世界发展银行-Using-Poverty-Maps-to-Improve-the-Design-of-Household-Surveys---The-Evidence-from-Tunisia_27页_1mb

Summary of "Using Poverty Maps to Improve the Design of Household Surveys: The Evidence from Tunisia"

Core Content

This paper introduces a new method for improving the design effect of household surveys by incorporating poverty maps into the selection of Primary Sampling Units (PSUs). The method leverages previously conducted poverty mapping exercises, which provide fine-grained spatial poverty estimates, to implement implicit stratification in the survey design. This approach enhances the precision of survey estimates and reduces the required sample size, thereby lowering survey costs.

Main Points

• Improving Survey Design:

  • The proposed method uses two-stage sampling with implicit stratification based on poverty mapping.
  • By using systematic sampling within strata defined by poverty estimates, the method increases the efficiency of the survey design.
  • This results in smaller standard errors and narrower confidence intervals, or reduced sample size needed for the same level of precision.

• Design Effect:

  • The design effect (deft) is a measure of how much the variance of an estimate is affected by the sampling design compared to simple random sampling (SRS).
  • The method aims to reduce the design effect by introducing stratification based on poverty levels, thus improving the accuracy of poverty and inequality estimates.

• Poverty Mapping Methodology:

  • The poverty mapping is based on regression models that use census data and household survey data (EBCNV 2015).
  • The H3-CensusEB estimator is used, which has been shown to be more accurate than the ELL method in terms of bias and mean square error (MSE).
  • The method includes three steps:
    1. Selection of common variables between census and survey.
    2. Estimation of a regression model on survey data to predict per capita consumption expenditure for census households.
    3. Calculation of poverty and inequality indicators using the imputed consumption data, along with their standard errors.

• Implementation in Tunisia:

  • The method is applied to the 2021 Household Budget Survey (EBCNV 2021) in Tunisia.
  • It is based on 24 governorates as explicit strata and uses PSUs as the first stage sampling units.
  • Three sampling scenarios are evaluated:
    1. Simple Random Sampling (SRS) within governorates.
    2. Stratified sampling with systematic sampling based on economic and geographical variables.
    3. Stratified sampling with systematic sampling based on poverty mapping-derived per capita consumption.

• Results and Implications:

  • The poverty mapping method produces more precise estimates at the national level.
  • At regional and governorate levels, the poverty mapping method significantly reduces standard errors and coefficients of variation (cv) compared to the EBCNV 2015 survey.
  • At the PSU level, direct survey estimates are less reliable due to high sampling errors, while poverty mapping estimates remain precise and robust.
  • The method enables more efficient sampling designs, which are essential for cost-effective and accurate data collection.

Key Information

• Poverty mapping is used to disaggregate poverty estimates to small geographic units.
• Implicit stratification is introduced by ordering PSUs according to per capita consumption expenditure.
• Systematic sampling is employed to maximize design effect improvement.
• Variance estimation techniques such as JRR (Jack-knife Repeated Replications) and BRR (Balanced Repeated Replications) are used to evaluate the precision of the estimates.
• The H3-CensusEB estimator is highlighted as superior in terms of bias and MSE to other poverty mapping methods.
• The paper's original contribution is the application of implicit stratification based on poverty mapping in a developing country context.

Structure of the Paper

1. Introduction:

   • Emphasizes the need for accurate and reliable poverty measures.
   • Discusses the importance of design effect in improving survey precision and cost-efficiency.

2. Theory of Design Effects, Variance Estimation, and Resampling Techniques:

   • Explains variance decomposition in two-stage sampling designs.
   • Introduces variance estimation methods like JRR and BRR for complex surveys.
   • Defines design effect (deft) and its components: stratification, clustering, and weighting.

3. Theory of Poverty Mapping and Its Application in Tunisia:

   • Describes the poverty mapping methodology used in Tunisia.
   • Highlights the H3-CensusEB estimator and its advantages.
   • Presents key findings from poverty mapping at the national, regional, and governorate levels.

4. Implementation of the New Sample Design in the 2021 EBCNV:

   • Describes the three sampling scenarios.
   • Demonstrates the benefits of the third scenario (based on poverty mapping) in terms of variance reduction and cost efficiency.

5. Conclusion:

   • Summarizes the main results and methodological contributions.
   • Suggests variations of the method for cases where census or household survey data is not available.

Conclusion

This paper provides a practical and innovative approach to improve the design of household surveys using poverty maps. The method is versatile, applicable to various poverty estimation frameworks, and cost-effective, as it allows for more precise estimates with smaller sample sizes. The Tunisia case study demonstrates the effectiveness of the method, particularly in reducing sampling variance and enhancing the reliability of sub-national poverty estimates.