Skip to content
Back to search results

Atkinson Index

Introduced by
Anthony Barnes Atkinson
Year
1970
Type
Welfare-based, parameterized inequality index
Family
Inequality measurement
Font size

What does it do?

The Atkinson index is a welfare-based measure of inequality that incorporates an explicit, analyst-chosen parameter for how much society dislikes inequality.

Introduced by Anthony Atkinson in 1970, it asks what fraction of total income could be discarded, under an equal distribution, while leaving social welfare unchanged — making the ethical judgement behind any inequality comparison transparent rather than hidden.

The intuition behind the method

Most inequality indices bury their value judgements. Atkinson made them explicit. Imagine you could redistribute income perfectly equally; how much smaller could the total pie be and still make society just as well off? If a society is willing to give up 20 percent of its income for the sake of equality, its Atkinson index is 0.20. The key dial is the inequality-aversion parameter epsilon: at zero you do not care about distribution at all (the index is zero regardless), and as it rises you weight the welfare of the poorest ever more heavily, so the same income distribution registers as more unequal.

History

Anthony Barnes Atkinson introduced the index in his landmark 1970 Journal of Economic Theory paper 'On the measurement of inequality', which reframed inequality measurement around explicit social-welfare functions and proved the equivalence between Lorenz dominance and unanimous welfare rankings. The work, foundational to modern inequality economics, gave the field a principled normative basis and inspired the broader axiomatic literature on inequality and poverty measurement, much of it synthesized in Frank Cowell's textbook Measuring Inequality.

When to use it

Use the Atkinson index when you want an inequality measure whose normative content is explicit and tunable — to show how inequality comparisons depend on how much weight is placed on the poor, or to compute the welfare cost of inequality. It is the natural choice for welfare-economic and policy analysis and is subgroup-decomposable. It requires choosing the aversion parameter epsilon, which is a value judgement, not a statistical estimate; results should be reported across a range of epsilon. It is defined for positive incomes (zeros or negatives break the power mean for epsilon ≥ 1) and, like other scalar measures, can rank distributions whose Lorenz curves cross differently depending on epsilon.

Limitations and considerations

  • Requires the analyst to choose the aversion parameter, which is a normative rather than empirical decision.
  • Defined only for positive incomes; zeros and negatives are problematic, especially for epsilon ≥ 1.
  • Less familiar and intuitive to general audiences than the Gini coefficient.
  • Different epsilon values can reverse the inequality ranking of distributions with crossing Lorenz curves.

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

Strengths

  • Makes the ethical inequality-aversion judgement explicit and adjustable through the parameter epsilon.
  • Has a direct welfare interpretation as the share of income society would forgo to eliminate inequality.
  • Subgroup-decomposable, allowing between- and within-group inequality to be separated.
  • By varying epsilon, can be made sensitive to whichever part of the distribution matters for the analysis.

Applications

  • Quantifying the welfare cost of income inequality at varying degrees of inequality aversion.
  • Comparing inequality across countries while making the underlying value judgements explicit.
  • Decomposing inequality into between- and within-group components.
  • Evaluating the distributional impact of tax and transfer policies in welfare terms.
  • Measuring inequality in health, education, or other welfare-relevant outcomes.

How it works

How it works

1

The index starts from a social-welfare function that sums a concave utility of each person's income. Concavity encodes inequality aversion: an extra dollar adds more welfare to a poor person than to a rich one. Different concave utilities correspond to different ethical stances on inequality.

Formula
W = \frac{1}{n}\sum_{i=1}^{n} U(x_i),\quad U\ \text{concave (inequality-averse)}
2

From that welfare function one derives the equally-distributed-equivalent (EDE) income — the income that, if everyone received it, would yield the same social welfare as the actual unequal distribution. For the standard constant-relative-inequality-aversion utility this is a generalized (power) mean of incomes with exponent governed by epsilon.

Formula
x_{\mathrm{EDE}} = \left(\frac{1}{n}\sum_{i=1}^{n} x_i^{\,1-\varepsilon}\right)^{\!\frac{1}{1-\varepsilon}},\ \varepsilon \neq 1
3

The Atkinson index is one minus the ratio of the EDE income to the actual mean income. Because the EDE is at most the mean (and equals it only under perfect equality), the index lies between 0 and 1 and reads directly as the share of mean income society would sacrifice to achieve equality without welfare loss.

Formula
A_\varepsilon = 1 - \frac{x_{\mathrm{EDE}}}{\bar{x}}
4

The parameter epsilon ≥ 0 sets the strength of inequality aversion. At epsilon = 0 the index is always zero (no aversion); as epsilon grows the index puts increasing weight on the bottom of the distribution, and in the limit it reflects only the poorest. Reporting the index at several epsilon values shows how conclusions depend on the ethical stance.

Formula
\varepsilon \ge 0:\ \text{larger } \varepsilon \Rightarrow \text{more weight on the poor}

Common pitfalls

  • Reporting the index at a single epsilon without noting that the chosen value embeds a strong ethical assumption.
  • Applying it to data with zero or negative incomes without adjustment, producing undefined or distorted values.
  • Comparing Atkinson values computed at different epsilon as if they were on the same scale.
  • Interpreting the index as a purely descriptive statistic when it is explicitly normative.

Frequently asked questions

What does the inequality-aversion parameter epsilon control?

Epsilon sets how strongly the index weights the lower end of the distribution. At epsilon = 0 distribution is ignored and the index is zero; as epsilon increases, transfers to poorer people count for more and the same distribution registers as more unequal. Common reported values are 0.5, 1, and 2. Because epsilon encodes an ethical stance, best practice is to report the index across several values.

How is the Atkinson index interpreted as a welfare cost?

An Atkinson value of A means that society could achieve the same level of social welfare with only (1 − A) of its current mean income if that income were distributed perfectly equally — equivalently, inequality is 'wasting' a fraction A of potential welfare. So A = 0.25 says a quarter of income is, in welfare terms, lost to inequality at the chosen aversion level.

How does the Atkinson index relate to the generalized entropy and Gini measures?

The Atkinson index is ordinally equivalent to a member of the generalized entropy (Theil) family for each epsilon, so they rank distributions identically though on different scales, and both are subgroup-decomposable. Unlike the Gini, the Atkinson index has an explicit welfare parameter and clean subgroup decomposition; unlike the Gini's pairwise-difference interpretation, it is grounded in a social-welfare function.

Related methods

Sources

  1. Atkinson, A. B. (1970). On the measurement of inequality. Journal of Economic Theory, 2(3), 244–263.

    Open DOI reference
    Copied.
  2. Cowell, F. A. (2011). Measuring Inequality (3rd ed.). Oxford University Press.

    ISBN: 978-0-19-959403-0
    Copied.