2010年-世界发展银行全球_Poverty_Maps_of_Bangladesh_2010___Technical_Report_56页_2mb

Summary of the 2010 Poverty Maps of Bangladesh - Technical Report

Core Content

This technical report presents the results of a poverty mapping exercise conducted in Bangladesh using data from the 2010 Household Income and Expenditure Survey (HIES) and the 2011 Population Census. The objective was to estimate poverty at the sub-national level, specifically at zila (district) and upazila (sub-district) levels, to support targeted policy and development interventions.

The poverty mapping methodology employed the ELLS method (Elbers, Lanjouw, and Lanjouw, 2003), which utilizes Small Area Estimation (SAE) techniques to predict consumption data for census households based on a consumption model derived from HIES data. This approach enables more granular poverty estimates than the national and division-level estimates traditionally provided by HIES.

Main Points

1. Poverty Mapping Methodology

• The ELLS method is a globally recognized technique for sub-national poverty estimation.
• It combines data from two sources:
  • HIES: A nationally representative survey with detailed expenditure and income data.
  • Population Census: A comprehensive survey covering all households, but with limited socio-economic variables.
• The consumption model derived from HIES is used to simulate consumption for census households, allowing for poverty rate estimation at zila and upazila levels.
• The PovMap2 software was used for the analysis, which supports up to two layers of error estimation (household and aggregation unit).

2. Data Sources

• HIES 2010: Conducted every 4-5 years, it is the primary source of poverty-related data in Bangladesh.
• 2011 Population Census: Provides data on all households, including demographic, employment, and infrastructure information.
• Both data sets share a common set of explanatory variables, facilitating the integration of data for poverty mapping.

3. Technical Challenges

• A major challenge was the potential change in consumption patterns between the HIES and the census due to the time gap.
• The ELL method assumes that the relationship between consumption and its correlates is consistent across households within a domain, which may not always hold.
• Error correlation at higher levels (e.g., union, upazila) was not fully addressed by the current software configuration, which may lead to understatement of standard errors.
• To mitigate this, only time-invariant variables were selected for the consumption model, and validation exercises were conducted based on indirect empirical evidence.

4. Production of Poverty Maps

• The poverty maps were created using GIS software by the BBS and WFP teams, led by Faizuddin Ahmed.
• The poverty estimates were derived using the upper and lower poverty lines from the HIES 2010 report.
• The maps highlight spatial variation in poverty incidence at the upazila level, which is more detailed than at the division level.

Key Results

1. Poverty Estimates at Division Level

• Poverty rates vary significantly across divisions.
• Rangpur has the highest poverty incidence at 42.3% (using upper poverty line), while Chittagong has the lowest at 26.2%.
• Extreme poverty (lower poverty line) ranges from 13.1% in Chittagong to 27.7% in Rangpur.
• The overall poverty headcount rate for Bangladesh is 30.7%, close to the HIES 2010 estimate of 31.5%.

2. Spatial Variation at Zila and Upazila Levels

• Poverty is highly concentrated in certain areas.
• In Dhaka division, the 10 poorest upazilas have poverty rates of 55% or higher, while the 10 richest have less than 4%.
• In Chittagong division, the 6 poorest upazilas have poverty rates of 50% or higher, and the 6 richest have less than 4%.
• In Rangpur division, the 7 poorest upazilas have poverty rates exceeding 60%, which is twice the national average.
• Sylhet division, despite being one of the more affluent regions, has upazilas with poverty rates over 50%.
• Khulna division has 3 upazilas with poverty rates of 50% or higher.

3. Standard Errors and Disaggregation Levels

• Standard errors of poverty estimates are generally small at the zila and upazila levels, with the median being around 1.2–2.2%.
• At the 95th percentile, standard errors are 4.5–7.7%, which is still acceptable for most policy applications.
• However, the maximum standard error at the union level reaches 21.5%, indicating that poverty estimates at this level may be unreliable.
• Therefore, it is recommended that poverty mapping estimates should not be disaggregated beyond the upazila level to maintain statistical reliability.

Key Correlates of Poverty Incidence

• The poverty map is correlated with educational attainment, agricultural wage rates, and perceptions of poverty.
• These correlations suggest that poverty is influenced by a range of socio-economic factors, which can be further analyzed to inform targeted interventions.

Conclusion

• The poverty mapping exercise provides detailed sub-national poverty estimates, enhancing the ability of policymakers to identify and address poverty hotspots.
• The results show significant spatial variation in poverty incidence, with some upazilas experiencing rates far above the national average.
• The methodology, while robust, has limitations related to error correlation and assumptions about consumption patterns, which should be considered when interpreting the results.
• The maps and estimates are a valuable tool for improving targeted poverty reduction strategies in Bangladesh.