2013年-世界发展银行全球_Nepal___Small_Area_Estimation_of_Poverty_2011_26页_2mb

Summary of Nepal Small Area Estimation of Poverty, 2011

Core Content

This report presents the results of the Nepal Small Area Estimation of Poverty (2010/11), focusing on poverty incidence, poverty gap, and poverty severity at the district, ilaka, and target area levels. The methodology combines data from the Nepal Living Standards Survey (NLSS3) and the 2011 Population Census, enriched with auxiliary geographic and socioeconomic variables from the World Food Program and other sources. The estimates aim to provide detailed, statistically reliable poverty information at a granular level, particularly at the Village Development Committee (VDC) level, which is crucial for local development planning.

Main Points

1. Purpose of Small Area Estimation

• Small area estimation (SAE) is a statistical technique used to improve the accuracy of poverty estimates for smaller geographic areas.
• It allows for more targeted poverty reduction efforts by revealing spatial variations in poverty that are not captured at the national level.
• The poverty map provides a visual representation of these variations, aiding in both technical and non-technical communication.

2. Methodology

• The ELL method (Elbers, Lanjouw, and Lanjouw) is used to combine survey and census data.
• The method involves three main steps:
  1. Selection of common variables between the NLSS3 and the Census.
  2. Regression modeling of household expenditure on these variables.
  3. Prediction of expenditure and poverty indicators using Monte-Carlo simulations and variance modeling.
• The report includes both beta models (for regression) and alpha models (for variance modeling) to improve accuracy and precision.

3. Data Sources

• NLSS3 (2010/11): A nationally representative survey with 5,988 households and 28,474 individuals.
• 2011 Population Census: Provides comprehensive population data at the VDC level.
• GIS data: From the World Food Program's Vulnerability Analysis and Mapping unit, including variables like elevation, slope, population density, and distance to district headquarters.
• Kosh variable: A new accessibility measure, representing the time it takes to walk from a VDC to the district headquarters.

4. Estimation Levels

• The report produces poverty estimates at VDC, ilaka, and district levels.
• Target areas are defined to ensure reliable estimates, especially in areas with small sample sizes.
• In mountainous regions, the ilaka level is used as the target area. In hill and terai regions, VDCs are used when feasible, and combined VDCs are used when precision is low.

5. Key Findings

• Poverty incidence, gap, and severity are mapped at the target area level.
• Poverty is highest in the hilly areas of Far West and parts of Mid-West, reaching up to 75% in some areas.
• In contrast, parts of Kathmandu have negligible poverty.
• Nearly half of the small areas have poverty rates above the national average of 25.2%, containing two-thirds of the poor population in Nepal.
• Poverty concentration has declined in the East and Central regions but increased in the West and Mid-West.
• The poverty maps reveal geographic disparities and help identify areas where poverty persists.

6. Reliability and Precision

• The report includes confidence bands for poverty estimates where statistical reliability is uncertain.
• The ELL method is validated using random subsamples of the data to test for over-fitting.
• The use of Empirical Best estimation improves the precision of poverty estimates by drawing errors from observed distributions.
• Standard errors are calculated and presented to ensure the accuracy of the estimates.

Key Information

• Total target areas: 2,344
• Population size of target areas: Ranges from 257 to 973,559 households
• Poverty line: 19,261 Nepali Rupees per person per year
• Geographic levels:
  • District level: 75 districts
  • Ilaka level: 967 ilakas
  • Target area level: 2,344 target areas
• Variables used:
  • Household characteristics (education, housing, ethnicity, etc.)
  • Geographic and infrastructure variables (elevation, slope, population density, road length, river length, Kosh)
• Methodological improvements:
  • Use of recent data (NLSS3 and 2011 Census)
  • Refinements in standard error modeling
  • Inclusion of more detailed geographic data

Conclusion and Suggestions

• The poverty maps offer a more detailed understanding of poverty distribution in Nepal.
• They provide statistically reliable estimates at the VDC level, which is essential for local development planning.
• Future work should consider expanding the poverty maps to include other welfare indicators such as nutrition and food security.
• Randomized experiments may be useful in areas with similar poverty levels to identify the most effective development interventions.
• The usability-certainty trade-off is important: more detailed estimates come at the cost of lower precision, especially in sparsely populated areas.

Figures and Tables

• Figure 1: Shows district boundaries and ecological belts (Mountains, Hills, Terai) in Nepal.
• Table 1: Provides population size statistics for different geographic levels (VDC, ilaka, target area, district).
• Table 2: Compares predicted poverty rates at the stratum level with observed rates from NLSS3.
• Table 3: Summary statistics of predicted poverty rates at the district, ilaka, and target area levels.
• Table 4: Over-fitting test comparing observed and predicted FGT(0) in a subsample of the NLSS3.
• Tables A5-A9: Present statistics for the selected variables in the three regional consumption models.

References

• Ghosh, M. and Rao, J. N. K. (1994)
• Rao, J. N. K. (2003)
• Elbers, C., Lanjouw, P. and Lanjouw, J. (2002, 2003)
• Bedi, S., et al. (2007)
• Molina, O. and Rao, J. N. K. (2010)
• Elbers, C., Lanjouw, P. and Lanjouw, J. (2008)
• CBS et al. (2006)

Appendices

• Appendix I: Includes detailed data on household variables, selected models, and definitions of target areas.
• Appendix II: Contains tables with small area estimations of poverty at the district, ilaka, target area, and VDC levels.