Survey gaps leave poverty predictions on shaky ground, World Bank researchers say

The World Bank, together with Universidad de Los Andes and Vrije Universiteit Amsterdam, warns that survey-to-survey imputation, a statistical shortcut for estimating poverty when fresh data is lacking—can be dangerously misleading if underlying assumptions fail. The report urges caution, transparency, and renewed investment in reliable household surveys to avoid mismeasuring global poverty.

The World Bank’s Development Data Group, in collaboration with Universidad de Los Andes and Vrije Universiteit Amsterdam, has published a revealing policy research paper that confronts one of the most stubborn obstacles in global development: how to measure poverty when reliable household survey data is missing. The report, Stress Testing Survey-to-Survey Imputation: Understanding When Poverty Predictions Can Fail (August 2025), authored by Paul Corral, Andres Ham, Peter Lanjouw, Leonardo Lucchetti, and Henry Stemmler, dissects both the promise and the pitfalls of a statistical shortcut increasingly used by researchers and governments. Poverty measurement, the authors argue, is not just about numbers; it is about credibility, policy direction, and the ability to target resources where they are needed most.

Why Reliable Surveys Are Hard to Find

The paper begins with an unsparing account of the weaknesses in global poverty monitoring. Household consumption and income surveys, the backbone of poverty statistics, are expensive, logistically demanding, and easily disrupted. Many countries, especially those with high poverty rates, go years without conducting or releasing reliable surveys. India did not publish a nationally representative consumption survey between 2011 and 2022, while Nigeria’s comparable data stopped in 2018. These long gaps create serious blind spots. Without fresh data, governments cannot properly calibrate social programs, and international organizations misjudge progress toward global goals. The COVID-19 pandemic magnified these vulnerabilities, as lockdowns halted survey fieldwork and revealed the fragility of systems that depend on door-to-door data collection.

Survey-to-Survey Imputation: A Statistical Shortcut

To fill these voids, economists increasingly rely on survey-to-survey (S2S) imputation. The technique links household traits, such as demographics or education levels, to consumption in one “source” survey, then applies that model to a “target” survey that lacks consumption data. By doing so, researchers generate fresh poverty estimates without conducting a new consumption survey. The appeal is obvious: it is faster, cheaper, and can keep statistics flowing in between survey rounds. S2S stems from earlier poverty-mapping methods but is broader in ambition, seeking to estimate national poverty levels rather than local variations. Yet, the authors caution, the method is perilous if its underlying assumptions do not hold. It presumes that the relationship between household characteristics and welfare remains stable across time, space, and surveys. Whenever shocks, structural changes, or survey inconsistencies intervene, the results can be badly distorted.

Mixed Global Evidence on Its Reliability

The study examines past applications of S2S and finds both successes and failures. In Vietnam and China, poverty estimates generated through imputation proved reasonably accurate, diverging only slightly from full survey benchmarks. In contrast, experiences in Malawi and Jordan revealed serious flaws, with shifts in survey design or changes in household economies leading to misleading estimates. India offers perhaps the most consequential example: for more than a decade, every poverty estimate for the country relied on S2S in the absence of new official data. That reliance underscores its importance but also highlights the danger of treating imputation as a replacement for proper surveys. The review demonstrates that different model specifications often produce different results, and there is no surefire way to know which will perform best in advance.

Stress Testing the Models

To move from case studies to rigorous evidence, the authors conducted detailed simulations, effectively stress-testing S2S under controlled scenarios. The findings are sobering. Standard corrections such as reweighting or covariate adjustments frequently fail when source and target surveys differ substantially. In practice, S2S tends to reproduce the welfare distribution of the source survey, making it unreliable for detecting genuine changes in inequality or poverty trends. Omitting rapidly changing variables, such as remittances, sudden employment shocks, or conflict exposure, can lead models to misrepresent poverty dynamics entirely. Moreover, statistical quirks like non-normal error distributions or heteroskedasticity introduce distortions that conventional methods cannot easily fix. Even advanced machine learning models, though somewhat more flexible, risk overfitting and generating misleading outputs.

The simulations underscore a blunt truth: S2S is not a neutral or fail-safe tool. It carries built-in biases that can mislead policymakers, especially in periods of economic upheaval. Without careful validation against other indicators such as GDP growth, labor market statistics, or administrative data, the results may offer a false sense of precision.

Caution, Transparency, and Investment in Data

The authors urge practitioners to use S2S with humility and restraint. It should be treated as an interim solution during crises, not as a replacement for properly conducted household surveys. Whenever S2S results are presented, they must be accompanied by clear caveats, transparency about assumptions, and disclosure of uncertainty ranges. Combining S2S with structural models that allow coefficients to vary over time could help capture real economic shifts, while statistical refinements such as bootstrapping may reduce error. Still, the paper’s core recommendation is not technical but political: the international community must invest in better survey systems rather than rely excessively on imputation.

The bigger message is that poverty statistics are only as trustworthy as the data behind them. While S2S can help bridge temporary gaps, leaning on it too heavily risks distorting the story of poverty reduction. Inaccurate numbers can mislead governments, misallocate resources, and misinform the global community about progress toward ending extreme poverty. The report warns against complacency: sophisticated methods cannot substitute for the hard but essential task of regularly gathering high-quality, representative household data.