Growth-Poverty Linkage and Multidimensional Poverty: What do we Already Know?

2.1 Poverty, Inequality, and Growth in Asia

The relationship between income growth and poverty reduction has been shown to be particularly stronger in developing Asia. Analysis of data from 51 developing countries around the world led to the observation that growth of 1% in average income is associated with a 1.5% decline in the incidence of US$1-a-day poverty on average, with growth explaining 57% of the variation in changes in poverty (Asian Development Bank [ADB] 2004). Interestingly, when the sample is limited to countries in East, Southeast, and South Asia, each 1% income growth is associated with 2% decline in poverty incidence, with 65% explanatory power. In other words, the data suggest that growth has served the poor better in Asia than elsewhere in the developing world. Ferreira and Ravallion (2008), in reviewing the evidence on levels and recent trends in global poverty and income inequality, similarly pointed to the dominant role of Asia in accounting for the bulk of the world's poverty reduction since 1981.

This observation masks a wide variation in experience among Asian countries, however. Official data in the Philippines, for example, indicate a perverse growth-poverty reduction experience: poverty incidence actually rose by 3% from 2003 to 2006, a period when the economy was reported to have enjoyed historically high rates of growth (National Statistics Office 2006, 2008). In this case, the growth-poverty reduction elasticity is actually positive, where rising GDP is associated with rising poverty. The Philippine experience, while unusual in the region, is not necessarily unique. Data show that Mongolia and Sri Lanka had also experienced rising poverty incidence within the 2002–2008 period.³ These perverse trends in Mongolia, Philippines, and Sri Lanka are indicative of the wide range that actual experience in countries even from within the same region can span.

The economic development literature is already replete with studies that have examined the linkages among poverty, inequality, and growth (referred to in the literature with the acronym PIG; see Sumner 2003), and the poverty elasticity of growth (PEG). There is also a growing body of literature on multidimensional poverty and its measurement. This study draws from both threads of work as it seeks to enrich the PEG analysis in the context of developing Asia.

2.1.1 PIG and PEG: Past Assessments

Dollar and Kraay (2001), in two related pieces of work (Dollar and Kraay 2001 and 2001a) provoked wide debate on the supposed empirical relationship they found between income and poverty reduction. The debate was not so much on the linkage itself, but on the inferred reasons for the linkage. They found, based on data from 92 countries spanning four decades that average incomes of the poorest fifth of society rise proportionately with average incomes. They also found that several determinants of growth—such as good rule of law, openness to international trade, and developed financial markets—benefit the poorest fifth of society as much as everyone else. They further found little evidence of the effects of several factors commonly thought to disproportionately benefit (i.e., be “biased” for) the poorest in society. All this led to the conclusion declared unequivocally in their title, i.e., that growth is good for the poor.

Numerous other critics (most of whom tended to focus on their other work relating trade openness to growth; see for example Rodrik 2000, Amann et al. 2002) pointed out the lack of theoretical structure supporting the specification of the Dollar/Kraay equations. They also questioned the validity of using income of the lowest quintile as indicator of poverty, and cited the difficulties in drawing conclusions from large cross-section samples, with attendant problems of data quality. They found the strong correlation between average per capita income and income of the lowest quintile to be robust, but warned that (i) a similarly strong result is also found for the higher quintiles, and (ii) the significance of the other Dollar/Kraay regressors changes dramatically under different samples and equations. Thus, they argued that the policy prescriptions associated with the Dollar and Kraay regressions cannot be sustained.

Azis (2002, 2008) has been similarly critical of studies on growth and poverty that fail to explain the mechanisms of how the former affects the latter. Using a computable general equilibrium analysis, he focused on the poverty impacts of the 1997–1998 Asian financial crisis in Indonesia (Azis 2002). Among his observations derived from the analysis, he found that the impacts of prices on poverty were far more significant than the impacts of income changes during the crisis. He also undertook a combined supply-aggregate demand analysis on Indonesia and Thailand to examine the impact of macroeconomic policies on poverty, and found differing poverty responses to positive fiscal shocks between the two countries.

Ferreira and Ravallion (2008) examined the statistical relationships between growth, inequality, and poverty and the correlation between inequality and the growth elasticity of poverty reduction. From an extensive examination of international datasets, they observed: (i) the absence of a correlation between growth rates and changes in inequality among developing countries, (ii) the strong (positive) correlation between growth rates and rates of poverty reduction, and (iii) the importance of inequality to that relationship.

In an extensive review of the growth, inequality, and poverty linkages in Asia, Quibria (2002) derived a number of empirical regularities: (i) a robust association between sustained growth and poverty reduction; (ii) no robust correlation between inequality and aggregate growth; (iii) rapid capital accumulation was the most important proximate cause of the “East Asian miracle”; (iv) initial conditions varied (widely) among the miracle economies, and were thus not the crucial factors for the economic dynamism of the region; and (v) regardless of conditions of political freedom (i.e., whether autocratic or more democratic), provision of critical economic freedoms and a structure of market-supporting institutions were common to the miracle economies.

Bourguignon (2002a) pointed out that many empirical cross-country studies on the growthpoverty linkage are based on linear regression models that are ill-specified because they fail to recognize the identity that links the rate of economic growth, the speed of poverty reduction and changes in the distribution of income, as follows:

where H is the headcount poverty index, F is the cumulative distribution function, z is the poverty line, y is income per adult equivalent (ỹ being the mean income), and subscripts t and t' refer to two distinct points in time. It is an identity because it simply restates the definition of the change in poverty ΔH = H_t’–H_t = F_t’(z/y_t’)–F_t(z/y_t), wherein Ft(z/y_t’) was simply added and subtracted on the right hand side. In this form, the first term on the right hand side of expression (1) gives a proportional change in all incomes that leaves the distribution of relative income unchanged (the “growth effect”), and the second term is change in the distribution of relative incomes, which is independent of the mean (the “distributional effect”). The identity implies that income redistribution reduces poverty in two ways. First, a permanent redistribution of income reduces poverty instantaneously through the above “distribution effect.” But in addition, it also contributes to a permanent increase in the poverty elasticity of growth, and therefore to an acceleration of poverty reduction at any given rate of economic growth. This is to be distinguished from findings in the literature that growth tends to be faster where there is less inequality. Such findings would suggest that redistribution policy offers a ”double dividend” of accelerating both growth itself, and the speed at which such growth leads to poverty reduction.

2.1.2 Explaining Cross-Country Differences in Outcomes

Much has been written about the Asia-Pacific region's success in poverty reduction amid rapid economic growth. In Chaterjee's (2005) survey of the literature, two broad classes of factors were examined: those explaining the phenomenal increase in economic growth and its relation to poverty reduction, and policies directly aimed at fostering inclusiveness of the development process. The experience in East and Southeast Asia differs from that of South Asia, with the latter having reaped the poverty reduction dividend of growth somewhat later, and having experienced less employment growth than the former. Chaterjee observed that labor-absorbing growth, land reform, microfinance, control of inflation, and human capital investments are important elements in pursuing inclusive growth.

It has been argued that the sectoral composition of output and source of output growth in the economy has an important bearing on the inclusiveness of growth, i.e., the poverty reduction effect of such growth. Because rural poverty tends to dominate the poverty scene in most countries, it is widely presumed that growth in the agricultural sector is key to attaining poverty-reducing growth. Hasan and Quibria (2004) showed that the sectoral growth effects in the growth-poverty linkage vary considerably across regions of the developing world. They thus cautioned against misplaced “agricultural fundamentalism,” or the argument that economic growth biased for agriculture will promote poverty reduction most rapidly. In their findings, the strong poverty-reducing effect of agricultural growth vs. industry and services growth is true mainly for East Asia, whereas the opposite is true in South Asia, especially India, where manufacturing growth has historically had the strongest poverty-reducing effect. Thus, while the sectoral composition of growth would have an important influence on poverty reduction outcomes of economic growth, the sectoral growth driver that matters could vary across regions.

Public expenditures, particularly on health and education, are also widely expected to have a major bearing on human development, and therefore on poverty outcomes in growing economies. However, ADB (2006) indicated that the empirical evidence on this is mixed. Anand and Ravaillon (1993), Bidani and Ravaillon (1997), and Self and Grabowski (2003), among others, found public expenditures to be a significant determinant of health outcomes, especially for the poor, and particularly in low- to middle-income countries. Baldacci, Guin- Siu, and De Mello (2003) similarly established a strong link between public spending and education outcomes.

On the other hand, Carrin and Politi (1996) and Filmer and Pritchett (1999) found no significant impact of public spending on health outcomes. Landau (1986) and Al-Samarrai (2002) likewise ascertained the weak correlation between public spending and education outcomes. The latter suggested that levels of household spending, the effectiveness of the public expenditure management system, and the composition of public education spending are important factors explaining this weak link. Notwithstanding these mixed findings, ADB (2006) warned against dismissing the importance of public expenditures for poverty reduction, and explored methodological reasons why the link appears weak in some past analyses. It is pointed out that these results should not be taken to imply that resources are unnecessary, but that increasing resources alone is unlikely to be sufficient. The composition of resources and institutions that govern the use of these resources plays a key role in translating resources into better health and education outcomes.

To derive implications for aid policy, Agenor, Bayraktar, and El Aynaoui (2005) developed a macroeconomic framework to capture linkages between aid, public investment, growth, and poverty. Public investment is disaggregated into education, infrastructure, and health, and affects both aggregate supply and demand. In their application of the model to Ethiopia, they concluded that the required increase in foreign assistance could be sizable if the elasticity of poverty with respect to growth is small, despite the positive externalities generated by aid.

Nature or quality of governance is another factor expected to have an effect on the growthpoverty reduction relationship. Quibria (2006) examined the relationship between governance and economic growth using the World Bank governance indicators developed by Kaufman and Kraay (2002, 2008) and associates. He found a seemingly paradoxical result that for developing Asia, countries that exhibit deficits in their governance indicators register on average a much higher growth on a sustained basis compared to those that exhibit a surplus. He conjectured that either the link between governance and economic performance is not as strong or immediate as is widely presumed, or the Kaufman-Kraay composite governance index fails to capture the nuances of governance-growth interactions. Quibria's analysis was confined to the governance-growth relationship, and stopped short of examining the relationship with poverty reduction aspect or inclusiveness of growth.

A measurable indicator for empowerment, as an aspect of governance, was devised by Alsop (2005), who examined evidence of the relationship between empowerment and poverty outcomes from five country case studies. She concluded, however, that while a unifying analytic framework can work in different settings, there is a need for contextual sensitivity when attempting to measure empowerment. Thus, considerable caution is required in developing a measure of empowerment that would permit cross-country comparisons.

As a related concept, Bjornskov (2007) found that the political ideology of incumbent governments influences the link between growth and inequality. That is, under left-wing governments, inequality is negatively associated with growth, while the association is positive under right-wing governments.

Son and Kakwani (2008) analyzed pro-poor growth in 80 countries and proposed a new measure of pro-poor growth that captures gains and losses of growth rates due to changes in distribution of consumption (with gains implying pro-poor growth, losses anti-poor growth). They found regional location of countries to have a bearing on the degree of “pro-poorness” of growth. A low inflation rate was found to be a significant contributor to pro-poor growth, while the effects of share of agriculture to GDP, openness to trade, and rule of law were found to be insignificant.

In their examination of poverty and income across provinces in the People's Republic of China (PRC), Chambers, Wu, and Yao (2008) found an inverted U-shaped relationship (i.e., poverty rises with income at lower income levels, but the opposite is true at higher income levels). Prior growth performance was found to be a dominant factor influencing this relationship, with many traditional poverty explanatory variables found to have weaker explanatory power after taking account of prior growth.

2.2 Measuring Multidimensional Poverty

Discussions on inclusive growth have increasingly moved toward a broader definition of poverty to reflect its widely acknowledged multidimensionality. Beyond lack of income (exemplified in the common yardsticks of US$1 per day—recently updated by the World Bank to US$1.25 per day—and US$2 per day), non-income quality of life indicators such as health and education are rightfully receiving as much attention in more recent poverty analyses.

Sumner (2003), in his stocktaking of almost 50 years of literature on poverty, inequality, and growth spanning Lewis (1954) to Dollar and Kraay (2002), cited a number of studies that examined non-income poverty dimensions in relation to economic growth. Barro and Sala-IMartin (1995) and Pritchett and Summers (1995) argued that growth is good for the improvement of health, while Barro and Lee (1997) showed that growth is good for education. Thomas et al. (2000) argued that countries with average annual GDP per capita growth of over 2.3% have had faster poverty reduction by various measures. However, several authors have been more cautious. Easterly (1999) noted that in only 10 of 81 cases were quality of life indicators positively linked to economic growth. Foulkes (2003) grouped countries into five types by GDP per capita and found that the countries with the fastest growth were not the same as those with the fastest improvement in life expectancy. In fact, life expectancy improved at the same rate (an average of 0.5% a year) in both those countries with the fastest economic growth (an average of 5.0% a year) and those with the lowest (negative) growth rates (an average of -0.3% a year). The United Nations Development Programme (UNDP) (1996) argued that 1% of redistribution was seven times more successful in improving the infant mortality rate than 1% of growth.

The UNDP's Human Development Index (HDI) is probably the most widely recognized and used composite measure embodying other welfare dimensions (i.e., on health and education) apart from income. Starting in 1998, UNDP also began releasing estimates of a Human Poverty Index, a measure closely related to the HDI.⁴ The approach to measurement of the HPI is illustrated in Figure 1 [ PDF 54.1KB | 1 page ]. Since 1998, ADB has routinely reported HPI for its developing member countries in its annual publication Key Indicators of Developing Asian and Pacific Countries. Apart from HPI, there is a growing body of literature on the measurement of multidimensional poverty, which potentially provides possibilities for even richer alternative measures that capture more of the poverty dimensions as available data may permit.

The economic literature has advanced considerably in the formulation and application of multidimensional poverty measures (Bourguignon and Chakravarty 2002, 2003). Silber (2007) provides a comprehensive survey of the conceptual approaches to measuring poverty with a multidimensional perspective (see also Kakwani and Silber 2008).

A prior question concerns the choice of poverty dimensions to assess. Alkire (2008) suggested five possible bases for selecting the dimensions:

• Data availability or an authoritative convention
• Implicit or explicit assumptions made by the researcher
• “Public consensus” (e.g., list of Millennium Development Goals or MDGs)
• Deliberative participatory processes
• Empirical evidence concerning people's values

While the UNDP's HDI combines education (basic literacy, and later, school enrollment rates) and health (life expectancy) dimensions with the economic dimension (per capita GDP), Allardt (1993) took a slightly different perspective and identified three dimensions simply defined as Having, Loving, and Being. Ramos and Silber (2005) attempted to implement the Allardt approach using the British Household Panel Survey, further defining subdimensions and corresponding indicators for the three dimensions as follows:

• Having – reflected in economic resources, housing, employment, working conditions, health, and education
• Loving – reflected in degree of satisfaction with social life (family, friends, etc.)
• Being – reflected in self-determination, political empowerment, leisure activities, opportunities to enjoy nature, and work satisfaction

Asselin (2005) applied multiple correspondence analysis (MCA) to the measurement of multidimensional poverty involving two steps. A composite indicator is first constructed from multiple primary poverty indicators, before proceeding with computation of poverty indices with the composite indicator. MCA is resorted to after explaining the limitations of the more popular approach of Principal Components Analysis (PCA).

However, introducing multidimensionality gives rise to more practical challenges in defining who are poor and nonpoor. One approach is to define a poverty threshold for the various non-income dimensions, then aggregating the dimensions, and finally aggregating across individuals (Chakravarty, Mukherjee, and Ranade 1998; Bourguignon and Chakravarty 2003; Chakravarty, Deutsch, and Silber 2005). The other way is to reverse the latter two steps, i.e., first define a poverty threshold for the various non-income dimensions, then aggregate across individuals before aggregating the dimensions. This is the approach adopted by the Fuzzy Approach to multidimensional poverty assessment, applying the mathematical theory of fuzzy sets (Zadeh 1965) to address the difficulty of defining who belong or do not belong to the “poor” category upon aggregating the poverty dimensions. Betti and Lemmi (2006) surveyed various implementations applying such approach, including the work of Deutsch and Silber (2006) applying the method to Israeli Census data.

Apart from those already cited above, various other authors have attempted to devise measures of multidimensional poverty and apply them to actual data from different countries. Costa (2003) compared a unidimensional approach based on the traditional income yardstick with a multidimensional one that incorporates economic, social, demographic, and cultural factors using data from 12 European countries.⁵ She concludes that an incomebased evaluation of poverty misses substantial insights that may be gained from a multidimensional assessment. Similarly, Dekkers (2003) applied a multidimensional measure using the European Community Household Panel (ECHP) data and found, among other things, that poverty rates based on income alone often overstate poverty (i.e., are often higher than rates based on a multidimensional measure). However, he also finds that for certain groups, particularly single parents and those with precarious health situation, their poverty risk is underestimated by the unidimensional measure based on income alone.

Notwithstanding the growing body of literature that has emerged, implementing a multidimensional poverty measure remains inherently difficult and data intensive. While it has been possible to devise and implement more comprehensive multidimensional poverty indicators for individual countries or groups of countries where detailed survey data are available, the UNDP's HPI is by far the only multidimensional poverty indicator available for use for cross-country analysis.

One can thus determine the numerical relationship between economic growth and such a broader poverty measure, thereby enriching the usual PEG analyses. The practical value of employing a more holistic poverty measure in such analysis lies in its policy usefulness to governments and development partners. Among other things, such enriched analysis could provide more focused guidance for strategies, policies, public investments, and operational frameworks for interventions to achieve inclusive growth and poverty reduction.⁶ It is also interesting to examine and compare differences in results obtained from using the limited income/expenditure-based definition of poverty and those obtained from using a more holistic one. With the latter being considered as a more appropriate basis for assessing inclusiveness of growth, this comparison would give an indication of the inadequacy of the income/expenditure-based poverty yardstick in assessing inclusive growth and pursuing it effectively.

2.3 Sources of Data

The analysis undertaken for this study requires compilation of relevant data across Asian countries and across time, to permit examination of trends in economic growth and poverty reduction, and in factors that influence their relationship. Of particular interest were data and indicators that could permit formulation and/or use of a multidimensional measure of poverty.

The following data series were useful sources of cross-country time series data for use in the analyses:

1. Country Tables in the ADB annual publication Key Indicators of Developing Asian and Pacific Countries.
2. Statistical Appendices to the ADB's annual Asian Development Outlook.
3. Statistical Appendices to the UNDP's annual Human Development Reports
4. Statistical Appendices to the World Bank's annual World Development Reports

In addition, the World Bank's Worldwide Governance Indicators data set has been used in examining the effects of governance on the growth-poverty linkage. Because these international data sets take time to compile and process into a form suitable for crosscountry comparison, there is a time lag of 2–3 years in the reported data sets. Hence, the most recent year for income and poverty data has been 2006, while the country governance indicators are available up to 2007. Qualitative information on the Asian economies studied have also been obtained from various ADB publications and other relevant publications.

While it would have been ideal to construct an Asian panel data set that would include a combination of time series and cross-country data, the varying frequencies of observations for different relevant variables precluded doing so within the time constraints of this study. For example, while the World Bank governance indicators are estimated annually, the HPI is not and is updated at varying frequencies across countries. Still, a consistent panel data set that includes multiple (if not annual) observations per country could have permitted much fuller regression analyses than has been possible here, including in the estimation of the PEGs, rather than the arc elasticity approach undertaken here. Such fuller quantitative analysis is left for future work.

Download this Paper [ PDF 324KB| 59 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.