2024-09-01-世界银行-在没有消费数据的情况下估算贫困指标_探索性分析（英）_79页_1mb

Survey-to-Survey Imputation: The paper proposes using existing household surveys to estimate poverty indicators without direct consumption data, addressing data scarcity issues in poorer countries.

Model Performance: Adding household utility or food expenditures to basic demographic/employment models significantly improves the accuracy of poverty estimates for most indicators (e.g., Model 9 using utility expenditures shows higher precision than basic models).

Geospatial Variables: Incorporating geospatial data (e.g., soil quality, nightlights) enhances imputation accuracy, especially for the headcount poverty index.

Utility vs. Food Expenditures: Including household utility consumption (electricity, water, garbage) often proves more effective for certain indicators like near-poverty rates compared to adding food expenditures alone.

Machine Learning: Alternative methods like LASSO and Random Forest do not consistently improve imputation accuracy beyond traditional linear models.

Cross-Country and Income Variances: Model performance varies by country and indicator; however, the findings generalize across different regions and income levels.

Time Intervals: Larger time gaps between base and target surveys reduce imputation accuracy, irrespective of model goodness-of-fit (R-squared).

Cost-Effectiveness: The approach provides cost-saving inputs for future survey design while maintaining reliable poverty estimates for policy decisions.