Summary

  • Orlando E. Martínez-Durive and four co-authors compared mobile-service demand with vote shares in 4,102 urban and suburban French communes around the 2019 and 2024 European Parliament elections. In 2024, adding service demand produced a 28.23% relative increase in mean correlation with vote shares, averaged across parties, according to the original IMC 2026 paper.
  • The measurements came from one mobile operator whose raw traffic remains under NDA; its curated research dataset and code are public. At RIPE 93, Martínez-Durive invited other network operators and researchers to help build a platform-independent measurement instrument.

A platform can withdraw an API. A phone still needs a network. Orlando E. Martínez-Durive’s French election study followed the second route: count demand for mobile services at network antennas, then ask whether the pattern across towns adds information to the vote totals already reported by commune.

The result rests on one major French mobile operator. Its two brands shared a backbone and a single unsampled dataset representing 31.9% of the mobile market in 2019 and 31.5% in 2024. The researchers counted bytes for 22 services before the first election and 27 before the second, including social media, news, messaging and streaming. Network probes assigned byte counts to services once a minute using proprietary Deep Packet Inspection classifiers. They combined those measurements with income, unemployment and age data, then modeled party vote shares with Dirichlet regression.

The classifiers run in the operator’s production network, but their internal method and per-service ground-truth accuracy remain under commercial secrecy and NDA, so the authors could not validate that layer independently.

Geography mattered. The team mapped antenna demand to commune boundaries because the carrier did not disclose each antenna’s exact coverage area. It restricted the analysis to 4,102 urban and suburban communes, which held about 66% of registered voters and more than 70% of mobile demand. Measurements came from the week before each vote; the weekday window ran from 20:00 to 07:00, when residents were more likely to be at home than passing through.

In 2024, adding mobile-service demand to the socioeconomic-only baseline produced a relative 28.23% gain in mean correlation with party vote shares, averaged across parties. The abstract connects higher Facebook and TikTok demand with right-wing support, while online news, Twitter and Instagram aligned with centrist or progressive vote. Those are commune-level relationships. No person-level vote record enters the model, and a town’s service mix cannot identify how any one resident voted.

Aggregated byte counts leave foreground use, background syncing and in-app content unknown; identifying whether service use or political preference came first would require an external change in service availability.

Martínez-Durive’s path to this work runs through both sides of the antenna. His IMDEA Networks biography says he studied computer science at the University of Havana, where his master’s thesis used phone data to examine human mobility in Cuba. His doctorate in telematics engineering at Carlos III University of Madrid joined societal phenomena and network operations in one research programme; the official thesis record preserves that arc in its title. The biography also lists work with Telefónica’s mobile-network team and research on mobile data as a sensor for social events.

The attraction is clear: network measurements can offer an independent view when a platform’s API changes or disappears. The catch is just as clear. The paper says raw mobile traffic remains under a nondisclosure agreement; a curated dataset and analysis code are public. The sensor sits outside the platform, but access to its raw signal still runs through the carrier.

At RIPE 93, Martínez-Durive framed that as an invitation to mobile operators, ISPs, IXPs and researchers. His talk argues that several network vantage points could form a platform-independent instrument for observing digital society. It does not yet specify who would hold a shared dataset, how operators would align service labels, or who could query and release its results. That missing access protocol is the next fact that would change the judgment.

A workable first version would need a small common measurement rulebook: what counts as a service, which time window and geographic unit are comparable, how each operator describes coverage, and who may publish an output. Networks could keep local extensions and choose whether to participate. The decisive test is another network joining under rules readers can inspect.

The APIs may belong to the platforms. The map belongs to whoever controls the probes, labels and release terms. Martínez-Durive has shown what one operator’s view can add. The next question is whether more operators will share enough of that view to make the instrument accountable.