085 085 This paper is a product of the Poverty and Equity Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. The authors may be contacted at dnewhouse@worldbank.org. The Poverty & Equity Global Practice Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. ‒ Poverty & Equity Global Practice Knowledge Management & Learning Team This paper is co-published with the World Bank Policy Research Working Papers. Who Are the Poor in the Developing World? Andrés Castañeda Dung Doan David Newhouse Minh Cong Nguyen Hiroki Uematsu João Pedro Azevedo World Bank Data for Goals Group1 JEL Classification: I32, O10 Keywords: Global poverty, Poverty Measurement 1 Poverty and Equity Global Practice, World Bank Group. This report is a background paper for the Poverty and Shared Prosperity Report 2016. It is based on data harmonized by the Data for Goals Group, which comprises approximately 20 staff and consultants from the World Bank’s Poverty and Equity Global Practice and is led by Andrew Dabalen, João Pedro Azevedo, and Nobuo Yoshida. Pablo Suarez-Becerra provided invaluable assistance in data processing and validation. The team thanks Kathleen Beegle, Jose Cuesta, Andrew Dabalen, Samuel Freije- Rodriguez, Pedro Olinto, and Prem Sangraula for constructive comments, and Ana Revenga and Carlos Silva- Jauregui for their support. Correspondence should be directed to dnewhouse@worldbank.org. The views expressed here do not necessarily reflect the views of the World Bank Group or its executive board, and should not be interpreted as such. 1. Introduction The world has made remarkable progress during the past two decades in raising the living standards of the poorest. In 1990, approximately two billion people, or 37 percent of the global population, lived on less than the current international poverty line of $1.90 a day. By 2013, the year for which the latest global poverty estimates are available, the number of extremely poor persons had fallen by over 60 percent. During the same period, the proportion of the global population living in extreme poverty fell even faster, from 37 to 11 percent. The Millennium Development Goal of halving extreme poverty in developing countries between 1990 and 2015 was met in 2010, five years ahead of time. Despite these impressive achievements, the latest estimate is that 770 million persons remained in extreme poverty as of 2013, a figure in the ballpark of the combined population of the European Union and the United States. Eradicating extreme poverty is a critical priority of the international development community. Ending poverty in all its forms is the first of the 17 Sustainable Development Goals adopted by the United Nations, and the World Bank has set an ambitious goal of reducing the rate of extreme poverty to 3 percent by 2030. Achieving this goal poses a formidable challenge. Economic growth is a key driver for poverty reduction, but several studies conclude that maintaining the pace of economic growth of the recent past will not be sufficient to meet the target.2 For example, based on current projections of GDP growth from 2005 to 2015, global poverty is projected to be 4.2 percent by 2030, falling shy of the World Bank’s goal by over a percentage point (Ferreira et al. 2015). Global growth has slowed in recent years, and it is far from clear that the high rates of economic growth observed in the developing world during the past decade can be sustained for the next 15 years. Therefore, the pace of poverty reduction depends critically on engendering growth that reflects broadly shared prosperity and improves the living standards of the poorest. Evidence on where the extreme poor live, in which sectors they work, their demographic characteristics, and how they differ from the non-poor can help inform the strategies of country governments, multilateral development organizations and non-government organizations committed to reducing extreme poverty. Serious knowledge gaps remain, however, about the characteristics of the extreme poor in the developing world, mainly due to the lack of globally harmonized household survey data. Country- specific poverty profiles are valuable inputs into national poverty reduction strategies, but are not internationally comparable because each country adopts a unique national poverty line. Studies that combine data from multiple countries therefore convert welfare to a common currency using Purchasing Power Parity (PPP) exchange rates, which despite their many conceptual and practical shortcomings remain the consensus method to compare welfare across countries.3 Cross-country studies based on PPP exchange rates, however, often have limited geographic coverage. Banerjee and Duflo (2010), for example, provide an extensive analysis of the poor’s economic behavior, including levels of saving, and investment in education and health, based on data for 13 low- and middle-income countries. Olinto et al. (2013) provide a preliminary analysis of the characteristics 2 Ferreira et al. 2015, Jolliffe et al. 2014, WDI 2016, Yoshida et al 2014. 3 See Deaton (2016) for a detailed discussion of the issues associated with using PPPs to make welfare measures comparable across countries. 2 of the poor using a much more comprehensive sample, based on household data from 73 low- and middle-income countries in the 2000s. That analysis primarily focused on analyzing historical trends in poverty at the country level, however, and only scratched the surface by profiling a few key characteristics of the poor.4 Furthermore, little is known about those living just above the extreme poverty line, who account for a considerable share of the population. In the 89 countries considered in this study, 20%, or around a billion people, are estimated to live on more than the extreme poverty line of $1.90 per person per day (in 2011 PPP terms) but less than $3.10 per day in 2013. Better understanding the characteristics of this group, which we term the “moderate poor”, is important because many of them are materially deprived and considered poor by national standards. Identifying the characteristics that distinguish the moderate poor from the extreme poor may also provide a measure of insight into key factors that drive reductions in extreme poverty. This paper presents a comprehensive demographic profile of the extreme and moderate poor by age, gender, household composition, educational attainment, urban/rural residence, and employment in the agricultural sector. The profile draws on the richest and most updated collection of household survey data on poverty assembled to date – the Global Micro Database (GMD). From this database, surveys were pooled across 89 countries, each collected since 2009. The results are based on the same welfare aggregates that are used to compute the regional and global poverty estimates published by the World Bank, which are often used by the countries themselves to estimate national poverty.5 This paper is also be the first to compare the extreme poor and moderate poor for a large number of countries, and to document patterns of missing data in this type of global analysis. The analysis “lines up” survey-based poverty estimates to a common year, 2013, and shows that this procedure has mild impacts on the profile of the poor. There are five main findings. First, the poor tend to be rural and young, slightly more so than previously thought. About 80 percent of the extreme poor and 76 percent of the moderate poor live in areas classified, according to national definitions, as rural. These shares are likely overestimated, because in many cases the welfare aggregates have not been adjusted to account for the lower cost of living in rural areas. Data from South Asia, however, suggest that the magnitude of this rural bias is modest.6 With respect to age, nearly 45 percent of the extreme poor are children under 15 years old, and over 60 percent of the extreme poor live in households with three or more children. Second, gender differences in poverty rates are muted. This is because poverty status is identified at the household level, whereas gender disparities are most apparent in individual-level indicators such as education, autonomy in decision making and labor market outcomes. Third, a primary school education is not sufficient to exit poverty. A sizeable minority of the extreme poor -- about 4 In addition, as discussed below, the welfare aggregates used to profile the poor in that study were not the same ones that were used to construct the World Bank’s official poverty estimates. 5 Countries in Europe and Central Asia and Latin America and the Caribbean are exceptions, where the income aggregate used for global poverty monitoring has been standardized across the region and may differ from the aggregates used for national poverty estimates. In other regions, the welfare aggregates used are not consistent across countries, and vary for example in their treatment of housing and health expenditures, as well as their use of spatial deflators as discussed below. 6 See section 3.7 below. 3 39 percent – graduated primary school, and over a quarter of those who completed primary school but not secondary school live on less than $3.10 per day. Fourth, the moderate poor, despite having similar profiles in terms of age and household composition, are considerably better educated and are less likely to work in the agriculture sector than the extreme poor. Finally, when conditioning on other observed characteristics, having two or fewer children, completing secondary education, and living in an urban area are strongly and positively associated with economic welfare within countries. Taken together, these findings emphasize the central importance of policies and programs that benefit households in rural areas and those with large numbers of children in reducing extreme poverty. This includes not only direct support, but also efforts to reduce the future prevalence of these types of households, such as speeding the demographic transition by increasing female education levels. The results also suggest that continued improvement in educational attainment and quality, as well as more rapid urbanization and increased non-agricultural employment, can further hasten movement from extreme to moderate poverty. The rest of this paper is organized as follows. Section 2 describes the data and the methodology used to harmonize and calculate poverty estimates across different national surveys. Section 3 reports and analyzes the demographic profiles of the poor in comparison with those of the non- poor. Section 4 considers the robustness of key results to alternative line-up methods, spatial deflation, and varying samples due to missing data. Section 5 concludes. 2. Data and methodology 2.1. Data The World Bank’s procedure for estimating global poverty rates is an extraordinarily data- intensive exercise. Global poverty estimates are derived from a collection of nationally representative survey data on household welfare – either income or consumption per capita – from 142 economies in the developing world. This collection of survey data is then combined with complementary data on population, inflation, real economic growth, and Purchasing Power Parity (PPP) exchange rates. Estimating poverty rates for different types of households requires additional data on individual characteristics, comparable across countries and regions, from the same household surveys used to calculate poverty. Poverty profiles typically utilize a set of variables that are relatively straightforward to obtain such as age, gender, education, and sector of work. Nonetheless, compiling these variables from diverse household surveys, which differ in the quality of their data and the nature of the questionnaires, and harmonizing variable names and codes across surveys, is a remarkable achievement. These heavy data requirements are the main reason why only a few empirical studies have examined patterns across such a large number of countries. This analysis is based on the September 2016 vintage of the Global Micro Database (GMD), a collection of globally harmonized household survey data recently developed by the Data for Goals 4 group of the World Bank’s Poverty and Equity Global Practice.7 The GMD is an ongoing initiative and new surveys are added each year. As of June 2016, it contained 443 household surveys from 134 countries. The surveys are as old as 1993, but more than 90 percent of the surveys are from 2005 or later, and surveys from 2009 to 2013 account for 56 percent of the collection of surveys. The GMD is the richest collection of nationally representative, globally harmonized household survey data on poverty.8 The figures on poverty are based on a measure of household welfare, which is either household per capita income or household per capita consumption, depending on which concept is used to measure national poverty in a particular country. Of the 89 countries in our sample, income is used in 30 of them, most of which are located in Europe, Central Asia, Latin America, and the Caribbean. The GMD is particularly suitable for profiling the global poor because its welfare aggregates are identical to those used to compute the global poverty estimates published by the World Bank, except for the notable exception of China.9 China is a special case because the World Bank does not have access to the individual level records of the Chinese Household Budget Survey (HBS), which is the source of official Chinese poverty statistics. The World Bank’s international poverty estimates for China are instead based on an approximate distribution derived from grouped data, which cannot be used to profile the characteristics of the poor. This study therefore utilizes household level data from the 2013 Chinese Household Income Project Survey (CHIPS), made available to the public by Beijing Normal University. The CHIPS is drawn from the same sample frame as the HBS, and an analysis of an earlier round of the survey, collected in 2007, yielded similar poverty rates as the official HBS-based estimates (Zhang et al, 2014). The poverty rate for urban and rural China, derived from the 2013 HBS, is applied to the CHIPS data to generate profiles of the extreme and moderate poor in China. The data from India also deserve special mention. In general, the results presented below are based on schedule one of the 2011 National Sample Survey (NSS), which is the primary source underlying both the estimates of poverty reported by the Indian government and the international poverty rate reported by the World Bank. The schedule one survey, however, does not collect information on labor market outcomes. Therefore, all information on sector of work is taken from schedule ten of the NSS, which collects both labor market information and sufficient information on household expenditure to construct an unofficial consumption aggregate. To calculate the poverty status of Indian workers by sector, the World Bank’s urban and rural headcount poverty rates, which are derived from the schedule one survey, are applied to the corresponding percentiles of the urban and rural distribution of schedule ten’s per capita consumption measure. Thus, the shares of agricultural workers that are below the $1.90 and $3.10 thresholds in India are estimated using the unofficial welfare aggregate collected in schedule ten. 7 Due to the nature of the license agreements between the World Bank and National Statistical Offices, data for most countries cannot be made publicly available. 8 Only one survey is not nationally representative: Argentina’s household consumption survey, the Encuesta Permanente de Hogares, which is not nationally representative and covers only about two-thirds of the country’s urban population instead. Given that the urban population accounted for about 90 percent of Argentina’s total population in 2013, the survey effectively only represents 61 percent of the national population. 9 For more information on how the welfare aggregates are constructed, see Ferreira et al. (2015). 5 For the purpose of this study, we utilize data from as many as 89 countries for which a nationally representative welfare survey from 2009 or later was conducted and obtained by the World Bank, and for which 2013 poverty rates have been published.10,11 Figure 1 presents the number of surveys in the final sample, according to their survey year. The final sample contains 104 surveys. In 15 of the 89 countries there are two available surveys bracketing 2013, which are both utilized as described below. The year 2009 was selected as the cut-off year to balance the competing goals of maximizing geographical coverage of the sample and minimizing error when “lining up”, or extrapolating, country-level poverty figures to 2013 as discussed in Section 2.2. Figure 1: Number of surveys by survey year  15 10 15 2009 2010 24 2011 2012 10 30 2013 2014 In total, the 89 countries in the sample used in this paper represent 84% of the developing world’s population in 2013 and all geographical regions (Figure 2).12 The 650 million in the sample that are classified as extremely poor amount to 85.4% of the estimated number of extreme poor worldwide in 2013, according to the World Bank.13 Table 1 lists the number of countries and the share of the 2013 population represented in the sample, by region and income classification. Except for the Middle East and North Africa (MENA) and Sub-Saharan Africa (SSA) regions, the sample used in this study provides excellent coverage of the global population and the extreme poor. While the extremely low data coverage in the MENA region is a concern, it is not unique to this study. In fact, PovcalNet has for several years suppressed regional poverty estimates for the MENA region due to low survey coverage. But since the region has traditionally had a much lower incidence of poverty than the average for global and developing countries, the omission of most countries from the region has a limited impact on the characteristics of the poor.14 The omission of over one-quarter of the population of Sub-Saharan Africa is a greater concern, since the region has the highest number of extreme poor (389 million) 10 In order to maintain consistency with the figures reported by the World Bank, we only include countries classified as “developing countries”. This excludes most of Western Europe, Canada, the United States, Australia, Japan, and the Republic of Korea. 11 Not every survey is fully nationally representative, due to the exclusion of rural areas or conflict areas in particular countries. 12 Population figures are based on population projections by the United Nations Department of Economic and Social Affairs (UNDESA). 13 Global poverty estimates are available on PovcalNet (http://iresearch.worldbank.org/PovcalNet). We refer to PovcalNet as a key reference point to assess the extent of our sample coverage and the validity of our estimates. As shown in Figure 1, our sample includes almost all countries covered by PovcalNet. 14 In 2008, the most recent year MNA estimates are reported, the region accounted for 0.7 percent of the total number of extreme poor. 6 as well as the highest extreme poverty rate in 2013 (41 percent). A total of 22 Sub-Saharan African countries are excluded from the sample in this study, either because data were not available or because their latest available survey is from before 2009.15 These countries combined represent 185 million people, with an average extreme poverty rate of 32 percent as of 2013.16 Figure 2: Geographical coverage of the GMD sample    Table 1: Population distribution in GMD by region and income group  Number of  Number of  Population   Share of sample  Share of developing  countries  countries  (millions)  population (%)  world population    included  excluded      represented in        sample (%)  Total  89  101  5,249.1                 100.0  84.2%  Income Group            Low income  21  14   628.5                      12.0   75.0%  Lower‐middle income  29  17  2,281.2                      43.5   87.8%  Upper‐middle income  26  30  2,071.1                      39.5   86.7%  High income  13  40  267.5                        5.1   65.2%  Region  .      .    .      East Asia & Pacific  11  23  1,889.3                      36.0   90.8%  Europe and Central Asia  24  14  424.7                        8.1   87.6%  Latin America & Caribbean  18  23  550.3                      10.5   88.8%  Middle East & North Africa  3  17  16.3                       0.3   4.1%  South Asia  7  1  1,667.1                      31.8   98.2%  Sub‐Saharan Africa  26  22  701.4                      13.4   73.9%  Welfare measure  .      .    .      Income  59  N/A  4,497.1                      85.7   N/A  Consumption  30  N/A  752.0                      14.3   N/A  15 They are Angola, Burundi, Cameroon, Cabo Verde, the Comoros, Côte d'Ivoire, Gabon, Ghana, Kenya, Liberia, Mauritania, Mozambique, and the Seychelles. 16 According to the 2013 poverty estimates based on 2011 PPP in PovcalNet. 7 Finally, the surveys in the GMD also contain a substantial amount of missing data for particular characteristics, either because of non-response by respondents, or because questions were not asked in particular surveys. If the variable is missing for a fraction of the national sample, we assume this results from non-response and/or errors during the data collection and cleaning process, and include that country when generating the global profile, in order to maximize the geographical coverage of our sample. In the latter case, when a question is entirely missing in a survey, the country is excluded when generating poverty statistics for that characteristic. Therefore, although the analysis draws on data from as many as 89 countries, the number of countries in the sample varies when considering different characteristics. Missing data, however, is less prevalent in more populous countries. Therefore, both missing and unreported data generally account for a small share of the weighted global sample, as shown in Table 2. The main exception is sector of work, which is only collected in surveys from 64 of the 89 countries. Consequently, data exist on only about 82 percent of the employed adult population, with 8.7 percent missing due to non-response and 9.6 percent missing due to the complete absence of information in the survey. The poverty profile by employment sector reported in this paper, thus, should be interpreted with appropriate caution.   Table 2: Missing data in key profiling variables  Share of  Share of  Number of  Share of  Share of      extreme  moderate  countries  non‐poor  population  poor  poor  Reported  87  100.0  100.0  99.9  99.9  Sector of  Non‐response  .  0.0  0.0  0.0  0.0  residence   Not in survey  2  0.0  0.0  0.1  0.1  Employment  Reported  64  68.5  78.2  84.6  81.7  sector  Non‐response  .  18.2  15.4  5.6  8.7  of working  Not in survey  25  13.2  6.4  9.8  9.6  adults aged 15+  Reported  89  100.0  100.0  100.0  100.0  Age   Non‐response  .  0.0  0.0  0.0  0.0  Not in survey  0  0.0  0.0  0.0  0.0  Reported  89  100.0  100.0  100.0  100.0  Gender   Non‐response  .  0.0  0.0  0.0  0.0  Not in survey  0  0.0  0.0  0.0  0.0  Educational  Reported  83  94.3  94.1  95.3  95.0  Attainment  Non‐response  .  4.3  5.2  2.3  3.0  of adults aged  Not in survey  6  1.4  0.7  2.3  1.9  15+  Educational  Reported  70  92.3  93.8  93.2  93.2  attainment of  Non‐response  .  6.3  5.3  1.9  3.5  children aged  Not in survey  19  1.4  0.9  4.9  3.3  12‐14  8 2.2.Methodology The major technical challenge in generating global poverty profiles is combining surveys from different countries and years. Since the availability and frequency of household surveys differ significantly from one country to the next, it is not feasible to produce a global profile that maintains adequate geographical coverage of the poor using data from one particular year. This raises the challenge of how to reliably describe the characteristics of the poor based on national surveys collected in different years. PovcalNet, the primary source of international poverty estimates maintained by the Research Department of the World Bank, reports global and regional poverty rates for particular reference years. For the reference year 2013, for example, PovcalNet uses approximately 50 surveys conducted that year. For the rest of the countries, the available survey data are “lined up” to 2013 using a complex procedure. First, the welfare measure in each country is multiplied by a constant scale factor to account for changes in welfare between the survey year and 2013. In most countries, the scale factor is the real growth between the survey year and 2013 in per capita household final consumption expenditure (HFCE), which come from the World Development Indicators Database. For the others, mostly in Sub-Saharan Africa, HFCE is either not available or has not been used, so real growth in GDP per capita in local currency units is used instead. Second, after this scale factor is applied to the welfare aggregate, extreme poverty in 2013 is calculated as the share of the sample population whose welfare falls below $1.90 per day in PPP terms. Two assumptions underpin this line-up method. The first is that the relative distribution of welfare across households remains constant over time; and the second is that HFCE or GDP growth provides a reasonable approximation to the growth in survey consumption means. World Bank researchers have attempted to validate these assumptions in both Africa and India.17 In India, the relative distribution of welfare across households changes little across years. However, in both India and Africa, GDP growth significantly exceeded the growth in household consumption collected in surveys, meaning that the line-up procedure overstates poverty reduction.18 The inconsistency between macro and survey-based measures of consumption growth further justifies setting a minimum year threshold of 2009 for surveys to be included in the analysis. Using the lined up estimates to generate a poverty profile, instead of an estimated poverty rate, has two additional drawbacks. First, the procedure assumes that it is those households closest to the poverty line that either escape or fall into poverty when economic growth or decline occurs. In reality, households enter and exit poverty from a variety of points in the welfare distribution. Second, the line-up procedure does not account for changes in individual or household characteristics, such as increased urbanization or educational attainment that occurred between the survey year and the reference year. This means that the key variables that describe the poor, such as location, sector of work, and education, are not lined up to a common year. 17 See Joliffe et al, 2014, p. 250-254, and Beegle et al, p. 43. 18 In addition, HFCE is typically calculated as a residual, meaning that its quality can vary greatly across countries and subsequent revisions can be substantial (Jerven, 2013). 9 Despite these issues, failing to line up poverty and population estimates is also problematic. In particular, pooling estimates from different years would bias the global profile towards characteristics of the poor in older surveys, many of whom might have since escaped poverty. For example, the most recent survey from India dates from 2011/12. Survey-based estimates show a fall in rural poverty from 36.3% to 24.8% between 2009/10 and 2011/12, and the “lined up” estimate for 2013 is 19.9%. Even though this latter estimate may overestimate poverty reduction, failing to adjust for this dramatic decline would over-represent poor rural Indians in the global poverty profile. At the same time, failing to account for population growth between 2010/11 and 2013 would underrepresent India relative to other countries. Although we know of no systematic evidence that using this line-up method improves the accuracy of the poverty profile, our main specification is based on the “lined up” figures available on the PovcalNet website. In other words, we take the 2013 poverty rates published by the World Bank and use the corresponding percentile value to set poverty lines in the GMD sample. This ensures consistency with the published estimates and eases interpretation of the results. It also maintains continuity with the approach adopted in Olinto et al. (2013), except that we implement an additional population adjustment. In particular, the sample weight is adjusted to match population projections for each gender and age group in each country as of 2013, taken from the United Nations Department of Economic and Social Affairs (UNDESA).19 In cases where population estimates by age group and gender are not available from UNDESA, we simply re-scale the sample weight to match the total population reported in the World Development Indicators. Section 3 below reports estimates based on this method. In section 4, we test the sensitivity of key results to the choice of line-up method. In particular, we compare key results to an alternative approach that simply takes the population weighted average of country-level profiles, without any attempt to line-up data from different survey years to a common year, and find minor changes. Even though the results are generally similar under this approach, they cannot be referenced to a specific year and thus must be interpreted with caution. Appendix 1 describes the line-up procedures in greater detail. Once the household surveys have been lined up, it is straightforward to use the pooled household survey data to construct poverty profiles. The figures displayed below are based on the international poverty lines of $1.90 and $3.10 per person per day in 2011 PPP as currently reported by the World Bank. The $1.90 line is the average national poverty lines from the 15 poorest countries originally used by Chen and Ravallion (2010) to establish the $1.25 a day poverty line in 2005 PPP terms.20 As mentioned earlier, statistics are reported separately for the extreme poor, the moderate poor, and the non-poor. The moderate poor are defined as individuals whose household per capita consumption (or income) lies between $1.90 and $3.10 19 This reweighting procedure multiplies household weights by a scale factor in order to maintain a constant weight within each household. However, in cases where that is not sufficient to match the age group and gender- disaggregated UNDESA population estimates, weights are rescaled according to individuals’ age and gender groupings to exactly match these totals, meaning that weights for different members of the household may vary. 20 See Ferreira et al (2015) for more details on how these lines are derived. 10 per person per day, while the non-poor have consumption or income greater than $3.10 per person per day.21 3. Findings 3.1. Poverty is disproportionally rural As shown in Figure 3, 18.2% of rural residents subsist on less than $1.90 a day, and 45.6 percent of rural residents are either extremely or moderately poor, and therefore live on less than $3.10 per person per day. The corresponding rates for urban residents, in contrast, are 5.5% and 16.2%. This large gap in poverty rates, combined with the concentration of the general population in rural areas, translates into a startling disparity in the number of poor people. About 80% of the extreme poor and 76% of the moderate poor live in rural areas, as compared to only 44% of the non-poor. Given that there are 655 million extreme poor and a little over a billion moderate poor people represented in the sample, the rural sector accounts for a total of 525 million extreme poor and additional 789 million moderate poor. Of course, these absolute numbers underestimate the true global figures, since the sample of 89 countries is only representative of 86.5% of the population of the developing world, and 73% of the population of the globe, in 2013. Figure 3: Poverty rate by residential sector           Figure 4: Share of population by residential sector  90 83.8 100% 80 67.7 70 80% 44.2 60 54.4 55 60% 80.1 75.7 50 40 27.4 40% 30 18.2 19.9 20 55.8 10.7 12.5 20% 45 10 5.5 19.9 24.3 0 0% Urban Rural Total Extreme Moderate Non‐poor Total poor poor population Extreme poor Moderate poor Non poor Urban Rural This finding suggests a slightly higher level of concentration of the poor in rural areas than previously thought. The study most comparable to this one is Olinto et al. (2013), which applied the $1.25 extreme poverty line (2005 PPP) in 2010 to a different set of harmonized household 21 These figures are generally calculated using the PPPs from 2011. But as of June 2016, The 2011 PPP conversion factors had not yet been adopted for 5 countries: Bangladesh, Cabo Verde, Cambodia, the Lao People’s Democratic Republic, and Jordan. In these cases, we use the $1.25 and $2.00 dollar-a-day poverty lines (measured in 2005 PPP), which are roughly equivalent to the $1.90 and $3.10 lines in 2011 PPP terms. (Ferreira et al. 2015). 11 surveys, from 2000 to 2009 for 73 low- and middle-income countries. That analysis reported that 58% of the total population and 78% of the extreme poor lived in the rural sector, as compared to 55% and 80%, respectively, in our sample. The slightly lower share of the total population living in rural areas in this analysis likely reflects increased urbanization in recent years, as this paper draws on newer household surveys and lines up the sample to 2013 instead of 2010. Although the two studies are not directly comparable, the higher share of the extreme poor in rural areas in the more recent data is suggestive that the rural sector has lagged in reducing poverty since 2010.22 As mentioned above, the extent of poverty in rural areas is subject to two countervailing biases. On the one hand, the failure to spatially deflate the welfare aggregate in several countries overestimates the extent of rural poverty, by not adjusting for the lower cost of living in rural areas. On the other hand, the use of outdated urban/rural definitions in national surveys understates rural poverty by continuing to classify newly urbanized suburbs, which tend to be less poor than more remote areas, as rural. The former is mitigated by the use of separate urban and rural PPP exchange rates in China, India, and Indonesia. This, combined with evidence from South Asia presented below, suggests mild effects on the overall share of rural poor of less than five percentage points. It is not possible to assess the magnitude of the second source of bias with the data at hand.23 In line with the high rate of rural poverty, however, poverty is also deeply ingrained in the agriculture sector. Nearly two-thirds of extremely poor workers aged 15 and above reported that their primary job is in the agricultural sector; and extreme poverty rates among these workers is more than four times higher than among non-agricultural workers (Figures 5 and 6).24 The significant gap in poverty rates between the agriculture and non-agriculture sectors is consistent with the well-documented earnings penalty faced by agriculture labor.25 While the proportions of the extreme poor and moderate poor living in rural areas are roughly similar (80.1% vs. 76.0%, or a 4.1 percentage point difference), the moderate poor are much less likely than the extreme poor to work in the agriculture sector (64.7 vs. 52.0 percent, a 12.7 percentage point difference). This is consistent with growth in non-agricultural employment facilitating movement from extreme to moderate poverty. Non-agriculture employment, however, is far from sufficient to escape poverty. Of the extremely poor workers in rural areas, nearly 24 percent work in non-agriculture jobs, and just under 40 percent of moderately poor workers in rural areas work outside of agriculture (Figure 7). This suggests further analysis to better understand why these non-agricultural workers in rural areas remain poor, and whether either facilitating migration to urban areas or greater investment in infrastructure would raise their productivity. Further analysis of the GMD, for example, can 22 Future work analysis will examine trends in the poverty profile using a comparable sample of countries. 23 An additional source of bias, if the population of interest is the entire developing world, results from the non- random selection of countries into the sample. In general, results can be interpreted as representing the 89 sample countries, which as noted above represent 86.5 percent of the developing world population. 24 The total poverty rate in Figure 6 is lower by 2.4 percentage points than in Figure 4. This is primarily due to the smaller number of surveys for which the variable on the sector of work is available. 25 A longstanding literature documents the central role of low returns to agriculture in development (Lewis (1954), Harris and Todaro (1970)). Alvarez-Cuadrado and Poshke (2012) and Gindling and Newhouse (2014) provide more recent empirical evidence on the earnings penalties faced by agricultural workers in developing countries. 12 explore in more detail how changes in local employment patterns relate to changes in economic welfare, to better understand the types of jobs that reduce extreme poverty in the rural sector. Figure 5: Poverty rate by employment sector      Figure 6: Share of population by employment sector  90 83.6 100% 80 73.8 90% 80% 35.4 70 48.5 70% 60 70.4 50.4 60% 79.8 50 50% 40 40% 29.8 30 30% 64.6 19.8 17.1 51.5 20 11.8 20% 9.1 29.6 10 4.6 10% 20.2 0 0% Agriculture Non‐Agriculture Total Extreme Moderate Non‐poor Total poor poor population Extreme poor Moderate poor Non poor Agriculture Non‐Agriculture Figure 7: Share of workers in agricultural sector by urban/rural  100% 23.8 80% 39.8 60% 40% 76.2 60.2 20% 46.1 32.5 36.3 17.0 0% 4.0 5.5 Extreme Moderate Non‐poor All urban Extreme Moderate Non‐poor All rural poor poor poor poor Urban Rural Agriculture Non‐agriculture 3.2.The poor tend to be young Over one in five children under 15 lives in an extremely poor household, and children under 15 make up 44% of all extreme poor (Figures 8 and 9). Extreme poverty rates are 8.2 percentage points higher for children 0 to 14 than those for young adults 15 to 24, and over 13 percentage points higher than adults aged 65 and above. This variation in poverty rates across age groups is 13 striking, but not new. Batana et al. (2013), for instance, find substantial gaps in poverty headcount rates between children and adults, and children and elderly 65 and above. These gaps amount to 14.4 and 19.5 percentage points, respectively, when defining children to be below 12 years old. Similarly, Olinto et al. (2015) find that 34% of the extreme poor but only 20% of the non-poor are children between 0 and 12 years old. In results not reported below, we find somewhat larger disparities when using these age cut-offs -- 39.7% of the extreme poor are children 0 to 12, as compared to 19% of the non-poor and 24 percent of the sample population Furthermore, these estimates assume that resources are equally distributed within the household, and relaxing that assumption may further raise child poverty.26 High rates of child poverty have serious implications for child mortality, morbidity, malnutrition, physical development, psychological health and education, which compromises both their long-term earnings potential and the growth prospects of the countries in which they live. Figure 8: Poverty rate by age group           Figure 9: Share of population by age group     100% 100% 3.8 5.2 7.7 6.7 90% 80% 80% 70% 60% 60% 50% 17 40% 18 40% 17.4 30% 25.4 17.2 20.6 19 20% 44.2 20% 17.2 15.5 15.5 15.5 34.6 27 21.6 10% 20.4 12.2 10.5 9.4 7.5 7.1 7.1 0% 0% Extreme Moderate Non‐poor Total 0‐14 15‐24 25‐34 35‐44 45‐54 55‐64 65+ poor poor population Age group 0‐14 15‐24 25‐34 35‐44 Extreme poor Moderate poor Non poor 45‐54 55‐64 65+ High rates of child poverty are also reflected in the household composition of poor households. Extremely poor households, on average, have 7.9 members, 3.5 of which are children under 15. This substantially exceeds the average number of children in moderately poor and non-poor households, which is 2.3 and 0.9 (see Table 3). Put another way, children 0 to 14 make up 44 percent of extremely poor households, 35 percent of moderately poor households, and only 21 percent of non-poor households. 26 Bargain et al. (2014). 14 Table 3: Household composition  Average number of members per household  Age group  Extreme poor  Moderate poor  Non‐poor  0‐14  3.5  2.3  0.9  15‐24  1.3  1.2  0.7  25‐34  1.1  1  0.7  35‐44  0.8  0.8  0.6  45‐54  0.5  0.6  0.5  55‐64  0.4  0.4  0.4  65+  0.3  0.3  0.3  Total  7.9  6.6  4.3   In contrast to children, the poverty rate among the elderly (aged 65 and above) is the lowest among the age groups considered. This is evident by both the low share of elderly in poor households – adults 65 and over account for less than 4 percent of the extreme poor and 5.2 percent of the moderate poor, as compared with 7.7 percent of the non-poor. (Figure 9). The youthful nature of extreme poverty is also reflected in the large share of the poor – over 58 percent – that live in households with three or more children (Figure 10). While 36 percent of the poor live in larger households with more than two children and more than two adults, a sizeable 22 percent have three or more children and two or fewer adults. Of the extreme poor, less than one in ten have no children under the age of 15. Figure 10: Share of the extreme poor by number of children and adults in household  40 36 Share of extreme poor 35 30 25 22.2 22 20 15 11.4 10 6.3 5 2.2 0 Number 0 1 to 2 More 0 1 to 2 More of kids than two than two Number Two or less More than two of adults Note: Children defined as less than age 15 Poverty rates vary sharply according to the age of the household head as well. (Figures 11 and 12). On the one hand, 57 percent of those living with a household head below 15 years old are either extremely or moderately poor, highlighting the deprivation and vulnerability of households in which an adult head is not present. Households with young heads are extremely rare, however, as 15 they make up 0.5 percent of the population, and even households with heads from 15 to 24 only comprise 3 percent of the population. Compared to households with heads aged 15 to 24, poverty rates are higher for households with heads between the ages of 25 and 34, who are more likely to be burdened with children. As the age of the head exceeds 34, the poverty rate declines slightly, reflecting the greater earnings power of older workers. Figure 11: Poverty rate by age of HH head               Figure 12: Share of population by age of HH head  100% 100% 10.7 11.3 12.1 11.8 80% 42.7 80% 15.7 17.5 17.7 17.4 68.7 62.4 64.9 69 69.9 69.9 60% 23.9 24.5 60% 25.6 25.2 40% 28.9 40% 29 27.6 26 26.7 20% 17.5 22.3 21.1 19.2 19.3 19.1 20% 28.4 18.2 17.2 15.6 16.3 13.7 15.2 14 11.8 10.8 10.9 0% 0% Extreme Moderate Non‐poor Total 0‐14 15‐24 25‐34 35‐44 45‐54 55‐64 65+ poor poor population Age group of head 0‐14 15‐24 25‐34 35‐44 Extreme poor Moderate poor Non poor 45‐54 55‐64 65+ 3.3.Gender gaps in poverty are modest Tallying the share of men and women that live in poor households shows modest gender inequality in measured poverty (Figures 13 and 14), but this reflects the fact that poverty status is measured at the household level. By assumption all household members are classified as either in or out of poverty, and the ratio of males to females is roughly 50/50 in both poor and non-poor households. Recently, two more sophisticated studies have used differential patterns of consumption to attempt to measure individual levels of poverty, with diverging results. Attempting to account for differences in consumption within households greatly increased poverty for women in Malawi, but had little effect in Côte d’Ivoire.27 But given the strong assumptions underpinning these estimates, future work can fruitfully examine gender disparities in labor market outcomes, intra-household allocation, and autonomy. These indicators of female empowerment are not only central to the gender agenda but can also have powerful indirect effects on poverty, by for example reducing the prevalence of households with large numbers of children. 27 Dunbar et al, 2013, Bargain, et al (2014). 16 Figure 13: Poverty rate by gender                  Figure 14: Share of population by gender  80 100% 67.5 67.9 90% 70 80% 50 50.6 50.8 50.7 60 70% 60% 50 50% 40 40% 30 30% 19.9 19.8 50 49.4 49.2 49.3 20% 20 12.7 12.3 10% 10 0% 0 Extreme Moderate Non‐poor Total poor poor population Extreme poor Moderate Non poor Male poor Female Male Female There are larger differences when considering gender of the household head, as poverty rates on average are moderately higher for male headed households than for female headed households. Male headed households are 3.4 percentage points more likely to be poor than female heads, and the share of the extreme poor living in male headed households is 4.4 percentage points higher than the population average (see Figures 15 and 16). There are a number of potential explanations for this result, which may at first glance appear to be counter-intuitive. A household is likely to report a female head if the usual male head is a migrant working out of town, in which case the household may benefit from remittances that make them less likely to be poor. Second, households are more likely to identify a female head if the woman is the breadwinner, which tends to occur in less poor households. Finally, single or divorced women may also be more likely to be financially independent. Figure 15: Poverty rate by gender of HH head          Figure 16: Share of population by gender of HH head  80 75.3 100% 65.7 16.2 15.5 22.9 20.6 70 80% 60 50 60% 40 83.8 84.5 40% 77.1 79.4 30 21.1 20 14.9 20% 13.2 9.8 10 0% 0 Extreme Moderate Non‐poor Total Extreme poor Moderate poor Non poor poor poor population Male Female Male Female 17 The moderately greater poverty rates among male headed households are driven mainly by lower middle income countries (Table 4). In these countries, which account for 58 percent of the extreme poor, the poverty rate for female headed households is 3 percentage points lower than that for male headed households. The remainder of the difference is due to female headed households being slightly more prevalent in upper middle income countries, which have much lower rates of extreme poverty. But even in low income countries, the differences in poverty rates between male and female headed households is minor. Table 4: Poverty rates for female head households are similar within income groups.  Country Income  Share of  Share of  Extreme poverty  Extreme poverty  Total  Group  population  in  extreme  rate  for female  rate for male  extreme  female headed  poor   headed  headed  poverty  households  households   households  rate   Low Income  16.6  33.6  35.6  34.9  35.0  Lower Middle‐Income   16.6  58.1  14.3  17.2  16.7  Upper Middle‐Income  21.4  8.2  3.3  2.4  2.6  High Income  57.5  0.1  0.2  0.4  0.3  Total   20.6  100.0  9.8  13.2  12.5  Finally, we see little evidence that poverty exacerbates gender discrimination in the educational attainment of older children. Table 5 shows the distribution of attainment among children 12 to 14 years old, by gender and poverty status. The results are nuanced: A slightly higher share of girls than boys have not completed any education, but a slightly higher share have already completed some secondary school. Compared to the total population, boys that are extremely poor are 2.7 times more likely to have no education. The same is true for girls, suggesting that the mild female penalty at the bottom of the education distribution is, if anything, less pronounced among the poor. Table 5: Distribution of highest educational attainment by gender and poverty status, children 12‐14   Moderate     Extreme poor  Non‐poor  Total  poor  Male           No education   13.7  6.7  1.7  5.0  Incomplete primary   45.2  38.3  44.8  43.2  Complete primary or  38.5  54.2  52.8  50.7  incomplete secondary  Secondary or above  2.6  0.8  0.8  1.1  Female          No education   15.5  7.6  1.9  5.8  Incomplete primary   41.7  34.9  41.4  39.9  Complete primary or  40.3  56.4  55.5  52.9  incomplete secondary  Secondary or above  2.6  1.1  1.2  1.4    18 3.4.A sizeable minority of the extreme poor have attended secondary school Poor adults tend to be poorly educated, and there is a strong link between an individual’s educational attainment and his or her economic well-being.28 As shown in Figure 17, nearly a quarter of all those with no education are extremely poor, and 58 percent live on less than $3.10 per day. Despite these high rates of poverty, however, those with no education are now a distinct minority of the population, and only constitute 39 percent of the extreme poor and 28 percent of the moderate poor (Figure 18). Headcount poverty rates, not surprisingly, decline sharply as education increases. Most strikingly, those with at least one year of completed tertiary education are very unlikely to be poor, as a mere 1.5% of these adults are extremely poor, and only 5.2% live on less than $3.10 per day. Figure 17: Adult poverty rate by education          Tertiary 1.3 4 94.6 Complete secondary 6.3 10.7 82.9 Complete primary or 7.8 20.3 71.9 Incomplete secondary Incomplete primary 10.5 18.7 70.8 No education 24.8 33.4 41.9 0% 20% 40% 60% 80% 100% Extreme poor Moderate poor Non poor Figure 18: Share of adults by education  Total 14.9 19.7 32.7 14.7 18 No education population Non‐poor 8.6 19.1 32.3 16.7 23.3 Incomplete primary Moderate Complete primary or 28.3 20.9 37.7 9 4.1 Incomplete secondary poor Complete secondary Extreme 39.1 21.8 26.8 9.8 2.5 poor Tertiary 0% 20% 40% 60% 80% 100% 28 An individual is defined as having no education if she/he has never attended any formal school. The individual is defined as having primary, secondary, or tertiary education is she/he has attended at least one year within that education level. A five category measure of educational attainment was utilized because it maintains strong country coverage while also allowing for disaggregation between those that did not and did complete primary school. 19 Poor children age 12 to 14 appear to be having mixed success, in terms of educational achievement and attendance. On the positive side, a full 85 percent of extremely poor children aged 12 to 14 have completed at least some primary school, and 40 percent of these have graduated primary school. But household poverty clearly depresses children’s educational attainment. Of the 5.4 percent of children who never completed any schooling, half are extremely poor and 80 percent live on less than $3.10 per day (Figures 19 and 20). The corresponding poverty rates for children whose educational attainment is more likely to be on track – having at least completed primary school – is only 15 and 41 percent. Furthermore, nearly 15 percent of extremely poor children aged 12 to 14 have never attended school (Figure 20), likely consigning them to an adult life of at least moderate poverty. Efforts to reduce the tuition and travel cost of school attendance can help break the intergenerational cycle between household poverty and children’s educational attainment.  Figure 19: Child poverty rate by education   Complete secondary 45.4 22.9 31.7 Complete primary or 15.1 26.2 58.7 Incomplete secondary Incomplete primary 23.2 24.3 52.5 No education 49.7 30.8 19.4 0% 20% 40% 60% 80% 100% Extreme poor Moderate poor Non poor   Figure 20: Share of children by education and poverty status   Total population 5.4 41.6 51.7 1.3 No education Non‐poor 1.8 43.2 54.1 1 Incomplete primary Moderate poor 7.1 36.7 55.2 1 Complete primary or Incomplete secondary Extreme poor 14.6 43.5 39.4 2.6 Complete secondary 0% 20% 40% 60% 80% 100%     20 The analysis presented so far has only considered aggregates for the entire sample, and this section briefly touches on regional differences in the characteristics of the poor.29 The share of the poor according to key characteristics are displayed in Figures 21 and 22. The predominantly rural nature of poverty is apparent in all regions except for Latin America and the Caribbean. Regions with high poverty incidence, such as South Asia and Sub-Saharan Africa, have a greater share of rural residents. Within each region, the extreme and moderate poor tend to have a similarly higher share of rural residents, while the non-poor have a much lower share of rural residents. The only exception is Sub-Saharan Africa, where the rural share among the extreme poor is notably higher than that of the moderate poor. As expected, working in agriculture is closely related to poverty status in each region, and poorer regions have higher shares of adults working in agriculture. In all regions, both extremely and moderately poor adults are much more likely to work in agriculture than non-poor adults are. Differences are starkest in Europe and Central Asia, Latin America, and East Asia and Pacific. Similar patterns hold for poverty among children. Except for Europe and Central Asia and Latin America and Caribbean, where poverty incidence is low, the share of children 14 years or younger is the highest for the extreme poor and lowest for the non-poor. Sub-Saharan Africa, the poorest region, has a particularly high share of children among extreme poor (50 percent) and moderate poor (44 percent). With respect to adult education, regional differences are clearer. While about 39 percent of extremely poor adults (15 years or above) have no formal education overall, less poor regions (East Asia and Pacific, Europe and Central Europe and Latin America and Caribbean) have a lower share of adults with no formal education (5 to 14 percent). In contrast, South Asia and Sub-Saharan Africa have a much higher incidence of adults with no formal education, 48 percent and 41 percent, respectively. In these two regions, educational attainment is closely correlated with poverty status; extremely poor adults are much more likely to have no education than moderately poor adults, who are in turn much more likely to have no education than the non- poor. Figure 21: Share of extreme poor by region and selected characteristics    29 Profiles for Middle East and North Africa are withheld due to low data coverage. 21 Figure 22: Share of population by region and poverty status   3.5. Rural residence, household composition, and educational attainment remain strongly correlated with welfare after controlling for other observed characteristics The analysis has so far looked at unconditional correlations between poverty and demographic characteristics. This section examines conditional correlations, estimated from a simple regression of welfare on demographic characteristics. Because the relationships between poverty and these characteristics are complex and interwoven, the results should not be interpreted as causal relationships. Nevertheless, when interpreted as descriptive correlations, they can provide useful insight into whether the patterns observed in the profiles remain, after controlling for multiple household and individual-level characteristics.30 The first set of regressions uses individual-level data and regresses the log of the welfare aggregate on household and individual characteristics. The regressions are estimated both with and without country fixed effects, which control for all characteristics common to each country (Table 6). The coefficients from the fixed effects regressions represent a pooled average of within-country relationships; for example, the coefficient on rural location in the fixed effect regression represents the penalty that rural residents face relative to their urban compatriots in the same country, averaged over all countries in the sample. This estimate, therefore, is not influenced by the greater prevalence of rural residents in poorer countries. The second set of regressions uses household-level data, and regresses the same dependent variable on characteristics of the households and their heads (Table 30 The regressions are not exactly comparable, as the dependent variable is the log of welfare rather than the poverty rate. But given their close inverse relationship, using log welfare instead of the poverty rate as the dependent variable does not alter the main conclusions drawn from the regression results. 22 7). Because the dependent variable is the log of welfare, each coefficients can be interpreted as the approximate percentage difference in per capita consumption associated with a particular category, compared to the omitted category, holding the other included variables fixed. The regressions confirm that, conditional on other characteristics, welfare is strongly and positively correlated with urban residence, having two or fewer children in the household, and educational attainment. When examining the fixed effects regression, persons living in urban areas on average consume or earn approximately 30 percent more per person than those in rural areas. Compared to those living in households with three or more kids, individuals living in households with one to two kids enjoy a 37 percent welfare premium if there are two or fewer adults, and a 24 percent premium if there are three or more adults.31 With respect to educational attainment, those who completed primary school but not secondary consume or earn 24 percent more than those with no education, while the attainment premiums rise to 43 and 68 percent, respectively, for those that complete secondary or tertiary education. These key results also hold when limiting the sample to households and examining how household and head characteristics relate to welfare. When comparing within country and conditioning on the other independent variables, welfare is on average 23 percent higher for urban households. In addition, the welfare of households whose head works outside of agriculture is another 23 percent greater on average than those that work in agriculture.32 For households with two or more adults, having two or more kids is associated with a 50 percent welfare penalty, and heads who have primary, secondary, and tertiary education enjoy a 20, 42, and 74 percent premium in welfare. While the results on gender inequality generally confirm the small differences found in the profile, the female head premium falls significantly when controlling for country fixed effects. In the individual regression, the premium falls from 9 to 1 percent, while in the household regressions, it falls from 19 to 10 percent. These reductions confirm the findings in Table 4, namely that a sizeable portion of the female head premium results from female headship being more prevalent in less poor countries. Meanwhile, the smaller premium in the individual regression suggests that the female head premium can largely be explained by differences in the observed characteristics of household members other than head, such as their gender, age, and educational attainment.   31 These are obtained by subtracting the coefficient for 1-2 kids from those for 3 or more kids in the right column of table 6, which gives -0.24 for more than two adults and -0.37 for two or fewer adults. 32 The estimate of 14 percent is obtained by subtracting -0.04 from 0.19 in Table 7. 23 Table 6: Individual‐level regression  Independent variables  Category  Coefficients  With country      Without country FE  FE  Residence  Urban          Rural  ‐0.30  ‐0.30  Demographic structure        Two or less adults  0 kids          1‐2 kids  ‐0.4  ‐0.36    More than two kids  ‐0.87  ‐0.73  Two or more adults  0 kids  ‐0.28  ‐0.27    1‐2 kids  ‐0.53  ‐0.49    More than two kids  ‐0.87  ‐0.73  Marital Status of adults  Married          Never married  ‐0.02  ‐0.02    Divorced  ‐0.06  ‐0.13    Living together  ‐0.03  ‐0.07    Widowed  0  0.02  Gender of household head  Man          Woman  0.09  0.01  Characteristics of individual          Age group  0‐14            15‐24    ‐0.22  ‐0.20    25‐34    ‐0.12  ‐0.14    35‐44    ‐0.05  ‐0.08    45‐54    ‐0.03  ‐0.07    55‐64    ‐0.02  ‐0.08    65 and up  ‐0.05  ‐0.13  Gender  Man          Woman  0.02  0.02  Educational attainment  No education        Incomplete primary  0.15  0.09  Complete primary or    0.31  0.24  some secondary    Complete secondary  0.52  0.43    Tertiary  0.86  0.68  Constant    2.07  2.19  Adjusted R2     0.43  0.50  Number of countries    78  78  Number of observations     5,971,267  5,971,267  Weighted number of observations ( ‘000s)  4,734,522  4,464,869     Note: Standard errors and significance is not reported because of the absence of PSU identifiers for many counties. In a subsample of 38 countries, all coefficients are statistically significant. Grey shading indicates that the category is excluded from the regression specification. 24 Table 7: Household‐level regression  Independent variables  Category  Coefficients      Without country FE  With country FE  Residence  Urban        Rural  ‐0.22  ‐0.23  Demographic structure        Two or less adults  0 kids        1‐2 kids  ‐0.49  ‐0.42    More than two kids  ‐1.00  ‐0.80  Two or more adults  0 kids  ‐0.29  ‐0.27    1‐2 kids  ‐0.57  ‐0.50    More than two kids  ‐1.04  ‐0.82  Marital Status of head  Married        Never married  ‐0.09  ‐0.02    Divorced  ‐0.10  ‐0.17    Living together  ‐0.18  ‐0.15    Widowed  ‐0.09  ‐0.01  Characteristics of  household    ‐0.08  0.00  head  Age of household head  0‐14      .    15‐24    0.10  0.26    25‐34    0.10  0.23    35‐44    0.15  0.26    45‐54    0.17  0.29    55‐64    0.23  0.33    65 and up  0.24  0.33  Gender of household  Man      head    Woman  0.19  0.08  Educational attainment of  No education      head  Incomplete primary  0.20  0.11  Complete primary or    0.33  0.26  some secondary    Complete secondary  0.48  0.45    Tertiary  0.89  0.73  Type of work of head  Head not working      Head working in    ‐0.15  ‐0.04  agriculture  Head working outside    0.19  0.19  of agriculture    Head 0‐14      Constant    .  .  Adjusted r2     1.84  1.68  Number of countries    0.42  0.5  Number of observations    76  76  Weighted number of     1,718,873  1,718,873  observations  ( ‘000s)  Note: All coefficients are statistically significant at 1 percent confidence level. Standard errors are not reported because they are not adjusted for sample survey design due to missing PSU identifiers for many counties. In a subsample of 38 countries, all coefficients are statistically significant. Grey shading indicates that the category is excluded from the regression specification or omitted due to multicollinearity. 25 4. Robustness checks 4.1. Sensitivity to lining up methods The results reported above are based on the selected line-up method (Method 1), which adjusts the population using the UNDESA projections and redraws the poverty line in each survey to match the World Bank’s published poverty estimates for 2013, as described in section 2.2 above. This method aims to reduce biases that arise from pooling data collected in different years and ensures consistency with the existing published estimates. However, the method relies on a line-up procedure that imposes strong assumptions. These assumptions have not to our knowledge been systematically validated across several countries, and there is no hard evidence that applying the procedure provides a more accurate poverty profile for 2013 than simply pooling data for multiple years. Therefore, as an initial robustness check, we report selected key results under two other line- up methods: one that adjusts both survey weights and the welfare aggregate as described in section 2.2 above, and uses the international poverty lines $1.90 and $3.10 (Method 2); and another that adjusts only the sample survey weights to match the UNDESA population projections (Method 3). In general, the choice of method makes little difference to the estimates. The key results, with respect to the share of the extreme poor belonging to different groups, changes by at most one to two a percentage points (Table 8). The results that are mildly sensitive to the choice of lineup method are the share of the poor working in agriculture and the share of the poor with no education, but even in these cases the changes barely exceed 2 percentage points and do not alter the qualitative nature of the results. This comparison does, however, illustrate that replicating the line- up procedure used to generate the published estimates is not necessarily straightforward, and that the nature of the line-up method usually has small but noticeable impacts on the profile. Table 8: Sensitivity to line‐up method  Method 1 (Use published  Method 2 (Adjust welfare  Method 3 (Adjust    estimates)  and population)   population only)   Percent of extreme poor in  80.1  80.0  80.3  rural areas  Percent of poor working  64.6  62.7  63.1  adults in agriculture  Percent of extreme poor  44.2  43.0  42.5  being 0‐14 years old  Percent of poor adults with no  39.1  37.4  37.7  education     Methodological details         Year   2013  2013  2009‐2014 mixed  Derived from PovcalNet  Poverty lines  $1.90 and $3.10  $1.90 and $3.10  poverty rates  Adjusted using national  Adjusted using national  Welfare aggregate  account data on real  account data on real  None  adjustment  consumption or GDP  consumption or GDP growth  growth  Population        26 4.2.Sensitivity to spatial deflation As mentioned above, the headline result that 80 percent of the extreme poor live in rural areas may be an overestimate. In Africa and much of South Asia, where the majority of the extreme poor live, welfare aggregates are not consistently spatially deflated when estimating international poverty rates. Since the cost of living is typically lower in rural areas, failure to account for regional price differences will overstate the share of poor that are rural. How much would using nominal welfare aggregates affect the share of the poor that is rural? To get a sense of this, we analyze data from countries in South Asia.33 For the purposes of this exercise, urban and rural India are considered to be separate countries, because each uses a different PPP exchange rate. As expected, using spatially deflated aggregates lowers poverty in rural areas, and within urban and rural India, reverses the urban-rural gap in the poverty rate (Figure 23). However, using deflated rather than nominal aggregates only moderately decreases the share of the poor in rural areas (Figure 24). In Bangladesh and Nepal, for example, deflating the welfare aggregate using regional poverty lines would cause the share of the poor that are rural to fall 6 to 7 percentage points. Figure 23: Sensitivity of urban/rural poverty rate to spatial price adjustment  Extreme poverty rate (%) 60 Urban Nominal Urban Spatially deflated 51.6 50 46.3 Rural Nominal Rural Spatially deflated 40 29.9 30 24.8 23.4 20.8 20.0 18.9 20 17.0 16.3 10.2 11.7 10.0 9.1 8.9 10 5.8 4.7 4.8 3.43.1 3.9 4.3 2.3 0.20.2 0.00.0 0.00.0 0.30.4 2.0 0 Bangladesh Bhutan India Rural India Urban Sri Lanka Maldives Nepal Pakistan 33 Data on spatial deflation was conveniently available for South Asia. Figures for Bangladesh are based on 2005 PPPs, and spatial deflators for Bangladesh are based on district level poverty lines. 27 Figure 24: Sensitivity of the urban‐rural gap to spatial price adjustment  Share of the extreme poor by urban‐rural 100 80 60 97.3 96.9 97.1 95.9 95.5 87.4 88.6 84.6 40 81.3 80.7 66.7 54.7 20 0 Nominal Spatially Nominal Spatially Nominal Spatially Nominal Spatially Nominal Spatially Nominal Spatially deflated deflated deflated deflated deflated deflated Bangladesh Bhutan Sri Lanka Maldives Nepal Pakistan 2010 2012 2012 2009 2010 2013 Rural Urban How much could this bias affect the global estimate of the share of the poor living in rural areas? An important consideration is that a sizeable share – about 39 percent -- of the extreme poor live in India, China, and Indonesia. These three countries use separate urban and rural PPP conversion factors for calculating regional and international poverty estimates. Because of this, the share of poor that is rural in these countries is not overestimated, even when using nominal welfare aggregates, because any differences in prices between urban and rural areas should be captured by the PPP conversion factor. An upward bias of about 6 to 7 percentage points that applies to 60 percent of the extreme poor would lead the estimate to overestimate the share of extreme poor by about 4 percentage points, modestly reducing the share of the extreme poor in rural areas only from about 80 to 76 percent. 4.3.Sensitivity to missing data The degree of missing data in the GMD varies by variable. (Table 9). The variable most severely affected by missing data is sector of work. This variable is reported in only 64 out the 89 countries in the sample, representing 2.1 billion of the full sample’s 5.25 billion people. For the purposes of this exercise, we consider education across all ages, which is reported in 83 out the 89 countries. Education is not asked for very young children, however, with the cutoff age varying across countries. The education profile, therefore, only represents 4.7 of the 5.25 billion people covered by the full sample. The results reported above use a separate sample for each profiling variable. To see if using a fixed sample would significantly alter the results, we construct key profile statistics based on two fixed 28 samples and comparing them with the results from the profile-specific samples reported above. The first fixed sample includes only observations that have full information on three key variables, urban/rural, age, and education. This sample includes 6.8 million of the original 7.7 million observations, and covers 81 countries. The second fixed sample additionally drops all observations that are missing sectoral employment, which includes all children under 15 and all adults that do not work. Because of this, the sample falls to 1.6 million observations, covering 59 countries (see Table 9). Table 9: Profiling‐variable specific samples and fixed samples    Number of  Number of  Weighted population  countries  observations  Full sample   89  5,249,087,488  7,657,672  Variable‐specific sample        Urban/rural residence   87  5,245,339,648  7,603,967  Sector of work    64  2,087,739,904  2,876,159  Age   89  5,249,087,488  7,657,672  Education (all ages)   83  4,701,530,624  6,859,233  Fixed samples  .  .  .  For urban/rural, age, and  81  4,698,225,664  6,810,288  education   For urban/rural, age, education  59  1,557,865,216  2,613,824  and sector of work    The concentration of poverty in rural areas is robust to the choice of samples (Table 10). The proportion of the extreme poor living in rural areas varies by 0.6 or 1.7 percentage points in absolute terms, which is 0.7 or 1.9 percent in relative terms, when one switches from the variable- specific sample to the fixed samples. It is particularly reassuring that the share of the poor in the rural sector changes only slightly (from 80.1 to 81.8 percent) even when the sample size is drastically reduced from 89 to the 59 countries for which all variables, including sector of work, are available. Table 10: Sensitivity to missing data  Fixed sample 2: for  Fixed sample 1: for  Varying  urban/rural, age,    urban/rural, age,  samples    education, and  and education   employment sector  Percent of extreme poor in rural areas  80.1  79.5  81.8  Percent of extreme poor working adults in  64.6  N/A  67.9  agriculture  Percent of extreme poor adults 0‐14 years old   44.2  40.9  N/A  Percent of extreme poor of all ages with no  43.7  43.7  31.5  formal education   29 The share of the poor that work outside agriculture, that are children, and have no formal education are all noticeably lower in the fixed samples. For example, the share of the poor working outside agriculture drops 3.3 percentage points, from 35.4 to 32.1, when limiting to observations where urban/rural and education are non-missing. The share of the poor under the age of 15 drops by 3.3 percentage points in the first fixed sample, because the sample differentially excludes poor children that did not report educational attainment. The proportion of poor adults with no schooling decreases 12 percentage points from 43.7 percent to 31.5 percent in the second fixed sample, which excludes all children and non-working adults. This sensitivity indicates that the pattern of missing variables is not independent across variables. Conducting the analysis on a fixed subsample for which all data are present would therefore distort the findings.34 5. Conclusions Using harmonized household data from 89 countries, this paper provides an overarching demographic profile of the global extreme poor and moderate poor in 2013, shedding light on where they live and who they are, and the extent to which they work in the agricultural sector. Not only is this the most updated and most comprehensive profile of the poor in terms of global coverage, but the analysis breaks new ground by examining the characteristics of the moderate poor, and presenting conditional correlations between demographic variables and household welfare. Five main conclusions emerge. First, both the extreme and moderate poor are rural and young, and mostly live in larger households with more children. More than four in five of the extreme poor live in rural areas, which is slightly more than previously thought. Moreover, 44 percent of the extreme poor are children under 15, and households with three or more children comprise nearly 60 percent of the extreme poor. The prevalence of child poverty raises the prospect of long-term consequences on the physical and intellectual development of poor children, which could in turn impede their future earning capacity. The gender gap in poverty is not apparent in traditional poverty measures, because poverty is measured based on household per capita welfare, whereas gender disparities are most apparent in individual-level indicators such as education, decision making power and labor market outcomes. Male headed households, however, are disproportionately likely to be poor, at least in the lower middle-income countries that contain the majority of the extreme poor. In general, these findings emphasize the potential benefits of programs that directly or indirectly support children, large households, and rural households, as well as indirect measures to reduce the future incidence of child poverty. Shifting out of low productivity agricultural work, while important, is not sufficient to escape poverty. Consistent with the clustering of the poor in rural areas, poor workers are much more likely than non-poor workers to make their living from agricultural work. Yet a substantial proportion of the extreme and moderate poor who live in rural areas – 24 and 40 percent, respectively – work outside the agricultural sector. Within the scope of this study, it is unclear why 34 Multiple imputation methods, which can address problems created by non-random patterns of missing survey data, can be explored in the future (Rubin, 2004). 30 non-agricultural labors in rural areas remain poor, and more broadly, what determines inequality within rural areas. Further research is therefore needed to better understand, for example, how industry of work relates to economic welfare, and which types of jobs have led to larger reductions in rural extreme poverty. Third, poverty and educational attainment are strongly and negatively correlated, among both children and adults. Most notably, although only 15 percent of adults have no formal education, nearly 25 percent of them live in extreme or moderate poverty, and another 33 percent live in moderate poverty. At the same time, a sizable proportion of extremely and moderately poor adults, 27 percent and 38 percent, respectively, have at least some secondary education. Not surprisingly, those with tertiary education are almost exclusive non-poor. The different education profiles of extremely poor, moderately poor, and non-poor adults highlight the role of education in driving poverty reduction. Nonetheless, graduating from primary school, while important, far from ensures an escape from poverty. Fourth, despite similarities in terms of age, household composition, and residential sector, there are two noticeable differences between the extreme poor and the moderate poor. First, the moderate poor are much less likely to make their living from agriculture. Second, they are significantly more likely to have graduated from primary school, but not secondary school. If one considers moderate poverty as a transition stage between extreme poverty and the absence of deprivation, these differences point to non-farm employment and basic education as potential pathways to improve living standards of the extreme poor. Finally, many of the most striking differences in the demographic profiles of the poor remain when controlling for differences in various characteristics as well as country fixed effects. Conditional on other characteristics, living in an urban area, having fewer than three children, and having greater educational attainment have a particularly strong and positive association with economic welfare within countries. The main findings on the nature of poverty are generally robust to a variety of methods. The share that is rural seems to vary little regardless of the lineup method or sample used. Furthermore, calculations based on South Asian data suggest that the share of the poor in rural areas would remain high even if welfare aggregates were spatially deflated in all countries. On the other hand, the shares of the poor that are young and less educated, and working in agriculture are mildly sensitive to how the welfare aggregates for each country are lined up to 2013, with differences of one to two percentage points. Therefore, further research could usefully document in detail the exact procedure used to line up the estimates and explore whether lining up improves the accuracy of profiles. As with all household survey data, there are missing values because information is occasionally not reported or processed. Because patterns of missing values vary across different variables, restricting the analysis to a common subsample would distort the key findings. This study is a first step towards exploiting the World Bank’s unique inventory of household survey data to better understand the poor’s living conditions, earning capacity, and economic constraints at a large scale. The database can also help fill in other important knowledge gaps on global poverty. These include a more detailed look at the relationship between labor market 31 outcomes and welfare, and how different types of households contribute to economic inequality. Furthermore, when past surveys are added to the GMD for all regions, additional analysis can document which groups of people have exited extreme poverty during the past decade, and how changes in labor market outcomes, educational attainment, and urbanization have contributed to recent reductions in extreme poverty. 32 References Banerjee, AV & Duflo, E 2007, The economic lives of the poor, Journal of Economic Perspectives, vol. 21, no. 1, pp.141–167. Bargain, Olivier, Olivier Donni, and Prudence Kwenda. "Intrahousehold distribution and poverty: Evidence from Cote d'Ivoire." Journal of Development Economics 107 (2014): 262-276. Batana, Y, Bussolo, M & Cockburn, J 2013, Global extreme poverty rates for children, adults and the elderly, Economics Letters, vol. 120, pp. 405-407. Beegle, Kathleen, Christiaensen, Luc, Dabalen, Andrew, and Gaddis, Isis. (2016). Poverty in a rising Africa. World Bank Publications. Chen, Shaohua, and Martin Ravallion. "The developing world is poorer than we thought, but no less successful in the fight against poverty." Quarterly Journal of Economics 125.4 (2010). Deaton, Angus. "Measuring and understanding behavior, welfare, and poverty." American Economic Review 106.6 (2016): 1221-43. Dunbar, Geoffrey R., Arthur Lewbel, and Krishna Pendakur. "Children's resources in collective households: identification, estimation, and an application to child poverty in Malawi." The American Economic Review 103.1 (2013): 438-471. Ferreira, F, Chen, S, Dabalen, A, Dikhanov, Y, Hamadeh, N, Jolliffe, D, Narayan, A, Prydz, EB, Revenga, A, Sangraula, P, Serajuddin, U & Yoshida, N 2015, ‘A global count of the extreme poor in 2012 - Data issues, methodology and initial results’, Policy Research Working Paper No. 7432, Development Research Group, The World Bank, Washington DC.  Gindling, T. H., and David Newhouse. "Self-employment in the developing world." World Development 56 (2014): 313-331. Harris, John R., and Michael P. Todaro. "Migration, unemployment and development: a two- sector analysis." The American economic review (1970): 126-142. Jerven, Morten. "Comparability of GDP estimates in Sub‐Saharan Africa: The effect of revisions in sources and methods since structural adjustment." Review of Income and Wealth 59.S1 (2013): S16-S36. Jolliffe, Dean. A measured approach to ending poverty and boosting shared prosperity: concepts, data, and the twin goals. World Bank Publications, 2014. Lewis, W. Arthur. "Economic development with unlimited supplies of labour." The Manchester School 22.2 (1954): 139-191. Newhouse, David, Pablo Suarez-Becerra, Martin C. Evans, and Data for Goals Group, 2016. “New Estimates of Extreme Poverty for Children.” Policy Research Working Paper no. 7845, World Bank, Washington, DC. 33 Olinto, P, Beegle, K, Sobrado, C & Uemastu, H 2013, ‘The state of the poor: Where are the poor, where is extreme poverty harder to end, and what is the current profile of the world’s poor?’, Economic Premise Note Series No. 125, Poverty Reduction and Economic Management Network, The World Bank, Washington DC. Ravallion, Martin. "On testing the scale sensitivity of poverty measures." Economics Letters 137 (2015): 88-90. Rubin, Donald B. Multiple imputation for nonresponse in surveys. Vol. 81. John Wiley & Sons, 2004. WDI 2016, ‘World Development Indicators 2016: Featuring the Sustainable Development Goals’, World Bank, Washington DC. Zhang, Chunni, et al. "Are poverty rates underestimated in China? New evidence from four recent surveys." China Economic Review 31 (2014): 410-425. 34 Appendix 1: List of countries and survey years Country name  Region  Income group  Survey year  Welfare measure  Inflator factor  Cambodia  East Asia and Pacific  Low income  2012  Consumption  HHFCE  China  East Asia and Pacific  Upper middle income  2013  Consumption  HHFCE  Indonesia  East Asia and Pacific  Lower middle income  2011, 2014  Consumption  HHFCE  Lao PDR  East Asia and Pacific  Lower middle income  2012  Consumption  HHFCE  Mongolia  East Asia and Pacific  Upper middle income  2012  Consumption  HHFCE  Papua New Guinea  East Asia and Pacific  Lower middle income  2009  Consumption  PCGDP  Philippines  East Asia and Pacific  Lower middle income  2012  Income  HHFCE  Thailand  East Asia and Pacific  Upper middle income  2012  Consumption  HHFCE  Tonga  East Asia and Pacific  Upper middle income  2009  Consumption  PCGDP  Vanuatu  East Asia and Pacific  Lower middle income  2010  Consumption  HHFCE  Vietnam  East Asia and Pacific  Lower middle income  2012, 2014  Consumption  HHFCE  Albania  Europe and Central Asia  Upper middle income  2012  Consumption  HHFCE  Armenia  Europe and Central Asia  Lower middle income  2013  Consumption  HHFCE  Bulgaria  Europe and Central Asia  Upper middle income  2013  Income  HHFCE  Croatia  Europe and Central Asia  High income  2013  Income  HHFCE  Czech Republic  Europe and Central Asia  High income  2013  Income  HHFCE  Estonia  Europe and Central Asia  High income  2013  Income  HHFCE  Georgia  Europe and Central Asia  Upper middle income  2013  Consumption  HHFCE  Hungary  Europe and Central Asia  High income  2013  Income  HHFCE  Kazakhstan  Europe and Central Asia  Upper middle income  2013  Consumption  HHFCE  Kosovo  Europe and Central Asia  Lower middle income  2013  Consumption  HHFCE  Kyrgyz Republic  Europe and Central Asia  Lower middle income  2012  Consumption  HHFCE  Latvia  Europe and Central Asia  High income  2013  Income  HHFCE  Lithuania  Europe and Central Asia  High income  2013  Income  HHFCE  Moldova  Europe and Central Asia  Lower middle income  2013  Consumption  HHFCE  Montenegro  Europe and Central Asia  Upper middle income  2013  Consumption  HHFCE  Poland  Europe and Central Asia  High income  2013  Income  HHFCE  Romania  Europe and Central Asia  Upper middle income  2013  Income  HHFCE  Russian Federation  Europe and Central Asia  High income  2012  Consumption  HHFCE  Serbia  Europe and Central Asia  Upper middle income  2013  Consumption  HHFCE  Slovak Republic  Europe and Central Asia  High income  2013  Income  HHFCE  Slovenia  Europe and Central Asia  High income  2013  Income  HHFCE  Tajikistan  Europe and Central Asia  Low income  2009  Consumption  HHFCE  Turkey  Europe and Central Asia  Upper middle income  2012  Consumption  HHFCE  Ukraine  Europe and Central Asia  Lower middle income  2013  Consumption  HHFCE  Argentina  Latin America and Caribbean  High income  2012, 2014  Income  HHFCE  Bolivia  Latin America and Caribbean  Lower middle income  2012, 2014  Income  HHFCE  Brazil  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  Chile  Latin America and Caribbean  High income  2013  Income  HHFCE  Colombia  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  35 Costa Rica  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  Dominican Republic  Latin America and Caribbean  Upper middle income  2013  Income  HHFCE  Ecuador  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  El Salvador  Latin America and Caribbean  Lower middle income  2012, 2014  Income  HHFCE  Guatemala  Latin America and Caribbean  Lower middle income  2011, 2014  Income  HHFCE  Haiti  Latin America and Caribbean  Low income  2012  Income  HHFCE  Honduras  Latin America and Caribbean  Lower middle income  2013  Income  HHFCE  Mexico  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  Nicaragua  Latin America and Caribbean  Lower middle income  2009, 2014  Income  HHFCE  Panama  Latin America and Caribbean  Upper middle income  2012  Income  HHFCE  Paraguay  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  Peru  Latin America and Caribbean  Upper middle income  2012, 2014  Income  HHFCE  Uruguay  Latin America and Caribbean  High income  2012, 2014  Income  HHFCE  Djibouti  Middle East and North Africa  Lower middle income  2012  Consumption  HHFCE  Tunisia  Middle East and North Africa  Upper middle income  2010  Consumption  HHFCE  West Bank and Gaza  Middle East and North Africa  Lower middle income  2009  Consumption  HHFCE  Bangladesh  South Asia  Low income  2010  Consumption  HHFCE  Bhutan  South Asia  Lower middle income  2012  Consumption  HHFCE  India  South Asia  Lower middle income  2011  Consumption  HHFCE  Maldives  South Asia  Upper middle income  2009  Consumption  PCGDP  Nepal  South Asia  Low income  2010  Consumption  HHFCE  Pakistan  South Asia  Lower middle income  2013  Consumption  HHFCE  Sri Lanka  South Asia  Lower middle income  2012  Consumption  HHFCE  Botswana  Sub‐Saharan Africa  Upper middle income  2009  Consumption  PCGDP  Burkina Faso  Sub‐Saharan Africa  Low income  2009  Consumption  PCGDP  Chad  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  Congo, Dem. Rep.  Sub‐Saharan Africa  Low income  2012  Consumption  PCGDP  Congo, Rep.  Sub‐Saharan Africa  Lower middle income  2011  Consumption  PCGDP  Ethiopia  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  Guinea  Sub‐Saharan Africa  Low income  2012  Consumption  PCGDP  Guinea‐Bissau  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  Lesotho  Sub‐Saharan Africa  Lower middle income  2010  Consumption  PCGDP  Madagascar  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  Malawi  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  Mali  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  Mauritius  Sub‐Saharan Africa  Upper middle income  2012  Consumption  PCGDP  Niger  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  Nigeria  Sub‐Saharan Africa  Lower middle income  2010  Consumption  PCGDP  Rwanda  Sub‐Saharan Africa  Low income  2010  Consumption  PCGDP  São Tomé and Príncipe  Sub‐Saharan Africa  Lower middle income  2010  Consumption  PCGDP  Senegal  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  Sierra Leone  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  South Africa  Sub‐Saharan Africa  Upper middle income  2010  Consumption  PCGDP  36 Sudan  Sub‐Saharan Africa  Lower middle income  2009  Consumption  PCGDP  Swaziland  Sub‐Saharan Africa  Lower middle income  2009  Consumption  PCGDP  Tanzania  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  Togo  Sub‐Saharan Africa  Low income  2011  Consumption  PCGDP  Uganda  Sub‐Saharan Africa  Low income  2012  Consumption  PCGDP  Zambia  Sub‐Saharan Africa  Lower middle income  2010  Consumption  PCGDP  37 Appendix 2: Lining up surveys to 2013 The exact implementation of the method used to line up surveys to 2013 varies across countries, depending on the availability of data from different years. If a country’s latest survey was conducted in 2013, that survey was used without adjusting the welfare aggregate. If a country’s most recent survey is before 2013, the latest survey is lined up forward to 2013. To do this, the World Bank poverty rates for 2013, taken from PovcalNet, are applied to the household survey data.35 If surveys are available both from before and after 2013, the surveys closest to 2013 on either side are lined up, backward and forward, to reduce potential extrapolation bias. Disaggregated poverty rates and the number of poor are then calculated as the weighted average of the estimates, where the weights are the distance between the survey years and 2013. This sandwiching procedure mirrors the procedure used generate the poverty estimates reported on PovcalNet, and was applied to 15 of the 89 countries in the sample.36 We examine two alternative approaches to lining up the surveys. The first approach brings population to its 2013 level but does not adjust the welfare aggregate, thus providing a poverty profile that pools data from different years. As noted in the text, this tends to give more weight to countries with older surveys. The second alternative approach is a variant of the official line-up methodology used by the World Bank. First, the welfare aggregate in each country is multiplied by a scale factor to account for changes in welfare between the survey year and 2013. For 60 of the 89 countries, the scale factor is the real growth between the survey year and 2012 in per capita household final consumption expenditure (HFCE), taken from the World Development Indicators Database. For the others, mostly in Sub-Saharan Africa, where HFCE is not available, real growth in GDP per capita in local currency units is used instead (see Appendix 1). Even after applying this line-up procedure, there remain discrepancies in the 2013 poverty rates between the lined-up GMD sample and the poverty rates published by PovcalNet. As displayed in Figure 25, the magnitude of the difference is less than one percentage point for 55 countries, and between 1 and 3 percentage points for 20 countries. Larger discrepancies are rarer, in only seven cases was the difference between 3 and 7 percentage points, and a further five countries have large differences of 7 percentage points or greater. These discrepancies warrant further investigation and likely arise from the use of different vintages of inflation and real economic growth data obtained from external sources such as the World Development Indicators, which are updated quarterly. The exercise illustrates however that it is not necessarily straightforward to replicate the World Bank’s official estimates with publicly available macroeconomic data, even when using the same household data. 35 For India, China, and Indonesia, national poverty rates are used, rather than urban and rural specific poverty rates. 36 These are almost all in Latin America. Namely, Argentina, Bolivia, Brazil, Colombia, Costa Rica, Ecuador, Mexico, Peru, Paraguay, El Salvador, and Uruguay contain surveys from 2012 and 2014, as does Vietnam. Guatemala and Indonesia contain surveys from 2011 and 2014. 38 Figure 25: Distribution of the discrepancy in national poverty rates between the second alternative  approach and PovcalNet  39 Poverty & Equity Global Practice Working Papers (Since July 2014) The Poverty & Equity Global Practice Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. This series is co‐published with the World Bank Policy Research Working Papers (DECOS). It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. For the latest paper, visit our GP’s intranet at http://POVERTY. 1 Estimating poverty in the absence of consumption data: the case of Liberia Dabalen, A. L., Graham, E., Himelein, K., Mungai, R., September 2014 2 Female labor participation in the Arab world: some evidence from panel data in Morocco Barry, A. G., Guennouni, J., Verme, P., September 2014 3 Should income inequality be reduced and who should benefit? redistributive preferences in Europe and Central Asia Cojocaru, A., Diagne, M. F., November 2014 4 Rent imputation for welfare measurement: a review of methodologies and empirical findings Balcazar Salazar, C. F., Ceriani, L., Olivieri, S., Ranzani, M., November 2014 5 Can agricultural households farm their way out of poverty? Oseni, G., McGee, K., Dabalen, A., November 2014 6 Durable goods and poverty measurement Amendola, N., Vecchi, G., November 2014 7 Inequality stagnation in Latin America in the aftermath of the global financial crisis Cord, L., Barriga Cabanillas, O., Lucchetti, L., Rodriguez‐Castelan, C., Sousa, L. D., Valderrama, D. December 2014 8 Born with a silver spoon: inequality in educational achievement across the world Balcazar Salazar, C. F., Narayan, A., Tiwari, S., January 2015 Updated on December 2016 by POV GP KL Team | 1 9 Long‐run effects of democracy on income inequality: evidence from repeated cross‐sections Balcazar Salazar,C. F., January 2015 10 Living on the edge: vulnerability to poverty and public transfers in Mexico Ortiz‐Juarez, E., Rodriguez‐Castelan, C., De La Fuente, A., January 2015 11 Moldova: a story of upward economic mobility Davalos, M. E., Meyer, M., January 2015 12 Broken gears: the value added of higher education on teachers' academic achievement Balcazar Salazar, C. F., Nopo, H., January 2015 13 Can we measure resilience? a proposed method and evidence from countries in the Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 14 Vulnerability to malnutrition in the West African Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 15 Economic mobility in Europe and Central Asia: exploring patterns and uncovering puzzles Cancho, C., Davalos, M. E., Demarchi, G., Meyer, M., Sanchez Paramo, C., January 2015 16 Managing risk with insurance and savings: experimental evidence for male and female farm managers in the Sahel Delavallade, C., Dizon, F., Hill, R., Petraud, J. P., el., January 2015 17 Gone with the storm: rainfall shocks and household well‐being in Guatemala Baez, J. E., Lucchetti, L., Genoni, M. E., Salazar, M., January 2015 18 Handling the weather: insurance, savings, and credit in West Africa De Nicola, F., February 2015 19 The distributional impact of fiscal policy in South Africa Inchauste Comboni, M. G., Lustig, N., Maboshe, M., Purfield, C., Woolard, I., March 2015 20 Interviewer effects in subjective survey questions: evidence from Timor‐Leste Himelein, K., March 2015 21 No condition is permanent: middle class in Nigeria in the last decade Corral Rodas, P. A., Molini, V., Oseni, G. O., March 2015 22 An evaluation of the 2014 subsidy reforms in Morocco and a simulation of further reforms Verme, P., El Massnaoui, K., March 2015 Updated on December 2016 by POV GP KL Team | 2 23 The quest for subsidy reforms in Libya Araar, A., Choueiri, N., Verme, P., March 2015 24 The (non‐) effect of violence on education: evidence from the "war on drugs" in Mexico Márquez‐Padilla, F., Pérez‐Arce, F., Rodriguez Castelan, C., April 2015 25 “Missing girls” in the south Caucasus countries: trends, possible causes, and policy options Das Gupta, M., April 2015 26 Measuring inequality from top to bottom Diaz Bazan, T. V., April 2015 27 Are we confusing poverty with preferences? Van Den Boom, B., Halsema, A., Molini, V., April 2015 28 Socioeconomic impact of the crisis in north Mali on displaced people (Available in French) Etang Ndip, A., Hoogeveen, J. G., Lendorfer, J., June 2015 29 Data deprivation: another deprivation to end Serajuddin, U., Uematsu, H., Wieser, C., Yoshida, N., Dabalen, A., April 2015 30 The local socioeconomic effects of gold mining: evidence from Ghana Chuhan-Pole, P., Dabalen, A., Kotsadam, A., Sanoh, A., Tolonen, A.K., April 2015 31 Inequality of outcomes and inequality of opportunity in Tanzania Belghith, N. B. H., Zeufack, A. G., May 2015 32 How unfair is the inequality of wage earnings in Russia? estimates from panel data Tiwari, S., Lara Ibarra, G., Narayan, A., June 2015 33 Fertility transition in Turkey—who is most at risk of deciding against child arrival? Greulich, A., Dasre, A., Inan, C., June 2015 34 The socioeconomic impacts of energy reform in Tunisia: a simulation approach Cuesta Leiva, J. A., El Lahga, A., Lara Ibarra, G., June 2015 35 Energy subsidies reform in Jordan: welfare implications of different scenarios Atamanov, A., Jellema, J. R., Serajuddin, U., June 2015 36 How costly are labor gender gaps? estimates for the Balkans and Turkey Cuberes, D., Teignier, M., June 2015 37 Subjective well‐being across the lifespan in Europe and Central Asia Bauer, J. M., Munoz Boudet, A. M., Levin, V., Nie, P., Sousa‐Poza, A., July 2015 Updated on December 2016 by POV GP KL Team | 3 38 Lower bounds on inequality of opportunity and measurement error Balcazar Salazar, C. F., July 2015 39 A decade of declining earnings inequality in the Russian Federation Posadas, J., Calvo, P. A., Lopez‐Calva, L.‐F., August 2015 40 Gender gap in pay in the Russian Federation: twenty years later, still a concern Atencio, A., Posadas, J., August 2015 41 Job opportunities along the rural‐urban gradation and female labor force participation in India Chatterjee, U., Rama, M. G., Murgai, R., September 2015 42 Multidimensional poverty in Ethiopia: changes in overlapping deprivations Yigezu, B., Ambel, A. A., Mehta, P. A., September 2015 43 Are public libraries improving quality of education? when the provision of public goods is not enough Rodriguez Lesmes, P. A., Valderrama Gonzalez, D., Trujillo, J. D., September 2015 44 Understanding poverty reduction in Sri Lanka: evidence from 2002 to 2012/13 Inchauste Comboni, M. G., Ceriani, L., Olivieri, S. D., October 2015 45 A global count of the extreme poor in 2012: data issues, methodology and initial results Ferreira, F.H.G., Chen, S., Dabalen, A. L., Dikhanov, Y. M., Hamadeh, N., Jolliffe, D. M., Narayan, A., Prydz, E. B., Revenga, A. L., Sangraula, P., Serajuddin, U., Yoshida, N., October 2015 46 Exploring the sources of downward bias in measuring inequality of opportunity Lara Ibarra, G., Martinez Cruz, A. L., October 2015 47 Women’s police stations and domestic violence: evidence from Brazil Perova, E., Reynolds, S., November 2015 48 From demographic dividend to demographic burden? regional trends of population aging in Russia Matytsin, M., Moorty, L. M., Richter, K., November 2015 49 Hub‐periphery development pattern and inclusive growth: case study of Guangdong province Luo, X., Zhu, N., December 2015 50 Unpacking the MPI: a decomposition approach of changes in multidimensional poverty headcounts Rodriguez Castelan, C., Trujillo, J. D., Pérez Pérez, J. E., Valderrama, D., December 2015 51 The poverty effects of market concentration Rodriguez Castelan, C., December 2015 52 Can a small social pension promote labor force participation? evidence from the Colombia Mayor program Pfutze, T., Rodriguez Castelan, C., December 2015 Updated on December 2016 by POV GP KL Team | 4 53 Why so gloomy? perceptions of economic mobility in Europe and Central Asia Davalos, M. E., Cancho, C. A., Sanchez, C., December 2015 54 Tenure security premium in informal housing markets: a spatial hedonic analysis Nakamura, S., December 2015 55 Earnings premiums and penalties for self‐employment and informal employees around the world Newhouse, D. L., Mossaad, N., Gindling, T. H., January 2016 56 How equitable is access to finance in turkey? evidence from the latest global FINDEX Yang, J., Azevedo, J. P. W. D., Inan, O. K., January 2016 57 What are the impacts of Syrian refugees on host community welfare in Turkey? a subnational poverty analysis Yang, J., Azevedo, J. P. W. D., Inan, O. K., January 2016 58 Declining wages for college‐educated workers in Mexico: are younger or older cohorts hurt the most? Lustig, N., Campos‐Vazquez, R. M., Lopez‐Calva, L.‐F., January 2016 59 Sifting through the Data: labor markets in Haiti through a turbulent decade (2001‐2012) Rodella, A.‐S., Scot, T., February 2016 60 Drought and retribution: evidence from a large‐scale rainfall‐indexed insurance program in Mexico Fuchs Tarlovsky, Alan., Wolff, H., February 2016 61 Prices and welfare Verme, P., Araar, A., February 2016 62 Losing the gains of the past: the welfare and distributional impacts of the twin crises in Iraq 2014 Olivieri, S. D., Krishnan, N., February 2016 63 Growth, urbanization, and poverty reduction in India Ravallion, M., Murgai, R., Datt, G., February 2016 64 Why did poverty decline in India? a nonparametric decomposition exercise Murgai, R., Balcazar Salazar, C. F., Narayan, A., Desai, S., March 2016 65 Robustness of shared prosperity estimates: how different methodological choices matter Uematsu, H., Atamanov, A., Dewina, R., Nguyen, M. C., Azevedo, J. P. W. D., Wieser, C., Yoshida, N., March 2016 66 Is random forest a superior methodology for predicting poverty? an empirical assessment Stender, N., Pave Sohnesen, T., March 2016 67 When do gender wage differences emerge? a study of Azerbaijan's labor market Tiongson, E. H. R., Pastore, F., Sattar, S., March 2016 Updated on December 2016 by POV GP KL Team | 5 68 Second‐stage sampling for conflict areas: methods and implications Eckman, S., Murray, S., Himelein, K., Bauer, J., March 2016 69 Measuring poverty in Latin America and the Caribbean: methodological considerations when estimating an empirical regional poverty line Gasparini, L. C., April 2016 70 Looking back on two decades of poverty and well‐being in India Murgai, R., Narayan, A., April 2016 71 Is living in African cities expensive? Yamanaka, M., Dikhanov, Y. M., Rissanen, M. O., Harati, R., Nakamura, S., Lall, S. V., Hamadeh, N., Vigil Oliver, W., April 2016 72 Ageing and family solidarity in Europe: patterns and driving factors of intergenerational support Albertini, M., Sinha, N., May 2016 73 Crime and persistent punishment: a long‐run perspective on the links between violence and chronic poverty in Mexico Rodriguez Castelan, C., Martinez‐Cruz, A. L., Lucchetti, L. R., Valderrama Gonzalez, D., Castaneda Aguilar, R. A., Garriga, S., June 2016 74 Should I stay or should I go? internal migration and household welfare in Ghana Molini, V., Pavelesku, D., Ranzani, M., July 2016 75 Subsidy reforms in the Middle East and North Africa Region: a review Verme, P., July 2016 76 A comparative analysis of subsidy reforms in the Middle East and North Africa Region Verme, P., Araar, A., July 2016 77 All that glitters is not gold: polarization amid poverty reduction in Ghana Clementi, F., Molini, V., Schettino, F., July 2016 78 Vulnerability to Poverty in rural Malawi Mccarthy, N., Brubaker, J., De La Fuente, A., July 2016 79 The distributional impact of taxes and transfers in Poland Goraus Tanska, K. M., Inchauste Comboni, M. G., August 2016 80 Estimating poverty rates in target populations: an assessment of the simple poverty scorecard and alternative approaches Vinha, K., Rebolledo Dellepiane, M. A., Skoufias, E., Diamond, A., Gill, M., Xu, Y., August 2016 Updated on December 2016 by POV GP KL Team | 6 81 Synergies in child nutrition: interactions of food security, health and environment, and child care Skoufias, E., August 2016 82 Understanding the dynamics of labor income inequality in Latin America Rodriguez Castelan, C., Lustig, N., Valderrama, D., Lopez‐Calva, L.‐F., August 2016 83 Mobility and pathways to the middle class in Nepal Tiwari, S., Balcazar Salazar, C. F., Shidiq, A. R., September 2016 84 Constructing robust poverty trends in the Islamic Republic of Iran: 2008‐14 Salehi Isfahani, D., Atamanov, A., Mostafavi, M.‐H., Vishwanath, T., September 2016 85 Who are the poor in the developing world? Newhouse, D. L., Uematsu, H., Doan, D. T. T., Nguyen, M. C., Azevedo, J. P. W. D., Castaneda Aguilar, R. A., October 2016 86 New estimates of extreme poverty for children Newhouse, D. L., Suarez Becerra, P., Evans, M. C., October 2016 87 Shedding light: understanding energy efficiency and electricity reliability Carranza, E., Meeks, R., November 2016 88 Heterogeneous returns to income diversification: evidence from Nigeria Siwatu, G. O., Corral Rodas, P. A., Bertoni, E., Molini, V., November 2016 89 How liberal is Nepal's liberal grade promotion policy? Sharma, D., November 2016 90 CPI bias and its implications for poverty reduction in Africa Dabalen, A. L., Gaddis, I., Nguyen, N. T. V., December 2016 91 Pro-growth equity: a policy framework for the twin goals Lopez-Calva, L. F., Rodriguez Castelan, C., November 2016 92 Building an ex ante simulation model for estimating the capacity impact, benefit incidence, and cost effectiveness of child care subsidies: an application using provider‐level data from Turkey Aran, M. A., Munoz Boudet, A., Aktakke, N., December 2016 93 Vulnerability to drought and food price shocks: evidence from Ethiopia Porter, C., Hill, R., December 2016 94 Job quality and poverty in Latin America Rodriguez Castelan, C., Mann, C. R., Brummund, P., December 2016 Updated on December 2016 by POV GP KL Team | 7 For the latest and sortable directory, available on the Poverty & Equity GP intranet site. http://POVERTY WWW.WORLDBANK.ORG/POVERTY Updated on December 2016 by POV GP KL Team | 8