely driven by just a few countries, explaining the sudden jumps in the curve

occurring when observations from these countries are excluded.

In general, the shape of the coefficient curves in Figures 5.3–5.5 reveals that the

globalization–health relationship varies with income level in a way too complex to be

captured by interaction effects or sample divisions only.

Figure 5.3.

Coefficients relating economic globalization to life expectancy at different levels of GDP per capita.

19 Figures 5.3, 5.4, and 5.5 do not include coefficient estimates based on the 40 observations from

countries with the lowest GDP per capita.

173


Chapter 5

Figure 5.4.

Coefficients relating social globalization to life expectancy at different levels of GDP per capita.

Figure 5.5.

Coefficients relating political globalization to life expectancy at different levels of GDP per capita.

174


Good for Living? On the Relationship between Globalization and Life expectancy

5.5 CONCLUSION

This paper examines the relationship between dimensions of globalization and life

expectancy. Our choice of dependent variable means that we differ from the

mainstream debate concerning the consequences of globalization, in which the effect

on GDP levels and growth has attracted much attention—for obvious reasons.

However, especially when it comes to the effects of globalization in low-income

countries, we should acknowledge that there are substantial measurement problems

in GDP data, and that results consequently should be interpreted with care. We do

not claim that life expectancy data are free from measurement errors, but we do argue

that our attempt to analyze the relationship between globalization and health is an

important complement to existing studies using other dependent variables.

Among our results, the most robust finding is the positive relationship

between economic globalization and life expectancy. While the effect of the KOF

Index on life expectancy has not been systematically analyzed before, our finding is in

line with previous findings, such as those of Owen and Wu (2007), who find that

trade/GDP has a positive effect on life expectancy. We find no evidence that this

positive effect is driven by rich countries: In fact, excluding the observations from

countries with the highest income increases the estimated effect until all observations

with incomes higher than 7300 PPP dollars are excluded. After that, the effect

decreases and is sometimes insignificant, though in the poorest part of our sample,

the effect is again positive, and both economically and statistically significant. In any

case, our analysis indicates that including only interaction terms will not fully capture

how globalization effects depend on income level.

To put the size of our estimated effect into perspective, note that Uganda, for

example, increased its KOF value for economic globalization from 22 to 46 (almost

two standard deviations) over the 1970–2005 period, thereby increasing life

expectancy by two to three years, according to our estimates. This effect is about as

great as a one-standard-deviation increase in nutritional intake, which increases life

expectancy by roughly two years. 20 Such calculations are only for illustrative purposes,

20 Assuming a coefficient for economic globalization of approximately 3 to 4, and a nutrition coefficient of

11 (taken from Tables 5.3 and 5.4).

175


Chapter 5

but they do indicate that the effects are economically and politically relevant. Closer

economic integration is clearly related to improved population health.

As for social and political globalization, life expectancy tends to be positively

related to social globalization and negatively related to political globalization, but

these effects are not robust to the various sensitivity tests performed. In particular,

the effect of political globalization seems to be very dependent on country-specific

circumstances.

Finally, it should be stressed that the globalization effects we find hold when

controlling for the four factors that other studies have found to be important for life

expectancy, i.e., nutrition, literacy, income, and public health (proxied by physicians

per capita). This suggests the beneficial channels leading from globalization to

improved life expectancy go beyond the suggested indirect effects. Those parts of the

globalization effect that work through other mechanisms, however, might be hard to

measure, such as knowledge transfer or changes in relative prices. Further research is

needed to learn more about the relevant mechanisms at work in the relationship

between globalization and health.

176


Table 5A The KOF Index of Globalization

A. Economic Globalization

Good for Living? On the Relationship between Globalization and Life expectancy

APPENDIX

i) Actual Flows

Trade (percent of GDP)

Foreign direct investment, flows (percent of GDP)

Foreign direct investment, stocks (percent of GDP)

Portfolio investment (percent of GDP)

Income payments to foreign nationals (percent of GDP)

ii) Restrictions

Hidden import barriers

Mean tariff rate

Taxes on international trade (percent of current revenue)

Capital account restrictions

B. Social Globalization

i) Data on Personal Contacts

Outgoing telephone traffic

Transfers (percent of GDP)

International tourism

Foreign population (percent of total population)

International letters (per capita)

ii) Data on Information Flows

Internet hosts (per 1000 people)

Internet users (per 1000 people)

Cable television (per 1000 people)

Trade in newspapers (percent of GDP)

Radios (per 1000 people)

iii) Data on Cultural Proximity

Number of McDonald’s restaurants (per capita)

Number of Ikeas (per capita)

Trade in books (percent of GDP)

C. Political Globalization

Embassies in country

Membership in international organizations

Participation in U.N. Security Council missions

177


Chapter 5

Table 5B. Sample coverage

Low-income countries

Bangladesh, Benin, Burundi, Central African Republic, Chad , Congo, Dem. Rep., Cote

d'Ivoire , Ghana, Guinea-Bissau, Haiti, India, Kenya, Madagascar , Malawi, Mali, Myanmar

Nepal, Niger, Nigeria, Pakistan, Rwanda, Senegal, Sierra Leone, Tanzania, Togo, Uganda,

Zambia, Zimbabwe

Middle-income countries

Albania , Algeria, Argentina, Belize , Bolivia, Botswana, Brazil, Bulgaria , Cameroon, Chile,

China, Colombia, Congo, Rep., Costa Rica, Croatia , Dominican Republic, Ecuador, Egypt,

El Salvador, Fiji, Gabon , Guatemala, Guyana, Honduras, Hungary, Indonesia, Iran, Islamic

Rep., Jamaica, Jordan, Latvia, Lithuania , Malaysia, Mauritius, Mexico, Morocco , Namibia

Nicaragua, Panama, Paraguay, Peru, Philippines, Poland, Romania , Russian Federation, Slovak

Republic , South Africa, Sri Lanka, Syrian Arab Republic, Thailand, Tunisia, Turkey, Ukraine

Uruguay, Venezuela RB

High-income countries

Australia, Austria, Bahamas , Barbados, Belgium, Canada, Cyprus, Czech Republic ,Denmark,

Estonia , Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Korea, Rep.,

Kuwait, Luxembourg , Malta , Netherlands, New Zealand, Norway, Portugal, Slovenia ,Spain,

Sweden, Switzerland, Trinidad and Tobago, United Arab Emirate s, United Kingdom, United

States

Countries in italics are only included in the regressions in the sensitivity analysis when

we allow the sample size to vary across specifications.

178


Table 5C Correlation matrix: Full sample

KOF1 KOF2 KOF3

GDP

per capita Dependency

Urban

population

Average

years

of education

Physicians Nutrition

Income

inequality

Government

consumption

Urbanization

rate

Alternative

education measure

(+15 years)

Alternative

education measure

(+25 years)

KOF1 1

KOF2 0.840 1

KOF3 0.245 0.374 1

GDP per capita 0.764 0.840 0.394 1

Dependecy -0.608 -0.651 -0.415 -0.767 1

Urban population 0.652 0.699 0.424 0.774 -0.655 1

Average years of education 0.683 0.786 0.527 0.840 -0.790 0.718 1

Physicians 0.655 0.694 0.447 0.789 -0.762 0.733 0.806 1

Nutrition 0.654 0.640 0.481 0.757 -0.708 0.657 0.755 0.747 1

Income inequality -0.291 -0.395 -0.428 -0.481 0.628 -0.350 -0.588 -0.593 -0.572 1

Government consumption 0.065 0.007 -0.162 -0.083 0.196 -0.179 -0.142 -0.040 -0.159 -0.124 1

Urbanization rate -0.450 -0.440 -0.281 -0.495 0.442 -0.422 -0.480 -0.554 -0.410 0.316 -0.023 1

Alternative education measure (+15 years) 0.689 0.786 0.383 0.816 -0.750 0.716 0.926 0.816 0.678 -0.560 -0.041 -0.546 1

Alternative education measure (+25 years) 0.684 0.784 0.380 0.808 -0.755 0.703 0.921 0.810 0.672 -0.571 -0.035 -0.544 0.995 1


Table 5D Correlation matrix: Low-income countries in 1970

KOF1 KOF2 KOF3

GDP

per capita Dependency

Urban

population

Average

years

of education

Physicians Nutrition

KOF1 1

KOF2 0.772 1

KOF3 0.108 0.174 1

GDP per capita 0.592 0.666 0.318 1

Dependecy -0.338 -0.383 -0.412 -0.675 1

Urban population 0.543 0.559 0.331 0.754 -0.456 1

Average years of education 0.500 0.602 0.299 0.720 -0.604 0.594 1

Physicians 0.377 0.406 0.351 0.729 -0.591 0.693 0.658 1

Nutrition 0.442 0.355 0.463 0.632 -0.536 0.555 0.486 0.652 1

Table 5E Summary statistics: Low-income countries in 1970

Variable Mean Std. Dev. Min Max n N

Life expectancy at birth (years) 58.75 9.21 27.72 77.59 47 307

Globalization - Kof* 3.41 0.33 2.31 4.19 47 307

Economic globalization - Kof1* 3.54 0.44 2.05 4.34 43 282

Social globalization - Kof2* 3.18 0.43 1.90 4.15 46 303

Political globalization - Kof3* 3.64 0.54 0.76 4.53 47 307

GDP per capita (PPP)* 7.19 0.87 5.07 9.62 47 307

Years in education (population 15+)* 1.20 0.64 -1.34 2.38 47 307

Number of physicians (per 1000 people)* -1.58 1.24 -4.17 0.89 47 307

Nutritional status (average calorie intake per capita)* 7.75 0.15 7.38 8.16 47 307

Dependency ratio 0.83 0.16 0.39 1.14 47 307

Urban population 35.50 16.56 4.07 80.44 47 307

Low-income country 0.46 0.50 0 1 47 307

Middle-income country 0.52 0.50 0 1 47 307

High-income country 0.02 0.15 0 1 47 307

* indicates that the variable is logged


Good for Living? On the Relationship between Globalization and Life Expectancy

REFERENCES

Arribas, I., Pérez, F., Tortosa-Ausina, E. 2009. Measuring globalization of international

trade: Theory and evidence. World Development, 37, 127–145.

Babones, S. J. 2008. Income inequality and population health: Correlation and causality.

Social Science & Medicine, 66, 1614–1626.

Barro, Robert J. and Jong-Wha Lee. 2000. "International Data on Educational

Attainment: Updates and Implications." CID Working Paper No. 42, April 2000.

Beck, Nathaniel and Jonathan N. Katz. 1995. “What to do (and not to do) with Time-

Series Cross-Section Data.” American Political Science Review 89:634–647

Bergh, Andreas and Therese Nilsson. 2008. "Do economic liberalization and

globalization increase income inequality?" Paper presented at the Annual Meeting of the

American Public Choice Society, San Antonio, 6–9 March 2008.

Biglaiser Glen. 2002. “The internationalization of Chicago's economics in Latin

America.” Economic Development and Cultural Change 50:269–286

Bjørnskov, C., Dreher, A., & Fischer, J. (2007). The bigger the better? Evidence of the

effect of government size on life satisfaction around the world. Public Choice

130:267–292.

Braumoeller, B. F. 2004. “Hypothesis Testing and Multiplicative Interaction Terms

International Organization 58(4): 807-820

Bussmann, Margit. 2009. “The Effect of Trade Openness on Women's Welfare and

Work Life.” World Development 37: 1027–1038.

Cutler, David, Angus Deaton, and Adriana Lleras-Muney. 2006. "The Determinants of

Mortality." The Journal of Economic Perspectives 20:97–120.

Deaton, A. 2004. "Health in an Age of Globalization " NBER Working Paper No. 10669.

Dreher, Axel. 2006. "Does Globalization Affect Growth? Empirical Evidence from a

new Index." Applied Economics 38:1091–1110.

Dreher, Axel and Noel Gaston. 2008. "Has Globalization Increased Inequality?" Review of

International Economics 16:516–536.

Dreher, Axel, Noel Gaston, and Pim Martens. 2008. Measuring Globalisation: Gauging Its

Consequences: Springer.

FAO (2009). FAOSTAT, Food and Agriculture Organization of the United Nations

http://faostat.fao.org/site/345/default.aspx

Fayissa, Bichaka and Paulos Gutema. 2005. "Estimating a health production function for

Sub-Saharan Africa (SSA)." Applied Economics 37:155–164.

Gravelle, Hugh. 1998. "How much of the relation between population mortality and."

BMJ: British Medical Journal 316:382–386.

Grosse, R. N. and B. H. Perry. 1982. "Correlates of life expectancy in less developed

countries." Health Policy and Education 2:275–304.

Gwartney, James, Robert Lawson, and Seth Norton. 2008. "Economic Freedom of the

World: 2008 Annual Report." The Fraser Institute. Data retrieved from

www.freetheworld.com.

Heston, A., Summers R. and B. Aten. 2006. Penn World Table Version 6.2, Center for

International Comparisons of Production, Income and Prices at the University of

Pennsylvania, September 2006.

Hoechle, Daniel. 2007. “Robust standard errors for panel regressions with cross–

sectional dependence.” Stata Journal 7:1–31

Husain, AR. 2002. "Life Expectancy in Developing Countries: A Cross-Section

Analysis." The Bangladesh Development Studies 28.

181


Chapter 5

Kabir, Mahfuz. 2008. "Determinants of life expectancy in developing countries.” The

Journal of Developing Areas 41:185–198, 201–204.

Katzenstein, Peter. 1985. Small states in world markets: Industrial policy in Europe. Ithaca,

N.Y.: Cornell University Press.

Kawachi, I and S Wamala. 2007. Globalization and Health. Oxford: Oxford University

Press.

Lee Ha, Yan, Antonio Ricci Luca, and Roberto Rigobon. 2004. "Once again, is openness

good for growth?" Journal of Development Economics 75:451–472.

Lichtenberg, Frank R. 2003. "The impact of new drug launches on longevity: Evidence

from longitudinal, disease-level data from 52 countries, 1982–2001." NBER

Working Paper 9754.

Lindbeck, Assar. 1975. "Business cycles, politics, and international economic

dependence." Skandinaviska Enskilden Bank Quarterly Review 2:53–68.

Lutz, W, Anne Goujon, K. C. Samir, and Warren Sanderson. 2007. "Reconstruction of

populations by age, sex and level of educational attainment for 120 countries for

1970–2000." Vienna Yearbook of Population Research 2007:193–235.

Medez, A and B.M Popkin. 2004. "Globalization, Urbanization and Nutritional change in

the Developing World." Journal of Agricultural and Development Economics 1:220–241.

Mellor, Jennifer M. and Jeffrey Milyo. 2002. “Income inequality and individual health:

Evidence from the Current Population Survey.” Journal of Human Resources

37:510–539

Nissanke, Machiko and Erik Thorbecke. 2006. "Channels and Policy Debate in the

GlobalizationInequality–Poverty Nexus." World Development 34:1338–1338.

Ovaska, Tomi and Ryo Takashima. 2006. "Economic policy and the level of selfperceived

well-being: An international comparison." Journal of Socio–Economics

35:308–325.

Owen, Ann L. and Stephen Wu. 2007. "Is Trade Good for Your Health?" Review of

International Economics 15:660–682.

Papageorgiou, Chris, Andreas Savvides, and Marios Zachariadis. 2007. "International

medical technology diffusion*." Journal of International Economics 72:409–409.

Pritchett, L. and L. H. Summers (1996). “Wealthier is healthier.” Journal of Human

Resources, 31(4): 841–868

Ravallion, Martin. 2006. "Looking Beyond Averages in the Trade and Poverty Debate."

World Development 34:1374–1374.

Reed, B., & Ye, H. 2009.“Which Panel Data Estimator Should I Use?.” forthcoming in

Applied Economics

Rodrik, Dani. 1998. "Why do more open economies have bigger governments?" The

Journal of Political Economy 106:997–1032.

Rodriguez, F and R Rodrik. 2000. "Trade policy and economic growth: a skeptic’s guide

to the cross-national evidence." NBER Macroeconomics Annual:261–325.

Schwartz, Barry. 2004. The Paradox of Choice. Why more is less. New York: Harper & Collins.

Shaw, James W., William C. Horrace, and Ronald J. Vogel. 2005. "The Determinants of

Life Expectancy: An Analysis of the OECD Health Data." Southern Economic

Journal 71:768–783.

Sinn, Hans Werner. 1997. "The selection principle and market failure in systems

competition." Journal of Public Economics 66:247–274.

Soares, Rodrigo. 2007. "On the Determinants of Mortality Reductions in the Developing

World." Population and Development Review 33:247–287.

182


Good for Living? On the Relationship between Globalization and Life Expectancy

Solt, Frederick. 2008. "Standardizing the World Income Inequality Database." Social

Science Quarterly Forthcoming.

Stark, Oded. 2004. "Rethinking the Brain Drain." World Development 32:15–22.

Stroup, Michael D. 2007. "Economic Freedom, Democracy, and the Quality of Life."

World Development 35:52–52.

Tsai, Ming-Chang. 2007. "Does globalization affect human well-being?" Social Indicators

Research 81:103–126.

Veenhoven, R. 2005. "Review of Barry Schwartz ``The Paradox of Choice. Why more is

less'' Harper & Collins, New York, USA 2004, ISBN 0-06-000568-8, 265 pages."

Journal of Happiness Studies 6:93–96.

Wiggins, Vince. 2001 “Comparing XTGLS with Regress Cluster ()” (STATA

Corporation), http://www.stata.com/support/faqs/stat/xtgls_rob.html

Wilkinson, Richard G. 1996. Unhealthy Societies: The Afflictions of Inequality. London:

Routledge.

Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data.

Cambridge, MA: MIT Press

World Bank. 2008. World Development Indicators. Washington DC: World Bank.

Yach, D, H Wipfli, R Hammond, and S. Glantz. 2007. "Globalization and Tobacco." in

Globalization and Health, edited by I. Kawachi and S. Wamala. Oxford: Oxford

University Press.

Acknowledgements

We thank Christian Bjørnskov, Jesper Roine, Carl Hampus Lyttkens, seminar participants

at the Ratio institute, and participants at the Annual Meeting of the American Public

Choice Society in Las Vegas, March 2009, for useful comments on earlier versions of this

paper. Therese Nilsson gratefully acknowledges financial support from the Swedish Council

for Working life and Social research (Health Economic Program, Lund University).

183


LUND ECONOMIC STUDIES

1. Guy Arvidsson Bidrag till teorin för verkningarna av räntevariationer, 1962

2. Björn Thalberg A Trade Cycle Analysis. Extensions of the Goodwin Model,

1966

3. Bengt Höglund Modell och observationer. En studie av empirisk anknytning och

aggregation för en linjär produktionsmodell, 1968

4. Alf Carling Industrins struktur och konkurrensförhållanden, 1968

5. Tony Hagström Kreditmarknadens struktur och funktionssätt, 1968

6. Göran Skogh Straffrätt och samhällsekonomi, 1973

7. Ulf Jakobsson och Göran Inkomstbeskattningen i den ekonomiska politiken. En kvanti-

Norman

tativ analys av systemet för personlig inkomstbeskattning 1952-

71, 1974

8. Eskil Wadensjö Immigration och samhällsekonomi. Immigrationens ekonomiska

orsaker och effekter, 1973

9. Rögnvaldur Hannesson Economics of Fisheries. Some Problems of Efficiency, 1974

10. Charles Stuart Search and the Organization of Marketplaces, 1975

11. S Enone Metuge An Input-Output Study of the Structure and Resource Use in the

Cameroon Economy, 1976

12. Bengt Jönsson Cost-Benefit Analysis in Public Health and Medical Care, 1976

13. Agneta Kruse och

Ann-Charlotte Ståhlberg

Effekter av ATP - en samhällsekonomisk studie, 1977

14. Krister Hjalte Sjörestaureringens ekonomi, 1977

15. Lars-Gunnar Svensson Social Justice and Fair Distributions, 1977

16. Curt Wells Optimal Fiscal and Monetary Policy - Experiments with an

Econometric Model of Sweden, 1978

17. Karl Lidgren Dryckesförpackningar och miljöpolitik - En studie av styrmedel,

1978

18. Mats Lundahl Peasants and Poverty. A Study of Haiti, London, 1979

19. Inga Persson-Tanimura Studier kring arbetsmarknad och information, 1980

20. Bengt Turner Hyressättning på bostadsmarknaden - Från hyresreglering till

bruksvärdesprövning, Stockholm 1979

21. Ingemar Hansson Market Adjustment and Investment Determination. A Theoretical

Analysis of the Firm and the Industry, Stockholm 1981

22. Daniel Boda Ndlela Dualism in the Rhodesian Colonial Economy, 1981

23. Tom Alberts Agrarian Reform and Rural Poverty: A Case Study of Peru, 1981

24. Björn Lindgren Costs of Illness in Sweden 1964-75, 1981

25. Göte Hansson Social Clauses and International Trade. An Economic Analysis

of Labour Standards in Trade Policy, 1981


26. Noman Kanafani Oil and Development. A Case Study of Iraq, 1982

27. Jan Ekberg Inkomsteffekter av invandring, 1983

28. Stefan Hedlund Crisis in Soviet Agriculture?, 1983

29. Ann-Marie Pålsson Hushållen och kreditpolitiken. En studie av kreditrestriktioners

effekt på hushållens konsumtion, sparande och konsumtionsmönster,

1983

30. Lennart Petersson Svensk utrikeshandel, 1871-1980. En studie i den intraindustriella

handelns framväxt, 1984

31. Bengt Assarsson Inflation and Relative Prices in an Open Economy, 1984

32. Claudio Vedovato Politics, Foreign Trade and Economic Development in the

Dominican Republic, 1985

33. Knut Ödegaard Cash Crop versus Food Crop Production in Tanzania: An

Assessment of the Major Post-Colonial Trends, 1985

34. Vassilios Vlachos Temporära lönesubventioner. En studie av ett arbetsmarknadspolitiskt

medel, 1985

35. Stig Tegle Part-Time Employment. An Economic Analysis of Weekly

Working Hours in Sweden 1963-1982, 1985

36. Peter Stenkula Tre studier över resursanvändningen i högskolan, 1985

37. Carl Hampus Lyttkens Swedish Work Environment Policy. An Economic Analysis,

1985

38. Per-Olof Bjuggren A Transaction Cost Approach to Vertical Integration: The Case

of Swedish Pulp and Paper Industry, 1985

39. Jan Petersson Erik Lindahl och Stockholmsskolans dynamiska metod, 1987

40. Yves Bourdet International Integration, Market Structure and Prices. A Case

Study of the West-European Passenger Car Industry, 1987

41. Krister Andersson and Capital Taxation and Neutrality. A study of tax wedges with

Erik Norrman

special reference to Sweden, 1987

42. Tohmas Karlsson A Macroeconomic Disequilibrium Model. An Econometric

Study of the Swedish Business Sector 1970-84, 1987

43. Rosemary Vargas-Lundius Peasants in Distress. Poverty and Unemployment in the

Dominican Republic, 1989

44. Lena Ekelund Axelson Structural Changes in the Swedish Marketing of Vegetables, 1991

45. Elias Kazarian Finance and Economic Development: Islamic Banking in Egypt,

1991

46. Anders Danielson Public Sector Expansion and Economic Development. The

Sources and Consequences of Development Finance in Jamaica

1962-84, 1991

47. Johan Torstensson Factor Endowments, Product Differentiation, and International

Trade, 1992


48. Tarmo Haavisto Money and Economic Activity in Finland, 1866-1985, 1992

49. Ulf Grönkvist Economic Methodology. Patterns of Reasoning and the

Structure of Theories, 1992

50. Evelyne Hangali Maje Monetization, Financial Development and the Demand for

Money, 1992

51. Michael Bergman Essays on Economic Fluctuations, 1992

52. Flora Mndeme Musonda Development Strategy and Manufactured Exports in Tanzania,

1992

53. Håkan J. Holm Complexity in Economic Theory. An Automata Theoretical

Approach, 1993

54. Klas Fregert Wage Contracts, Policy Regimes and Business Cycles. A

Contractual History of Sweden 1908-90, 1994

55. Per Frennberg Essays on Stock Price Behaviour in Sweden, 1994

56. Lisbeth Hellvin Trade and Specialization in Asia, 1994

57. Sören Höjgård Long-term Unemployment in a Full Employment Economy,

1994

58. Karolina Ekholm Multinational Production and Trade in Technological

Knowledge, 1995

59. Fredrik Andersson Essays in the Economics of Asymmetric Information, 1995

60. Rikard Althin Essays on the Measurement of Producer Performance, 1995

61. Lars Nordén Empirical Studies of the Market Microstructure on the Swedish

Stock Exchange, 1996

62. Kristian Bolin An Economic Analysis of Marriage and Divorce, 1996

63. Fredrik Sjöholm R&D, International Spillovers and Productivity Growth, 1997

64. Hossein Asgharian Essays on Capital Structure, 1997

65. Hans Falck Aid and Economic Performance - The Case of Tanzania, 1997

66. Bengt Liljas The Demand for Health and the Contingent Valuation Method,

1997

67. Lars Pålsson Syll Utility Theory and Structural Analysis, 1997

68. Richard Henricsson Time Varying Parameters in Exchange Rate Models, 1997

69. Peter Hördahl Financial Volatility and Time-Varying Risk Premia, 1997

70. Lars Nilsson Essays on North-South Trade, 1997

71. Fredrik Berggren Essays on the Demand for Alcohol in Sweden - Review and

Applied Demand Studies, 1998

72. Henrik Braconier Essays on R&D, Technology and Growth, 1998

73. Jerker Lundbäck Essays on Trade, Growth and Exchange Rates, 1998

74. Dan Anderberg Essays on Pensions and Information, 1998


75. P. Göran T. Hägg An Institutional Analysis of Insurance Regulation – The Case of

Sweden, 1998

76. Hans-Peter Bermin Essays on Lookback and Barrier Options - A Malliavin Calculus

Approach, 1998

77. Kristian Nilsson Essays on Exchange Rates, Exports and Growth in Developing

Countries, 1998

78. Peter Jochumzen Essays on Econometric Theory, 1998

79. Lars Behrenz Essays on the Employment Service and Employers’ Recruitment

Behaviour, 1998

80. Paul Nystedt Economic Aspects of Ageing, 1998

81. Rasha M. Torstensson Empirical Studies in Trade, Integration and Growth, 1999

82. Mattias Ganslandt Games and Markets – Essays on Communication, Coordination

and Multi-Market Competition, 1999

83. Carl-Johan Belfrage Essays on Interest Groups and Trade Policy, 1999

84. Dan-Olof Rooth Refugee Immigrants in Sweden - Educational Investments and

Labour Market Integration, 1999

85. Karin Olofsdotter Market Structure and Integration: Essays on Trade,

Specialisation and Foreign Direct Investment, 1999

86. Katarina Steen Carlsson Equality of Access in Health Care, 1999

87. Peter Martinsson Stated preference methods and empirical analyses of equity in

health, 2000

88. Klas Bergenheim Essays on Pharmaceutical R&D, 2000

89. Hanna Norberg Empirical Essays on Regional Specialization and Trade in

Sweden, 2000

90. Åsa Hansson Limits of Tax Policy, 2000

91. Hans Byström Essays on Financial Markets, 2000

92. Henrik Amilon Essays on Financial Models, 2000

93. Mattias Lundbäck Asymmetric Information and The Production of Health, 2000

94. Jesper Hansson Macroeconometric Studies of Private Consumption,

Government Debt and Real Exchange Rates, 2001

95. Jonas Månsson Essays on: Application of Cross Sectional Efficiency Analysis,

2001

96. Mattias Persson Portfolio Selection and the Analysis of Risk and Time

Diversification, 2001

97. Pontus Hansson Economic Growth and Fiscal Policy, 2002

98. Joakim Gullstrand Splitting and Measuring Intra-Industry Trade, 2002


99. Birger Nilsson International Asset Pricing, Diversification and Links between

National Stock Markets, 2002

100. Andreas Graflund Financial Applications of Markov Chain Monte Carlo Methods,

2002

101. Therése Hindman

Persson

102. Göran Hjelm

Economic Analyses of Drinking Water and Sanitation in

Developing Countries, 2002

Macroeconomic Studies on Fiscal Policy and Real Exchange

Rates, 2002

103. Klas Rikner Sickness Insurance: Design and Behavior, 2002

104. Thomas Ericson Essays on the Acquisition of Skills in Teams, 2002

105. Thomas Elger Empirical Studies on the Demand for Monetary Services in the

UK, 2002

106. Helena Johansson International Competition, Productivity and Regional Spillovers,

2003

107. Fredrik Gallo Explorations in the New Economic Geography, 2003

108. Susanna Thede Essays on Endogenous Trade Policies, 2003

109. Fredrik CA Andersson Interest Groups and Government Policy, A Political Economy

Analysis, 2003

110. Petter Lundborg Risky Health Behaviour among Adolescents, 2003

111. Martin W Johansson Essays on Empirical Macroeconomics, 2003

112. Joakim Ekstrand Currency Markets - Equilibrium and Expectations, 2003

113. Ingemar Bengtsson Central bank power: a matter of coordination rather than money

supply, 2003

114. Lars Pira Staples, Institutions and Growth: Competitiveness of

Guatemalan Exports 1524-1945, 2003

115. Andreas Bergh Distributive Justice and the Welfare State, 2003

116. Staffan Waldo Efficiency in Education - A Multilevel Analysis, 2003

117. Mikael Stenkula Essays on Network Effects and Money, 2004

118. Catharina Hjortsberg Health care utilisation in a developing country -The case of

Zambia, 2004

119. Henrik Degrér Empirical Essays on Financial Economics, 2004

120. Mårten Wallette Temporary Jobs in Sweden: Incidence, Exit, and On-the-Job

Training, 2004

121. Tommy Andersson Essays on Nonlinear Pricing and Welfare, 2004

122. Kristian Sundström Moral Hazard and Insurance: Optimality, Risk and Preferences,

2004

123. Pär Torstensson Essays on Bargaining and Social Choice, 2004

124. Frederik Lundtofte Essays on Incomplete Information in Financial Markets, 2005


125. Kristian Jönsson

Essays on Fiscal Policy, Private Consumption and Non-

Stationary Panel Data, 2005

126. Henrik Andersson Willingness to Pay for a Reduction in Road Mortality Risk:

Evidence from Sweden, 2005

127. Björn Ekman Essays on International Health Economics: The Role of Health

Insurance in Health Care Financing in Low- and Middle-Income

Countries, 2005

128. Ulf G Erlandsson Markov Regime Switching in Economic Time Series, 2005

129. Joakim Westerlund Essays on Panel Cointegration, 2005

130. Lena Hiselius External costs of transports imposed on neighbours and fellow

road users, 2005

131. Ludvig Söderling Essays on African Growth, Productivity, and Trade, 2005

132. Åsa Eriksson Testing and Applying Cointegration Analysis in

Macroeconomics, 2005

133. Fredrik Hansen Explorations in Behavioral Economics: Realism, Ontology and

Experiments, 2006

134. Fadi Zaher Evaluating Asset-Pricing Models in International Financial

Markets, 2006

135. Christoffer Bengtsson

136. Alfredo Schclarek

Curutchet

Applications of Bayesian Econometrics to Financial Economics,

2006

Essays on Fiscal Policy, Public Debt and Financial

Development, 2006

137. Fredrik Wilhelmsson Trade, Competition and Productivity, 2006

138. Ola Jönsson Option Pricing and Bayesian Learning, 2007

139. Ola Larsson Essays on Risk in International Financial Markets, 2007

140. Anna Meyer Studies on the Swedish Parental Insurance, 2007

141. Martin Nordin Studies in Human Capital, Ability and Migration, 2007

142. Bolor Naranhuu Studies on Poverty in Mongolia, 2007

143. Margareta Ekbladh Essays on Sickness Insurance, Absence Certification and Social

Norms, 2007

144. Erik Wengström Communication in Games and Decision Making under Risk,

2007

145. Robin Rander Essays on Auctions, 2008

146. Ola Andersson Bargaining and Communication in Games, 2008

147. Marcus Larson Essays on Realized Volatility and Jumps, 2008

148. Per Hjertstrand Testing for Rationality, Separability and Efficiency, 2008

149. Fredrik NG Andersson Wavelet Analysis of Economic Time Series, 2008

150. Sonnie Karlsson Empirical studies of financial asset returns, 2009


151. Maria Persson From Trade Preferences to Trade Facilitation, 2009

152. Eric Rehn Social Insurance, Organization and Hospital Care, 2009

153. Peter Karpestam Economics of Migration, 2009

154. Marcus Nossman Essays on Stochastic Volatility, 2009

155. Erik Jonasson Labor Markets in Transformation: Case Studies of Latin

America, 2009

156. Karl Larsson Analytical Approximation of Contingent Claims, 2009

157. Therese Nilsson Inequality, Globalization and Health, 2009

More magazines by this user
Similar magazines