2024-09-17-世界银行-通过整合调查和地理空间数据对四个西非国家的贫困进行小面积估算（英）_33页_1mb

Small Area Estimation of Poverty in West Africa Using Survey and Geospatial Integration

• Purpose and Context: This paper presents a methodology for small area estimation (SAE) of poverty in four West African countries (Chad, Guinea, Mali, Niger) where recent census data are unavailable. It leverages geospatial data and survey information to generate experimental estimates at finer administrative levels, addressing data scarcity and improving the timeliness and precision of poverty assessments. Burkina Faso (with recent census data) is used for an evaluation.

• Methodology: The approach employs an empirical best predictor (EBP) under a nested error regression model, using geospatial covariates derived from satellite, infrastructure, or population data (e.g., nightlights, elevation, building density) aggregated to grids. Direct survey estimates are combined with these covariates for model-based predictions, avoiding reliance on outdated census data. This is contrasted with area-level models and direct estimation, with adjustments for sample weights.

• Key Findings:

  • Geospatial data enables feasible SAE even without censuses, with efficiency gains demonstrated through reduced coefficient of variation (CV) in sampled areas (reductions up to ~68%).
  • In Burkina Faso, geospatial estimates correlated highly with census-based ones in sampled areas but less so in unsampled regions, indicating context-dependent reliability.
  • The unit context model (household-level with grid covariates) outperformed area-level models in accuracy and precision.
  • Challenges include potential biases, lower predictive power in unsampled zones, and the need for careful weight handling.

• Conclusion: Integrating geospatial data offers a pragmatic and efficient alternative for interim poverty mapping in data-scarce regions. While yielding significant improvements over direct survey estimates, the method requires further validation, especially for out-of-sample predictions, under future censuses. This approach supports targeted interventions and timely monitoring but should be complemented with caution.