Methodological Innovations in Multidimensional Poverty Measurement Oxford Poverty & Human Development Initiative (OPHI) University of Oxford Rabat, 4 June 2014
Why such interest? Ethics Human lives are battered and diminished in all kinds of different ways. Amartya Sen Overview While assessing quality-of-life requires a plurality of indicators, there are strong demands to develop a single summary measure. Stiglitz Sen Fitoussi Commission Report Effectiveness Acceleration in one goal often speeds up progress in others; to meet MDGs strategically we need to see them together. UNDP 2010 50-country study Management Track progress towards national plan; M&E. Feasibility Surveys; measure deprivations directly; computations
Empirical Motivation: A Better Picture of Poverty EU2020-Europe s Multidimensional measure Income poverty Material Deprivation All 3 deprivations Joblessness Atkinson, A. B., E. Marlier, F. Monatigne, and A. Reinstadler (2010) Income poverty and income inequality, in Income and Living Conditions in 3 Europe, Atkinson and Marlier (eds), Eurostat.
Recent Developments in Multidimensional Poverty Measurement Introduction: The Global Multidimentional Poverty Index (MPI) Robustness tests of multidimensional poverty rankings Leaving no one behind: Inequality among the Poor and Destitution Changes over time of multidimensional poverty.
GLOBAL MPI 2013 METHODOLOGY
What is the MPI? The global MPI is an index of acute multidimensional poverty for over 100 developing countries. It has been reported and updated in UNDP s Human Development Report since 2010. OPHI compute estimates. National MPIs use different specifications and priorities. The MPI methodology was developed by Alkire and Foster (Journal of Public Economics 2011) Robustness tests for the global MPI 2010 are in Alkire and Santos (World Development 2014) Systematic presentation: Multidimensional Poverty: Measurement &Analysis (Oxford University Press, 2015) A new global MPI-2015+ may be used alongside $1.25/day
1. Data: Surveys Note: in 2014 we have updated MPI for 30 countries and 2.5B people. Demographic & Health Surveys (DHS - 51) Multiple Indicator Cluster Surveys (MICS - 30) World Health Survey (WHS 17) Additionally we used 6 special surveys covering urban Argentina (ENNyS), Brazil (PNDS), Mexico (ENSANUT), Morocco (ENNVM), Occupied Palestinian Territory (PAPFAM), and South Africa (NIDS) Constraints: Data are 2002-2011. Not all have precisely the same indicators.
2. MPI Dimensions, Weights & Indicators Note: there are no PPPs for multidimensional poverty as deprivations are measured directly.
3. Identification: Who is poor? People are multidimensionally poor if they are deprived in 33% of the dimensions. Endah s deprivations: 73% 33%
4. What is the MPI? The MPI is one implementation of the first measure of the Alkire & Foster family, M 0. The MPI is the product of two components: MPI = H A 1) Incidence ~ the percentage of people who are disadvantaged, or the headcount ratio H. 2) Intensity of people s deprivation ~ the average share of dimensions in which disadvantaged people are deprived A.
The MPI: High Resolution The MPI can be broken down in different ways: 1. By Headcount to show how many are poor 2. By Dimension to show how people are poor 3. By Intensity to show who has greatest intensity 4. By Sub-group to show how groups vary (in headcount, intensity, and composition) In fact, it is the MPI Plus a dashboard (a set) of consistent subindices that unpack the AF analysis and supply powerful analysis.
GLOBAL MPI 2013 SOME RESULTS
104 Developing Countries: ~ 29 Low Income Countries, (681M), 86% ~ 67 Middle Income Countries, (4634), 93%: ~ 41 Lower Middle Income (2433M) 98% ~ 26 Upper Middle Income (2201M) 89% ~ 8 High Income Countries (43M), of which: ~ 5 OECD (29M) ~ 3 non-oecd (13M) Total Population: 5.4 Billion people Which is 78% of the world s population (population figures from 2010; data from 2002-2011).
Total Population in 104 MPI countries Sub- Saharan Africa 14,3% Europe and Central Asia 7,5% Arab States 4,2% Latin America and Caribbean 9,5% Half of the world s MPI people live in South Asia, and 29% in Sub- Saharan Africa South Asia 29,8% East Asia and Pacific 34,6% MPI poor people by region Europe and Central Asia 0.7% Sub-Saharan Africa 28.90% South Asia 51.3% Arab States 2.12% Latin America & Caribbean 2.2% East Asia & Pacific 14.9%
Most poor people live in middle-income countries. 72% of MPI poor people live in Middle Income Countries Total Population by Income Category High Income 0,8% Low Income 12,7% Lower Middle Income 45,4% Upper Middle Income 41,1% 2010 Population Data MPI Poor Population High Income 0,1% Low Income 27,5% Upper Middle Income 12,3% Lower Middle Income 60,1%
Niger Ethiopia Mali Burundi Burkina Faso Liberia Guinea Somalia Mozambique Sierra Leone Senegal DR Congo Benin Uganda Rwanda Timor-Leste Madagascar Malawi Tanzania Zambia Chad Mauritania Cote d'ivoire Gambia Bangladesh Haiti Togo Nigeria India Cameroon Yemen Pakistan Kenya Lao Cambodia Nepal Republic of Congo Namibia Zimbabwe Lesotho Sao Tome and Principe Honduras Ghana Vanuatu Djibouti Nicaragua Bhutan Guatemala Indonesia Bolivia Swaziland Tajikistan Mongolia Peru Iraq Philippines South Africa Paraguay China Morocco Suriname Guyana Estonia Turkey Egypt Trinidad and Tobago Belize Syrian Arab Republic Colombia Sri Lanka Azerbaijan Maldives Kyrgyzstan Dominican Republic Hungary Croatia Viet Nam Mexico Czech Republic Argentina Tunisia Brazil Jordan Uzbekistan Ecuador Ukraine Macedonia Moldova Uruguay Thailand Latvia Montenegro Palestinian Territories Albania Russian Federation Serbia Bosnia and Herzegovina Georgia Kazakhstan United Arab Emirates Armenia Belarus Slovenia Slovakia 100% Comparing the Headcount Ratios of MPI Poor and $1.25/day Poor 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Intensity 69.4% & More Intensity 50-69.4% Intensity 44.4-50% Intensity 33.3-44.4% $1.25 a day
MPI varies greatly within income categories High Income
Average Intensity of Poverty (A) MPI varies: High and Upper Middle Income 75% 70% 65% 60% 55% High Income Upper-Middle Income Lower-Middle Income Low Income Poorest Countries, Highest MPI 50% China 45% Brazil Namibia 40% 35% Hungary Turkey Czech Republic Peru 30% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percentage of People Considered Poor (H) The size of the bubbles is a proportional representation of the total number of MPI poor in each country
Average Intensity of Poverty (A) MPI varies: including Lower Middle Income 75% 70% 65% 60% High Income Upper-Middle Income Lower-Middle Income Low Income Pakistan India Nigeria Cote d'ivoire Poorest Countries, Highest MPI Senegal 55% 50% China Philippines Indonesia Honduras Zambia 45% 40% 35% Brazil Egypt Hungary Turkey Czech Republic Bhutan Ghana Namibia 30% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percentage of People Considered Poor (H) The size of the bubbles is a proportional representation of the total number of MPI poor in each country
Average Intensity of Poverty (A) MPI varies: including Low Income 75% 70% 65% 60% 55% 50% 45% 40% 35% Brazil Egypt China Kyrgyzstan Hungary Czech Republic High Income Upper-Middle Income Lower-Middle Income Low Income Philippines Indonesia Tajikistan Turkey Bhutan Honduras Ghana Pakistan Nepal India Nigeria Namibia Cambodia Zimbabwe Kenya Cote d'ivoire Bangladesh Madagascar Benin Tanzania Mozambique Burkina Faso Senegal Guinea Liberia Sierra Leone DR Congo Uganda 30% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percentage of People Considered Poor (H) Poorest Countries, Highest MPI Niger Burundi Ethiopia The size of the bubbles is a proportional representation of the total number of MPI poor in each country
Robustness tests
Robustness tests Household Composition Changes in indicators Changes in deprivation cutoffs Changes in Poverty cutoff Changes in weights Alkire and Santos 2014
Robustness of Poverty cutoff k= 20% to 40% 93% to 96% of pairwise comparisons are robust overall. 90% or more pairwise comparisons are robust in categories of SA, SSA, AS, EAP, DHS, MICS, 10 indicators, and low and lower-middle income countries. For China, 95 to 97% of the significant pairwise comparisons obtained with the baseline MPI hold when we vary the poverty cutoff from 33.33% to either 20% or 40%. These results suggest that across poverty cutoff from 20-40%, rankings are quite stable and robust, particularly for poorer countries and regions Alkire and Santos 2014
Robustness across weights Weighting each dimension 50% and the other two at 25%, and comparing the rankings with 33% weights, we obtain: High Rank Correlation (Spearman): 0.95 and above High Rank Correlation (Kendall): 0.83 and above 85% of all possible pair-wise comparisons are robust Alkire and Santos 2014
Robustness to weights (Kendall) Table X: Correlation coefficients between MPI using alternative weighting structures 50% Education 50% Health Equal Weights 25% Health 25% Education 33% each 25% LS 25% LS 50% Education 25% Health 0.889 25% LS 50% Health 25% Education 0.925 0.835 25% LS 50% LS 25% Health 0.901 0.852 0.863 25% Education Note: LS: Living Standard. In all cases 104 countries were considered. The Spearman rank correlation coefficients are 0.95 and higher. Alkire and Santos 2014
LEAVING NO ONE BEHIND: Inequality among the Poor
Inequality Among the Poor. We ve done inequality measures for each of the MPI2014 countries and for each of the 780 subnational region for which we have data, to show disparities across countries and regions. Empirical results will be published with MPI on 16 June 2014. The policy goal is to end poverty, not inequality among the poor. Yet inequality measures help to visualize horizontal inequalities, and capture the variance in deprivation scores. Seth and Alkire 2014 27
LEAVING NO ONE BEHIND: Ethnic Groups
MPI over time by groups This year we release a study of how MPI has changed over time for 34 countries, covering 2.5 Billion people. We analyse over 330 subnational regions of these countries, to see where the poorest are being left behind and where the policies are most strongly pro-poor. We also study changes over time by ethnic groups. Alkire, Roche and Vaz 2014. 29
In this country, the poorest ethnic group saw no change in MPI over time. They are being left behind. 30
In this country, the poorest ethnic group reduced MPI the fastest. They are catching up. 31
LEAVING NO ONE BEHIND: Destitution
Destitution: A subset of the poor This year we release a study of destitutes people who are MPI poor but are extremely deprived experiencing severe malnutrition, losing 2 children, open defecation, no one has more than 1 year of school, kids out of primary school. We analyse over 49 countries with this measure. We find that a sad and high percentage of MPI poor are also destitute yet that countries vary a lot in eliminating destitution. Alkire Conconi and Seth 2014 33
Deprivation cutoffs: Destitute Indicator Deprivation Cutoff Schooling No one completed at least one year of schooling (>=1) Attendance At least one child not attending school up to the age at which they should finish class 6 Nutrition Severe Undernourishment of any adult (BMI<17kg/m 2 ) or any child (-3 standard deviations from median) Mortality 2 or more children died in the household Electricity The household has no electricity (No change) Sanitation Water Floor Cooking fuel Assets There is no facility/bush, or other (open defecation) The household does not have access to safe drinking water, or safe water is more than a 45-minute walk (round trip) The household has a dirt, sand, or dung floor (No change) The household cooks with dung or wood (coal/lignite/charcoal are now non-deprived) The household has no assets (radio, mobile phone, etc) and no car 34
Share of Destitue to MPI Poor (HD/H) What % of MPI poor are destitute? 90% 75% 60% 45% 30% 15% 0% Upper and Lower circles have similar MPI values, but a larger share of MPI poor are destitute in Upper. Can we learn from Lower? 0.000 0.130 0.260 0.390 0.520 0.650 Where MPI is high, a higher MPI Countries with similar MPI have 35 share of poor are destitute. different % of destitutes.
In Sum
The MPI is like a high resolution lens
The MPI is like a high resolution lens You can zoom in
The MPI is like a high resolution lens You can zoom in and see more