One country, multiple portraits: representativeness in GPS-based mobility data is source-specific and spatially dependent

2026-06-22

AI Summary

This study addresses the pervasive issue of non-representative GPS mobility data in low- and middle-income countries, where coverage biases across data sources remain poorly understood. Integrating Facebook and Veraset GPS datasets with census data from 2,478 Mexican municipalities, the authors employ interpretable machine learning and spatial statistical models to systematically dissect the origins, spatial structure, and drivers of coverage bias. Findings reveal that coverage bias exhibits strong data-source specificity and spatial dependence: Facebook data demonstrate more uniform coverage, whereas multi-app aggregated data disproportionately represent wealthier, more digitally connected areas. Explicitly modeling spatial autocorrelation substantially improves explanatory power for these biases. The results underscore the necessity of tailoring bias-correction strategies to specific data sources and highlight that a portion of spatial variation in coverage cannot be fully accounted for by observable covariates.

Abstract

Anonymised GPS-based mobile phone data are increasingly used to estimate population distribution and human mobility, supporting applications across disaster response, public health, urban planning and migration research. Yet whether these data fairly represent the populations they describe, particularly outside high-income countries, remains poorly understood. We quantify coverage bias for 2,478 municipalities in Mexico by comparing population estimates from a single-platform source (Facebook) and a multi-app aggregator (Veraset) against the 2020 Mexican Population Census. We find that the magnitude and spatial distribution of coverage bias differ substantially across sources. Facebook provides higher and more evenly distributed coverage, whereas the multi-app data concentrate users in larger, wealthier and more digitally connected places. Coverage bias is also spatially structured, with neighbouring municipalities showing similar levels of over- or under-coverage. Using explainable machine learning, we show that digital access and material resources are the dominant drivers of bias for the multi-app data, while demographic and population structure dominate for Facebook. Explicitly modelling spatial dependence improves the performance of statistical models for explaining bias and reveals that an appreciable share of spatial variation remains unexplained by observed covariates. These findings show that coverage bias is source-specific and spatially dependent, and provide a foundation for adjustments that improve the representativeness of mobile phone data in unequal, data-scarce settings.

Problem

• coverage bias
• representativeness
• mobility data
• spatial dependence
• population distribution

Innovation

• coverage bias
• spatial dependence
• mobile phone data
• representativeness
• explainable machine learning