Datt-Ravallion Decomposition
- Introduced by
- Gaurav Datt & Martin Ravallion
- Year
- 1992
- Type
- Poverty-change decomposition
- Family
- Poverty dynamics analysis
What does it do?
The Datt-Ravallion decomposition, introduced by Gaurav Datt and Martin Ravallion in 1992, separates the observed change in a poverty measure between two dates into a growth component — the change attributable to a shift in mean income holding the relative distribution fixed — and a redistribution component — the change attributable to a shift in the Lorenz curve holding mean income fixed.
A residual captures the interaction between the two. It became the standard way to ask whether falling poverty was driven by rising average incomes or by changes in inequality, and underlies the empirical literature on pro-poor growth.
The intuition behind the method
When poverty falls, two very different things could be happening: the whole income distribution could be sliding upward (growth lifting everyone), or its shape could be changing so that the poor capture a larger relative share (redistribution). Datt and Ravallion's idea is to construct counterfactual distributions to isolate each channel. Hold the relative distribution (the Lorenz curve) fixed and let only the mean move: the resulting poverty change is the growth effect. Hold the mean fixed and let only the Lorenz curve move: that change is the redistribution effect. Because the two effects interact — growth at a higher inequality level reduces poverty differently than at a lower one — a residual term picks up the part that cannot be cleanly assigned to either, unless one uses a reference-date-symmetric or Shapley version that eliminates it.
History
Gaurav Datt and Martin Ravallion published the decomposition in 1992 in the Journal of Development Economics, applying it to Brazil and India in the 1980s to ask whether observed poverty changes reflected growth or distributional shifts. The method built on the recognition that poverty measures can be written as functions of the mean and the Lorenz curve, allowing counterfactual experiments that move one while fixing the other. Kakwani (2000) and others later proposed alternative, residual-free formulations, and the approach was subsumed as a special case of Shorrocks's Shapley decomposition framework. It remains the canonical starting point for analyzing the growth-versus-distribution drivers of poverty and for the pro-poor growth literature.
When to use it
Use the Datt-Ravallion decomposition when you have comparable income or consumption distributions at two points in time, a fixed real poverty line, and you want to know whether a change in poverty was driven by economic growth or by changes in distribution. It is the foundational tool of the pro-poor growth literature and is widely used in country poverty assessments to interpret observed poverty trends. The growth elasticity of poverty and the diagnosis of whether growth was 'pro-poor' both flow from it. Be aware that the basic version leaves a residual whose interpretation is awkward; for a clean, base-independent split use the symmetric (Shapley) variant. The decomposition is descriptive accounting, not causal — it does not explain why the mean or the Lorenz curve moved.
Limitations and considerations
- The standard one-reference-date version leaves a residual that is difficult to interpret and depends on the choice of base period.
- It is descriptive accounting, not causal: it does not explain why the mean or distribution changed, nor attribute changes to policies.
- Requires the poverty line to be held fixed in real terms and the two distributions to be genuinely comparable in welfare definition and price deflation.
- Sensitive to survey comparability problems — changes in questionnaire, sampling, or deflators can masquerade as growth or redistribution effects.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Cleanly separates the two policy-relevant channels of poverty change — average income growth versus distributional change.
- Applies to any poverty measure expressible in terms of the mean and the Lorenz curve, including the full FGT family.
- Provides the empirical foundation for measuring pro-poor growth and growth elasticities of poverty.
- Simple to compute from two survey distributions and a fixed line, and widely implemented in poverty software.
Applications
- Diagnosing whether a country's observed poverty reduction was driven by growth or by changes in inequality.
- Measuring pro-poor growth and constructing growth-incidence-curve-based and elasticity-based summaries.
- Comparing the growth and redistribution contributions to poverty change across countries and time periods.
- Informing policy debates on whether to prioritize aggregate growth or redistribution for poverty reduction.
- Serving as a benchmark special case within broader Shapley decompositions of distributional change.
How it works
How it works
The decomposition treats any poverty measure as a function of just two arguments: the ratio of the poverty line to mean income (z/mu), which summarizes the level of average living standards relative to the line, and the Lorenz curve L, which summarizes the relative distribution. The observed change in poverty between dates t and t+n is then the change in this function as both the mean and the Lorenz curve move from their initial to their final values.
Formula
P_t = P\!\left(\frac{z}{\mu_t},\, L_t\right), \qquad \Delta P = P_{t+1} - P_tThe growth component is the change in poverty that would occur if the mean shifted from its initial to its final value while the Lorenz curve were held fixed at a reference distribution L_r. It captures the distribution-neutral effect of rising (or falling) average income: every income scaled proportionally, so the relative shape is unchanged. Holding the Lorenz curve at the initial date gives the conventional Datt-Ravallion growth term.
Formula
G(t, t+n; r) = P\!\left(\frac{z}{\mu_{t+n}}, L_r\right) - P\!\left(\frac{z}{\mu_t}, L_r\right)The redistribution component is the change in poverty that would occur if the Lorenz curve shifted from its initial to its final shape while the mean were held fixed at a reference value mu_r. It isolates the effect of changing inequality — a more equal Lorenz curve reduces poverty for a given mean, a more unequal one raises it. Holding the mean at the initial date gives the conventional redistribution term.
Formula
D(t, t+n; r) = P\!\left(\frac{z}{\mu_r}, L_{t+n}\right) - P\!\left(\frac{z}{\mu_r}, L_t\right)The observed poverty change equals the growth component plus the redistribution component plus a residual R, all evaluated with the initial date as reference. The residual exists because the growth and redistribution effects are not separable when evaluated at a single reference date — it measures the difference between using the initial versus the final period as the base. Choosing the final period as reference relabels the residual; averaging the two (or applying a Shapley decomposition) yields a symmetric, residual-free split.
Formula
\Delta P = G(t, t+n; t) + D(t, t+n; t) + R(t, t+n; t)
Common pitfalls
- Reporting the growth and redistribution components without the residual, or silently folding the residual into one of them, which biases the interpretation.
- Comparing decompositions that use different reference dates without recognizing that the split (and the residual) is base-dependent.
- Using a poverty line that drifts in real terms or distributions deflated inconsistently, so spurious growth/redistribution effects appear.
- Reading the components as causal effects of growth policies versus redistribution policies rather than as an accounting identity.
Frequently asked questions
What is the residual term and how do I get rid of it?
The residual arises because the growth and redistribution effects interact: the poverty impact of a given growth depends on the level of inequality, and vice versa, so when both move, the change cannot be split exactly using a single reference date. The residual equals the difference between evaluating the components at the initial versus the final period. To eliminate it, use a symmetric formulation — average the initial-base and final-base decompositions, equivalent to applying the Shapley decomposition — which assigns the interaction evenly and leaves no residual.
Does the decomposition tell me whether growth was pro-poor?
It is the main input to that judgment. If the growth component accounts for most of an observed poverty reduction and the redistribution component is small or adverse, growth was the dominant but not distribution-improving driver. Pro-poor growth analysis builds on this by comparing the actual poverty reduction with what distribution-neutral growth would have delivered, or by examining the full growth-incidence curve. The decomposition itself is descriptive: it quantifies the two channels but does not by itself define a pro-poor threshold.
Can it be applied to any poverty measure?
Yes, to any poverty measure that can be written as a function of the ratio of the poverty line to mean income and the Lorenz curve, which includes the entire FGT family (headcount, poverty gap, squared gap) and the Watts index. The counterfactual experiments — shifting the mean with the Lorenz curve fixed, and shifting the Lorenz curve with the mean fixed — are well defined for all such measures, so the same growth/redistribution split can be reported for each.
Related methods
Sources
Datt, G., & Ravallion, M. (1992). Growth and redistribution components of changes in poverty measures: a decomposition with applications to Brazil and India in the 1980s. Journal of Development Economics, 38(2), 275–295.
Open DOI referenceCopied.