Skip to content
Back to search results

Multidimensional Poverty Index

Introduced by
Sabina Alkire & James Foster
Year
2011
Type
Counting-based multidimensional poverty measure
Family
Multidimensional poverty measurement
Font size

What does it do?

The Multidimensional Poverty Index applies the Alkire-Foster method, introduced by Sabina Alkire and James Foster in 2011, to measure poverty as the joint deprivation of individuals across several dimensions such as health, education, and living standards.

Its signature is a dual-cutoff identification: a person is deprived in an indicator if they fall below that indicator's cutoff, and they are counted as multidimensionally poor only if their weighted count of deprivations crosses a cross-dimensional cutoff k. The headline measure is the adjusted headcount ratio M0 = H times A, the product of the share of people who are poor (incidence) and the average breadth of their deprivations (intensity).

The intuition behind the method

Income alone misses much of what it means to be poor: a household may earn enough to clear an income line yet have no schooling, poor sanitation, and a malnourished child. The Alkire-Foster method counts deprivations directly. First decide, indicator by indicator, who is deprived. Then weight and add up each person's deprivations into a single deprivation score. A person counts as poor only if that score reaches a threshold k — this is the second cutoff, which ensures the measure captures joint, overlapping deprivation rather than a single isolated shortfall. The headline number then multiplies how many people are poor by how broadly deprived they are on average, so that helping a poor person escape one deprivation lowers the index even if they remain poor.

History

Sabina Alkire and James Foster developed their counting-based method in work at the Oxford Poverty and Human Development Initiative (OPHI), with the definitive statement in their 2011 Journal of Public Economics paper 'Counting and Multidimensional Poverty Measurement'. The approach generalized the FGT logic to many dimensions by adding a second, cross-dimensional cutoff to the standard within-dimension cutoffs, resolving the long-standing identification problem in multidimensional poverty. In 2010 UNDP's Human Development Report adopted the global MPI built on this method, and dozens of countries have since constructed official national MPIs, making the Alkire-Foster framework the dominant approach to multidimensional poverty measurement.

When to use it

Use the Alkire-Foster MPI when poverty is genuinely multidimensional and you have individual- or household-level data on several deprivation indicators with defensible cutoffs and weights. It is the standard tool behind the global MPI published by UNDP and OPHI and behind many national multidimensional poverty measures, and it is appropriate whenever income data alone would misrepresent deprivation. Report incidence H, intensity A, and the adjusted headcount M0 together, and use the indicator decomposition to identify which deprivations drive poverty in each group. The method requires explicit normative choices — which dimensions, what cutoffs, what weights, and the poverty cutoff k — so transparency and robustness checks over these choices are essential. When indicators are cardinal, the related M1 and M2 measures add depth and severity.

Limitations and considerations

  • Requires explicit, contestable normative choices — dimensions, indicator cutoffs, weights, and the cross-dimensional cutoff k — that drive the results.
  • The headline M0 uses only binary deprivation status, discarding information on how far below each indicator cutoff a person falls (depth within indicators).
  • Equal or expert-set weights are hard to justify and the index is sensitive to them, especially for the identification cutoff k.
  • Comparability across countries and over time is fragile when indicators, cutoffs, or data sources differ.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Captures the joint distribution of deprivations across dimensions, which separate single-dimension indices and dashboards miss.
  • The adjusted headcount M0 satisfies dimensional monotonicity, falling when any poor person escapes a deprivation, unlike the raw multidimensional headcount.
  • Decomposable by population subgroup and breakable down by indicator, enabling targeted policy and clear poverty profiles.
  • Works with ordinal data (the common case for health and education indicators), where many alternative measures require cardinality.

Applications

  • The global MPI published annually by UNDP and OPHI to compare multidimensional poverty across developing countries.
  • National multidimensional poverty measures used as official statistics and policy targets.
  • Targeting and monitoring of social programs by identifying which deprivations drive poverty in specific regions or groups.
  • Tracking progress on the Sustainable Development Goals' multidimensional poverty targets.
  • Decomposing multidimensional poverty by subgroup and indicator to inform sectoral budget allocation.

How it works

How it works

1

For each person i and indicator j, an indicator-specific cutoff z_j determines deprivation: g0_ij equals one if the achievement x_ij falls below z_j and zero otherwise. The weighted sum of these deprivation indicators, c_i = sum of w_j times g0_ij, is the person's deprivation score, where the nonnegative weights w_j sum to one and encode the relative importance of each indicator. The deprivation matrix and weighting scheme together define what deprivation means before any poverty cutoff is applied.

Formula
g^0_{ij} = \mathbf{1}\!\left(x_{ij} < z_j\right), \qquad c_i = \sum_{j=1}^{d} w_j\, g^0_{ij}, \quad \sum_j w_j = 1
2

The second, cross-dimensional cutoff k identifies the poor: a person is multidimensionally poor only if their deprivation score c_i is at least k (for example, poor if deprived in at least one third of weighted indicators). This dual-cutoff step is the heart of the method — the first cutoff is within each dimension, the second across dimensions. Applying the focus axiom, the censored score c_i(k) keeps the deprivation score for the poor and sets it to zero for the non-poor, so only the joint deprivations of the poor count.

Formula
\text{poor}_i \iff c_i \ge k, \qquad c_i(k) = c_i \cdot \mathbf{1}(c_i \ge k)
3

Two summary numbers follow. The multidimensional headcount H is the share of people identified as poor — the incidence of poverty. The intensity A is the average censored deprivation score among the poor — the average breadth of deprivation, that is, the average share of weighted indicators in which a poor person is deprived. H answers 'how many are poor?' and A answers 'how poor, in breadth, are they?', and the headcount alone (like its income counterpart) is blind to intensity.

Formula
H = \frac{1}{N}\sum_{i=1}^{N} \mathbf{1}(c_i \ge k), \qquad A = \frac{\sum_{i} c_i(k)}{\sum_i \mathbf{1}(c_i \ge k)}
4

The adjusted headcount ratio M0 is the product H times A, equivalently the population mean of the censored deprivation scores. Because it multiplies incidence by intensity, it satisfies dimensional monotonicity — it falls when any poor person escapes a deprivation, which the raw headcount does not — and it is decomposable by population subgroup and breakable down by indicator. The contribution of each indicator is its weight times its censored deprivation headcount, so M0 can be read as a weighted sum of indicator contributions, supporting policy targeting.

Formula
M_0 = H \times A = \frac{1}{N}\sum_{i=1}^{N} c_i(k) = \sum_{j=1}^{d} w_j\, \text{CH}_j

Common pitfalls

  • Reporting the multidimensional headcount H alone, ignoring intensity A and so missing changes in the breadth of deprivation among the poor.
  • Treating the weights and the cutoff k as objective; the index can shift substantially under reasonable alternative choices, so robustness analysis is required.
  • Mixing indicators measured inconsistently across surveys or countries and then comparing M0 as if it were harmonized.
  • Forgetting that M0 ignores within-indicator depth; for cardinal indicators the M1 (gap) and M2 (severity) extensions should be considered.

Frequently asked questions

What are the two cutoffs in the dual-cutoff method?

The first cutoff is within each indicator: a person is deprived in indicator j if their achievement falls below that indicator's threshold z_j (for example, no household member completed primary school). The second cutoff is across dimensions: a person counts as multidimensionally poor only if their weighted deprivation score reaches the poverty cutoff k (for example, deprived in at least one third of weighted indicators). The first defines deprivation indicator by indicator; the second defines who is poor overall, ensuring the measure reflects joint deprivation.

Why use the adjusted headcount M0 instead of the simple headcount H?

The simple multidimensional headcount H counts who is poor but is insensitive to how broadly each poor person is deprived — if a poor person escapes one deprivation but remains above the cutoff k, H does not move. The adjusted headcount M0 = H times A multiplies incidence by intensity, so it falls whenever any poor person reduces their deprivations. This dimensional monotonicity is the key axiomatic advantage and the reason M0, not H, is the headline measure of the Alkire-Foster method.

How are the weights and the poverty cutoff k chosen?

Both are normative choices made by the analyst or institution. The global MPI, for example, uses three equally weighted dimensions (health, education, living standards) with equal weights within each, and sets k at one third of weighted indicators. These choices are defensible but not unique, and the index can be sensitive to them, so good practice is to justify the choices transparently and report robustness over alternative weights and a range of k values rather than presenting a single configuration as definitive.

Related methods

Sources

  1. Alkire, S., & Foster, J. (2011). Counting and multidimensional poverty measurement. Journal of Public Economics, 95(7–8), 476–487.

    Open DOI reference
    Copied.