Skip to content
Back to search results

Poverty Dominance Analysis

Introduced by
Anthony Atkinson (1987); James Foster & Anthony Shorrocks (1988)
Year
1987
Type
Robust distributional ordering
Family
Robust poverty comparison
Font size

What does it do?

Poverty dominance analysis asks whether one distribution has unambiguously less poverty than another for a whole class of poverty measures and a whole range of poverty lines, rather than for a single index and a single line.

Building on Anthony Atkinson's 1987 stochastic-dominance treatment of poverty and the Foster-Shorrocks 1988 poverty-orderings results, it compares cumulative distribution functions (poverty incidence curves) and their successive integrals (poverty deficit and severity curves). When the curve for one distribution lies everywhere below another, that distribution has less poverty for every measure in a corresponding class and every line in the range — a robust conclusion immune to the index-and-line arbitrariness that bedevils single-number comparisons.

The intuition behind the method

Two ordinary criticisms of poverty comparisons are that the answer depends on which poverty line you pick and on which poverty measure you use. Dominance analysis sidesteps both by checking whether the ranking holds for all reasonable lines and a whole family of measures at once. Plot the headcount as a function of the poverty line — the poverty incidence curve. If country A's curve lies entirely below country B's, then A has fewer poor people than B no matter where you draw the line: that is first-order dominance, and it guarantees less poverty for every sensible measure. If the curves cross, integrate them once to get cumulative poverty deficits and check again (second order), which orders the more distribution-sensitive measures. Each integration buys agreement among a larger but more restricted class of measures.

History

Anthony Atkinson's 1987 Econometrica paper 'On the Measurement of Poverty' reframed poverty comparison as a stochastic-dominance problem, showing that rankings robust to the poverty line and to the choice among classes of poverty measures correspond to dominance of the cumulative distribution and its integrals — directly importing the welfare-dominance results of Atkinson, Shorrocks, and others from inequality analysis. James Foster and Anthony Shorrocks's 1988 'Poverty Orderings', also in Econometrica, established the formal equivalence between dominance of the FGT curves at successive orders and unanimous poverty rankings by the corresponding classes of measures. The TIP curve (Jenkins and Lambert, 1997) later gave the second-order results an intuitive graphical form, and the approach is now standard practice for robust poverty comparison.

When to use it

Use poverty dominance analysis whenever a poverty comparison must be robust to the choice of poverty line and poverty measure — for example, ranking two countries, two time periods, or two policy scenarios when reviewers could object to your specific line or index. Check first-order dominance first; if it holds you have an exceptionally strong, near-uncontestable ranking. If incidence curves cross, move to second-order (poverty deficit) and then higher-order dominance, accepting that each step narrows the class of measures over which agreement holds. Use TIP curves to visualize incidence, intensity, and inequality together. The limitation is that dominance often yields no verdict — curves cross and orderings are incomplete — and that statistical testing of curve dominance over a continuum of points requires care with multiple comparisons and sampling error.

Limitations and considerations

  • Often inconclusive: when curves cross, no unambiguous ordering exists at that order and one must move to a higher (more restrictive) order or accept indeterminacy.
  • Provides an ordinal ranking only, not a quantitative statement of how much less poverty one distribution has.
  • Statistical inference must test dominance over a continuum (or grid) of points, raising multiple-comparison and power issues with sampling data.
  • Results depend on the chosen upper bound z-plus of admissible poverty lines, which is itself a judgment.

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

Strengths

  • Delivers poverty rankings that are robust to the choice of poverty line over an entire admissible range, not just one line.
  • A single dominance check establishes agreement across a whole class of poverty measures at once, avoiding index cherry-picking.
  • Nested orders (first, second, higher) trade breadth of measures for strength, giving a principled hierarchy of robustness.
  • TIP and incidence/deficit curves provide intuitive visual diagnostics of incidence, depth, and inequality of poverty.

Applications

  • Establishing robust poverty rankings between countries or regions that hold for any reasonable poverty line.
  • Determining whether poverty fell unambiguously over time without committing to a single index or line.
  • Evaluating whether a reform reduced poverty robustly across the class of distribution-sensitive measures.
  • Visualizing the incidence, intensity, and inequality of poverty simultaneously via TIP curves.
  • Stress-testing single-index poverty conclusions for sensitivity to the line and measure before publication.

How it works

How it works

1

The starting object is the poverty incidence curve, which is simply the cumulative distribution function F(z): the headcount ratio plotted as a function of the poverty line z. Reading it as a poverty measure, F_A(z) is the share of population in A below any candidate line z. Comparing whole curves rather than single points is what lets the analysis range over all admissible poverty lines up to an upper bound z-plus, beyond which no one would set a poverty line.

Formula
F_A(z) = P_A^{0}(z) = \frac{1}{N}\#\{\, i : y_i \le z \,\} \quad \text{(headcount as a function of } z)
2

Distribution A first-order dominates B (has less poverty) if its incidence curve lies at or below B's at every poverty line in the admissible range. By the equivalence between stochastic dominance and classes of welfare/poverty measures, first-order dominance implies that A has no more poverty than B for every poverty measure that is non-increasing in incomes — essentially every measure satisfying the focus and monotonicity axioms — and for every poverty line up to z-plus. It is the strongest and most widely applicable ranking, but it fails whenever the curves cross.

Formula
A \succeq_1 B \iff F_A(z) \le F_B(z) \;\; \forall z \in [0, z^{+}]
3

When incidence curves cross, integrate them to obtain the poverty deficit curve D2(z), the area under the incidence curve up to z, which corresponds to the poverty gap accumulated up to each line. Second-order dominance holds if A's deficit curve lies at or below B's everywhere. It implies A has less poverty for the smaller class of measures that are also sensitive to the depth of poverty (those satisfying the transfer axiom), again for all admissible lines. Higher-order dominance integrates once more to order severity-sensitive measures such as the squared gap and the Watts index.

Formula
D_A^{2}(z) = \int_0^{z} F_A(t)\, dt \le D_B^{2}(z) \;\; \forall z \in [0, z^{+}] \iff A \succeq_2 B
4

The TIP (Three I's of Poverty) curve offers a complementary graphical device: it cumulates the ordered poverty gaps from the poorest upward against the population proportion. Its length along the horizontal axis shows the incidence of poverty, its height shows the intensity (aggregate normalized gap), and its concavity shows the inequality of gaps among the poor. Non-intersecting TIP curves deliver dominance results equivalent to second-order poverty dominance, so the TIP curve visualizes incidence, intensity, and inequality and the robust ordering simultaneously.

Formula
TIP_A(p) = \frac{1}{N}\sum_{i=1}^{\lceil pN \rceil} \max(z - y_{(i)}, 0), \qquad p \in [0,1]

Common pitfalls

  • Declaring dominance from eyeballing two curves without a formal statistical test over the grid of lines, ignoring sampling uncertainty.
  • Concluding 'no difference' when curves merely cross, when a higher-order dominance check might still yield a robust ordering.
  • Testing dominance at many points without controlling for multiple comparisons, inflating the chance of a spurious dominance verdict.
  • Comparing distributions with inconsistent welfare definitions or deflators, so curve differences reflect data artifacts rather than real poverty differences.

Frequently asked questions

What is the difference between first- and second-order poverty dominance?

First-order dominance compares the poverty incidence curves (the CDFs): if A's lies below B's at every admissible line, A has less poverty for every measure that is non-increasing in incomes and for every line. Second-order dominance compares the integrals of those curves (the poverty deficit curves): it applies when incidence curves cross and yields a robust ranking for the smaller class of depth-sensitive measures satisfying the transfer axiom. Each higher order integrates once more, ordering a more restrictive but more distribution-sensitive class of measures.

What do I do when the curves cross?

A crossing of incidence curves means there is no first-order dominance: the ranking reverses for some poverty lines, so no robust verdict holds for the broad class of measures. The standard response is to test a higher order of dominance — integrate to the poverty deficit (second order) and then severity (third order) curves — which may still order the distributions for the narrower class of distribution-sensitive measures. If dominance fails at every reasonable order, the comparison is genuinely ambiguous and one must restrict the range of lines or accept that the ranking depends on the measure.

How does the TIP curve relate to dominance analysis?

The TIP (Three I's of Poverty) curve cumulates ordered poverty gaps against the population share, so its horizontal extent shows poverty incidence, its height shows intensity, and its curvature shows the inequality of gaps among the poor. Non-intersecting TIP curves are equivalent to second-order poverty dominance: if one TIP curve lies everywhere above another, the corresponding distribution has more poverty for all depth-sensitive measures and all admissible lines. The TIP curve thus gives a single picture that both summarizes the three I's and signals the robust ordering.

Related methods

Sources

  1. Atkinson, A. B. (1987). On the measurement of poverty. Econometrica, 55(4), 749–764.

    Open DOI reference
    Copied.
  2. Foster, J. E., & Shorrocks, A. F. (1988). Poverty orderings. Econometrica, 56(1), 173–177.

    Open DOI reference
    Copied.