Project topics and materials
Get Complete Project Material File(s) Now! »
Data issues and Stylized facts
Inequality
When examining inequality, it is important to answer the question: inequality of what? There are many dimensions of inequality: inequality of outcomes – for instance inequality of individual earnings, family income, wealth, consumer spending, or individual economic well-being; and inequality of opportunities. As most studies, this dissertation focuses on inequality of income or consumption within countries, which remains the level at which most policies operate. Besides “within country” income inequality, it exists inequality between countries and global1 inequality which represents a combination of the two former types of inequality.
Measuring inequality is not easy or straightforward (World Bank, 2016). Cross-country comparisons can be challenging, particularly when it comes to study income inequality within countries. First, the main source of inequality measures – household surveys covers either consumption or income expenditure. National statistical offices in industrialized countries and Latin America tend to compile statistics on household income while national statistical offices in South Asia, Sub-Saharan Africa (SSA), and the Middle East collect data on consumption expenditure. The World Bank (2016) underlines the fact that data understate the level of inequality in countries where consumption data are used. Second, data from household surveys are likely to underreport incomes (or consumption) at the top of the distribution. Other measurement challenges of inequality include data availability and survey comparability, significant differences in sampling unit used, as well as the definition of income (net or gross) or the time coverage of expenditure and income data.
The following subsections present some stylized facts on different dimensions of inequality and on poverty around the world.
Inequality within countries
The Gini coefficient is the most widely used measure of inequality within countries. It is derived from the Lorenz curve, which displays the cumulative share of the population versus the cumulative proportion of income (or expenditure). The Gini coefficient considers the entirety of distribution rather than just the extremes. Besides the World Bank Gini coefficient, it exists a number of Gini coefficients from different sources. First, the Gini index from the Luxembourg Income Study (LIS) is based solely on income surveys and its goal is to reach the highest level of harmonization. In addition, it does not include imputed data. However, the LIS database covers relatively few countries and years, mainly high and middle-income countries. The Gini coefficients after taxes and transfers or post-fiscal (net) and pre-fiscal (market) from the Standardized World Income Inequality Database (SWIID) were assembled by Frederik Solt using the LIS data as baseline. The SWIID Gini coefficients have the advantage to (i) maximize the comparability of data on income inequality and (ii) maintain the largest possible coverage across countries and over time. However, it also generates values using model-based imputation for the missing observations.
While the Gini coefficient is over-sensitive to changes in the middle of the distribution, it is less sensitive to changes at the top and bottom of the income distribution. The World Bank PovCalnet database provides detailed information on income distributions. It reports mean income and income shares by decile and quintile from national household surveys. Data on income shares have the advantage to capture income distribution at the tails.
Figure 1.1 shows the evolution of income inequality and top and bottom income shares across the world by income level. Data suggest that since the 1980s income inequality as measured by the World Bank Gini coefficient has increased in most advanced and many emerging and low-income countries (Figure 1.1a). Data on the LIS Gini index cover mainly advanced economies and show the same trend (Figure 1.1b). Disposable income (post-tax and post-transfer) inequality displays a similar upward trend (Figure 1.1c). However, there is a wider disparity across countries, mostly due to different degrees of progressivity in (income) tax systems and spending policies (Woo et al., 2017, Bastagli et al., 2012 among others). Pre-fiscal income inequality exhibits an upward trend for most advanced economies, but changes in market Gini estimates for many developing countries strikingly differ (Figure 1.1d).
Data suggest that the income share of the top decile in most advanced economies have increased between the period 1985-1995 and 2000-2010, with the exception of Slovenia. In contrast, the bottom 10 percent income share has decreased in most advanced economies (Figure 1.1e and Figure 1.1f). In developing countries, changes in the income share of the top and bottom deciles vary across countries.
Inequality between countries
Remarkable economic transformation in parts of the world, especially in East Asia has led to global economic convergence. However, since the 1960s convergence with the United States has been reversed for the Sub-Saharan African region on average, despite substantial progress in some countries, especially this century (Figure 1.2). While Rwanda and Ethiopia have substantially improved their average incomes since 2000, even more than South Asia on average, the income gap with South Asia and East Asia has widened substantially for most Sub-Saharan African countries.
Assessing between-country inequality rests on properly measuring the wealth of Nations, which can be captured by Gross Domestic Product (GDP). However, GDP is subject to various issues since it is computed by different institutions, including the World Bank and the International Comparison Program (ICP). The ICP compiles detailed expenditure values of GDP for countries around the world. It also estimates purchasing power parities (PPP) exchange rate of countries around the world. The measurement uncertainty in purchasing power parity exchange rate is related to the fact that “relative prices are much more dissimilar between Tajikistan and the US, or China or India and the US, than between Canada and the US.” (Deaton, 2012). This raises questions regarding differences in PPP-adjusted GDP per capita between three reference databases: ICP, World Development Indicators (WDI) and Penn World Table (PWT).
Following the release of the ICP round 2011 by the World Bank (2014, 2015) and the PWT version 9 that incorporates estimates of PPP from ICP round 2011, this section computes differences in PPP-adjusted GDP per capita estimates between the three reference databases: International Comparison Program (ICP) round 2011, World Development Indicators (WDI) version 2016, and Penn World Table (PWT) version 9. This section sheds some light on the sources of these differences.
The release of the ICP round 2011 by the World Bank (2014, 2015) in April 2014 led to surprising results (Dykstra, Kenny, and Sandefur, 2014). According to Deaton and Aten (2017): “Until the publication of these results, the World Bank in its World Development Indicators provided extrapolated PPP exchange rates. For most countries, these were based on the 2005 round of the ICP, updated using relative inflation rates for each country. The new estimates for 2011 from ICP 2011 are quite different from these extrapolations. In particular, most poor countries of the world are estimated to be larger relative to the United States and other rich countries than was estimated from the extrapolations. This aspect of the results has attracted a good deal of attention, particularly the fact that the aggregate Chinese economy is much closer to the United States than previously estimated, and also that the Indian economy is now estimated to be larger than the Japanese economy. The new results also sharply reduce previous estimates of international inequality.” Deaton and Aten (2017) and Inklaar and Rao (2017) confirmed that the new measures of PPP prices of the ICP round 2011 are more accurate than the measures of the previous round 2005. They emphasize that regions were linked in ICP 2005. In both rounds of ICP, the methodology consists in collecting and comparing prices across countries within a region and then linking the regions (low-income, middle or high-income) to allow international price comparisons. This suggests that changes of prices of lower-income economies relative to higher-income countries are sensitive to changes of the linking approach.
Although impressive progress has been made in computing accurate PPP exchange rates from 2005 to 2011, this section shows that there is not much progress in harmonization among the three main data sources for the benchmark year 2011 with respect to the previous comparison done by Ram and Ural (2013) for the benchmark year 2005 using ICP round 2005, WDI and PWT. Yet, countries that present widely distinct estimates across ICP, WDI and PWT differ between 2011 (this dissertation) and 2005 (Ram and Ural 2013 paper).
Countries with significant differences in PPP-adjusted GDP per capita between ICP, PWT and WDI
Table 1.1 shows that between 9% (Columns 1 and 3) and 15% (Column 2) of countries have over 10% absolute differences in PPP-adjusted GDP per capita estimates between two databases. Between 24% and 40% of countries have over 5% absolute differences between two databases. Uzbekistan (UZB, 38%) and Comoros (COM, 57% and 52%) have the highest relative differences in PPP-adjusted GDP per capita estimates across the three databases.
While both versions of PWT 9.0 and WDI 2016 use the information of prices found in ICP 2011 as key input, their PPP-adjusted GDP per capita estimates markedly differ from those of ICP for several countries such as Mali. Some countries, for instance Iraq, have measures that differ markedly for each of the three databases. Conversely, other countries have very little differences in PPP-adjusted GDP per capita estimates between ICP and PWT and at the same time significant differences between ICP and WDI or vice versa. Conflict-affected countries, African countries, islands and oil-exporters are over-represented in countries showing large differences between the three data sources. On average, differences in PPP-adjusted GDP per capita estimates between WDI and PWT are smaller than differences between ICP and PWT, and between WDI and ICP. Figure 1.3a illustrates these differences and Figure 1.3b zooms in on some selected countries around the center of Figure 1.3a. These figures include only countries for which the three measures are available. The vertical axis represents (WDI-PWT)/PWT. Countries close to the vertical axis have minor differences in PPP GDP per capita estimates between ICP and PWT. This is the case of Mali (MLI). Countries such as Comoros (COM) and Nigeria (NGA) that are very far from the vertical axis have significant differences between ICP and PWT. For Comoros, estimates of PPP GDP per capita from WDI are 52% larger than PWT estimates. For Nigeria, estimates of PPP GDP per capita from WDI are 40% larger than PWT estimates.
Table 1.2 presents 20 countries for which estimates of PPP GDP per capita exhibit the largest differences. Comoros is the country having the largest difference in GDP per capita between ICP and PWT in percentage of PWT. For this country, GDP per capita estimates from ICP are 57% smaller than PWT relative to PWT while the percentage difference between WDI and PWT is only -5%. More than half of the 20 countries displaying the largest percentage differences in PPP GDP per capita between ICP and PWT are African and less developed countries. Yet, a few emerging countries such as Kuwait, Georgia and Ukraine also show significant disparities in PPP-adjusted GDP per capita between ICP and PWT.
The horizontal axis represents (ICP-PWT)/PWT. Countries close to the horizontal axis have small differences in PPP-adjusted GDP per capita estimates between WDI and PWT. This is the case of Comoros (COM) and Nigeria (NGA). For these two countries, PPP-adjusted GDP per capita estimates from WDI widely differ from ICP estimates while differences in PPP-adjusted GDP per capita estimates between PWT and ICP are very small. The first panel of Table 1.3 displays the top 10 countries for which PPP-adjusted GDP per capita estimates from WDI are larger than the ones from PWT. Out of these 10 countries, eight are African countries. For Mali, while WDI PPP-adjusted GDP per capita estimates are larger by about 25% than PWT estimates (relative to PWT), WDI estimates are about 24% larger than ICP estimates but the difference in PPP-adjusted GDP per capita between ICP and PWT is only 1%. This suggests that for this country, corrections made in PWT relative to ICP are minor compared to adjustments made in WDI relative to PWT. The second panel shows countries for which PPP-adjusted GDP per capita estimates from PWT are larger than the ones from ICP. In addition to African countries, some European and Central Asian countries such as Uzbekistan, Turkmenistan, Ukraine and Kyrgyz Republic show significant differences. For instance, PPP-adjusted GDP per capita estimates for Uzbekistan are 38% smaller in WDI than in PWT (in percentage of PWT). However, for countries such as Swaziland (0%), Nigeria (1%) and Liberia (1%) the gaps between WDI and PWT are negligible.
The deviation with respect to the line y=x represents y-x=(WDI-ICP)/PWT. Countries on the line y=x are such that estimates of PPP-adjusted GDP per capita in WDI and ICP are identical. This is the case of Tajikistan (TJK) in Figure 1.3b. WDI PPP-adjusted GDP per capita estimates differ from ICP estimates for countries such as Mali (close to the vertical axis) and Comoros (close to the horizontal axis) – that are far from the line y=x. Table 1.4 presents 20 countries for which estimates of PPP-adjusted GDP per capita in WDI and ICP show the largest differences. 12 African countries are listed in this table. The estimates of PPP-adjusted GDP per capita for Comoros, Nigeria and Tanzania are sizably larger in WDI than in ICP (relative to PWT). For countries such as Chad, Qatar, Bahrain, Nepal, Angola, Kuwait ICP estimates are larger (from 7% to 10%) than WDI estimates.
In Figure 1.3b, Iraq is far from the vertical and horizontal axes as well from the line y=x. This means that fort this country, the three measures of WDI, PWT and ICP are markedly different. Finally, several countries such as the United States, at the center of the graph have similar measures for the three databases WDI, PWT and ICP.
Understanding differences in PPP-adjusted GDP per capita between ICP, PWT and WDI
As GDP per capita is one of the most widely used measures of a country’s standard of living, understanding differences between these three sources is an important matter for researchers and policy makers. For example, Ciccone and Jarociński (2010) and Johnson et al. (2013) highlighted that the variability of vintages of PWT impacts on results of growth studies.
One of the causes of the problem may be weak statistical capacity in these countries to collect data (Devarajan, 2013). This dissertation finds correlation coefficients that are rather weak, below 0.1 in absolute value and are not significant. Differences in PPP-adjusted GDP per capita across databases do not seem to be related to the weak statistical capacity of source countries (Table 1.5).
Figure 1.4, Figure 1.5 and Figure 1.6 present correlations between the absolute differences in real GDP per capita across WDI, ICP and PWT (relative to PWT) and the statistical capacity indicator. These figures are zoomed in on countries for which the percentage differences in absolute value are below 20%. Correlations are not significant. However, one observes that some countries (Ukraine, Armenia) with higher levels of statistical capacity indicator report larger measurement errors in PPP-adjusted GDP per capita between WDI and PWT, and between ICP and PWT (relative to PWT) but a slighter difference between WDI and ICP (0.1% for Ukraine and 1% for Armenia).
Major differences in PPP-adjusted GDP estimates may be due to differences in nominal GDP and/or Purchasing Power Parity (PPP) estimates as different agencies use diverse sources of data and different methodologies. For instance, while the World Bank estimates GDP by collecting data from National Statistics Institutes, national accounts and the Organization for Economic Co-operation and Development – OECD’s national accounts data files, PWT relies mainly on data on GDP at current and constant prices, in local currency units from National Accounts. The main source of the National Accounts’ data is the United Nation Main Aggregates Databases. In addition, different countries rely on different definitions, methods and reporting standards in estimating GDP. Differences between the three databases in GDP at local currency can be a result of national accounts revisions. Table 1.6 reports absolute differences of nominal GDP estimates across the three databases respectively in percentage of PWT and ICP. For 3% (difference between ICP and PWT) to 75% (difference between WDI and PWT) of countries, percentage differences are higher than 50 percent.
Another cause of significant differences in PPP-adjusted GDP per capita between WDI version 2016, PWT 9.0 and ICP round 2011 may be the different aggregation methods used to compute the Parity Purchase Power (PPP). According to Feenstra, Inklaar and Timmer (2015), low-income countries with lower relative prices of nontraded goods will seem poorer if their expenditures are simply converted at the nominal exchange rate. For example, in 2011 nominal GDP per capita of Cambodia was 1.9% of that of the United States while its real (PPP-adjusted) GDP per capita was 5.9% of that the United States in 2011. To get an estimate of real GDP, PWT statistical agencies rely on an econometrically estimated expenditure function while ICP statistical agencies prefer index-number methods2.
Table 1.7 summarizes absolute differences of PPP estimates between ICP and PWT respectively in percentage of PWT and ICP for the year 2011. 22% of countries have a difference in absolute value of at least 10 percent while three to four percent of countries have more than 50 percent in terms of percentage differences of PPP estimates between ICP and PWT. Countries registering the largest percentage differences include Zambia, Cayman Islands and Kyrgyzstan.
Differences in PPP-adjusted GDP per capita may also be due to discrepancies across databases in population estimates. Devarajan (2013) states that in many countries GDP accounts use old methods and population censuses are out of date.
Since differences in PPP-adjusted GDP per capita are significant between the three sources especially for some countries, conclusions of research may vary according to which estimates and which sample of countries are used. Table 1.8 reports simple correlations of the World Bank control of corruption index with real (PPP-adjusted) GDP per capita across PWT 9.0, ICP round 2011 and WDI version 2016. Differences in correlations are more pronounced for countries with the largest percentage differences in real GDP per capita between the three sources and low and low-middle income countries than the average country (all the sample). For instance, results show that control of corruption is positively and significantly associated with development in countries with the largest differences in the three GDP estimates when using the ICP real GDP per capita. However, this result does not hold anymore when the real GDP estimates from WDI and PWT are used.
While undertaking international comparison studies, it is important for researchers and policy makers to exercise caution when forming critical recommendations as differences in data sources can impact results.
Inequality of opportunities
In addition to income inequality, most countries around the world face significant inequality of opportunities. Inequality in outcomes and inequality of opportunities are strongly associated (Lefranc, Pistolesi and Trannoy, 2008 among others). This subsection discusses the evolution of various aspects of inequality of opportunities, including access to education, health and financial services, in different parts of the world.
Health. There is a significant gap between the wealthiest and the poorest in terms of access to healthcare, particularly in developing economies. Figure 1.7 displays the coverage of reproductive, maternal, newborn, and child health interventions by income shares. It shows that there is a major difference in health coverage between rich and poor people, particularly in Sub-Saharan Africa. In Nigeria, 68.4 percent of the richest 20 percent have access to these health interventions as compared to only 13 percent for the poorest quintile of income distribution.
Education. There are large disparities in access to education between the richest and poorest 20 percent. As presented in Figure 1.8, the average years of education for individuals aged between 20 and 24 are much higher for the richest quintile than for the poorest quintile. The gap is larger in Nigeria where the average years of schooling for the 20-24 years old are 12 for the top quintile and below 2 for the bottom quintile. Such a gap is smaller in Kazakhstan.
Financial services. Access to finance is important for households, particularly low-income individuals. Households, especially poor ones may need credit to afford appropriate health services, adequate education and nutrition. The lack of adequate financial services can thus generate income inequality. There are some disparities in access to financial services across the income distribution. The share of adults in the top 60 percent of the income distribution with a bank account is higher as compared to the share for adults in the bottom 40 percent in the income distribution (Figure 1.9). At the global level, 66 percent of adults in the top 60 percent of income distribution have accounts at a financial institution as compared with 53.3 percent of adults in the bottom 40 percent of the income distribution. The gap is wider in Sub-Saharan Africa than in other regions.
Poverty
International comparisons of poverty data entail many issues. First, poverty measurements require household survey data for frequent measures of income or consumption and its distribution across households. However, household surveys are not conducted on each year. Therefore, one needs additional data to produce a reliable length of time for poverty data. Second, sampling weights, which ensure the survey is nationally representative are needed to produce poverty estimates. However, census data, which are required to produce sampling weights are often of low quality or outdated. Third, population data are also required to produce poverty rates and poverty counts. Typically, population censuses are conducted every 10 years (World Bank, 2015). In addition, to estimate the number of extreme poor in the world it is essential to rely on a poverty line that is comparable across countries. Purchasing Power Parity (PPP) index numbers from the International Comparison Program (ICP) are used to adjust differences in the cost of living across different countries. It is worth noting that poverty estimates tend to be sensitive to changes in the PPP data (World Bank, 2015).
Table 1.9 (upper panel) below presents the number of people living on less than $1.25 per day by region. One observes that across the world, the poverty headcount at $1.25 declined from 1990 to 2008, but it remains high essentially in Sub-Saharan Africa and South Asia. Table 1.9 (lower panel) shows the evolution of poverty worldwide and by region. Extreme poverty declined worldwide over the last two decades particularly in East Asia and Pacific where poverty rates dropped from more than 60 percent in 1990 to less than 4 percent in 2013. The other regions except Europe and Central Asia also experienced reductions in poverty but to a lesser extent. Even with these recorded progress, the number of extreme poor was estimated at 768.51 million in 2013. More than a half of extreme poor lived in Sub-Saharan Africa in 2013.
Outline and main results
This dissertation provides some evidence-based policy lessons in alleviating poverty and attaining pro-poor and inclusive growth. It comprises three main chapters, in addition to the overview chapter. The second chapter examines the role of institutions in promoting pro-poor and inclusive growth at the macroeconomic level. Using a novel empirical model, it assesses the nonlinear relationship between good governance, pro-poor and inclusive growth. The findings show that growth has been pro-poor – that is, it has reduced poverty. However, growth has been not inclusive, as it did not significantly impact the growth of the income share held by the poorest 20 percent. All features of good governance are pro-poor but only government effectiveness and rule of law promote inclusive growth. While the impact of growth on the income of the poor is nonlinear and increases with the control of corruption, the relationship between government effectiveness and inclusive growth is linear. Education spending, infrastructure improvement, and financial development are key factors for promoting poverty reduction and inclusive growth.
Can a government reduce income inequality by changing the composition of public spending while keeping the total level fixed? This question is of particular relevance given the everlasting financing constraint most governments face, either because of already high public spending, elevated public debt, limited domestic resources, or the combination of the three. The third chapter examines the effects of public spending reallocation on income inequality. Using a newly assembled data on disaggregated public spending for 83 countries across all income groups, it shows that reallocating spending towards social protection and infrastructure is associated with reduced income inequality, particularly when it is financed through cuts in defense spending. However, the political and security situation matters. We do not find evidence that lowering defense spending to finance infrastructure and social outlays improves income distribution in countries with weak institutions and at higher risk of conflict. Reallocating social protection and infrastructure spending towards other type of spending tend to increase income inequality. Accounting for the long-term impact of health spending, and particularly education spending, help to better capture their equalizing effects.
The fourth chapter uses 1-2-3 survey data on the Democratic Republic of Congo to analyze heterogeneity in the informal sector. It empirically identifies three types of entrepreneurs in the sector. The first group of entrepreneurs—top performers—is growth oriented and enjoys greater access to capital. The second group—constrained gazelles—includes entrepreneurs who share many characteristics, especially management skills, with the top performers, but operate with less capital. The third group—survivalists—comprises firms struggling to grow. Based on logit and fixed effect ordinary least squares models, the results presented in this chapter show that poverty and income inequality are more common among constrained gazelles and survivalists. The chapter also shows that income inequality is explained mainly by educational disparities and lack of credit access among entrepreneurs. Additionally, the outcomes of a Blinder-Oaxaca decomposition show that the performance of firms is a key factor in explaining differences in income. Examining the drivers of performance, the chapter finds that human capital and managerial skills are important engines of performance.
Table of contents :
General Introduction and Overview
1.1. Context
1.2. Data issues and Stylized facts
1.2.1. Inequality
1.2.2. Poverty
1.3. Outline and main results
The Quest for Pro-poor and Inclusive Growth: The Role of Governance
2.1. Introduction
2.2. Literature Review
2.2.1. Growth, Poverty, and Income Distribution
2.2.2. Governance and Pro-poor Growth
2.3. Econometric Methodology
2.4. Data
2.4.1. Measuring Poverty and Inequality
2.4.2. Defining and Measuring Governance
2.4.3. Main explanatory variables
2.5. Pro-poor and Inclusive Growth: Empirical Evidence
2.5.1. Has growth been pro-poor and inclusive?
2.5.2. Pro-poor and inclusive growth: the role of governance
2.5.3. Other determinants of pro-poor and inclusive growth
2.6. Nonlinear and threshold estimations
2.6.1. Exogenous nonlinear estimation
2.6.2. Endogenous nonlinear estimation: Panel Smooth Transition Regression
2.7. Conclusion and discussion
Appendix A
Reallocating Public Spending to Reduce Income Inequality: Can it work?
3.1. Introduction
3.2. Data
3.2.1. Composition of public spending: A new dataset
3.2.2. Some Stylized Facts on Public Spending and Income Distribution
3.3. Econometric Analysis: Composition of Public Spending and Income Inequality
3.3.1. Estimated Model
3.3.2. Baseline Results
3.3.3. The Role of Conflict and Institutions
3.4. Further Robustness Checks
3.4.1. Long-Run Impact of Public Spending
3.4.2. Alternative indicators of inequality
3.4.3. Accounting for local government spending, the efficiency of public spending, and the use of debt to finance public outlays
3.5. Conclusion
Appendix B
Informal Sector Heterogeneity and Income Inequality: Evidence from the Democratic Republic of Congo
4.1. Introduction
4.2. Informal sector heterogeneity and inequality: Literature review
4.3. Data and descriptive statistics
4.3.1. 1-2-3 survey
4.3.2. Characteristics of the informal sector in DRC
4.4. Identification strategy: Informal firms
4.4.1. Defining a top performer
4.4.2. Sample selection bias
4.4.3. Identification of the constrained gazelles and survivalists
4.5. Heterogeneity in the informal sector
4.5.1. Individual entrepreneur characteristics
4.5.2. Firm typology and the choice of sector
4.5.3. Structural and behavioral factors
4.6. Urban poverty and income inequality in the informal sector
4.7. Drivers of the performance of informal firms
4.7.1. Explaining differences in income using Blinder-Oaxaca decomposition
4.7.2. Drivers of informal firms’ performance
4.7.3. Explaining differences in performance using Blinder-Oaxaca decomposition
4.8. Conclusion and Policy Recommendations
Appendix C
General Conclusion
Résumé général
References
