Multidimensional Poverty Index
The Multidimensional Poverty Index (MPI) is a composite metric that quantifies acute multidimensional poverty by assessing household deprivations in three dimensions—health, education, and standard of living—using ten weighted indicators derived from survey data, where individuals are deemed multidimensionally poor if deprived in at least one-third of the total weighted indicators.[1] Developed by economists Sabina Alkire and James Foster via their Alkire-Foster (AF) counting methodology, which computes both the incidence (headcount ratio) and intensity of poverty to yield an adjusted headcount measure ranging from 0 to 1, the MPI provides a flexible framework adaptable to national contexts while enabling cross-country comparisons.[2] First introduced by the United Nations Development Programme (UNDP) in its 2010 Human Development Report in collaboration with the Oxford Poverty and Human Development Initiative (OPHI), the global MPI has tracked acute poverty affecting over 1.1 billion people across more than 100 developing countries as of recent estimates, highlighting deprivations often overlooked by income-based metrics alone.[3] Its adoption in policy has facilitated targeted interventions, such as in India's national MPI aligned with sustainable development goals, though empirical critiques note methodological sensitivities including arbitrary deprivation cutoffs and equal weighting assumptions that can distort rankings, particularly inflating apparent poverty in low-income settings compared to correlation-adjusted alternatives.[4][5]History and Development
Origins and Theoretical Foundations
The theoretical foundations of the Multidimensional Poverty Index (MPI) originate from Amartya Sen's capabilities approach, articulated in works from the late 1970s onward, which reframed poverty as a shortfall in individuals' freedoms and capabilities to achieve valued functionings rather than mere income deficiency.[6] Sen argued that unidimensional monetary metrics, such as income poverty lines, inadequately capture poverty's causal structure by neglecting deprivations in non-income domains like nutrition, health, and education, where empirical data from household surveys reveal persistent overlaps uncorrelated with income alone.[7] This critique, echoed in Anand and Sen's 1997 analysis, emphasized that aggregating poverty solely via income overlooks the multidimensional nature of human deprivation, potentially misleading policy by ignoring how deprivations compound across life aspects.[8] Building on Sen's framework, Sabina Alkire and James Foster developed the MPI's core methodology in the mid-2000s at Oxford University, motivated by the need for a practical, data-driven tool to quantify these overlapping deprivations using counting-based techniques rather than utilitarian averaging.[9] Their seminal 2007 paper proposed an identification strategy that counts weighted deprivations across dimensions, extending axiomatic poverty measurement principles to respect the distinct, non-substitutable impacts of deprivations in areas like schooling and sanitation, as evidenced by discrepancies in cross-national datasets where monetary poor often escape non-monetary hardship and vice versa.[10] This approach prioritized empirical robustness over income-centric proxies, addressing Sen's call for measures sensitive to the intensity and breadth of poverty experienced by individuals.[11] The Oxford Poverty and Human Development Initiative (OPHI), co-founded by Alkire in 2007, institutionalized this work by integrating the Alkire-Foster method with Sen's capabilities paradigm, fostering research that grounded abstract theory in verifiable survey indicators to reveal poverty's true incidence beyond GDP correlations.[12] Unlike earlier multidimensional efforts reliant on composite indices that averaged dissimilar metrics, the MPI's foundations emphasized decomposability and subgroup analysis to trace causal pathways of deprivation, enabling targeted interventions informed by direct evidence of unmet basic needs.[10]Launch and Initial Adoption
The Multidimensional Poverty Index (MPI) was formally launched on July 14, 2010, through a collaboration between the United Nations Development Programme (UNDP) and the Oxford Poverty and Human Development Initiative (OPHI), marking its introduction in the 20th anniversary edition of the UNDP Human Development Report.[13][14] This debut complemented the report's Human Development Index by providing a measure of acute multidimensional poverty, drawing on the Alkire-Foster methodology to assess deprivations in health, education, and living standards.[15] The inaugural global MPI applied to 104 developing countries, encompassing roughly 5.2 billion people or 78% of the world's population at the time, with data sourced primarily from Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), and World Health Surveys conducted between 2000 and 2008.[16][17] It identified approximately 1.75 billion individuals—one-third of the covered population—as multidimensionally poor, highlighting the prevalence and intensity of overlapping deprivations across these nations.[16] Initial adoption occurred swiftly among governments seeking tools for poverty targeting beyond monetary metrics. Mexico pioneered an official national MPI in 2010, establishing it via normative framework under the National Council for the Evaluation of Social Development Policy (CONEVAL) to guide social program allocation and evaluation.[18][19] Bhutan similarly adapted the approach for national use shortly thereafter, with other countries like Chile exploring implementations to inform policy.[13][17] In India, the MPI gained traction through early analytical applications to 2011-12 survey data, enabling regional decompositions that supported evidence-based interventions in high-poverty areas.[20] These early integrations underscored the index's utility for disaggregated poverty analysis, though national variants often customized dimensions to local contexts.Evolution and Global Implementation
Following its launch in 2010 by the Oxford Poverty and Human Development Initiative (OPHI) in collaboration with the United Nations Development Programme (UNDP), the Multidimensional Poverty Index (MPI) has undergone annual updates to refine its scope and coverage.[21][22] These reports have progressively expanded data availability, drawing from household surveys such as Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS), to track acute multidimensional poverty across developing countries. By 2025, the global MPI encompassed estimates for 109 countries, representing a population of approximately 6.3 billion, with consistent indicators enabling assessments of changes over time.[21][23] The 2024 and 2025 global MPI reports estimated that 1.1 billion people—about 18.3% of the covered population—lived in acute multidimensional poverty, highlighting persistent challenges despite reductions in some regions.[24] To enhance cross-country comparability and longitudinal analysis, OPHI introduced harmonized estimates in 2025, restricting computations to indicators consistently available across survey waves and countries, thereby addressing variations in data quality and survey design that previously limited reliable trend comparisons.[25] This methodological refinement supports more robust policy evaluations, though it depends on the frequency and completeness of national surveys, which remain uneven in low-income settings.[25] Nationally, over 40 countries have adapted the MPI framework to local contexts, often customizing dimensions, indicators, and weights to reflect priorities such as housing quality or employment access.[26] Colombia implemented one of the earliest national versions in 2011, incorporating deprivations aligned with its social policy goals.[27] In Bhutan, the MPI draws from the 2022 Bhutan Living Standards Survey and integrates elements resonant with Gross National Happiness principles, emphasizing psychological wellbeing alongside standard health, education, and living standards indicators.[28] These adaptations facilitate targeted interventions but require ongoing data collection to maintain relevance. Since the adoption of the Sustainable Development Goals (SDGs) in 2015, the MPI has been integrated into global and national poverty monitoring, particularly for SDG Target 1.2, which calls for halving the proportion of people in multidimensional poverty by national definitions.[29] OPHI's expansions of the MPI database, leveraging MICS data from UNICEF, have enabled disaggregated analyses that complement monetary metrics and inform SDG progress tracking in over 100 countries.[23] This broader implementation underscores the MPI's role in evidence-based policymaking, contingent on sustained investments in survey infrastructure for timely, comparable data.[30]Methodology
Dimensions, Indicators, and Weights
The Multidimensional Poverty Index (MPI) employs three dimensions—health, education, and standard of living—each weighted equally at one-third of the total index to reflect deprivations in basic human capabilities.[3] These dimensions aggregate ten binary indicators, where households are deemed deprived or not based on specific cutoffs derived from international standards such as WHO nutritional guidelines and DHS survey definitions.[3] Weights are assigned equally within dimensions: the health and education dimensions, with two indicators each, allocate 1/6 per indicator, while the standard of living dimension, with six indicators, allocates 1/18 per indicator, ensuring dimensional parity despite differing numbers of metrics.[3] The indicators and their deprivation cutoffs are specified as follows:| Dimension | Indicator | Deprivation Cutoff | Weight |
|---|---|---|---|
| Health | Nutrition | Any adult (19–70 years) with BMI <18.5 kg/m², or any child under 5 with height-for-age or weight-for-age z-score <-2 SD, or child (5–19 years) with BMI-for-age z-score <-2 SD. | 1/6 |
| Health | Child Mortality | Any child under 18 died in the household in the past 5 years. | 1/6 |
| Education | Years of Schooling | No household member of school entrance age +6 years has completed 6 or more years of schooling. | 1/6 |
| Education | School Attendance | Any school-age child (up to the age of completing 8 years of schooling) not attending school up to grade 8. | 1/6 |
| Standard of Living | Cooking Fuel | Household uses dung, wood, charcoal, or coal for cooking. | 1/18 |
| Standard of Living | Sanitation | Household uses a sanitation facility not shared with other households and not improved (per WHO/UNICEF guidelines). | 1/18 |
| Standard of Living | Drinking Water | Household lacks access to improved drinking water or safe water is >30 minutes round-trip walk. | 1/18 |
| Standard of Living | Electricity | Household has no electricity. | 1/18 |
| Standard of Living | Housing | Household has at least one of: dirt, sand, dung, palm/bamboo, or similar floor; same for roof or walls. | 1/18 |
| Standard of Living | Assets | Household owns fewer than 1 of: radio, TV, phone, bike, or motorbike; and no car or truck. | 1/18 |
Alkire-Foster Counting Method
The Alkire-Foster counting method aggregates multidimensional deprivations by first identifying the poor through a deprivation cutoff and then computing a product of poverty incidence and intensity, distinguishing it from additive or geometric aggregation techniques that may dilute information on the extent of overlapping deprivations.[10] For each household or individual, a deprivation score c_i is calculated as the weighted sum of binary deprivations across indicators, where a deprivation d_{ij} = 1 if the unit falls below the indicator-specific cutoff and 0 otherwise, and weights w_j reflect relative importance (often equally distributed within dimensions, summing to 1 overall).[10] A unit is classified as multidimensionally poor if c_i \geq k, where k is a normative poverty cutoff representing the minimum weighted deprivations required (e.g., one-third of the total weighted indicators).[10] Poverty incidence, or headcount ratio H, measures the proportion of the population identified as poor: H = \frac{q}{n}, with q as the number of poor units and n the total population.[10] Intensity A captures the average deprivation share among the poor: A = \frac{\sum_{i: c_i \geq k} c_i}{q}, reflecting the depth of poverty beyond mere identification.[10] The adjusted headcount ratio, M_0 = H \times A, serves as the primary measure, satisfying axioms such as decomposability (allowing breakdown by subgroups or indicators), monotonicity (poverty rises with added deprivations), and robustness to dimensional inequality among the non-poor.[10] This counting approach enables precise targeting by decomposing M_0 into contributions from specific indicators or population subgroups, revealing where interventions might reduce either incidence or intensity most effectively.[10] For instance, consider a population of three households across three equally weighted indicators with k = 1/3. Household 1 is deprived in all three (c_1 = 1), Household 2 in two (c_2 = 2/3), and Household 3 in none (c_3 = 0). Households 1 and 2 are poor, yielding H = 2/3. The intensity A = (1 + 2/3)/2 = 5/6 \approx 0.833, so M_0 = (2/3) \times (5/6) \approx 0.556.[10] If Household 2 instead has deprivations totaling $2/3 (e.g., full deprivation in one and partial adjustment, but binary here), its intensity aligns directly with its score among the poor. Unlike Rawlsian metrics emphasizing the worst-off alone, the AF method balances breadth (via H) and depth (via A), avoiding undercounting moderate but widespread deprivations.[10]Computation and Dual Cutoffs
The computation of the Multidimensional Poverty Index (MPI) relies on the Alkire-Foster counting methodology, which applies dual cutoffs to identify deprivation and poverty at the household level before aggregating to population-level measures. For each household i, a deprivation score c_i is first calculated as the sum of weighted binary deprivations across the selected indicators: c_i = \sum_j w_j d_{ij}, where d_{ij} = 1 if the household is deprived in indicator j (based on its deprivation cutoff z_j) and $0 otherwise, and w_j are the normative weights (typically equal within dimensions, summing to 1 overall).[34][35] The second cutoff, the poverty line k (commonly set at k = 0.333), determines multidimensional poverty: a household is poor if c_i \geq k, meaning it experiences deprivations whose weighted sum reaches at least one-third of the total possible.[36][25] This dual-cutoff approach allows flexibility in parameter selection, with k reflecting a normative judgment on the breadth of deprivation required for poverty identification.[34] The national MPI is then the population-weighted average of censored deprivation scores: M_0 = \frac{1}{N} \sum_{i=1}^N g_i(k), where g_i(k) = c_i if c_i \geq k and $0 otherwise, equivalent to the product of the incidence of poverty H (share of population with c_i \geq k) and the average intensity A (mean c_i among the poor).[34][37] Computations incorporate survey design effects, such as clustering and stratification in Demographic and Health Surveys (DHS), using weighted estimators to ensure representativeness; for instance, Stata'ssvy commands adjust for primary sampling units and probability weights.[36]
Sensitivity to the poverty cutoff k is notable: increasing k (e.g., from 0.333 to 0.5) reduces H by excluding marginally poor households but raises A among the remaining poor, with the net effect on M_0 depending on the deprivation distribution; robustness tests by the Oxford Poverty and Human Development Initiative (OPHI) across k values like 0.2 and 0.4 often preserve country rankings, though headcounts vary substantially.[38][39] Early implementations, including 2011 revisions, validated parameter choices through such analyses to enhance replicability across datasets.[34]