1
Thesis
Name: Fulvia Fraizzoli
Student number: 11124954
Specialisation: Economics
Field: Development economics (Macroeconomics)
Number of credits: 12 EC
Title: The effect of improved water sources on income inequality.
2 Statement of originality
This document is written by Student Fulvia Fraizzoli who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table of contents
Abstract 4
1. Introduction 4
2. Theoretical framework 6
2.1 Other factors affecting income inequality 9
3. Data section 11
3.1 Inequality (Gini_disp) 11
3.2 Improved water source (Improved) 13
3.3 Control variables 14
4. Methods section 16
4.1 The formulation of the final regression 17
4.2 Fixed or random effects? 17
5. Results 19
6. Conclusion 24
Reference list 26
4 Abstract
This paper studies the relationship between income inequality and improved water sources, using a cross-country panel data analysis over the time period 2000-2015. It aims to investigate whether improved water sources can be considered a health level proxy, as an alternative to the more commonly used mortality rate. This was done by altering Dabla-Norris et al.’s (2015) research. Results show that improved water sources have a significant negative effect on income inequality, meaning that a more widespread availability of drinking water could significantly reduce the level of income inequality.
1. Introduction
Water is an essential element to the existence of humankind. Yet, approximately 10% of the world’s population does not have access to improved water sources, and 30% to safely managed drinking water services (WHO & UNICEF, 2017). In these areas, diseases that in the developed world are considered minor, cause millions of deaths. For example, the Guardian reports that “499,000 children under five (and 1.3 million people of all ages) died of diarrhoea in 2015, making it the fourth leading cause of mortality among young children. Diarrhoea was responsible for 8.6% of all deaths among under-fives.” (Lyons, 2017). Some of these diseases are provoked by a poor quality of water and could largely decrease with easy and cheap interventions on basic sanitation, such as chlorinating or piping the water sources (Fewtrell et al., 2005). Currently, many NGOs operating in the developing countries are conducting field experiments on this issue. A good example is Gram Vikas which installs sanitation and water systems in Indian villages. The changes resulted in over 80% reduction in incidences of waterborne diseases (Banerjee & Duflo, 2011).
Improving the basic health conditions with easy and cheap solutions not only would largely reduce the mortality rate, but also, according to Bonds et al. (2010), would give a way to exit the
so-5
called health-based poverty trap. The health-based poverty trap is a vicious cycle: if a population’s health conditions are poor, families will also run into debt, thus becoming even poorer, and leading to even worse health conditions. There is therefore a strong link between health and poverty. Moreover, this kind of poverty leads to a persistent inequality: those who are poor will remain in the same condition and will not be able to exit this trap.
Indeed, in most of these same areas the inequality of income is an enduring problem. For example, UNDP (2017) reports that, in Africa, there are seven outlier countries with exceptionally high Gini coefficients. Moreover, 10 out of the 19 most unequal countries globally are in Sub-Saharan Africa, where poverty rates peak as high as 41%.
Previous research showed a correlation between health level and income inequality (Rodgers, 1979; Kawachi & Kennedy, 1999; Dabla-Norris, Kochhar, Suphaphiphat, Ricka & Tsounta, 2015), where health was measured with the mortality rate. However, the link between health level measured as availability of drinking water and income inequality has never been studied. Therefore, such analysis could contribute to the academic research, as well as provide useful insights for governments, policymakers, and NGOs.
Therefore, this paper will investigate the relationship between income inequality and availability of improved water sources. This will be done with a panel data analysis over the year-span 2000-2015 in order to discover the latest developments. First, in section 2 a theoretical framework will be outlined, with the most important findings on the topic up to now; then, in section 3 the data used for this research will be described; in section 4 the methodology used for this paper will be explained; section 5 will contain the results, and, finally, in section 6 there will be a conclusion.
6 2. Theoretical framework
Income inequality and its implications have been studied for long time. In 1912 Corrado Gini came up with a numerical measure for it: the so-called Gini coefficient, Gini ratio or Gini index, a frequency distribution of incomes that ranges from 0 to 1. It has been revisited multiple times to include different social welfare factors, such as education, and also to make the calculation easier.
In 1955 Simon Kuznets started studying the relationship between income inequality and economic growth and came up with a new model featuring the Gini index, the Kuznets curve. Kuznets argues that as a market develops (i.e. GDP per capita increases) inequality increases too, until some point where it starts to decrease. Therefore, according to this theory, income inequality is a physiological feature of the growth of a country, that will follow a certain pattern and thus solve by itself, as time passes by. This gives an insight to the possible differences in predictions for the income inequality according to the level of development.
However, this pattern does not always apply. Taking tropical Africa into consideration, many discording patterns can be observed: it is a developing area, but between 1980 and 2000, it experienced negative growth in income per capita, with wide ranges of income throughout countries (Sachs at al., 2004). Nevertheless, there is no evidence that those countries are governed in a worse way than elsewhere, once the income levels are controlled. So, where does this large diversity between incomes come from? Sachs et al. (2004, pp. 121-122) argue from a macroeconomic point of view that tropical Africa is stuck in a so-called poverty trap, which means that those countries are “too poor to achieve robust, high levels of economic growth and, in many places, too poor to grow at all”. Other studies analyse those poverty-traps from a microeconomic point of view to try to find reasons and “ladders” to exit the traps. For example, Barham et al. (1995) identify education as way out of it. They contend that an investment in education could not only affect the economic growth, but also the distribution of welfare among households. Bonds et al. (2010), instead, hypothesise how improving the health conditions could help the poorest parts of the population to exit the
poverty-7
trap. In fact, they argue from a microeconomic point of view that the vicious cycle of poverty inevitably also includes the health conditions. For example, a family where the father suffers of a disease will get into debt, thus starting a vicious cycle for which cures will increase the debts, so that they will not be able to pay them back and live a prosperous life, so that more family members will get sick, and so on in a cycle. The authors argue that, to break it, an intervention for health and sanitation is needed. Breaking the cycle would cause those shares of populations to possibly exit extreme poverty and therefore creating a more equal society. In this sense Bonds et al. (2010, p.1186) argue that diseases in developing countries are a “barrier to economic development”.
Health could not only have an effect on economic growth, as Bonds et al. (2010) state, but also on inequality. This was the main object of study for Rodgers (1979), who conducted a cross-country study to investigate the relationship between mortality rate and income inequality. He found consistent results that lower life expectancy, and thus higher mortality, are correlated with higher inequality. The reversed causality (i.e. inequality affects health level), instead was the focus of Kawachi and Kennedy (1999, p.215). Their hypothesis was that there is a higher level of health where the distribution of income is more uniform, health being measured with life expectancy as a proxy. This was based on previous studies that showed a correlation between the absolute value of income and health level and was extended to the relationship between a relative value for income (i.e. its distribution) and the health level. They found three mechanisms that explain the results, which in the US were found to support the hypothesis. They explained that income inequality is linked to a disinvestment in human capital, as well as to the erosion of social capital, and that income inequality directly leads to sickness because of stressful social comparisons. The outcome to these mechanisms is that poverty has a significant effect on variations in total mortality, so a higher mortality level could lead to a higher income inequality. Therefore, even if this study was based on a developed nation such as the US, it could be argued that the mechanisms and patterns described by Kawachi and
8
Kennedy (1999) could be applied to developing countries too, as Rodgers’ (1979) cross-country research pictures.
However, these studies leave some questions unanswered. For example: what are the causes for the diseases leading to higher mortality levels, and thus higher income inequality? What kind of cheap and effective interventions could be implemented to fix the problem?
Cutler and Miller (2005) investigated the dramatic decrease of mortality rate in the US between 1900 and 1946. They attribute the reduction by a half of the overall mortality and the decrease by three quarters of the infant mortality rate to the increased availability of clean water and to a better sanitation. In fact, during that period some clean water technologies, such as chlorination and filtration, were adopted. Additional proof was given by the meta-study of Fewtrell et al. in 2005: they analysed the effect of water, sanitation and hygiene interventions on the reduced diffusion of diarrhoea, while focussing on developing countries. The results were that water quality interventions (i.e. piping uncontaminated and chlorinated water) are actually the most effective, reducing incidences of the disease by as much as 95%. These interventions are indeed cheap and effective. Quick et al. (2002) conducted a field experiment in Zambia to improve the water sources. Piping water was too expensive, but chlorination was affordable (a bottle of chlorine costs about 800 kwachas, $0.18 USD and lasts a month) and effective (diarrhoea in young children dropped by up to 48%). This means that the quality and availability of water plays a role in determining the health level of a country. However, as suggested by Banerjee and Duflo (2011), the interventions that in theory work may not have such a positive and significant effect in reality. This is because at a microeconomic level, people not always behave rationally or are aware of the benefits of some new health measures, such as chlorinating or piping, as suggested by Fewtrell et al. (2005).
So far, the correlation between improved water sources and health condition has been researched extensively, and so has the one between health condition and income inequality, as well as the effect that one could have on the other. Therefore, it is interesting to ask how water could
9
directly affect income inequality, which as of today has not been the subject of specific investigation yet.
The nearest study was that of Dabla-Norris et al. (2015, p.24). With a wider scope, they ran a cross-country panel regression on the different factors affecting income inequality, to detect causes and consequences. Their study, conducted on a worldwide sample, used variables that can be classified as financial or economic (e. g. trade and financial globalization or domestic financial market development), or as demographic, such as years of education and mortality, depending on the country being an emerging one. The proxy for health was given by the female mortality aged 15-60 and turned out to be significant and positive, thus proving once again how health is a contributing factor in increasing income inequality.
Given the previous researches and outcomes, the present paper aims to find a possible direct relationship between availability of improved water sources and income inequality by modifying the study by Dabla-Norris et al. (2015). This will be done by adding one core variable: the percentage of population using improved water sources. The expectation is for this new variable to be significant and to be having a negative impact on income inequality: the more available improved water sources are, the better the general level of health, and therefore, the lower income inequality
2.1 Other factors affecting income inequality
Dabla-Norris et al. in 2015 use many variables to analyse the drivers of income inequality. For example, they use the measure of government spending as a share of GDP. This value is supposed to have a negative effect on upper-middle income shares, while having a positive effect on the lowest decile of income earners (Roine, Vlachos, & Waldenström, 2009). This would therefore milden the inequality, thus government spending would have a negative effect on income inequality. Another measure used by Dabla-Norris et al. (2015) was financial globalisation, measured as FDI assets and liabilities as a share of GDP. According to Freeman (2010 as cited in: Dabla-Norris et al., 2015), FDI and portfolio flows have been shown to have a positive effect on income inequality, both in emerging
10
and in advanced economies. As for labour market flexibility, Kahn (2012) argues that if it increases, there will be more room for new workers, possibly less skilled, in the labour market. As a result, a higher level of economic insecurity for those more skilled and for those who were already in the market, together with a higher level of income inequality would follow. Finally, the percentage of total population living in rural areas also has a positive impact on income inequality (Kuznets, 1955). Kuznets gives a thorough explanation based on his theory of relationship between GDP per capita and income inequality. According to him, this is a quadratic relationship. For this reason, both a simple and a quadratic term of GDP per capita will be analysed. The expectation is that the countries are on the right side of the parabola, so that an increase in GDP per capita would lead to a decrease in income inequality and that the quadratic term’s coefficient will be negative too, thus supporting Kuznets’ theory.
All in all, in the table below the expected effects of the variables included in the empirical regression which will be run in this paper, namely the one below, on income inequality are shown.
Inequalityit = β0 + β1 financialit + β2 labour flexibilityit + β3 female mortalityit + β4 govt spendingit + β5
GDP per capitait + α1 improved waterit + α2 rural populationit + εit.
Variable Expected sign
Improved water sources -
GDP per capita -
(GDP per capita)2 -
Financial globalisation +
Labour market flexibility +
Female mortality +
Government spending -
11 3. Data section
This study aims to improve the one of Dabla-Norris et al. (2015) by eliminating the variables found to be insignificant at a 5% significance level and adding new ones for a panel data analysis over the time span 2000-2015 for the developing countries. For this, several datasets were used and then merged to run the following regression:
Inequalityit = β0 + β1 financialit + β2 labour flexibilityit + β3 female mortalityit + β4 govt spendingit + β5
GDP per capitait + α1 improved waterit + α2 rural populationit + εit.1
In this section, first, information will be given about the data, background, rationale of using and manipulation done.
3.1 Inequality (Gini_disp)
The dependent variable, income inequality, was measured with the Gini coefficient2. This measure is calculated as a frequency distribution of incomes, where 0 indicates that all the values are the same (i.e. the distribution is uniform, thus everyone has the same income), and 1 expresses the maximal inequality (i.e. in a large sample one person has everything, and others have nothing).
For this paper, the data were retrieved from the SWIID (Standardised World Income Inequality Database). This is the dataset with the largest coverage of countries and years and maximises the comparability of data (Solt, 2016). Moreover, in the version 6.1 (i.e. the version used for this paper) the values are probability adjusted by drawing 100 series from the previous distribution, in order to overcome the problems of accuracy that could originate from the collection of data, as well as conversion by creating a confidence interval, as shown in the graph below. The more accurate the measures are, the thinner the confidence interval is (e.g. Italy in the graph below) and the opposite applies to countries whose data are less reliable (e.g. Gambia and Ethiopia below).
1 Where i stands for country variable and t for time.
2 To be noted is that there is some criticism concerning Gini coefficient as a measure for income inequality. However, since this is the measure used by Dabla-Norris et al. (2015), it was considered in this study as an acceptable measure.
12
First, the measure for Gini index of inequality in equivalized household disposable (post-tax, post-transfer) was chosen. Then, the data were manipulated as follows: the simple average per year and per country of the 100 values available was calculated. Finally, a time span between 2000 and 2015 was selected, because of larger availability of data on the focus of improved water sources. In a later stage the multiple imputation method was also implemented, in order to account for the uncertainty of the measures.
Over the time span, there is a trend of reduction of inequality on average, as it can be seen in the graph above and from the descriptive table below, where statistics on income inequality calculated for any type of country (Least Developed Countries or not) are grouped under “General”, and the ones focussing only on LDC are grouped under “LDC”.
GINI Year Average Minimum Maximum Observations
General 2000 0.39 0.226 0.61 131
General 2008 0.38 0.23 0.606 141
13
LDC 2000 0.40 0.29 0.50 27
LDC 2008 0.39 0.28 0.525 34
LDC 2014 0.389 0.31 0.51 11
3.2 Improved water source (Improved)
The main independent variable of interest, improved water source, refers to the percentage of the total population using an improved water source.
Before explaining why this measure was used, it is important to give some definition about the different types of water sources. WHO and UNICEF (2017) categorize water sources as non-improved or non-improved. The second ones can potentially deliver safe water (free from contamination)3,
given their supply structure (piped or non-piped). Water services can be classified as follows: • Limited: Improved water sources that require more than 30 minutes collection time; • Basic: Improved water sources that require less than 30 minutes collection time;
• Safely managed: Basic improved water sources that are accessible on premises and available when needed.
For this paper the whole of improved sources between 2000 and 2015 was used, while data on the percentage of rural population with access to improved water sources will be used for a robustness check. Both measures were divided by 100, in order for them to span from 0 to 1, for scaling reasons.
Data were retrieved from the WHO/UNICEF Joint Monitoring Programme (JMP) for Water Supply and Sanitationfor a time span 2000-2015. Over this time span the availability of improved
3To be noted is that data on improved water sources only measure the ready access to drinking water, but this does not
14
water sources has increased worldwide, as pictured in the descriptive table below, which displays statistics for LDC countries, as well as gives a general overview (General in the table).
IMPROVED Year Average Minimum Maximum Observations
General 2000 0.805 0.235 1 151 General 2008 0.845 0.309 1 152 General 2015 0.874 0.400 1 149 LDC 2000 0.55 0.235 0.827 37 LDC 2008 0.62 0.309 0.877 38 LDC 2015 0.69 0.479 0.916 37 3.3 Control variables
This study uses several control variables; some of them being significant at least at 5% in the original study of Dabla-Norris et al. (2015); and some of them being originally introduced by the author of this paper. In this section, however, all the control variables including the ones found not to be significant in the 2015 study are portrayed4.
• Trade: The sum of exports and imports of goods and services measured as a share of gross domestic product. This measure represents the trade globalisation of a country. Retrieved from World Bank national accounts data, and OECD National Accounts data files.
• Financial*: The sum of FDI assets and liabilities measured as a share of GDP and divided by 100 for scaling purposes. This measure represents the financial globalisation of a country over
4 The variables marked with * will be the ones included in the regression. The variables marked with ** will be the ones used for robustness checks.
If the variable is not marked, it means that the variable was not used in the final regression because found not to be significant (see section 4.1), most likely because of little or no significant variation considering the LDC sample or because they reduced the size of the sample largely.
15
time. Retrieved from the extended version of The External Wealth of Nations Mark II by the IMF.
• Credit: The domestic credit provided by the financial sector as a share of GDP. This measure represents the domestic financial market development. Retrieved from the International Monetary Fund, International Financial Statistics and data files, and World Bank and OECD GDP estimates.
• Skill Premium: Average years of education for the total population over 15. Retrieved from Barro- Lee Education Attainment data (2013).
• Labour Market Flexibility*: Index that measures labour market efficiency. Here, this means for the market to match workers with the most suitable jobs for their skillset and to incentivise the two parties of the market to increase the productivity of the human capital, which results in more efficient employees and in employers more capable to provide the right incentives. This index has been realised by the World Economic Forum in their Global Competitiveness Dataset, where the highest value is the highest efficiency.
• Education Gini: The Gini coefficient of average years of schooling for the population over 15. This measure represents the degree of inequality in years of schooling in a society, and it is similar to the Income Gini (it ranges from 0, perfect equality, to 100, perfect inequality). Retrieved from Demographic and Health Surveys (DHS).
• Female mortality*: The adult mortality rate for females in the age span 15-60 per 1000 female adults. This measure represented the health proxy in Dabla-Norris et al.’s study (2015). Retrieved from the United Nations Population Division.
• Government spending*: The general government final consumption expenditure as a percentage of GDP and divided by 100 for scaling purposes. Retrieved from the World Bank national accounts data, and OECD National Accounts data files.
16
• Emerging**: A dummy variable that takes the value of 1 if the country is part of the Least Developed Countries (LDC) and 0 otherwise. The values are announced by the United Nations every 3 years and over the time span they are assumed to remain constant. Retrieved from the United Nations.
• Water rural**: The percentage of rural population with access to improved water sources, then divided by 100 for scaling purposes. Retrieved from WHO/UNICEF Joint Monitoring Programme (JMP) for Water Supply and Sanitation.
• GDP per capita*: GDP per capita (PPP in current international $) divided by 1000 for scaling purposes. Retrieved from World Bank, International Comparison Program database.
• (GDP per capita)2: The measure above squared, to capture the pattern and dynamics described by Kuznets (1955).
• Rural population*: It refers to the amount of people living in rural areas as a percentage of the total population, divided by 100 for scaling purposes. Retrieved from World Bank staff estimates based on the United Nations Population Division’s World Urbanization Prospects. • Total population: The total amount of people living in a certain country. This measure was
used as a condition to select only countries that have a population larger than 1 million. This allowed to exclude very small countries, mainly islands, that had many missing observations and that often have peculiar dynamics, because of their size and location. Retrieved from United Nations Population Division, World Population Prospects.
4. Methods section
In this section it will be explained how the data were manipulated to get results. Moreover, explanations will be provided about how some analyses were carried out.
17 4.1 The formulation of the final regression
In order to formulate the final regression, I first tried to run a regression with all the control variables used by Dabla-Norris et al. (2015), as well as improved water, GDP per capita, GDP per capita squared, and rural population. However, this was not possible, because of a lack of observations, namely 6, and multicollinearity issues.
Therefore, the final regression, done on panel-data given by a cross-country sample of LDC countries with population larger than 1 million over the time-period 2000-2015, was formulated through a trial-and-error process. First, a simple regression with inequality (Gini_disp) as dependent and improved water (Improved) as independent variable was run to see whether a fixed or random effects analysis should be used (see section 4.2). Then, I tried to include one control variable at a time, to analyse the significance of each variable, while excluding possible interactions. The only control variables then included in a more comprehensive regression were the ones that gave significant results at a 5% at this preliminary stage. These were GDP per capita (GDPpc; p=.001), financial openness (Financial; p=.001), labour market flexibility (labour_flex; p<.001), female mortality (FemMortality; p<.001), government spending (GovtSpending; p=.012), and rural population (rural_pop; p=.002). The excluded ones, instead were trade (p=.939), credit (p=.643), skill premium (p=.796), GDP per capita squared (p=.204), and education Gini (p=.424). This led to the following preliminary regression:
Gini_dispit = β0 + β1 Improvedit + β2 GDPpcit + β3 Financialit + β4 labour_flexit + β5 FemMortalityit +
β6 GovtSpendingit + β7 rural_popit + εit. 4.2 Fixed or random effects?
The graph below shows the relationship between inequality and improved water in the year 2008 (halfway through the time span) for LDC countries with more than 1 million inhabitants. It is evident how diverse the situation is: some countries have very low inequality, but also low availability of water (e.g. Ethiopia and Afghanistan); some others have relatively high levels of health (i.e. high
18
improved sources), but also very high inequality (e.g. Lesotho). This diversity could be due to many unobservable characteristics. For example, it seems reasonable for a warzone like Afghanistan to have poor health conditions, together with low inequality, due to a lack of resources. The same applies to Ethiopia, where famine has been a persistent problem, especially in some areas which are even called “famine belt”, together with severe droughts (Dejene, 1990; van der Veen, 2000, 1984 as cited in: Little, 2008). This easily explains the lack of improved water sources. The fact that this is one of the poorest countries on Earth (the poverty rate was 30% in 2010 (World bank, 2017)) and the resulting lack of resources clarifies the low inequality as well. Instead, considering Lesotho, it probably has a higher embedded inequality, while the resources are allocated in different ways (e.g. in 2004 Lesotho spent 4.1% of its GDP in public health, a high percentage relatively to other Least Developed Countries, while it is somewhat low compared to more developed ones (Mackintosh, 2007)).
Due to these significant unobservable differences, a Hausman test was run on a simple regression of improved water sources on income inequality. Its results (p=.0001) backed up the choice to run a
Afghanistan Angola Bangladesh Benin Burkina Faso Burundi Cambodia
Central African Republic
ChadCongo, D.R. Ethiopia Gambia Guinea Guinea Bissau Haiti Lao Lesotho Liberia Madagascar Malawi Mali Mauritania Mozambique Nepal Niger Rwanda Senegal Sierra Leone Sudan Tanzania Togo Uganda Yemen Zambia .3 .3 5 .4 .4 5 .5 .5 5 G in i_ d is p 20 40 60 80 100 Value
19
regression with fixed effects. Therefore, the correct specification for the econometric model is the following:
Gini_dispit = β0 + β1 Improvedit + β2 GDPpcit + β3 Financialit + β4 labour_flexit + β5 FemMortalityit +
β6 GovtSpendingit + β7 rural_popit +µi + εit.
Here, µi represents a full set of LDC country dummies that allows to focus on within-country changes.
5. Results
The regression5 stated in section 4 (Methods) was run on LDC countries with more than 1 million inhabitants, using the average value of the Gini coefficient. In Table 1 are pictured the results of the analyses carried out using the average value: (1) is the original regression, (2) and (3) are robustness checks respectively run on an LDC sample and a general one, (4) and (5) are
additional analyses carried out in order to solve some minor issues, and (Exp) is the list of expected signs for the coefficients, based on the literature review (section 2).
5 Gini_dispit = β
0 + β1 Improvedit + β2 GDPpcit + β3 Financialit + β4 labour_flexit + β5 FemMortalityit + β6 GovtSpendingit
20 Table 1: Summary of results
(Exp) (1) (2) (3) (4) (5)
Original Robust 1 Robust 2 No LDC LDC
Gini_disp Gini_disp Gini_disp Gini_disp Gini_disp
Improved - -0.238*** -0.163*** (0.0293) (0.0261) rural_pop + 0.133 0.152* 0.0947** 0.184*** -0.0726 (0.0688) (0.0712) (0.0327) (0.0303) (0.0435) GDPpc +/- -0.00244 -0.00347 -0.000326* -0.000354* -0.00613* (0.00495) (0.00514) (0.000139) (0.000143) (0.00237) GovtSpen ding - -0.00827 -0.000232 -0.0327 -0.0176 0.0383 (0.0283) (0.0294) (0.0192) (0.0197) (0.0245) FemMort ality + -0.0000967*** -0.000105*** -0.0000128 0.0000266 0.0000423 (0.0000269) (0.0000283) (0.0000216) (0.0000213) (0.0000234) labour_fle x + 0.00961** 0.00833* 0.00237 0.00260 (0.00313) (0.00326) (0.00172) (0.00177) Financial + 0.322 0.314 0.00460 0.00273 -1.772*** (0.510) (0.531) (0.0123) (0.0127) (0.515) improved _rural - -0.218*** (0.0293) _cons 0.449*** 0.415*** 0.490*** 0.299*** 0.443*** (0.0635) (0.0647) (0.0335) (0.0144) (0.0322) N 109 109 754 754 379 Countries 24 24 116 116 36 R2 0.711 0.688 0.182 0.132 0.089
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
From column (1) in the table above, it can be seen the only coefficients significant at 5% were improved water sources (p<.001), female mortality (p=.001), and labour market flexibility (p=.003). How can these results be interpreted? First, it can be seen that, ceteris paribus, improved water sources have a significant negative impact on income inequality. This means that a more widespread availability of drinking water would be correlated with lower inequality within the
21
country. This is in line with the expectations: income inequality is indeed lower when the health level improves. Female mortality, as a second proxy for health, however, has an opposite result: from the regression, it appears that higher mortality rates are correlated with lower inequality. This is a surprising result, given the existing literature on the topic: not only Kawachi and Kennedy (1999), but also Rodgers (1979) claim that there is a consistent positive correlation between
mortality and inequality. As for the other control variables, all of them have the expected sign, thus confirming the hypotheses.
Given the fact that female mortality’s and improved water sources’ coefficients were both negative, some robustness checks were implemented (for the total results see Appendix D). For example, the same regression was run with percentage of rural population with access to improved water sources (improved_rural) substituting the share of total population using improved water sources (Improved) on LDC countries with a population larger than 1 million (Regression (2) Robust 1 in Table 1). The results were very similar to the ones of the original analysis, with the only difference being the significance of share of rural population. Similar results for what concerns the correlation between improved water sources and income inequality are obtained also when the initial regression is run worldwide for countries with more than 1 million inhabitants, without any distinctions between LDC and non-LDC countries (Regression (3) Robust 2 in Table 1). In this analysis, the signs of the coefficients are the same as in the original regression. However, GDP per capita is significant (p= .020), as well as rural population, while female mortality is not significant, as well as labour market flexibility (respectively, p=.555 and p=.168). Therefore, both improved water sources and female mortality keep their negative sign, thus the discordance, in all the robustness checks. Nevertheless, improved water sources seem to always be significant and in line with the previous literature, thus it could be considered a proxy for health level, thus causing a problem of multicollinearity if both proxies are included in the regression.
22
Indeed, correlation between improved water sources and female mortality is quite strong (ρ=-.6827). This correlation shows a plausible relationship, which is also in line with the literature found (see section 2): lower female mortality rates correspond to a more widespread availability of improved water sources, and vice versa. Even stronger evidence is shown if a regression is run with female mortality as dependent and improved water sources as independent variable. In fact, not only improved water sources were significantly negatively correlated with female mortality, but also its t-statistics were very high (t=-20.66; see Appendix E). In fact, if the improved water sources variable is omitted while regressing on both LDC and non-LDC countries, female mortality becomes positively correlated to income inequality, even if not significantly (p=.212) (Regression (4) No LDC in Table 1). Instead, in order to get a positive sign for female mortality in the original regression run on LDC countries with more than 1 million inhabitants, it is necessary to not only exclude improved water sources, but also labour market flexibility (Regression (5) LDC in Table 1). The most likely reason for excluding labour market flexibility is that this variable reduces significantly the size of the sample, as well as the amount of countries (i.e. respectively, from 379 observations without labour flexibility, to 109 including it, and from 36 to 24 countries available).
So far, the analyses were done on the average value of the Gini coefficient. However, if the SWIID database is used to its full potential with the multiple imputation method, the results are different. In Table 2 are the results of: (1) the original regression run with Multiple Imputation (MI from now on); (2) the same regression run on the whole population with MI, comparable to (3) in Table 1; (3) MI regression without Improved and Labour flexibility, comparable to (5) in Table 1.
23 Table 2: Results with multiple imputation
(Exp) (1) (2) (3)
MI Original MI no LDC MI no Impr
Gini_disp Gini_disp Gini_disp
Improved - -0.234 -0.16085* (0.15981) (0.072308) rural_pop + 0.13626 0.0952346 -0.074954 (0.38099) (0.091064) (0.130346) GDPpc +/- -0.0025 -0.000333 -0.006186 (0.02587) (0.0003724) (0.00656) GovtSpending - -0.009665 -0.0327696 0.038303 (0.129256) (0.052544) (0.059475) FemMortality + -0.000095 -0.000012 0.000043 (0.000162) (0.000068) (0.0000664) labour_flex + 0.009575 0.0023287 (0.014252) (0.004125) Financial + 0.3379314 0.0044562 -1.761568 (2.514104) (0.035704) (1.37055) _cons 0.4442378 0.487825*** 0.444839*** (0.3459691) (0.0903747) (0.0946) N 109 754 379 Countries 24 116 36
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
The first remarkable outcome is that most of the results turned out not to be significant any more. Probably this is caused by the fact that the uncertainty is taken into account, and that for LDC countries the confidence interval given by the SWIID is generally large (see the graph in section 3.1). Moreover, the problem of female mortality and improved water sources having the same sign is present. Again, this is solved by excluding improved water sources and labour market flexibility from the regression, but this causes financial globalisation to take a value higher than 1, which is not possible, given that the dependent variable ranges from 0 to 1.
24
All in all, results showed that if uncertainty is excluded improved water sources have a significant negative effect on income inequality, that is a more widespread availability of drinking water is correlated with lower inequality. This consistent result, together with not significant results for female mortality, could mean that the measure for improved water sources may be considered as a better proxy for health than mortality rate.
However, when the uncertainty is included in the calculations with the method of multiple imputation, results no longer are significant and take values that are not consistent with the model proposed. For this reason, further research on the topic, possibly with more accurate data, is suggested.
6. Conclusion
The present paper aimed to analyse the relationship between income inequality and improved water sources, as a proxy for the health level of a country. The unique contribution of this paper was to include a new proxy for health level, as an alternative to the more commonly used rate of mortality (here female mortality) or life expectancy, while slightly altering the study of Dabla-Norris et al. conducted in 2015. To do this, a cross-country panel data analysis over the time period 2000-2015 was carried out.
The study was mainly done on LDC countries with more than 1 million inhabitants, but robustness checks that led to the same results were done on both LDC and non-LDC countries.
The results showed that improved water sources have a significant negative effect on income inequality, thus a more widespread accessibility corresponds to a lower inequality in income. However, this outcome was conflicting with the one of female mortality, which, against all the literature on the topic, was negatively related to income inequality. For this reason, multicollinearity was hypothesised. Further analyses showed that there is indeed a strong negative correlation between
25
female mortality and accessibility to improved water sources, so that the latter can be used as a proxy for health. In fact, after some manipulation they both led to the same result: more accessible drinking water is linked to lower mortality rates, which is related with lower income inequality. The selection of improved water sources as health level proxy instead of mortality rate may cause some criticism, because only the quality of a resource would be taken into account, rather than the deaths per se. However, it would exclude the risk of biases caused by reasons for a certain mortality rate level, other than health, such as deaths caused by crimes.
Since improved water sources have a significant negative impact on inequality, and interventions on this issue would be cheap and easy, as showed by Quick et al. (2002), this study should be taken into account by governmental and non-governmental organisations for future policies. However, needless to say, more research is needed to further understand the relevance of these findings. For example, the microeconomic scope of this issue should be researched, as suggested by Banerjee and Duflo (2011), as well as reversed causality issues, which should be analysed with more advanced econometric models or by simply lagging the values. Such investigations would greatly improve the understanding of this issue and, hopefully, draw the attention of policymakers to this persistent problem.
26 Reference list
Banerjee, A. V., & Duflo, E. (2011). Poor economics: A radical rethinking of the way to fight
global poverty. Public Affairs.
Barham, V., Boadway, R., Marchand, M., & Pestieau, P. (1995). Education and the poverty trap. European Economic Review, 39(7), 1257-1275.
Barro, R., & Lee, J. (2013). A New Data Set of Educational Attainment in the World, 1950-2010." Journal of Development Economics, 104, pp.184-198.
Bonds, M. H., Keenan, D. C., Rohani, P., & Sachs, J. D. (2010). Poverty trap formed by the ecology of infectious diseases. Proceedings of the Royal Society of London B: Biological
Sciences, 277(1685), 1185-1192.
Cutler, D., & Miller, G. (2005). The role of public health improvements in health advances: the twentieth-century United States. Demography, 42(1), 1-22.
Dabla-Norris, M. E., Kochhar, M. K., Suphaphiphat, M. N., Ricka, M. F., & Tsounta, E.
(2015). Causes and consequences of income inequality: a global perspective. International Monetary Fund.
Demographic and Health Surveys (DHS) (n.d.). Education Gini [Data file]. Retrieved from https://data.worldbank.org/
Fewtrell, L., Kaufmann, R. B., Kay, D., Enanoria, W., Haller, L., & Colford, J. M. (2005). Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: a systematic review and meta-analysis. The Lancet infectious diseases, 5(1), 42-52.
International Monetary Fund, International Financial Statistics and data files, and World Bank and OECD GDP estimates (n.d.). Domestic credit provided by financial sector (% of GDP) [Data file]. Retrieved from https://data.worldbank.org/indicator/FS.AST.DOMS.GD.ZS
27
Kahn, L. M. (2012). Labor market policy: A comparative view on the costs and benefits of labor market flexibility. Journal of Policy Analysis and Management, 31(1), 94-110.
Kawachi, I., & Kennedy, B. P. (1999). Income inequality and health: pathways and mechanisms. Health services research, 34(1 Pt 2), 215.
Kuznets, S. (1955). Economic growth and income inequality. The American economic
review, 45(1), 1-28.
Lane, P. R., & Milesi-Ferretti, G. M. (2007). The external wealth of nations mark II: Revised and extended estimates of foreign assets and liabilities, 1970–2004. Journal of international
Economics, 73(2), 223-250.
Little, P. D. (2008). Food aid dependency in northeastern Ethiopia: myth or reality?. World
Development, 36(5), 860-874.
Lyons, K. (2017, June 2). Diarrhoea kills half a million children globally, shows Lancet study. The
Guardian. Retrieved from http://theguardian.com
Mackintosh, M. (2007). International migration and extreme health inequality: robust arguments and institutions for international redistribution in health care. The economics of health
equity, 151-173.
Quick, R. E., Kimura, A., Thevos, A., Tembo, M., Shamputa, I., Hutwagner, L., & Mintz, E. (2002). Diarrhea prevention through household-level water disinfection and safe storage in Zambia. The American Journal of Tropical Medicine and Hygiene, 66(5), 584-589.
Rodgers, G. B. (1979). Income and inequality as determinants of mortality: an international cross-section analysis. Population studies, 33(2), 343-351.
Roine, J., Vlachos, J., & Waldenström, D. (2009). The long-run determinants of inequality: What can we learn from top income data?. Journal of Public Economics, 93(7), 974-988.
28
Sachs, J., McArthur, J. W., Schmidt-Traub, G., Kruk, M., Bahadur, C., Faye et al. (2004). Ending Africa's poverty trap. Brookings papers on economic activity, 2004(1), 117-240.
Solt, F. (2016). The Standardized World Income Inequality Database. Social Science Quarterly
97(5):1267-1281. SWIID Version 6.1, October 2017. (2017-10-27).
United Nations Committee for Development Policy Secretariat (n.d.). Triennial review dataset 2000
– 2015 [Data file]. Retrieved from
https://www.un.org/development/desa/dpad/least-developed-country-category.html
United Nations Development Programme (2017). Income inequality trends in sub-Saharan Africa.
Divergence, determinants and consequences. New York, NY: Odusola, A., Cornia, G. A.,
Bhorat, H., & Conceicao, P.
United Nations Population Division (n.d.). Mortality rate, adult, female (per 1,000 female adults) [Data file]. Retrieved from https://data.worldbank.org/indicator/SP.DYN.AMRT.FE United Nations Population Division. World Population Prospects (n.d). Population, total [Data
file]. Retrieved from https://data.worldbank.org/indicator/SP.POP.TOTL
WHO/UNICEF Joint Monitoring Programme (JMP) for Water Supply, Sanitation and Hygiene (2017). Improved water sources (% of population with access) [Data file]. Retrieved from https://data.worldbank.org/indicator/SH.H2O.SAFE.ZS
WHO/UNICEF Joint Monitoring Programme (JMP) for Water Supply, Sanitation and Hygiene (2017). Improved water sources, rural (% of population with access) [Data file]. Retrieved from https://data.worldbank.org/indicator/SH.H2O.SAFE.RU.ZS
World Health Organization & United Nations Children’s Fund (2017). Progress on drinking water,
sanitation and hygiene:2017 update and SDG baselines. Geneva, Switzerland: WHO &
29
World Bank, Development Research Group (n.d.). Poverty headcount ratio at $1.90 a day (2011
PPP) (% of population) [Data file]. Retrieved from
https://data.worldbank.org/indicator/SI.POV.DDAY
World Bank, International Comparison Program database (n.d.). GDP per capita (PPP in current
international $) [Data file]. Retrieved from
https://data.worldbank.org/indicator/NY.GDP.PCAP.PP.CD
World Bank national accounts data, & OECD National Accounts data files (n.d.). Trade (% of
GDP) [Data file]. Retrieved from https://data.worldbank.org/indicator/NE.TRD.GNFS.ZS
World Bank national accounts data, & OECD National Accounts data files (n.d.). General
government final consumption expenditure (% of GDP) [Data file]. Retrieved from
https://data.worldbank.org/indicator/NE.CON.GOVT.ZS
World Bank staff estimates based on the United Nations Population Division’s World Urbanization Prospects (n.d.). Rural population (% of total population) [Data file]. Retrieved from https://data.worldbank.org/indicator/SP.RUR.TOTL.ZS
World Economic Forum Global Competitiveness Index (n.d.). Labor Market Efficiency [Data file]. Retrieved from https://tcdata360.worldbank.org/
30 APPENDIX
Appendix A – mini regressions that led to the main regression (summary of the variables kept, tables of the eliminated variables)
(1) (2) (3) (4) (5) (6)
Gini_disp Gini_disp Gini_disp Gini_disp Gini_disp Gini_disp
Improved -0.140*** -0.142*** -0.116*** -0.161*** -0.165*** -0.0976*** (0.0182) (0.0180) (0.0142) (0.0176) (0.0230) (0.0142) Control variables Rural_pop GDPpc GovtSpendi ng FemMortalit y Labour_flex Financial -0.110** 0.00658*** 0.0578* -0.0000948** * 0.0141*** -1.363*** (0.0350) (0.00189) (0.0229) (0.0000183) (0.00344) (0.403) _cons 0.557*** 0.471*** 0.458*** 0.521*** 0.443*** 0.459*** (0.0328) (0.00951) (0.00934) (0.0149) (0.0218) (0.00864) N 419 418 382 419 121 414 Countries 37 36 36 37 25 37 R2 0.144 0.149 0.172 0.180 0.433 0.159
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
32 Appendix B – Hausman test on Gini_disp Improved
Appendix C – Results of the original regression
34 Appendix E – Regression of Improved on female mortality
