2010年-世界发展银行全球_Small_Area_Estimation_of_Poverty_in_Rural_Bhutan_39页_1mb

Summary of "Small Area Estimation of Poverty in Rural Bhutan"

Core Content

This report outlines the process and findings of the Small Area Estimation (SAE) of poverty in rural Bhutan, conducted jointly by the National Statistics Bureau (NSB) and the World Bank. The goal of the project was to provide more detailed poverty estimates at the gewog (sub-district) level, enabling better targeted resource allocation and poverty alleviation strategies.

The methodology used the Elbers et al. (2003) approach, which combines census data with survey data to estimate poverty at a finer geographic level. The Bhutan Living Standard Survey (BLSS) 2007 and the Population & Housing Census (PHCB) 2005 were the primary data sources. The BLSS provided detailed consumption and socio-economic data, while the PHCB offered comprehensive demographic and geographic information.

Main Points

Poverty Mapping Methodology

• The SAE method allows for reliable poverty estimates at the gewog level, which traditional survey-based methods cannot achieve due to sampling errors.
• The method involves imputing consumption expenditures for census households using a regression model derived from BLSS data.
• The model includes explanatory variables (household and individual characteristics) that are available in both census and survey data.
• The model is estimated using Feasible Generalized Least Squares (FGLS) to account for regression disturbances and error structures.

Technical Challenges

• A short time interval between PHCB 2005 and BLSS 2007 (about two years) was considered, but there were concerns about changes in consumption patterns and population distribution over this period.
• The Tarrozi and Deaton (2008) critique raised concerns about error correlation at higher levels of disaggregation, such as gewog and district levels.
• The PovMap2 software was used for poverty mapping and training, and it accommodates two layers of errors (household and cluster level).

Data and Model Selection

• Three consumption models were developed: large urban, small/medium urban, and rural.
• The rural model had the lowest R-squared (0.50) and adjusted R-squared (0.49), indicating moderate explanatory power.
• The cluster effect (chiwog or town/city level) was found to contribute 12% to the total residual variance in the rural model, which is above the recommended 10%.
• The error structure was adjusted to include gewog-level variables, which helped reduce the risk of underestimating standard errors.

Key Findings

• The poverty headcount rate in rural Bhutan was 30.9%, with high poverty rates observed in Zhemgang, Samtse, Monggar, and Lhuntse dzongkhags.
• Poverty is strongly associated with limited market access, low electrification, and poor access to education.
• Densely populated but non-poor gewogs in Paro, Chukha, Thimphu, and Punakha have high electrification rates.
• Poor gewogs tend to have lower school attendance rates than non-poor ones.
• The poverty maps were validated using statistical tests, and the results were deemed reliable at the gewog level.
• The standard errors of poverty estimates increased significantly when the cluster level was moved from chiwog to gewog, with a ratio of 2.3–2.4.

Conclusion

• Poverty maps at the gewog level are an effective tool for targeted resource allocation and poverty reduction strategies.
• The project emphasizes the importance of capacity-building, with training sessions provided to RGOB staff.
• The integration of poverty maps with geographical data (e.g., market accessibility, education, electrification) can provide valuable insights for development planning.
• The methodology and results are intended to serve as a guide for future poverty mapping efforts in Bhutan.

Key Information

• Main data sources: BLSS 2007 and PHCB 2005.
• Methodology: Elbers et al. (2003) SAE method.
• Key findings: Rural poverty is high in certain dzongkhags, and poverty is linked to market access, electrification, and education.
• Technical challenges: Error correlation at gewog and district levels, changes in consumption patterns, and population distribution.
• Model selection: Three consumption models were used (urban and rural), with rural having the lowest predictive power.
• Standard errors: Increased significantly when moving from chiwog to gewog level.
• Quality checks: The poverty estimates at the gewog level were found to be reliable and transparent.
• Implications: Poverty maps can support better policy design, monitoring, and targeted interventions.

Annexes

• Annex I: Results of Rural Poverty Map.
• Annex II: Results of Consumption Models.
• Annex III: Accessibility Indicators.
• Annex IV: National poverty line, BLSS data, and Census specifics.
• Annex V: Standard errors and statistically distinguishable rankings of poverty rates at the gewog level.