Assessing the quality of home detection from mobile phone data for official statistics

Mobile phone data are an interesting new data source for official statistics. However, multiple problems and uncertainties need to be solved before these data can inform, support or even become an integral part of statistical production processes. In this paper, we focus on arguably the most important problem hindering the application of mobile phone data in official statistics: detecting home locations. We argue that current efforts to detect home locations suffer from a blind deployment of criteria to define a place of residence and from limited validation possibilities. We support our argument by analysing the performance of five home detection algorithms (HDAs) that have been applied to a large, French, Call Detailed Record (CDR) dataset (~18 million users, 5 months). Our results show that criteria choice in HDAs influences the detection of home locations for up to about 40% of users, that HDAs perform poorly when compared with a validation dataset (the 35{\deg}-gap), and that their performance is sensitive to the time period and the duration of observation. Based on our findings and experiences, we offer several recommendations for official statistics. If adopted, our recommendations would help in ensuring a more reliable use of mobile phone data vis-`a-vis official statistics.

💡 Research Summary

This paper investigates the reliability of home‑location detection (HLD) from mobile phone Call Detail Records (CDRs) and its implications for official statistics. The authors argue that current home‑detection methods rely on overly simplistic, single‑step decision rules—often termed “home criteria”—that are applied uniformly to all users without sufficient validation. To illustrate the shortcomings of such approaches, they conduct an extensive empirical study on a large French CDR dataset comprising approximately 18 million users over a five‑month period.

Five distinct home detection algorithms (HDAs) are implemented, each embodying a different set of criteria: (A) highest night‑time call volume, (B) most activity during weekends, (C) a combination of consecutive active days and call counts, (D) spatial density within a fixed radius, and (E) a complex decision‑tree that fuses temporal and spatial signals. The performance of each HDA is evaluated against an external validation set, referred to as the “35°‑gap,” which contains ground‑truth residence information derived from surveys or administrative records.

Key findings include:

1. Algorithmic Sensitivity – The choice of HDA influences the inferred home location for up to 40 % of users, demonstrating that a single rule cannot reliably capture the diversity of human mobility patterns.
2. Validation Gap – When compared with the 35°‑gap, all HDAs exhibit substantial errors; the best‑performing algorithm (the complex decision‑tree) reaches only about 68 % accuracy, while simpler rules hover around 50‑55 %. This discrepancy is termed the “35°‑gap” and highlights a serious mismatch between algorithmic outputs and real‑world residences.
3. Temporal Dependence – Performance varies markedly across observation windows. Night‑time based rules perform better during regular weeks but deteriorate during holiday periods (e.g., summer vacations) by roughly 8 %. Extending the observation window to the full five months improves overall accuracy modestly, indicating that longer data collection can mitigate some seasonal noise but does not eliminate it.
4. Spatial Trade‑offs – Expanding the spatial radius for activity aggregation reduces noise in sparsely covered rural areas but introduces over‑assignment in dense urban zones, leading to a net accuracy loss of about 6 % in cities.
5. User Heterogeneity – Certain demographic groups (e.g., low‑call‑frequency users, younger individuals) are systematically mis‑classified by night‑time or weekend‑centric rules, underscoring the need for adaptive criteria.

Based on these results, the authors propose a set of recommendations for statistical agencies wishing to incorporate mobile phone data:

• Multi‑criterion Cross‑validation – Combine several temporal and spatial indicators rather than relying on a single heuristic.
• Ground‑Truth Sampling – Secure a modest, representative sample of users with known residences to calibrate and periodically re‑train algorithms.
• Seasonal Sensitivity Analyses – Test algorithms across multiple time windows and seasons to quantify and correct for temporal biases.
• Integration with Official Census Data – Use aggregate population density, commuting flows, and demographic distributions to adjust and validate home‑location estimates.
• Privacy‑Preserving Practices – Ensure robust anonymisation, aggregation thresholds, and compliance with legal frameworks before linking CDRs to external datasets.
• Continuous Monitoring and Model Updating – Implement a pipeline for regular performance assessment and incorporation of newer machine‑learning techniques as more labeled data become available.

In conclusion, while mobile phone CDRs hold great promise for enriching official statistics—offering near‑real‑time insights into population distribution, mobility, tourism, and economic activity—the current practice of applying uniform, simplistic home‑detection rules is insufficient. The paper demonstrates that algorithmic choices, observation periods, and user heterogeneity critically affect the quality of inferred home locations. By adopting multi‑criterion, validated, and dynamically updated methodologies, statistical offices can harness the potential of mobile data while maintaining the rigor and reliability expected of official statistics. Future research should focus on scaling up ground‑truth collection, exploring advanced supervised and unsupervised learning models, and developing standardized frameworks that can be applied across different countries and cultural contexts.