Skip to content
Back to search results

Foster-Greer-Thorbecke Index

Introduced by
James Foster, Joel Greer & Erik Thorbecke
Year
1984
Type
Parametric class of poverty measures
Family
Income poverty measurement
Font size

What does it do?

The Foster-Greer-Thorbecke (FGT) index is a parametric class of poverty measures introduced by James Foster, Joel Greer, and Erik Thorbecke in 1984 that became the workhorse of applied poverty analysis.

A single parameter alpha tunes how much weight the measure places on the depth and distribution of poverty: alpha = 0 gives the headcount ratio (the share of people below the poverty line), alpha = 1 gives the poverty gap (the average normalized shortfall), and alpha = 2 gives poverty severity (which weights larger shortfalls more heavily). Its defining virtue is additive decomposability — total poverty is the population-weighted sum of subgroup poverty — which makes it ideal for profiling poverty across regions, sectors, and demographic groups.

The intuition behind the method

Imagine everyone whose income falls below a poverty line z, and for each poor person measure how far they fall short as a fraction of the line. If you simply count those people you get the headcount; if you average their fractional shortfalls you capture how deep poverty is; and if you average the squared shortfalls you give extra weight to the very poorest, so transfers from a destitute person to a slightly-less-poor one register as worsening. The single knob alpha controls this sensitivity: turn it up and the measure cares more about the poorest of the poor. Because each group's contribution can be summed with population weights, you can always ask how much each region or sector adds to the national total.

History

James Foster, Joel Greer, and Erik Thorbecke published the index in a brief but enormously influential 1984 Econometrica note, motivated by Amartya Sen's 1976 critique that the headcount and poverty-gap measures fail basic axioms while Sen's own index was not additively decomposable. By embedding a poverty-aversion parameter inside a decomposable functional form, FGT reconciled axiomatic desiderata (monotonicity and transfer for suitable alpha) with the practical need to apportion poverty across subgroups. The measure was rapidly adopted by the World Bank and national statistical offices and remains the default class of poverty measures in applied development economics; the squared poverty gap (alpha = 2) in particular is the standard distribution-sensitive money-metric measure.

When to use it

Use the FGT index when you have a credible welfare indicator (income or, preferably in developing-country settings, consumption) for individuals or households and a defensible poverty line, and you want a coherent family of measures that can describe not just how many people are poor but how poor they are. Report P_0, P_1, and P_2 together: the headcount communicates prevalence, the gap communicates depth and the cost of elimination, and the severity measure captures inequality among the poor and responds to transfers. The additive decomposability makes FGT the natural choice for poverty profiling — apportioning national poverty across regions, urban/rural sectors, or household types. Choose higher alpha when distributional concern among the poor matters; rely on the headcount alone only for the simplest communication, since it is insensitive to how far below the line people fall.

Limitations and considerations

  • Requires an exogenously chosen poverty line z, and rankings — especially of the headcount — can be sensitive to where that line is drawn.
  • The headcount member (alpha = 0) violates monotonicity and the transfer axiom: it ignores how far below the line the poor are and can fall when a poor person becomes destitute (if another crosses the line).
  • The choice of alpha is not data-driven; different alphas can rank distributions differently, so conclusions may hinge on a normative parameter.
  • Like all unidimensional money-metric measures it ignores non-income dimensions of deprivation and the quality of the underlying welfare and price data.

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

Strengths

  • Provides a single coherent family in which a tunable parameter alpha moves smoothly from prevalence (headcount) to depth (gap) to severity, so one framework answers several questions.
  • Additively decomposable: national poverty is the exact population-weighted sum of subgroup poverty, enabling poverty profiles and targeting.
  • For alpha greater than or equal to 1 the measure satisfies the monotonicity axiom, and for alpha greater than 1 it satisfies the transfer axiom, giving it sound axiomatic foundations.
  • Simple to compute from survey microdata and universally implemented in poverty-analysis software, making results comparable across studies.

Applications

  • Constructing national and regional poverty profiles that apportion total poverty across geographic and demographic subgroups for targeting.
  • Monitoring poverty trends over time, reporting headcount, gap, and severity side by side to distinguish changes in prevalence from changes in depth.
  • Evaluating the poverty impact of policies and transfers, since the gap and severity measures respond to changes that leave the headcount unchanged.
  • Comparing poverty across countries using internationally harmonized lines such as the World Bank's international poverty line.
  • Serving as the base measure in growth-redistribution decompositions and stochastic-dominance robustness checks of poverty comparisons.

How it works

How it works

1

First fix an exogenous poverty line z and identify the poor as everyone whose welfare indicator y_i (income or consumption) falls below z. For each individual compute the normalized poverty gap g_i = max((z − y_i)/z, 0), which equals zero for the non-poor and rises toward one as a poor person's income approaches zero. The count of the poor is q and the total population is N; only the gaps of the poor enter the index, embodying the focus axiom that nothing about the non-poor should affect the poverty measure.

Formula
q = \#\{\, i : y_i < z \,\}, \qquad g_i = \max\!\left(\frac{z - y_i}{z},\, 0\right)
2

The FGT index averages the alpha-th power of the normalized gaps over the entire population: P_alpha = (1/N) sum over the poor of ((z − y_i)/z)^alpha. The exponent alpha is a 'poverty aversion' parameter — the higher it is, the more the measure penalizes large individual shortfalls relative to small ones, so it shifts emphasis from the prevalence of poverty toward its intensity and inequality among the poor. Dividing by N rather than q is what makes the measure additively decomposable across subgroups.

Formula
P_\alpha = \frac{1}{N} \sum_{i=1}^{q} \left(\frac{z - y_i}{z}\right)^{\alpha}, \qquad \alpha \ge 0
3

Three integer values give the canonical members. With alpha = 0 every gap is raised to the zero power (equal to one for the poor), so P_0 = q/N is the headcount ratio. With alpha = 1, P_1 is the mean normalized shortfall across the whole population — the poverty gap, interpretable as the share of the poverty line needed (per capita) to lift everyone to z under perfect targeting. With alpha = 2, P_2 squares the gaps, so it satisfies the transfer axiom and is sensitive to inequality among the poor; it is often called the squared poverty gap or poverty severity.

Formula
P_0 = \frac{q}{N}, \quad P_1 = \frac{1}{N}\sum_{i=1}^{q} g_i, \quad P_2 = \frac{1}{N}\sum_{i=1}^{q} g_i^{2}
4

Because the index is a population-weighted average of individual gap powers, total poverty decomposes exactly as P_alpha = sum over subgroups k of (N_k/N) times P_alpha for that subgroup, where N_k is the subgroup population. This additive decomposability lets analysts compute each region's or sector's absolute contribution to national poverty and the percentage share it represents, which is the basis of poverty profiles used in policy targeting.

Formula
P_\alpha = \sum_{k=1}^{K} \frac{N_k}{N}\, P_\alpha^{(k)}

Common pitfalls

  • Reporting only the headcount ratio and treating it as 'the' poverty rate, ignoring that it is blind to the depth of poverty and to transfers among the poor.
  • Comparing FGT values across surveys that use different poverty lines, equivalence scales, or welfare definitions without harmonizing them first.
  • Forgetting that P_1 and P_2 are averaged over the whole population N (not over the poor q); dividing by q instead breaks decomposability and changes interpretation.
  • Drawing strong poverty rankings from a single line and single alpha when the ordering may reverse under reasonable alternatives — robustness via dominance analysis is then warranted.

Frequently asked questions

What does the alpha parameter actually control?

Alpha is a poverty-aversion parameter that governs how sensitive the measure is to the depth and distribution of poverty. At alpha = 0 only the count of the poor matters. At alpha = 1 the average size of shortfalls matters, so the measure captures depth. At alpha greater than 1 larger shortfalls are weighted disproportionately, so the measure becomes sensitive to inequality among the poor and responds (in the right direction) to transfers from poorer to less-poor individuals. Raising alpha thus shifts attention from the prevalence of poverty toward its intensity and concentration.

Why divide by total population N instead of by the number of poor q?

Dividing by the full population N is what makes the FGT class additively decomposable: total poverty becomes the population-weighted sum of subgroup poverty, so each subgroup's contribution to the national figure can be read off directly. If you instead divided by q you would obtain a conditional mean among the poor (such as the average normalized gap among the poor), which is interpretable but no longer aggregates across subgroups with simple population weights.

How does FGT relate to the separate poverty-gap and Watts indices?

The poverty-gap index is exactly the FGT member at alpha = 1, so it is a special case rather than a different method. The Watts index, by contrast, sits outside the FGT class: it uses the logarithm of the ratio of the line to income rather than a power of the normalized gap, which makes it a distinct distribution-sensitive measure. All three belong to the broader literature on decomposable, subgroup-consistent poverty measurement.

Related methods

Sources

  1. Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposable poverty measures. Econometrica, 52(3), 761–766.

    Open DOI reference
    Copied.