Unveiling Regional Disparities

Integrating Multiple Data Sources for Small-Area Income Inequality Estimates

Jinjing Li, University of Canberra
ANZRSAI Conference 2024

[email protected]

Introduction

• Income level and its distribution at small area level can provide important insights for evidence-based policymaking
• Multiple data sources are available but not trivial to combine; longitudinal data is rare
• This paper: integrating multiple data sources for small-area income inequality estimates in a new framework

Why Inequality at Small Area Level?

• Policy Implications: Small-area estimates are crucial for targeted policy interventions
• Peer Effects: Local conditions tend to have a strong effect on wellbeing
  • Health outcomes (Pickett & Wilkinson, 2015)
  • Economic mobility (Chetty et al., 2014)
  • Social cohesion, education outcomes etc.

Common Approaches

• Estimate inequality at a large-enough area with sufficient data
• Reweighting survey data to match selected census values
• Use a hierarchical model with a random effect (often normally distributed) to provide small area estimates (Rao & Molina, 2015)
• Impute values onto (unit-record) Census data, e.g. ELL (Elbers et al., 2003)

Key Challenges in Small-Area Inequality Estimation

Unified Data Integration Models

• Treat survey, census, and tax data as different samples in a "super" dataset with missing values
• Income distribution is modelled as a mixture distribution
  • Spatially correlated income generation opportunities ("clusters") defined by the area attributes, incorporating Census and other data
  • Income distribution from each cluster is defined by the population characteristics and the area they live in, incorporating taxation data
  • Survey can be considered as a sample of selected area, which is used to estimate the income distribution parameters

Theoretical and Empirical Consistency

• Aligns with microeconomic foundation on income generation
• Local administrative data helps to ground local estimates to reliable values
• Exploit spatial correlations in the underlying income opportunities (rather than arbitrary spatial smoothing on income or inequality)
• Links individual income to small area, to national income distribution

Methodology Overview

The mixture model framework

• (income) comes from survey data
• , (area attributes) come from census, tax and weather data at area level
• is the income distribution for the -th component based on a Singh-Maddala Distribution
• are the parameters of the income distribution

Methodology Overview

The mixture model framework

• are the mixing proportions
• is the number of mixture components
• Survey data doesn't need to cover all areas as long as Census and Tax data cover all areas
• Captures multi-modal income patterns

Advantages

• Respects the overall spatial correlation in income generation opportunities and income level
• Takes into account the spatial heterogeneity and demographic variations, including abrupt shifts
• Avoids strong parametric assumption on income distribution shape or modality
• Referencing administrative data
• Can be extended to arbitrary small area level with local administrative data
• Uncertainties quantified as s.e. in the estimation

Data Integration

Area standardised to SA3 level for this particular analysis

Results - National Estimates

Results

Gini distribution shift

Results

Results - Clusters

Results

Correlation between Area Characteristics and Cluster Membership

Results: Cluster Analysis

Temporal Trends

Results

Results

Areas with highest income inequality in 2021

Results

Areas with lowest income inequality in 2021

Further Research

• Identify and isolate the effect of different income sources and the inequality of opportunities
• Can extend to arbitrary small area level with local administrative data (e.g. SA1)
• Potential extension to other metrics of interests (e.g. wellbeing, health conditions, wealth, extended poverty etc.)
• International comparisons

Thank You

Questions?

Contact Information:
Jinjing Li
Email: [email protected]