Public Disclosure Authorized Pensions & Social Insurance Social Protection & Labor Policy Note June 2015 | Number 18 The note is a joint product of the Social Protection and Labor & Poverty and Equity Global Practices. Who is Poorer? Poverty by Age Public Disclosure Authorized Highlight Survey data from 62 developing countries in the Developing World is used that represents more than half the world’s Brooks Evans and Robert Palacios population. Poverty head count by age varies by the unit of analysis. Ranking Abstract1,2 by individuals, we find that children are the poorest This note shows that based on headcount poverty rates, at the household level, Public Disclosure Authorized households with elderly members are roughly equally poor to non-elderly households, group in all but one though with variation when using more detailed compositions, and the elderly are country, while households less poor than children in 98 percent of the countries sampled when comparing with and without elderly different age groups. Further, as a share of the poor, elderly average only 10 percent, are equally poor. children 36 percent, and adults 54 percent. Moderate equivalence adjustments result in a four percentage point change in the number of countries with children better off than elderly. (A separate note provides detailed sensitivity analysis.)3 These results can be seen as a starting point for further analysis that would look at the reasons behind differences between countries as well as age sub-groups within countries. The findings provide preliminary evidence that households with elderly, and especially elderly individuals are often not the most poor. The analysis is relevant to countries considering new policies or reforms of existing programs targeting particular age groups, such as the elderly or children. Public Disclosure Authorized I. Background Whether the elderly are poor relative to children and adults in society has important implications for social and economic policy. Large components of social protection systems can be traced to society’s concern about poverty among the elderly and children, often viewed as the most at-risk groups. Both contributory and non-contributory pension programs emerged in the wake of urbanization and industrialization as traditional patterns of old-age income support became less relevant for workers that depended 1 Citation guideline: Evans, B. and Palacios, P. 2015. “Who is Poorer? Poverty by Age in the Developing World”. World Bank. Washington D.C. JEL Codes: B410, D040, D600, H550, O150, O200 2 The note was written by Brooks Evans (bevans2@worldbank.org) and Robert Palacios (rpalacios@ worldbank.org). Peer review comments provided by Margaret Grosh, Phillippe Leite, and Kinnon Scott. Additional comments were provided during a May 5, 2015 Social Protection & Labor and Poverty co- sponsored discussion on “Who is Poorer? Age-based Poverty Patterns and Measurement: Findings & Implications for Bank Operations & Clients”. 3 See Evans, B. 2016. “Equivalence Sensitivity of Age-Based Poverty Measures”. World Bank. Washington D.C. 1 Policy Note: Pensions & Social Insurance June 2015 | Number 18 on wages. In the high income countries, this has resulted in differences between the instruments. These are discussed large transfer systems and generally low poverty rates for the in a separate methodological note.⁹ Age groupings are also elderly relative to the population. This has led some OECD harmonized to allow for comparability across countries. countries to question transfers from relatively poor households ‘Child’ is defined as age 0-14, ‘Adult’ 15-59, and ‘Elderly’ 60+. with adults often with children, to relatively well-off elderly households via payroll taxes. Welfare aggregate: While recognizing the tradeoffs inherent to welfare measurement, monetary based welfare per capita In most low and middle income countries,4 pension systems is used as a baseline, as this is among the most commonly have failed to achieve high coverage, and the majority of the used approaches.10 All regions in the sample use expenditure elderly continue to depend on traditional income support except for Latin America and the Caribbean, where only sources, especially family.⁵ Along with children and the income data is available. Welfare per capita is calculated by disabled, the elderly are often considered to be vulnerable taking a household’s consumption expenditure or income11 to poverty. The ILO recommends that its “Minimum Floor” and dividing by the number of household members.12 be applied specifically to the elderly as well as these other The formula, if consumption is used is: vulnerable groups. Per capita expenditure = household consumption expenditure houshehold size Previous evidence as to whether the elderly are poorer than other age groups generally shows this is often not the case, This calculation implicitly assumes 1) that there is no benefit with children most frequently the poorest. In his review of from living in in larger households, such as sharing housing, studies available at the time, Whitehouse (2000), found that cooking supplies or transportation costs (i.e. economies of “most studies show that the old are represented proportionally scale do not exist); and 2) all household members have equal or under-represented among the poor.” His analysis focused income or consumption needs and access, and so a five- mainly on middle and higher income countries, but similar year-old consumes the same as an adult in labor intensive findings have been presented for low income countries such as employment (i.e. no economies of composition exist.)13 India and many countries in Sub-Saharan Africa.⁶ Although these assumptions can and have been contested, This note provides comparative poverty rankings, mainly there is no agreement on the best approach. This note cross- using per capita welfare, of elderly and non-elderly references a detailed analysis of the sensitivity to different households and individuals in 62 developing countries. assumptions about composition and scale economies and Recognizing the shortcomings of per capita welfare, a presents a feasible adjustment derived from that analysis companion note investigates how the two most common (0.75 economies of scale and composition). parametric equivalence scale adjustments – economies of scale and composition, affect the distribution and measures of Poverty lines: Relative poverty rates are used here as this note poverty by age.7 focuses on comparing poverty rates of different age groups and household compositions in the sample countries. An absolute poverty line is not used, as poverty rates would be II. Methodology close to zero percent in higher income countries, such as in ECA and some in LAC, while there would be higher poverty The data are from the World Bank’s ASPIRE, ECAPOV, and rates in low income countries.14 The choice of any poverty line SEDLAC databases, which are based on household survey will affect the percentage of people that are poor. data.⁸ While over 70 surveys were available, only 62 are used since only these contained basic demographic individual Two measures of relative poverty are used: the share of information needed for age-based poverty analysis. The households and individuals that (i) fall within the poorest two surveys have been harmonized in terms of coding and to quintiles of the distribution and (ii) those that fall below half the extent possible, variable definitions. There are some of median per capita expenditure/income. The latter poverty line will tend to capture more of the distributional dynamics in a country. Poverty lines are generated both at the individual 4 The developing country regions included are Sub-Saharan Africa (AFR), and household level. East Asia and the Pacific (EAP), Eastern Europe and Central Asia (ECA), Latin America and the Caribbean (LAC), Middle East and Northern Africa (MNA), and South Asia (SAR). 5 See “Evans. B and Palacios, R. 2015. “An Examination of Elderly Co- 10 Sen (1985) and others define poverty based on concepts of capabilities and residence in the Developing World”. World Bank. Washington D.C.” for the functioning. extent of family support. 11 Consumption expenditure is generally seen as a more robust measure of 6 See Deaton (1997) and Pal and Palacios (2011) for India and Kakwani welfare (see for example Deaton 1997). Here, the term per capita welfare refers and Subbarao (2005) for Africa. to consumption unless only income is available. 7 The use of per capita welfare with its inherent assumptions, has clear 12 These surveys do not provide data that would allow us to look at intra- limitations as discussed in Deaton and Paxson (1997) and many others. household distribution of resources. 8 See www.worldbank.org/aspire, http://data.worldbank.org/data-catalog/ 13 Adult equivalent adjustment are often used, for example, children and/or sedlac, and http://povertydata.worldbank.org/poverty/region/ECA. elderly can be assumed to consume a fraction of adults, such as 50 percent, 9 See Evans, B. 2016. “Using Household Surveys for Age-based Poverty, and would each be 0.5 adult equivalents, which decreases the effective Pensions, and Social Assistance Analysis”. World Bank. Washington D.C., household size. which can be downloaded at www.worldbank.org/pensions. 14 Commonly used absolute poverty lines are $1.25 and $2.50 USD a day. 2 Social Protection & Labor | World Bank Group Poverty measure: Foster-Greer-Thorbecke (FGT) measures and 34 percent), while AFR is lowest (23 percent). Only five are used for calculating the poverty headcount (FGT0), gap percent of households have only elderly members. These (FGT1), and severity (FGT2). This measurement approach figures are highest in ECA at 10 percent, while averaging one is commonly used, and allows for sub-group decomposition, to three percent in the other regions. Households with adult such as analysis by differing household types or age groups. members and children are the largest group in all regions The note focuses primarily on poverty headcount. (averaging 53 percent of households in sample countries). Household-level poverty III. Results Poverty rates among elderly households. Figure 1a and 1b show per capita poverty rates for households with elderly The following section presents the findings of the population based on the two poverty lines. Each bar represents a country. shares of different household types, and poverty rankings of Figure 1a refers to the share of households with elderly households. This is followed by individual-level analysis. with less than half the median welfare per capita. Figure 1b refers to the share of households with elderly that can Household types be found in the bottom two quintiles of the distribution. There is significant cross-country variation: the range for the Table 1 shows the composition of households in the sample 50 percent of median per capita measure is between 4 percent of countries used for this analysis. The data are presented by and 26 percent. Between 29 percent and just over one half of region using simple averages with the important caveat that households with elderly are in the bottom 40 percent of the the regional samples vary in size and are therefore not fully distribution. On average, 13 percent of households with any representative. elderly are poor based on the 50 percent median measure, while an average of 41 percent of households with elderly are Table 1: Household types by region (%) found in the bottom 40 percent of the distribution.15 Figure 1a: Households with elderly, poverty rates using 50% of median AFR EAP ECA LAC MNA SAR Avg 30% Elderly only Share of households with elderly - lone 1 1 4 1 0 0 2 25% below poverty line Elderly only 20% - 2+ 1 1 6 2 0 1 3 15% Elderly & Adult 4 7 14 10 6 6 10 10% Elderly 5% & Children 1 1 0 0 0 0 0 0% 3-Genera- Source: Authors' calculations, poverty line set at 50% median. tion 16 18 16 14 19 27 17 Figure 1b: Households with elderly, poverty rates using 40th percentile Adult only 9 13 24 15 5 9 16 60% Share of households with elderly Adult & 50% Children 69 59 36 56 69 57 53 below poverty line 40% Total 100 100 100 100 100 100 100 30% 20% Source: Authors’ calculations. ‘Avg’ in final column refers to simple average for all sample countries. 10% 0% The share of households in our sample with any elderly Source: Authors' calculations, poverty line set at 40th percentile. members averages 32 percent. The difference between regions is not as high as might be expected given the shares of elderly people in the total population: Notably ECA and SAR have 15 Since 50 percent median poverty lines capture national distribution shapes, the highest share of households with elderly (40 percent high variation in poverty lines and rates is expected. 3 Policy Note: Pensions & Social Insurance June 2015 | Number 18 Poverty rates of elderly compared to other households. Figures Not surprisingly, households with only elderly and children 2a and 2b show the difference in poverty rates of households are poorer than other households as these are not prime with and without elderly. These figures hide significant working ages.16 This is followed by three-generation differences in poverty rates among differing household households, while adult-only households are the least poor. arrangements. In most countries, for example, households with only elderly members are not as poor as households with When considering population shares, adult with children both elderly and non-elderly members. The results show that households are the largest population group and represent the the share of countries with elderly and non-elderly households highest share of poor households (55%), followed by elderly poorest is roughly equal, though with some variation. In with children (19.6%), while the smallest shares are elderly Figure 2b, in over a third of the countries, the poverty rates with adult (0.2%) and elderly-lone households (0.8%). differ by two and a half percentage points or less. Poverty Individual-level poverty rates differ substantially in only about one in eight countries, depending on the cutoff applied. Ranking individuals rather than households allows for more a direct comparison of poverty rates among age groups.17 Elderly-only households are the least poor household type Figures 3a and 3b compares poverty rates for children, adults in the majority of the countries, and so most of the poor and elderly individuals in each country. The orange line shows households with elderly contain non-elderly members. The the poverty rates of adult individuals, and the x-axis moves combination of the financial ability to live alone, and the more from left to right starting with the lowest overall poverty mechanical explanation that household welfare is only divided rate. The higher line refers to children poverty rates while the by fewer people in elderly only households explains much of bottom line refers to elderly poverty rates. this result. Figure 2a: Difference in poverty rates at 50% median Figure 3a: Poverty rates by age using 50% median 45% 20.0% 40% Non-elderly minus elderly househld 15.0% 35% 30% 10.0% 25% poverty rate 5.0% 20% 15% 0.0% 10% -5.0% 5% 0% -10.0% Children Adults Elderly -15.0% Source: Authors' calculations, poverty line 50% median. Source: Authors' calculations with poverty line 50% median. Figure 2b: Difference in poverty rates, bottom two quintiles Figure 3b: Poverty rates by age using 40th percentile 25.0% 70% 20.0% Non-elderly minus elderly househld 60% 15.0% 50% 10.0% 40% poverty rate 5.0% 0.0% 30% -5.0% 20% -10.0% 10% -15.0% 0% -20.0% Children Adults Elderly -25.0% Source: Authors' calculations, poverty line 40th centile. Source: Authors' calculations with poverty line set at 40th percentile in each country. 16 Economies of scale also have an effect, as larger households are generally 17 Though still subject to measurement issues since data is collected at the poorer when using per capita, which assumes not resources are shared. household and extrapolated to the individual level, and per capita fails to capture scale or composition economies. 4 Social Protection & Labor | World Bank Group The figures shows that with only two exceptions (Iraq for 54 percent of the poor, which is mainly driven by their 50 percent median, and Bhutan for the 40th percentile), per population share that average just under 60 percent. capita, the poverty rates of children are always higher than for either adults or elderly. While not in the figure, older elderly, On a per capita basis, the findings show that elderly are the defined as age 75 and above, are poorer than younger elderly least poor and smallest share of the poor, while children in three-quarters of the countries sampled. It is interesting to have the highest poverty measures (headcount and gap) and note that the countries where the disparity between poverty are the second largest group among the poor. rates of children and the elderly are highest are all located in Eastern Europe and Central Asia, East Asia and the Pacific, Comparative poverty rates at the household and Latin America (in particular in the Southern Cone and individual level countries).18 In Brazil, for example, the relative poverty rate of children is more than three and a half times the rate of the Another approach of analysis is to focus on poverty rates elderly (61 percent and 17 percent respectively). The lowest of individuals compared to households. Figure 4, based differentials are found in South Asia and Sub-Saharan Africa. on the poverty line set at half of the median, shows that as expected, while the correlation between individual elderly Looking beyond the headcount poverty rates, the poverty and elderly household poverty rates is high, the latter are gap is on average the lowest for elderly and highest for generally higher. (The outlier is Iraq.) children. This means that on average, of the elderly that are poor, they are less poor, as they lie closer to the poverty Sensitivity of the results line than children. As measured by the poverty gap, in all regions, poor children are the poorer than poor elderly. As mentioned earlier, the results presented here will be affected by assumptions about the effects of household As has been shown, the poverty headcount of children economies of scale and composition on welfare. Economies is higher than adults and elderly in nearly all sample of scale, through the shared use of resources such as stoves countries, and this result is accentuated when accounting or bulk purchases, could justify adjustments to the welfare for population shares. Elderly average just over 12 percent used to compare households and individuals. Larger of the population, while children over two and a half this at household sizes appear better-off as greater scale economies 29 percent, and adults five times at 59 percent. The poverty are assumed. Lower consumption needs of children and the headcount of children averages 16 percentage points above elderly may also justify using equivalence scales in these elderly and 14 percentage points above adults. When looking comparisons. A companion note “Equivalence Sensitivity of the distribution among the poor, the highest headcount Age-Based Poverty Measures” reviews the relevant literature and second highest population share results in children and tests the results to a wide range of possible adjustments. representing 36 percent of the poor. Elderly by contrast A general finding regarding the comparison of children and have the smallest share of the poor at 10 percent due to both the elderly is that the most important impact appears to the lowest incidence and population size. Adults represent come from scale economy adjustments. Source: Authors’ calculations with poverty line 50% median. 18 Countries where child poverty is greater than 40 percentage points above elderly include Brazil (44 percent) and Uruguay (43 percent). 5 Policy Note: Pensions & Social Insurance June 2015 | Number 18 Figure 5: Illustration of sensitivity of poverty differentials (elderly & children) to 0.75 equivalence 50.0% 40.0% 30.0% Argentina Equivalized 0.75 20.0% India Hungary 10.0% 0.0% -10% 0% 10% 20% 30% 40% 50% Moldova Bhutan Bosnia Croatia -10.0% Per capita Source: Authors’ calculations using 0.75 equivalence. A positive percent means that the children poorer than elderly. With this in mind, Figure 5 highlights the differential In a small number of countries, the poverty rate differential between poverty rates of elderly and children based on the becomes negative. In other words, poverty rates are higher 40th percentile cutoff line with and without an equivalence for elderly than for children. This group is made up of ECA adjustments for scale and composition. Clearly, the effect favors countries (Bosnia and Herzegovina, Croatia and Moldova), children and reduces the difference between poverty among and elderly in Bhutan remain poorer than children. In a second children and elderly (as shown by the fact that all but one point group of countries, there is a large change in the differential, lies below the 45 degree line). Of note is that even with the but elderly still have lower poverty rates. A third group has moderate 0.75 equivalence adjustment, children remain poorest elderly poverty fall relative to children, which only includes in all but four countries, a slight decrease from children being India. The last and largest group of countries show only a small poorest in all by one country using per capita.19 effect as they remain close to the 45 degree line. As economies of scale have been observed at 0.5 and assuming children and elderly consume 50 percent of adults may be realistic, future research examines more extreme equivalence adjustments to demonstrate the age poverty ranking effect. 19 As shown in the companion note, with a more substantial economies of scale value of 0.5, elderly become poorer than children in most countries. 6 Policy Note: Pensions & Social Insurance June 2015 | Number 18 IV. Conclusions and Future Research These results raise a number of important policy questions and suggest areas for further research. The key results presented in this note can be summarized as follows: 1. Should countries with high levels of spending on programs for the elderly and high poverty rate 1. Poverty rates of elderly individuals and households with differentials with children poorer reconsider their overall elderly members vary widely across countries. social protection policies? 2. There is no strong pattern for if households with or 2. Given high co-residence rates among the poor (see note without elderly are poorer, though there is for certain on co-residence)20 , is there a case for targeting social household compositions. Adult with child households pensions to poor households in many countries?21 represent the largest share of poor households, while the smallest shares are elderly with adult and elderly-lone 3. To what extent does dependence on payroll tax based households. contributory pensions as the main source of income for the elderly lead to the observed welfare disparities 3. Per capita children are poorer than the elderly in nearly between the elderly and children? all countries both by headcount and share of the poor. In some cases, particularly in the demographically older 4. How does the provision of health services and insurance countries of Latin America, the difference is very high. affect this analysis given its importance for the elderly Elderly individuals average the smallest share of the poor and children? averaging 10 percent, children 36 percent, and adults with 5. How will falling pension coverage in Eastern Europe and the largest share of 54 percent. Central Asia change the relative poverty position of the 4. Old elderly are poorer than younger elderly in most elderly? countries. 6. How will aging populations affect the share of poor and 5. The relative position of children versus the elderly policies in different regions and countries? changes in a few countries, especially in ECA, if moderate equivalence adjustments are applied. 6. Even with these adjustments, the difference between the poverty rates of children and the elderly persists, with children poorer than elderly in over 90 percent of the sample countries. 20 See Evans, B. and Palacios, R. 2014. “An examination of elderly co-residence in the developing world”. World Bank. Washington D.C. 21 A related issue is the potential impact of transfers to the elderly on intra- household allocation of resources as shown by the evidence on transfers to women in other programs. 7 Policy Note: Pensions & Social Insurance June 2015 | Number 18 References Foster, J. Greer, J. Thorbecke E. 1984. “A Class of Decomposable Poverty Indices”. Econometrica 52, pp.761-766. Deaton, A. and C. Paxson, 1997. “Poverty Among Children and Lanjouw, J.O., P. Lanjouw, B. Milanovic and S. Paternostro, the Elderly in Developing Countries.” Research Programme in 2004. “Relative Price Shifts, Economies of Scale and Poverty Development Studies Working Paper No. 179, Princeton, N.J.: During Economic Transition.” Economics of Transition, Vol. Princeton University. 12(3), pp: 509-536. Deaton, Angus S & Muellbauer, John, 1986. “On Measuring Lanjouw, P. and M. Ravallion. 1995. “Poverty and Household Child Costs: With Applications to Poor Countries,” Journal of Size”. Economic Journal; 105: 1415-34. Political Economy, University of Chicago Press, vol. 94(4), pages 720-44, August. Pal S. and R. Palacios (2011), “Understanding Poverty among the Elderly in India: Implications for Social Pension Policy”, in Deaton, A. and C. Paxson, 1998. “Economies of Scale, Journal of Development Studies, August, 2011. (http://dx.doi.or Household Size and the Demand for Food.” Journal of Political g/10.1080/00220388.2010.509783) Economy, Vol. 106, pp: 897-930. Robalino, D. Rao, G.V. and Sluchynsky, O. 2008. “Preventing Evans, B. 2016. “Using Household Surveys for Age-based Poverty among the elderly in MENA countries: Role Poverty, Pensions, and Social Assistance Analysis”. World Bank. and Optimal Design of Old-Age Subsidies”. World Bank, Washington D.C. Washington, D.C. Evans. B 2016. “Equivalence Sensitivity of Age-Based Poverty Sen, A., 1985. Commodities and Capabilities. Oxford: Oxford Measures”. World Bank. Washington D.C. University Press Evans. B and Palacios, R. 2015. “An Examination of Elderly Co- Whitehouse. E. 2000 “How poor are the old? A survey of residence in the Developing World”. World Bank. Washington evidence from 44 countries”. World Bank. Washington, D.C. D.C. Zimmer and Das. 2013. “The Poorest of the Poor: Composition Gasparini et al. 2007. “Poverty Among the elderly in Latin and Wealth of Older Person Households in Sub-Saharan Africa”. America and the Caribbean”. CEDLAS. Universidad Nacional Research on Aging. de La Plata. La Plata, Argentina. Kakwani, N. and Subbarao, K. 2005. “Aging and Poverty and the Role of Social Pensions”. World Bank. Washington D.C. The findings, interpretations, and conclusions expressed herein are those of the author(s), and do not necessarily reflect the views of the International Bank for Reconstruction and Development/The World Bank and its affiliated organizations, or those of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. © 2015 International Bank for Reconstruction and Development/The World Bank 8