2005年-世界发展银行全球_The_Construction_and_Interpretation_of_Combined_Cross-Section_and_Time-Series_Inequality_Datasets_67页_647kb

Summary of "The Construction and Interpretation of Combined Cross-Section and Time-Series Inequality Datasets"

Core Content

This paper addresses the challenges of constructing and interpreting cross-section and time-series inequality datasets that are both nationally and internationally comparable. It critiques the widely used World Bank and UN datasets, such as the Deininger and Squire (DS) dataset and the World Income Inequality Dataset (WIID), highlighting issues of measurement error and inconsistency in their definitions. The authors propose a new methodology to reduce these errors and create a more reliable dataset for empirical research.

Main Points

1. Limitations of Existing Datasets

• The DS and WIID datasets are frequently used but criticized for their potential to introduce measurement error due to inconsistent definitions of income concepts, reference units, and data sources.
• These datasets often assume broad comparability of definitions, which can lead to misleading results, especially in OECD countries.
• The "glacial change" hypothesis, which suggests that inequality does not vary much over time, is rejected in favor of the existence of significant time trends.

2. New Dataset Construction

• A new combined dataset is created by applying a consistent grouping methodology to heterogeneous data from existing secondary datasets.
• This dataset includes six main inequality series, reducing the implicit measurement error.
• The dataset is based on the WIID, which provides more comprehensive data than DS, but also has its own limitations.

3. Methodological Improvements

• The authors use parametric non-linear estimation of Lorenz curves from grouped data to estimate the entire income distribution.
• This method allows for the calculation of Gini coefficients, Atkinson indexes, and poverty rates.
• They emphasize that no single inequality index is universally accepted, and using multiple indexes (e.g., Gini and Atkinson) provides a more robust analysis.

4. International and Temporal Patterns

• Between-country inequality variation is more significant than within-country variation.
• This suggests that country-specific characteristics play a larger role in explaining inequality levels than time trends.
• For OECD countries, the authors detect a U-shape pattern in inequality, confirming the "U-turn" hypothesis.
• For developing countries, inequality generally decreased over the analyzed period, with a slight increase in the 1990s.
• Country-specific time trends vary significantly, making it difficult to draw uniform conclusions.

5. Measurement Issues and Adjustments

• The use of fixed adjustments to reconcile different definitions (e.g., gross vs. net income, expenditure vs. income) is not sufficient to eliminate measurement error.
• The authors argue that dummy variables and fixed adjustments can introduce arbitrary noise into the data.
• The choice of income concept, reference unit, and data source significantly affects the results.

Key Information

• Data Sources: The paper uses the WIID as the main data source, which includes data from various national surveys and databases.
• Quality Criteria: The WIID includes quality ratings (OKIN and NOOK), with OKIN data being considered reliable.
• Inequality Definitions: The six main characteristics used to classify data include:
  • Concept measured (e.g., income or expenditure)
  • Reference unit (e.g., person or household)
  • Area coverage
  • Population coverage
  • Data sources
  • Quality ratings
• Data Challenges: Inconsistent definitions and sources across countries and time periods make it difficult to create a reliable cross-country dataset.
• Poverty Trends: The paper finds a decline in poverty ratios over time in most countries, with the African continent being the exception.

Conclusion

• The authors reject the "glacial change" hypothesis and support the existence of significant time trends in inequality.
• They argue that the use of multiple inequality indexes (e.g., Gini and Atkinson) is necessary for robust empirical conclusions.
• The new dataset allows for a more accurate exploration of international inequality patterns and poverty trends.

Table of Inequality Series for Chile (1968–1996)

Year	Gini Coefficient (DS-accept)
1968	45.64
1970	-
1971	46.00
1973	-
1974	-
1975	-
1976	-
1977	-
1978	-
1979	-
1980	53.21
1981	53.46
1982	56.98
1983	54.49
1984	55.85
1985	54.91
1986	55.69
1987	56.72
1988	54.50
1989	57.88
1990	54.70
1991	-
1992	52.19
1993	50.00
1994	56.49
1995	-
1996	56.37

Note: The values represent the Gini coefficient as reported in the DS-accept series.

Figures

• Figure 1: Spain's Gini coefficient series, showing differences between income and expenditure-based measures.
• Figure 2: Mexico's Gini coefficient series, highlighting differences between household and person-based measures.

These figures illustrate how the grouping of heterogeneous data can alter inequality trends and levels, raising concerns about the reliability of the DS dataset.

Methodological Contribution

• The authors introduce a consistent grouping methodology to address the limitations of existing datasets.
• This approach improves the accuracy of inequality estimates and provides more reliable poverty measures.
• It allows for non-linear parametric estimation of Lorenz curves, which in turn enables the computation of alternative inequality and poverty indices.

Final Remarks

• The new dataset emphasizes the importance of multiple definitions and sources in inequality measurement.
• It provides a richer source of information for empirical studies, enabling more nuanced analysis of inequality and poverty trends.
• The paper concludes that no single dataset is perfect, and that robust conclusions require careful consideration of data definitions, sources, and methodological adjustments.