Concentration Index
- Introduced by
- Adam Wagstaff, Pierfilippo Paci & Eddy van Doorslaer; Nanak Kakwani
- Year
- 1991
- Type
- Rank-based summary index of socioeconomic inequality in health
- Family
- Health-economics inequality measurement
What does it do?
The concentration index is the standard summary measure of socioeconomic inequality in health: it quantifies the degree to which a health outcome - illness, mortality, malnutrition, or healthcare use - is disproportionately concentrated among the poor or the rich.
It is built from the concentration curve, which plots the cumulative share of health against the cumulative share of the population ranked from poorest to richest, and the index is simply twice the area between that curve and the line of perfect equality. Wagstaff, Paci, and van Doorslaer's 1991 critique of inequality measures argued that, unlike the simple range or the Gini, the concentration index properly reflects the socioeconomic dimension of health inequality and the experience of the whole distribution. Kakwani, Wagstaff, and van Doorslaer's 1997 paper then supplied a computational formula, a convenient regression estimator, and the asymptotic variance needed for statistical inference. The index ranges from minus one to plus one, with zero meaning no socioeconomic gradient, a negative value meaning ill health concentrates among the poor, and a positive value the reverse. It has become the lingua franca of health-equity monitoring at agencies like the World Bank and WHO.
The intuition behind the method
Picture lining everyone up from the poorest person on the left to the richest on the right, then walking along the line and keeping a running tally of how much ill health you have accumulated. If illness were spread evenly across rich and poor, your tally would grow in lockstep with the share of people you had passed - a straight diagonal line. But suppose poverty and sickness travel together: then in the early part of the walk, among the poor, your tally of ill health races ahead of the share of people counted, and the curve bulges above the diagonal. The bigger that bulge, the more sharply disease is concentrated among the disadvantaged. The concentration index just measures the size of the bulge - twice the area between the curve and the diagonal - boiling a whole distribution of who-bears-what into one number whose sign tells you which end of the income ladder shoulders the burden. Because it is built on ranks rather than absolute income, it captures the social gradient itself rather than the spread of income.
History
Measuring inequality in health by analogy to income inequality matured in the early 1990s. Wagstaff, Paci, and van Doorslaer's 1991 Social Science & Medicine paper critically appraised the available measures and argued that the concentration index and the slope index of inequality alone adequately capture the socioeconomic dimension, displacing crude range and decile-ratio measures. Kakwani, Wagstaff, and van Doorslaer's 1997 Journal of Econometrics article then put the index on firm statistical footing, providing the covariance formula, the convenient-regression estimator, and the asymptotic variance, and clarifying the role of demographic standardization. The World Bank subsequently institutionalized the method in its analytic toolkits for monitoring health equity, and later work by Erreygers and Wagstaff resolved the difficulties posed by bounded outcomes. The index is now a cornerstone of global health-equity measurement.
When to use it
Use the concentration index when you want a single, comparable number for how strongly a health outcome is patterned by socioeconomic status and you have both an outcome and a ranking variable for SES. It is ideal for monitoring trends over time, comparing inequality across countries or regions, and feeding inequality into decomposition analyses that attribute it to determinants. Because it weights the whole distribution and uses SES ranks, it is preferable to crude rich-poor ratios or range measures when you care about the social gradient itself. It is less appropriate when you lack a meaningful SES ranking, when groups are nominal and unordered (where between-group measures like the Theil index fit better), or when the outcome is bounded and you neglect the necessary normalization. For binary or bounded outcomes, always pair it with an Erreygers- or Wagstaff-type correction so cross-population comparisons remain valid.
Limitations and considerations
- The standard index for bounded outcomes depends on the mean, so naive comparisons across populations with different prevalence are misleading.
- It captures only the relative, rank-based gradient and says nothing about absolute gaps between rich and poor.
- Results hinge on the choice and quality of the SES ranking variable, which is often a noisy asset index.
- A single index can mask non-monotonic patterns and crossing concentration curves that only the full curve reveals.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Reflects the socioeconomic dimension of inequality by using SES ranks, capturing the social gradient rather than mere dispersion.
- Summarizes the entire concentration curve in one signed number whose direction and magnitude are easy to interpret.
- Comes with a convenient regression estimator and closed-form variance, enabling standard errors and significance tests in any package.
- Extends naturally to decomposition, attributing measured inequality to underlying determinants of health.
Applications
- Quantifying income-related inequality in child malnutrition, immunization, or under-five mortality across developing countries.
- Monitoring whether socioeconomic inequality in access to and use of health services is widening or narrowing over time.
- Comparing the magnitude of socioeconomic gradients in chronic disease across nations using harmonized survey data.
- Decomposing measured inequality into contributions from determinants such as education, region, and insurance status.
- Assessing the distributional impact of health-system reforms by tracking concentration indices before and after policy change.
How it works
How it works
Everything starts with ranking individuals by socioeconomic status - typically income, consumption, an asset-based wealth index, or education - from the most to the least disadvantaged. Each person is assigned a fractional rank giving their relative position in the SES distribution, and when survey weights are present the ranks are computed on the weighted cumulative population. This ranking, not the raw level of income, is the axis along which inequality is measured, which is precisely why the index isolates the social gradient in health rather than mixing it with how unequal incomes happen to be. Ties and the choice of SES variable both matter here and are decisions to record explicitly.
Formula
R_i = \frac{i - 0.5}{n},\quad i = 1,\dots,n\ \text{(fractional rank by ascending SES)}With people ordered by SES, the concentration curve plots the cumulative proportion of the health variable held by the cumulative proportion of the population, starting from the poorest. If the curve lies above the equality diagonal, the health variable is concentrated among the worse-off; if below, among the better-off; and if it coincides with the diagonal, there is no socioeconomic gradient. The curve is the full graphical object the index summarizes, and inspecting it can reveal whether inequality is driven by the tails or spread across the distribution. Comparing two curves can establish dominance - one situation being unambiguously more unequal - in a way the single index alone cannot.
Formula
L(p) = \text{cumulative share of health among the poorest } p \text{ fraction of the population}The concentration index is defined as twice the area between the concentration curve and the diagonal, equivalently one minus twice the area under the curve. Kakwani and colleagues give the practical covariance formula, in which the index equals twice the population-weighted covariance between the health variable and the fractional rank, divided by the mean health level. This makes computation a matter of a single weighted sum once ranks and the mean are in hand. The resulting number lies between minus one and plus one, with the sign indicating the direction of the gradient and the magnitude its strength, so a value of, say, minus 0.15 signals appreciable concentration of ill health among the poor.
Formula
C = 1 - 2\!\int_0^1 L(p)\,dp = \frac{2}{n\mu}\sum_{i=1}^{n} h_i R_i - 1For inference, Kakwani, Wagstaff, and van Doorslaer derive a 'convenient regression' in which a simple transformation of the health variable is regressed on the fractional rank; the ordinary-least-squares slope coefficient equals the concentration index. This trick delivers the point estimate and, with a correction for the serial structure induced by the ranks, a valid standard error in any regression package. It also opens the door to decomposing the index into contributions from determinants of health, since the regression framework can be extended to attribute inequality to its sources. The regression route is why the concentration index is so easy to deploy in standard statistical software.
Formula
2\sigma_R^2\,\frac{h_i}{\mu} = \alpha + \beta R_i + \varepsilon_i,\qquad \hat{C} = \hat{\beta}A subtlety arises when the health variable is bounded - a binary indicator or a 0-100 score - because the standard index then depends on the mean and cannot reach its theoretical limits, making comparisons across populations with different prevalence unfair. Several corrections address this; the Erreygers normalization rescales the index by four times the mean over the variable's range so that it behaves consistently and satisfies desirable axioms for bounded outcomes, while the Wagstaff normalization offers an alternative for binary variables. Choosing and reporting the appropriate correction is essential whenever the outcome is not unbounded and continuous, otherwise apparent differences in inequality may be artifacts of differing average health.
Formula
E = \frac{4\mu}{h_{\max} - h_{\min}}\,C\ \text{(Erreygers correction for bounded } h)Common pitfalls
- Comparing standard concentration indices across populations with different mean health without applying a bounded-outcome correction.
- Treating the index as a measure of absolute inequality when it is purely relative and rank-based.
- Reporting a near-zero index without inspecting the curve, thereby missing a crossing curve that hides offsetting inequalities.
- Computing standard errors that ignore the dependence induced by the ranks, understating uncertainty.
Frequently asked questions
What do the sign and magnitude of the concentration index mean?
The index runs from minus one to plus one. A value of zero means the health variable is distributed independently of socioeconomic status - no gradient. A negative value means the outcome is concentrated among the poor, which for an adverse outcome like illness signals pro-rich inequality; a positive value means it concentrates among the better-off. The further from zero, the stronger the concentration. So a concentration index of minus 0.2 for child stunting indicates that stunting is substantially more prevalent among poorer children.
How does the concentration index differ from the Gini coefficient?
The Gini measures inequality in a single variable - say income - ranking people by that same variable. The concentration index measures inequality in one variable (health) while ranking people by a different variable (socioeconomic status). This bivariate, rank-by-SES construction is what lets it capture the social gradient in health specifically. If you ranked people by their health and measured health inequality, you would get the health Gini; ranking by income instead yields the concentration index, which is the quantity health-equity research cares about.
Why does the index need correction for binary or bounded outcomes?
For an unbounded continuous variable the concentration index ranges freely over minus one to plus one, but for a bounded variable - a 0/1 indicator or a capped score - its attainable range shrinks and depends on the mean. That means two populations with identical underlying inequality but different prevalence can show different indices, biasing comparisons. Normalizations such as the Erreygers correction rescale the index by the mean and range so it behaves consistently and satisfies sensible axioms, making it safe to compare across populations.
Related methods
Sources
Wagstaff, A., Paci, P., & van Doorslaer, E. (1991). On the measurement of inequalities in health. Social Science & Medicine, 33(5), 545-557.
Open DOI referenceCopied.Kakwani, N., Wagstaff, A., & van Doorslaer, E. (1997). Socioeconomic inequalities in health: Measurement, computation, and statistical inference. Journal of Econometrics, 77(1), 87-103.
Open DOI referenceCopied.