Watts Poverty Index
- Introduced by
- Harold W. Watts (1968); axiomatized by Buhong Zheng (1993)
- Year
- 1968
- Type
- Distribution-sensitive poverty measure
- Family
- Income poverty measurement
What does it do?
The Watts index, proposed by Harold Watts in 1968, was the first poverty measure to be sensitive to the distribution of income among the poor, anticipating the axiomatic poverty-measurement literature by nearly a decade.
It averages, over the whole population, the natural logarithm of the ratio of the poverty line to each poor person's income. Because the log gives ever-larger weight to incomes near zero, the Watts index satisfies the strong transfer principles that the headcount and the linear poverty gap fail, and Buhong Zheng's 1993 axiomatic characterization established it as the smallest distribution-sensitive measure satisfying a natural set of axioms.
The intuition behind the method
Think of how much worse it feels to lose a dollar when you already have almost nothing versus when you are close to the line. The logarithm captures exactly this: the ratio of the poverty line to income, ln(z/y), explodes as income falls toward zero, so a poor person who is destitute contributes far more than one who is just below the line. Averaging these log shortfalls over everyone gives a poverty measure that not only counts the poor and measures their average shortfall but also penalizes the concentration of severe deprivation. Move a dollar from a slightly-poor person to a desperately poor one and the Watts index falls, because the log curvature rewards reducing the worst deprivation.
History
Harold W. Watts introduced the index in a 1968 essay 'An Economic Definition of Poverty', arguing that poverty should be measured by a continuous, distribution-sensitive function of the shortfall rather than by a simple headcount. The idea lay relatively dormant until Amartya Sen's 1976 axiomatic program revived interest in distribution-sensitive measures; Anthony Atkinson, James Foster, and others then situated Watts within the modern axiomatic framework. Buhong Zheng's 1993 paper provided a formal axiomatic characterization, showing the Watts index to be a minimal measure satisfying continuity, monotonicity, the transfer axiom, and subgroup consistency, cementing its status as a benchmark distribution-sensitive poverty measure.
When to use it
Use the Watts index when distributional concern among the poor is central and you want a measure with strong axiomatic credentials — it satisfies monotonicity, the transfer axiom, transfer sensitivity, subgroup decomposability, and continuity, and it underlies clean poverty-dominance and growth-decomposition results. It is especially appropriate for evaluating policies that target the poorest of the poor, where a measure insensitive to the distribution below the line (such as the headcount or linear gap) would miss the point. Report it alongside the headcount and FGT measures for a fuller picture. The chief practical caution is that the index needs strictly positive welfare values, so it is typically computed on consumption rather than income, and recorded zeros or negatives must be censored at a small floor before logs are taken.
Limitations and considerations
- Undefined for non-positive incomes because it takes logarithms, requiring censoring at a positive floor that can materially affect the value.
- Highly sensitive to incomes near zero, so measurement error or misreporting at the bottom of the distribution can dominate the index.
- Expressed in log-welfare units, it is less directly interpretable to non-specialists than the headcount or the per-capita poverty gap.
- Like all unidimensional money-metric measures it depends on an exogenous poverty line and ignores non-income dimensions of deprivation.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Satisfies the strong transfer axiom and transfer sensitivity, so it genuinely reflects inequality and the severity of deprivation among the poor.
- Was the first distribution-sensitive poverty measure and has a clean axiomatic characterization (Zheng 1993) as a minimal such measure.
- Additively decomposable across subgroups, enabling poverty profiles, and central to Watts-based poverty-dominance and TIP-curve analysis.
- Underpins exact growth-redistribution and elasticity results, making it analytically convenient for policy decompositions.
Applications
- Evaluating anti-poverty policies whose value lies in helping the poorest of the poor, where distribution below the line matters most.
- Constructing TIP (Three I's of Poverty) curves to display the incidence, intensity, and inequality of poverty simultaneously.
- Serving as the welfare measure in poverty-dominance tests and in growth-redistribution decompositions with exact elasticity properties.
- Building decomposable poverty profiles across regions and demographic groups using its additive structure.
- Comparing the severity of poverty across countries and over time when distribution-sensitivity is required.
How it works
How it works
As with all money-metric measures, fix a poverty line z and identify the poor as everyone whose welfare y_i falls below z. Only the poor enter the sum (the focus axiom), and there are q of them out of a population of N. The Watts index requires strictly positive incomes for the poor because it takes logarithms; zero or negative recorded incomes must be handled by censoring at a small positive floor or by using consumption, which is rarely non-positive.
Formula
\text{poor} = \{\, i : y_i < z \,\}, \qquad q = \#\{\, i : y_i < z \,\}The index is the mean over the whole population of the log ratio of the line to income among the poor: W = (1/N) sum over the poor of ln(z/y_i), equivalently the sum of (ln z − ln y_i). Each term is the proportional shortfall expressed in log points, which grows without bound as income approaches zero. Averaging over N (not q) preserves additive decomposability, mirroring the FGT convention, and gives the index its per-capita interpretation in log-welfare units.
Formula
W = \frac{1}{N} \sum_{i : y_i < z} \ln\!\left(\frac{z}{y_i}\right) = \frac{1}{N} \sum_{i : y_i < z} \bigl[\ln z - \ln y_i\bigr]The second derivative of ln(z/y_i) with respect to income is strictly positive (1/y_i squared), so the contribution function is strictly convex in income. This convexity is precisely what makes the Watts index satisfy the (strong) transfer axiom: a mean-preserving transfer from a poorer to a less-poor person below the line strictly raises the index, and the reverse strictly lowers it. The headcount fails this axiom entirely and the linear poverty gap satisfies it only weakly, which is why Watts is called distribution-sensitive.
Formula
\frac{\partial^2}{\partial y_i^2}\,\ln\!\left(\frac{z}{y_i}\right) = \frac{1}{y_i^{2}} > 0 \;\Rightarrow\; \text{strict transfer axiom}Like the FGT measures, the Watts index is additively decomposable: total poverty equals the population-weighted sum of subgroup indices, W = sum of (N_k/N) W_k, supporting poverty profiles. It is also the basis of the TIP (Three I's of Poverty) curve and of Watts-based poverty-dominance results: cumulating the ordered log shortfalls traces a curve whose incidence, intensity, and inequality components can be read off, and second-order dominance of these curves implies unambiguous Watts (and broader) poverty rankings.
Formula
W = \sum_{k=1}^{K} \frac{N_k}{N}\, W_k, \qquad TIP_W(p) = \frac{1}{N}\sum_{i=1}^{\lceil pN \rceil} \ln\!\left(\frac{z}{y_{(i)}}\right)Common pitfalls
- Applying the log to recorded zero or negative incomes, which makes the index undefined or implausibly large; censoring at a floor is required and must be reported.
- Comparing Watts values without noting the censoring floor, since the choice of floor strongly influences the contribution of the very poorest.
- Treating the index's units as money; it is in log-welfare units, so its magnitude is not a per-capita dollar shortfall.
- Ignoring that extreme sensitivity at the bottom magnifies bottom-tail measurement error relative to the headcount or linear gap.
Frequently asked questions
How does the Watts index differ from the FGT squared poverty gap?
Both are distribution-sensitive and satisfy the transfer axiom, but they use different curvature. The squared poverty gap (FGT alpha = 2) weights the squared normalized shortfall, which is bounded between zero and one and grows quadratically. The Watts index weights the log of the line-to-income ratio, which is unbounded and grows ever faster as income approaches zero, so Watts gives stronger weight to extreme deprivation. Watts is also the natural welfare measure behind several exact decomposition and dominance results, whereas the squared gap is more directly interpretable as a normalized quantity.
Why must incomes be strictly positive for the Watts index?
Because the index takes the natural logarithm of income, ln(z/y_i), which is undefined at zero and undefined (or complex) for negative values, and which diverges to infinity as income approaches zero. Surveys frequently record zero or negative incomes (for example, business losses or no recorded earnings), so in practice analysts either use consumption — which is rarely non-positive — or censor income at a small positive floor before computing the index. The chosen floor should be reported, since it influences the contribution of the poorest.
Is the Watts index decomposable across population subgroups?
Yes. Because it is a population-weighted average of individual contributions, total poverty equals the sum over subgroups of the subgroup population share times the subgroup Watts index. This additive decomposability lets analysts apportion national poverty to regions or demographic groups, exactly as with the FGT family, and is one of the axioms in Zheng's characterization of the measure.
Related methods
Sources
Zheng, B. (1993). An axiomatic characterization of the Watts poverty index. Economics Letters, 42(4), 347–353.
Open DOI referenceCopied.