2006年-世界发展银行全球_Robust_Multidimensional_Spatial_Poverty_Comparisons_in_Ghana_Madagascar_and_Uganda_23页_234kb

Summary of "Robust Multidimensional Spatial Poverty Comparisons in Ghana, Madagascar, and Uganda"

Core Content

This paper investigates spatial poverty comparisons in three African countries—Ghana, Madagascar, and Uganda—using multidimensional indicators of well-being. It introduces a robust poverty dominance approach that applies to union, intersection, and intermediate poverty definitions. The study also emphasizes the importance of sampling variability in poverty comparisons and shows how bivariate poverty analysis can yield different conclusions than univariate methods.

Main Points

• Multidimensional Poverty: Poverty is not just a single-dimensional issue (e.g., income or expenditure). The paper considers two dimensions: household expenditures per capita and children's height-for-age z-scores (HAZ).

• Robust Comparisons: The methodology ensures that poverty comparisons are robust to the choice of poverty lines and indices, and to the aggregation rules used for multiple welfare variables.

• Union vs. Intersection Poverty Definitions:

  • Union: A person is poor if they are poor in either dimension.
  • Intersection: A person is poor only if they are poor in both dimensions.
  • Intermediate: A person is poor if they fall below a certain threshold in one dimension and are sufficiently poor in the other.

• Bivariate vs. Univariate Comparisons:

  • Univariate poverty comparisons (based on income or HAZ alone) often conclude that urban areas have lower poverty than rural ones.
  • Bivariate comparisons, however, reveal that poverty differences can vary depending on the correlation between the two dimensions.
  • In some cases, urban areas may have higher poverty if the correlation between income and health is stronger there.

• Statistical Validity:

  • The paper computes sampling distributions of poverty estimators to perform statistical tests of poverty differences.
  • This ensures that the poverty comparisons are statistically sound and not based on arbitrary assumptions.

• Methodology Overview:

  • The dominance approach is extended to the bivariate case, where poverty is compared using a surface rather than a line.
  • The paper introduces the $\Pi^{1,1}$ class of poverty measures, which are nondecreasing, anonymous, and additive in both dimensions, with the added condition that the two dimensions must be substitutes (i.e., better performance in one dimension reduces poverty more when the other is worse).

• Limitations of Aggregation Indices:

  • Indices like the Human Development Index (HDI), which aggregate multiple dimensions into a single scalar, are less general than the dominance approach.
  • The $\lambda(x, y)$ function defines the domain of dominance tests, and the weights assigned to each dimension affect the outcome of poverty comparisons.

Key Information

• Data Sources:

  • Ghana: 1988 Living Standards Study.
  • Madagascar: 1993 National Household Survey.
  • Uganda: 1999 National Household Survey.

• Welfare Indicators:

  • Household expenditures per capita: Standard for poverty analysis.
  • Children's height-for-age z-scores (HAZ): Measures nutritional status and general health.

• Key Findings:

  • In Madagascar, poverty in rural Toliara is higher than in urban Mahajanga/Antsiranana for a wide range of poverty lines and measures.
  • In Uganda, rural Central areas have higher poverty than urban Eastern regions, but this varies depending on the correlation between the two dimensions of well-being.
  • The correlation between well-being indicators significantly affects poverty rankings, especially in urban areas where it is often higher.

• Practical Implications:

  • The dominance approach allows for more nuanced and robust poverty comparisons than traditional univariate or aggregated indices.
  • It provides a theoretical and ethical basis for poverty measurement that avoids arbitrary choices in poverty line or index selection.

Conclusion

The paper demonstrates that spatial poverty comparisons using multidimensional indicators are more informative and robust than univariate or aggregated methods. It highlights the importance of correlation between dimensions and introduces a statistical framework for testing poverty differences. The methodology is applicable to a wide range of poverty measures and is particularly useful in developing economies where well-being is influenced by multiple factors.