2018年-世界发展银行全球_Household_Expenditure_and_Poverty_Measures_in_60_Minutes___A_New_Approach_with_Results_from_Mogadishu_25页_882kb

Summary of "Household Expenditure and Poverty Measures in 60 Minutes: A New Approach with Results from Mogadishu"

Core Content

This paper introduces a new methodology for estimating household consumption and poverty in fragile and insecure contexts where face-to-face interviews are limited in duration. The approach, called the rapid consumption survey methodology, is designed to reduce interview time while maintaining the accuracy of poverty measurements. It is particularly relevant for places like Mogadishu, Somalia, where security concerns restrict interviews to 60 minutes.

Main Viewpoints

• Traditional Methods Limit Applicability: Full household consumption surveys typically take 90–120 minutes, which is impractical in insecure environments.
• Reduced Consumption Methodology Flaws: It underestimates consumption, leading to overestimation of poverty.
• Multiple Visits Increase Costs and Security Risks: Attrition and the need for repeated visits make this approach less viable.
• Baseline Surveys Not Always Available: In some contexts, such as Mogadishu, baseline surveys are not feasible, limiting the use of existing methodologies.
• New Methodology Combines Design and Imputation: It uses a core module and optional modules to reduce time while employing imputation techniques to estimate missing data.

Key Information

Methodology Overview

The methodology consists of five main steps:

1. Core Module Selection: Core items are selected based on their contribution to overall consumption.
2. Partitioning into Optional Modules: Remaining items are split into optional modules using an algorithm that ensures orthogonality within modules and correlation between them.
3. Random Assignment of Modules: Each household is randomly assigned one optional module, stratified by enumeration areas.
4. Imputation of Missing Data: Techniques such as summary statistics, regression models, and multiple imputation are used to estimate consumption for non-assigned modules.
5. Poverty Estimation: The final consumption aggregates are used to estimate poverty indicators.

Consumption Estimation Techniques

• Summary Statistics: Uses average or median of module-specific consumption to estimate missing values.
• Module-wise Regression: Applies Ordinary Least Squares (OLS) and Tobit regression to estimate consumption based on household characteristics and core consumption.
• Multiple Imputation Chained Equations (MICE): Imputes missing values iteratively using regression models.
• Multi-Variate Normal Regression (MImvn): Uses an EM-like algorithm with a Data-Augmentation (DA) approach to estimate missing data.

Performance Evaluation

• Ex Post Simulation in Hargeisa: Used as a benchmark to evaluate the methodology.
• Results:
  • Summary Statistics (average and median) under-estimate or over-estimate consumption, leading to biased poverty estimates.
  • Regression Techniques show considerable upward bias.
  • Multiple Imputation Techniques (MICE and MImvn) perform best, with a bias below 1% and relative standard errors ranging from 1% to 3%.
  • FGT0 (poverty headcount) is estimated with virtually unbiased results using multiple imputation.

Application in Mogadishu

• CAPI Technology Used: To collect data in 60 minutes.
• Data Collection Time: Average of 40 minutes per visit, with 90% of interviews under 65 minutes.
• Data Retention: 675 households were retained after data cleaning.
• Welfare Model Performance:
  • Food consumption model: $ R^2 = 0.24 $
  • Non-food consumption model: $ R^2 = 0.16 $
• Module Assignment Refinement: Required manual adjustments due to differences in consumption patterns between Hargeisa and Mogadishu.

Key Findings

• Core Module Captures Majority of Consumption: In Hargeisa, the core module captures 92% of food and 88% of non-food consumption. In Mogadishu, it captures 64% and 62% respectively.
• Multiple Imputation Techniques Provide Accurate Estimates: These techniques are particularly effective in fragile contexts, delivering reliable poverty and inequality indicators.
• Bias and Standard Error: Multiple imputation techniques have lower bias and standard error compared to other methods, especially at the cluster and global levels.
• Consistency Checks: Applied to validate the methodology, showing that the approach can yield credible results even with limited data.

Conclusion

• Rapid Consumption Survey is Effective: It significantly reduces interview time while maintaining the reliability of poverty estimates.
• CAPI Enhances Implementation: The use of CAPI (Computer-Assisted Personal Interviewing) helps in reducing enumerator bias and improving data collection efficiency.
• Need for Further Research: The methodology requires further validation and refinement, especially in different contexts and with more diverse data sets.

Figures and Tables

• Figure 1: Illustration of the rapid consumption survey methodology.
• Figure 2: Average relative bias across six estimation techniques.
• Figure 3: Average relative standard error across six estimation techniques.
• Figure 4: Average bias for poverty and inequality indicators (FGT0, FGT1, FGT2, Gini coefficient).
• Figure 5: Average standard error for poverty and inequality indicators.
• Table 1: Number of items and consumption share per module in Hargeisa.
• Table 2: Module composition and consumption shares in Hargeisa and Mogadishu.
• Table 3: Performance metrics for different estimation techniques.

This approach offers a promising solution for rapid and reliable poverty measurement in insecure and fragile environments.