深度报告-2025-09-01-世界银行-压力测试调查到调查推断_了解贫困预测何时可能失败（英）页_46页_898kb

Summary of "Stress Testing Survey-to-Survey Imputation: Understanding When Poverty Predictions Can Fail"

Core Content

This paper explores the limitations and potential pitfalls of survey-to-survey (S2S) imputation in the context of poverty measurement. The authors emphasize that while S2S imputation is a practical tool for updating poverty estimates in the absence of direct welfare data, it is not without significant challenges, especially when applied across time or under economic and structural changes. The paper is a technical analysis using simulated data to examine the method's assumptions and performance under various conditions.

Main Viewpoints

1. Challenges in Poverty Measurement

• Traditional poverty measurement relies on household surveys that collect detailed consumption or income data.
• These surveys are costly and complex, often requiring detailed recall and long implementation periods.
• In many countries, especially those with high poverty rates, recent and comprehensive surveys are inaccessible or unavailable, leading to data gaps.

2. What is Survey-to-Survey (S2S) Imputation?

• S2S imputation uses a predictive model from a source survey (with welfare data) to estimate welfare in a target survey (without welfare data).
• It is used to maintain poverty monitoring during data collection disruptions, such as crises, conflicts, or natural disasters.
• The method is based on small area estimation (SAE) techniques and has been widely used in poverty mapping.

3. Limitations and Biases of S2S Imputation

• Sampling bias correction techniques, such as re-weighting, may not be sufficient when source and target surveys differ fundamentally.
• S2S tends to replicate the welfare distribution of the source survey, making it unreliable for measuring changes in poverty or inequality.
• Omitted variable bias is a major concern, especially when imputing across time periods with economic shocks or structural changes.

4. Statistical Foundations

• The method assumes a linear model for the welfare distribution, where:
  $$
  \ln y_i = x_i \beta + e_i; \quad e_i \sim N(0, \sigma_e^2)
  $$
• The predicted poverty probability is derived from:
  $$
  Prob(poor_i) = \Phi\left(\frac{\ln z - x_i \beta}{\sqrt{\sigma_e^2}}\right)
  $$
• Multiple imputation is a common approach, generating plausible values for missing welfare data to reflect uncertainty and improve estimation accuracy.

5. Alternative Methods and Considerations

• Bootstrap-based imputation and predictive mean matching (PMM) are alternative methods that are less sensitive to model misspecification.
• ML techniques, such as lasso regression, can be used in conjunction with S2S, but they are not directly compatible with the traditional multiple imputation framework.
• The number of neighbors in PMM is practitioner-determined and should be adaptive, based on sample size and model performance.

Key Information

• Countries with significant poverty gaps such as India and Nigeria often lack recent and comprehensive household survey data.
• S2S imputation is increasingly used in global poverty monitoring, but its reliability is context-dependent.
• Parameter stability is a critical assumption in S2S, and violations can lead to biased poverty estimates.
• Economic shocks and survey design changes can significantly affect the validity of S2S imputation.
• No foolproof method exists to identify the ideal model for imputation, raising concerns about the robustness of poverty estimates.

Practical Guidance

• S2S imputation should be used cautiously, especially when:
  • There are substantial changes in economic conditions.
  • Survey design or implementation has altered between data collection periods.
  • Time gaps are large, increasing the risk of parameter instability.
• Transparency in communicating limitations is essential for policy design and monitoring.
• Parsimonious models and careful selection of covariates are recommended, especially when training data is not updated.

Conclusion

The paper highlights the importance of understanding the assumptions and limitations of S2S imputation in poverty measurement. While it is a valuable tool in the face of data scarcity, its effectiveness is not guaranteed and depends heavily on the stability of model parameters and the consistency of covariates across surveys. The authors call for greater awareness and rigorous application of S2S to ensure accurate and reliable poverty estimates.