Health Inequality Gini Decomposition
- Introduced by
- Robert Lerman & Shlomo Yitzhaki; Adam Wagstaff & Eddy van Doorslaer
- Year
- 1985
- Type
- Inequality-measurement and decomposition pipeline
- Family
- Health-inequality measurement / total (inter-individual) inequality
What does it do?
The Gini coefficient is the most familiar single-number summary of inequality, and applied to a health variable it captures total, inter-individual health inequality — how unequally health is distributed across all people, regardless of their socioeconomic position.
Its real analytic power comes from decomposition. Robert Lerman and Shlomo Yitzhaki's 1985 covariance formulation rewrites the Gini as twice the covariance between health and its rank divided by the mean, which makes it decomposable into the contributions of separate sources or components, each weighted by its share, its own Gini, and its Gini correlation with the overall distribution. The same machinery supports a between-versus-within-group split. As Wagstaff and van Doorslaer's review of health-inequality measurement explains, this 'pure' inequality view complements socioeconomic measures like the concentration index: the Gini asks how unequal health is, while the concentration index asks how that inequality is patterned by income or rank.
The intuition behind the method
Line everyone up from worst to best health and draw the Lorenz curve — the cumulative share of total health held by the cumulative share of people. The Gini is the gap between that curve and the line of perfect equality: zero when everyone is equally healthy, larger as health concentrates in fewer people. Lerman and Yitzhaki's insight is that the Gini equals twice the covariance between a person's health and their rank in the distribution, divided by mean health. Writing it this way turns a geometric area into an algebraic object you can take apart. If health is the sum of components — say physical and mental health, or contributions of different conditions — you can ask how much each component adds to overall inequality, accounting not just for how unequal that component is on its own but for how it lines up with the overall ranking. A component that is unequal but uncorrelated with who is healthy overall contributes little; one that reinforces the existing ranking contributes a lot.
History
The Gini coefficient, introduced by Corrado Gini in 1912 for income inequality, was reframed for decomposition by Robert Lerman and Shlomo Yitzhaki, whose 1985 Review of Economics and Statistics paper expressed it as a covariance and derived its decomposition by income source, including the marginal-effect formula. Health economists imported these ideas to measure inequality in health itself, distinguishing total inter-individual inequality (the health Gini) from inequality patterned by socioeconomic position, a programme advanced by Richard Illsley, Julian Le Grand, and especially Adam Wagstaff and Eddy van Doorslaer. Their work, reviewed in Wagstaff and van Doorslaer's 2000 Annual Review of Public Health article, set out the conceptual and methodological foundations and clarified when the Gini versus the concentration index is the right tool, shaping the modern health-inequality measurement literature.
When to use it
Use the health Gini and its decomposition when you want to quantify total, inter-individual inequality in a cardinal or ratio-scaled health variable — such as a health-utility index, height, a biomarker, or a continuous self-rated-health score — and especially when you want to attribute that inequality to additive sources or to between- versus within-group differences. The source decomposition is appropriate when health (or a related quantity) is naturally a sum of components and you want each component's marginal contribution. Choose the concentration index instead when the question is how health is distributed across a socioeconomic ranking rather than how unequal it is in absolute terms; choose Theil or other entropy indices when perfect subgroup additivity is required. The Gini needs a meaningful zero and ratio scale, so it is ill-suited to purely ordinal categorical health measures without transformation.
Limitations and considerations
- Requires a ratio-scaled health variable with a meaningful zero; ordinal or bounded self-rated categories need transformation and the choice affects results.
- The subgroup decomposition leaves a residual overlap term, so the Gini is not cleanly additive across groups the way entropy indices are.
- Most sensitive to transfers around the middle of the distribution, which may not match the policy focus on the worst-off tail.
- As a pure inequality measure it ignores the social gradient; on its own it cannot reveal whether the poor or a disadvantaged group bear the burden.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Provides a single, bounded, widely understood measure of total inter-individual health inequality on a 0-1 scale.
- Lerman-Yitzhaki covariance form makes the Gini exactly decomposable into source contributions weighted by share, own-Gini, and Gini correlation.
- Yields the marginal effect of changing a source on overall inequality, directly informing policy on inequality-increasing versus reducing components.
- Distinguishes pure inequality (Gini) from socioeconomically patterned inequality (concentration index), clarifying which question is being answered.
Applications
- Measuring total inter-individual inequality in health-utility scores, biomarkers, or anthropometric outcomes across a population.
- Decomposing inequality in a composite health measure into contributions of its components or conditions.
- Estimating between- versus within-group shares of health inequality across regions, ethnic groups, or birth cohorts.
- Assessing whether a health-affecting transfer or program is inequality-increasing or reducing at the margin via the Lerman-Yitzhaki marginal effect.
- Benchmarking the health Gini against the concentration index to separate the magnitude of inequality from its social patterning.
How it works
How it works
The pipeline starts by expressing the Gini coefficient of the health variable in Lerman and Yitzhaki's covariance form: twice the covariance between each person's health value and their fractional rank in the distribution, divided by the mean. This is algebraically identical to the classic Lorenz-area definition but is far more tractable, because covariances decompose linearly. The output is total inter-individual health inequality on a 0-1 scale, a pure measure that does not reference income or any external ranking variable.
Formula
G = \frac{2\,\mathrm{cov}\bigl(h,\,F(h)\bigr)}{\bar{h}}When health is the sum of additive components or sources, the Gini decomposes into a sum over sources, each term being the product of three interpretable quantities: the source's share of mean health (S_k), the source's own Gini (G_k), and the Gini correlation between that source and the total (R_k). The Gini correlation measures whether a source is distributed in line with the overall health ranking. A source can be highly unequal yet contribute little to total inequality if it is weakly or negatively correlated with overall health, a distinction the source decomposition makes explicit.
Formula
G = \sum_{k} S_k\,G_k\,R_k,\qquad S_k = \frac{\bar{h}_k}{\bar{h}},\ \ R_k = \frac{\mathrm{cov}\bigl(h_k, F(h)\bigr)}{\mathrm{cov}\bigl(h_k, F(h_k)\bigr)}Beyond static shares, the Lerman-Yitzhaki framework yields the marginal effect of a small proportional change in one source on overall inequality. The relative marginal effect is the source's share of total Gini minus its share of the mean. This answers a policy-relevant question directly: would expanding a particular component of health (or a particular transfer or program affecting it) increase or decrease total inequality, holding everything else constant? A positive value means the source is inequality-increasing at the margin.
Formula
\frac{\partial G / \partial e_k}{G} = \frac{S_k\,G_k\,R_k}{G} - S_kThe same covariance machinery supports a between- and within-group decomposition. Total Gini splits into a between-group component (inequality that would remain if everyone within a group had the group mean), a weighted sum of within-group Ginis, and a residual overlap term that arises because, unlike variance, the Gini is not perfectly separable when group distributions overlap. This lets analysts ask how much health inequality is accounted for by differences between social groups versus differences among individuals within them.
Formula
G = G_{\text{between}} + \sum_{g} \pi_g\,\sigma_g\,G_g + R_{\text{overlap}}Finally, results are interpreted and their uncertainty quantified. The Gini lies in 0-1, twice the Gini equals the Gini mean difference scaled by the mean, and standard errors are typically obtained by jackknife or bootstrap because the rank-based statistic has a non-standard sampling distribution. As Wagstaff and van Doorslaer emphasize, the health Gini measures pure inter-individual inequality and should be reported alongside, not instead of, socioeconomic inequality measures such as the concentration index when the question concerns the social gradient in health.
Formula
\widehat{\mathrm{Var}}(G)\ \text{via jackknife};\quad G \in [0,1],\ \ 2G = \text{Gini mean difference}/\bar{h}Common pitfalls
- Applying the Gini to ordinal health categories as if they were ratio-scaled, producing a number with no defensible interpretation.
- Reporting a source's own Gini as its contribution to total inequality while ignoring the Gini correlation that determines its actual share.
- Confusing the health Gini with the concentration index and claiming a socioeconomic-gradient finding from a pure inequality statistic.
- Treating between-group Gini as the full social contribution and ignoring the overlap residual that the decomposition explicitly carries.
Frequently asked questions
How does the health Gini differ from the concentration index?
The Gini measures pure inter-individual inequality: how unequally health is distributed across all people, ranked by their own health. The concentration index measures socioeconomic inequality: how health is distributed across people ranked by income or another social variable. As Wagstaff and van Doorslaer stress, these answer different questions. A population can have substantial total health inequality (high Gini) with little socioeconomic gradient (near-zero concentration index), or vice versa. They are complements, and which to report depends on whether the concern is overall dispersion or the social gradient.
What does the Gini correlation in the source decomposition add?
It captures whether a component is distributed in line with the overall health ranking. In the Lerman-Yitzhaki decomposition each source contributes its share of the mean times its own Gini times its Gini correlation with the total. A component can be very unequal on its own yet contribute little to total inequality if it does not track who is healthy overall — for example if it is high for some healthy and some unhealthy people alike. Ignoring the Gini correlation and reading only the component's own Gini badly misattributes the sources of inequality.
Can I use the Gini on ordinal self-rated health?
Not directly. The Gini requires a ratio scale with a meaningful zero, because it is defined relative to mean health and rests on the Lorenz curve. Ordinal categories like excellent/good/fair/poor have no defensible ratios, so a raw Gini on coded categories is uninterpretable and the result depends arbitrarily on the coding. Options are to map categories to a cardinal health-utility scale with external valuation, or to use measures designed for ordinal data. For purely ordinal socioeconomic-gradient questions, rank-based indices such as the concentration index are generally more appropriate.
Related methods
Sources
Lerman, R. I., & Yitzhaki, S. (1985). Income inequality effects by income source: A new approach and applications to the United States. The Review of Economics and Statistics, 67(1), 151-156.
Open DOI referenceCopied.Wagstaff, A., & van Doorslaer, E. (2000). Income inequality and health: What does the literature tell us? Annual Review of Public Health, 21, 543-567.
Open DOI referenceCopied.