Inferring fine-grained migration patterns across the United States

2025-03-26

AI Summary

The U.S. lacks high spatiotemporal-resolution, fine-grained population migration data: census data suffer from coarse spatial and temporal granularity, while commercial data—though high-resolution—exhibit systematic biases. To address this, we propose a scalable Iterative Proportional Fitting (IPF) framework that, for the first time, integrates heterogeneous multi-source data—high-bias, high-resolution commercial data with low-resolution, unbiased census data—to produce unbiased annual migration matrices at the census block group level (47.4 billion origin–destination pairs) for 2010–2019. We publicly release the MIGRATE dataset. Validation against independent benchmarks shows strong correlation (r > 0.9) and substantial improvement over raw commercial data. The dataset successfully captures sub-county population redistribution induced by California wildfires. MIGRATE establishes the first high-accuracy, reproducible, and scalable fine-grained migration baseline for social, environmental, and public health research.

Technology Category

• Humans and AI: Crowd Sourcing and Human Computation
• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Data transparency and provenance

Abstract

Fine-grained migration data illuminate important demographic, environmental, and health phenomena. However, migration datasets within the United States remain lacking: publicly available Census data are neither spatially nor temporally granular, and proprietary data have higher resolution but demographic and other biases. To address these limitations, we develop a scalable iterative-proportional-fitting based method which reconciles high-resolution but biased proprietary data with low-resolution but more reliable Census data. We apply this method to produce MIGRATE, a dataset of annual migration matrices from 2010 - 2019 which captures flows between 47.4 billion pairs of Census Block Groups -- about four thousand times more granular than publicly available data. These estimates are highly correlated with external ground-truth datasets, and improve accuracy and reduce bias relative to raw proprietary data. We publicly release MIGRATE estimates and provide a case study illustrating how they reveal granular patterns of migration in response to California wildfires.

Problem

• Lack of fine-grained US migration data
• Reconciling biased proprietary and low-resolution Census data
• Creating accurate high-resolution migration datasets

Innovation

• Iterative-proportional-fitting reconciles biased and reliable data
• Produces highly granular annual migration matrices
• Improves accuracy and reduces bias in migration estimates