What do we mean by pro-poor growth and how do we measure it?
Why are some growth processes more pro-poor than others?
This component aims to provide a deeper understanding of:
the role played by initial conditions (including initial inequalities) in acheiving pro-poor economic growth, and
what drives the distributional changes we see in survey data, including churning (whereby there are gainers and losers at each level of living).
The following subcomponents will address this question from various perspectives, focusing on areas where there appears to be high value-added from new research.Income Dynamics, Risk and Vulnerability
Social Exclusion and Poverty
Pro-Poor Growth and Inequality at Country Level
Aggregate inequality is a strong predictor of differences in the growth elasticity of poverty across countries. In an economy where inequality is persistently low, one can expect that the poor will tend to obtain a higher share of the gains from growth than in an economy in which inequality is high. To put this another way, an important determinant of the rate of poverty reduction is the distribution-corrected rate of growth in average income, given by a measure of initial equality (100 minus the measure of inequality) times the rate of growth. Indeed, the distribution-corrected growth rate knocks out the ordinary growth rate when both are used in a regression for the rate of poverty reduction.
To draw out the implications for policy from this finding, we need to develop a deeper understanding of the specific ways in which inequality matters to how much impact a given rate of growth has on poverty. Three questions will motivate the planned research under this heading:What specific aspects of “distribution” matter?
The Gini index is a well-known summary statistic, but it is only one of many possible measures and there can be no presumption that it aggregates relative welfare levels consistently with the way they impact on the poverty-reducing impact of growth. It would be better to start from a more flexible representation of the distribution and how it is changing. Aggregation should also be consistent with the purpose at hand, in this case assessing impacts on poverty.What specific dimensions of “inequality” are likely to matter?
Possibly it is not income inequality per se that matters to the rate of poverty reduction at a given rate of growth. Some inequalities (such as in human capital) may matter more than others to how much the poor share in growth.What economic and non-economic factors underlie the measured differences in distribution, as relevant to why growth is more pro-poor in some settings than others?
There are conflicting results in the literature, notably on the effects of greater openness on inequality. There are also signs of important interaction effects between the diverse initial conditions of countries and the impacts of economic policy reforms on inequality.
Subnational Determinants of Pro-Poor Growth
Geographic data bases within countries offer the prospect of a cleaner identification of the factors underlying the heterogeneity in impacts of economic growth on poverty than is possible using cross-country data sets. A “benchmark” paper on this topic has been completed for India. That paper used a long time series of household surveys (spanning 20 years) at state level to construct comparable consumption-poverty measures and collate these with aggregate data on the sectoral composition of economic growth. The paper then tested for inter-state differences in the impact of growth on poverty and (when such differences were found) explained them in terms of state-level factors, such as the composition of growth, literacy, urban-rural dualism and rural infrastructure. By allowing for explicit interaction effects between these factors and growth the paper was able to test hypotheses about what specific aspects of “inequality” mattered to pro-poor growth in India. A copy of this paper is provided in the library (see Ravallion and Datt, “Why Has Economic Growth Been More Pro-Poor in Some States of India than Others?”). The program will aim to develop further country case studies along similar lines, for which the most promising cases appear to be Brazil, China and Indonesia. The time periods will vary according to data availability, and this will also take account of sample sizes. For example, in the case of Indonesia it will be possible to construct finer (sub-provincial) geographic units if we focus on the SUSENAS surveys since the early 1990s that have used greatly expanded sample sizes over those for the 1980s.
Questions are also being asked about the role played in the evolution of the living standards by geographic factors, such as poor (physical and human) infrastructure and weak social organization. Differences across areas in these factors have been found to play an important role in the longer-term progress in poverty reduction across states of India, notably in terms of how much impact growth has on poverty. There are opportunities for looking at the same questions in other countries, including Brazil, China and Indonesia.
Such questions have been stimulated by both operational experience and economic theory. But to date there has been only limited empirical work able to shed light on such processes. There are a number of new (data and analytic) tools that can help. One is the use of household panel data collated with geographic data bases to better understand how location influences the evolution of living standards, allowing for latent (individual and geographic) heterogeneity. Recent research has shown promise in developing such models in China, and there is interest in extending the approach to other settings with appropriate data.
Income Dynamics, Risk and Vulnerability
A number of panel data studies have suggested considerable churning under the surface of the aggregate distributional statistics. Some of this is inter-temporal measurement error. But economic theory and a body of empirical evidence suggest that risk market failures entail considerable exposure to uninsured risks. It has also been argued (though the evidence is as yet scant) that exposure to uninsured risks can create poverty traps and mean that initial inequalities in various dimensions can impede overall rates of economic growth and poverty reduction. An important challenge remains to properly test the links in these theoretical arguments and flesh out their implications for development policy. Household-level panel data sets for a number of countries could be used to study the features of household income dynamics. This could embrace a number of the concerns coming from the regions to better understand the (distinct?) causes of “persistent” vs “transient” poverty.
A key question for which there has been very little empirical work concerns the possibility of nonlinearity in income dynamics. Nonlinear dynamics in household incomes can yield poverty traps and distribution-dependent growth. The potential implications for policy are dramatic; effective social protection from transient poverty will be an investment with lasting benefits, and pro-poor redistribution will promote aggregate economic growth. It is proposed to test for non-linearity in the dynamics of household expenditures and incomes using panel data for China, Hungary, Pakistan and Russia.
A specific question raised in recent policy discussions concerns the extent to which cross-sectional information on household and group characteristics can be used to ascertain the likelihood of a future adverse outcome in the event of an adverse future event, aggregate or idiosyncratic. The term ‘vulnerability’ is increasingly being used to connote such ex-ante exposure to uninsured risk.
We propose to study how well household-level vulnerability can be predicted ex ante. Current work on transient poverty requires panel data sets which are typically not available and can be quite costly to collect. There is considerable interest in finding ways to identify the characteristics of vulnerable households and groups using cross-section data (this was raised by the Bank’s Social Protection Board in a presentation of DECRG work on social protection). Given that income risks are endemic, and households are likely to be differentially buffered against such risks, an ex-ante measure of vulnerability is likely to be quite useful, by allowing policy makers to identify the characteristics of the most vulnerable and thus devise policy levers that are pro-active and forward looking, i.e., focus on poverty prevention.
However, research is needed to test how well it is possible to predict vulnerability ex ante. The simplest version of the stochastic permanent income hypothesis predicts that consumption should behave as a martingale, such that current household circumstances would have no ability to predict future changes in consumption. This would not suggest much hope for predicting “vulnerability” ex ante, though tests of the PIH in developing country settings have not been supportive. Several recent papers have proposed vulnerability measures. The papers share a common definition of vulnerability but offer rather different measurement strategies.
The proposed research will test the measures of vulnerability proposed in the existing literature, and develop new measures if necessary. Our empirical strategy will be to exploit existing panel datasets, which allow us to view actual transitions in the data against proposed vulnerability measures. In short, we will see to what extent revealed vulnerability to down-side risk can be predicted on the basis of currently available data.
Social Exclusion and Poverty
Social exclusion refers to social arrangements or structures within a society that systematically exclude disadvantaged groups from economic opportunities for reasons other than their potential productivity. Social exclusion appears to be an important but under-researched impediment to pro-poor growth. One often finds high and resistant poverty amongst certain social groups, identified by gender, ethnicity or culture. There have also been claims that these groups are often being left out of the growth process (even when pro-poor overall). Support for such claims has been found in some recent research pointing to the instrumental importance of inequalities defined on social or ethnic dimensions to aggregate welfare outcomes.
There is a potentially large set of questions about what should be done to address this type of poverty. We have chosen four specific topics under this heading, as potential “benchmark” studies that might help stimulate further research.
The first of the proposed studies starts from the observation that, while there have now been several studies documenting the loss in income from belonging to a socially excluded group from many parts of the world, there has been less work trying to understand the strategies that excluded groups use to counter the discrimination they face. To understand some of these strategies, we propose to examine qualitative and quantitative data from slums in Delhi India, where 60 per cent of the population comes from a schedule caste or tribe and is therefore classified as disadvantaged. How do the social networks of the disadvantaged differ from other groups? Do these groups access a different mix of occupations? Are they less likely to face discrimination in the anonymity of the city rather than within the traditional constraints of the village?
The second subcomponent stems from arguments that social exclusion is best remedied by affirmative action. Affirmative action has taken various forms – usually by providing preferred admissions and subsidies for education and employment. One innovative strategy in India has been to try to break traditional power structures in rural India by “reserving” seats for schedule castes and women in Panchayats – village government. The effectiveness of this strategy has never been properly evaluated. We propose to supplement a recent project that uses a natural experiment comparing border regions of Indian states to tease out the extent to which this affirmative action strategy has resulted in concrete improvement in increased decision making power, and public service delivery for disadvantaged groups.
The third subcomponent concerns gender discrimination. This is pervasive all over the developing world and has important implications for the distribution of resources within the family. Several studies have now documented that women who have better outside options tend to have children who are better nourished and educated. What is less understood, however, is how gender discrimination actually works. What are the social and cultural mechanisms that exclude women from decision making? How do these social and cultural structure interact with economic choices to affect distribution within the family? We will use survey data from rural India and Indonesia that asks questions on who makes decisions on a set of important issues within the household.
The fourth subcomponent will provide a case study in using mixed (quantitative and qualitative) methods to study poverty and social exclusion in Brazil. Recent studies have sought to identify the structural sources (returns to education, access to government pensions, etc) of Brazil’s inequality. However, several important questions remain. What role is played by social exclusion in perpetuating inequality in Brazil? Which precise groups are excluded from full participation in Brazilian society and economy, and how does this exclusion manifest itself in rates and levels of poverty? What are the mechanisms (social, legal, economic, political) by which these groups become, and remain, excluded? What strategies do such groups employ to cope with, and in some cases overcome, their situation? How do high rates of exclusion and inequality affect the economic growth process in poor communities and poor regions? To what extent are the poor and excluded groups differentially affected during periods of economic expansion or crisis? How might policy (new initiatives, or reforms to existing ones) help?
The policy research working papers below are drawn from the World Bank's institutional archives. Each link opens a page with an abstract of the document and several download options. Choose the 'light-weight documents' option for easy download.
You can also download other related documents. These include content-rich current outputs (updated document versions, miscellaneous documents, and web pages).