Validating Urban Scaling Laws Through Mobile Phone Data: Analysis of Brazil's Largest Cities
Computational Urban Science, 6: 21 · DOI: 10.1007/s43762-026-00252-x
The Problem
Urban scaling theory predicts that larger cities exhibit disproportionately higher socioeconomic output and human interaction — a relationship captured empirically across many indicators, from patents to GDP to wages. Yet most evidence comes from wealthy, developed nations. Whether these superlinear scaling laws hold in rapidly urbanising, deeply unequal societies remains an open question.
Brazil offers an ideal stress test: its 100 most populous municipalities span São Paulo (over 12 million inhabitants, contributing ~32% of national GDP) to mid-sized cities of under two million, across vastly different climates, infrastructure levels, and income distributions. The country's Gini coefficient is among the world's highest, challenging the assumption that scaling mechanisms operate uniformly regardless of socioeconomic context.
Prior studies used indirect proxies or datasets from single countries with different contexts. What was missing was a national-scale, high-resolution test using direct behavioural data — one that measures both remote communication and physical mobility as complementary dimensions of urban interaction.
São Paulo residents maintain approximately 10 mobile phone contacts on average, while those in Belém (population ~1.3 million) maintain about 7. This ~40% difference over a tenfold population gap is precisely what superlinear scaling with β ≈ 0.19 per-capita would predict.
The Approach
The study develops a rigorous statistical framework to extrapolate from a single telecom operator's sampled data (~20% market share) to population-level interaction metrics, using two complementary proxies for human interaction intensity across all 100 cities.
-
Step 01
Home location inference
Each user's residence is estimated as the antenna location with the highest call frequency between 7 p.m. and 6 a.m. on weekdays and at any time on weekends. Only users with a clearly established residential location (at least 50% of qualifying calls from the inferred antenna) are included, anchoring all subsequent interaction metrics to the city of residence.
-
Step 02
Contact degree — measuring remote social ties
A contact network is constructed where nodes are users and edges denote at least one voice call between two users. The degree ki of each user is their number of distinct contacts. City-level cumulative degree K is extrapolated from the sampled users to the full population using a sampling-convergence framework validated empirically: the average sampled degree converges to the population average as sample size grows.
-
Step 03
Trip detection — measuring physical mobility
A trip is defined as movement between two antennas at least 2 km apart, inferred from consecutive calls made within a 30-minute to 4-hour window. This threshold focuses on medium- to long-range, deliberate travel — commutes, professional engagements, consumer activities — while filtering out pedestrian noise and antenna fluctuations. OD matrices are constructed for each city, and total trips per city are summed across all users.
-
Step 04
Scaling estimation via log-log regression
For each urban metric Y (contacts, trips, GDP, antennas, users), ordinary least squares regression is applied to the log-transformed data: log Y = β log N + const. The slope β is the scaling exponent. Values β > 1 indicate superlinear (increasing-returns) scaling; β < 1 indicates sublinear (economies of scale). Standard errors quantify uncertainty across the 100-city sample.
Key Findings
Superlinear social interactions
Cumulative voice-call degree scales as K ~ N1.19. Larger cities not only have more residents — they generate disproportionately more communication ties per person, confirming the core prediction of urban scaling theory in a developing-nation context for the first time at national scale.
Superlinear physical mobility
Total trips scale as τ ~ N1.25, slightly steeper than contact scaling. The consistency between two independent interaction proxies — phone contacts and physical trips — rules out measurement artefacts and suggests complementary mechanisms: digital connectivity and face-to-face engagement reinforce each other in larger cities.
GDP scales superlinearly, intermediate
Gross Domestic Product scales as GDP ~ N1.12, lying between linear and the interaction exponents. This is consistent with the hypothesis that enhanced connectivity creates conditions for economic productivity rather than being a simple byproduct of wealth — causality remains an open question for longitudinal study.
Sublinear infrastructure — economies of scale
The number of mobile antennas relative to active users scales sublinearly: NA ~ U0.84. Denser, more active urban users share antenna infrastructure more efficiently. This aligns with classical infrastructure scaling theory and introduces the concept of effective population as the appropriate denominator for infrastructure analysis.
Phone adoption grows faster than population
The number of mobile phone IDs (SIM cards) scales as U ~ N1.29 — faster than population growth. Larger, more economically dynamic cities show higher per-capita SIM card multiplicity, likely reflecting multi-device adoption, business use, and prepaid plan switching behaviour common in Brazil's mobile market.
Robust across inequality
Scaling exponents remain statistically stable across cities with a per-capita GDP difference of a factor of 2 (São Paulo vs. Manaus). This consistency across profound socioeconomic inequality suggests that the mechanisms driving superlinear interaction growth operate independently of local economic conditions — they appear to be fundamental urban properties.
Why It Matters
Confirming urban scaling laws with mobile phone data from 100 Brazilian cities matters beyond theory: it informs how cities invest in density and connectivity, how telecom networks are planned, and how researchers can study cities when complete data is not available.
- Urban Policy The superlinear scaling of social interactions suggests that investment in urban density and connectivity has amplified returns. Infrastructure that reduces friction for movement and communication in large cities generates outsized economic and social gains compared to equivalent investment in smaller centres.
- Global South These findings extend the empirical foundation of urban scaling theory to a rapidly urbanising, highly unequal nation. The robustness of scaling laws across Brazil's diverse cities suggests they may generalise to other Global South contexts — a key open question for future research across Africa, South Asia, and Latin America.
- Telecom Planning The sublinear scaling of antennas with active users (β = 0.84) implies that infrastructure costs per connected user decrease in larger cities. Telecom operators and regulators can use scaling exponents as benchmarks to identify cities where antenna deployment is disproportionate to actual usage — a signal of either over-investment or under-served demand.
- CDR Methodology The paper demonstrates a sampling framework for extrapolating city-level interaction metrics from a single operator's data (~20% market share), validated by the convergence of average degree with sample size. This methodology is applicable to any country where multi-operator data remains proprietary — offering a practical path to national-scale urban research from incomplete telecom data.
- Future Work The cross-sectional design limits causal inference. Whether heightened interaction drives economic growth or vice versa — or both — requires longitudinal data tracking how scaling exponents evolve as cities grow. Functional boundary definitions based on mobility rather than administrative units may also alter measured exponents and deserve further investigation.
Infographics
Paper & Authors
Authors
Cite This Work
Open access · CC BY 4.0