Dialing up Mobile Phone Data for Official Statistics

Esperanza frames the session around a long-running effort to use mobile phone data as a regular part of official statistics rather than as an occasional novelty. She presents this as an institutional mandate and a collaborative programme aimed at sustained country adoption over time.

She explains that the work was mandated by the UN Statistical Commission to use new data sources to complement or supplement official statistics, and that mobile phone data is one such source . She also notes that the joint project with the World Bank is already working with 25 countries and aims to reach at least 30 countries by 2030 that can sustainably use mobile phone data in producing official statistics .

Esperanza argues that successful use of mobile phone data depends on groundwork done before a crisis or production need arises. She stresses that governance discussions and stakeholder engagement are what make rapid and effective deployment possible later.

After Daniel's presentation, she highlights that in the DRC case mobile phone data could be used because discussions had already taken place before the Ebola emergency emerged . She adds that such long discussion is necessary and that the governance tools and stakeholder-engagement resources discussed in the session are useful when engaging relevant actors .

Esperanza states clearly that mobile phone data is not being promoted as a substitute for all existing statistical sources. Instead, it should work alongside traditional data collection to strengthen statistical production.

She says explicitly that the work on mobile phone data is not claiming it will replace traditional data sources, but that it will be used hand-in-hand with them . She further clarifies that it can serve as a complement or supplement to traditional sources .

Esperanza presents mobile phone data as especially useful for internet-use measurement when survey-based methods are not feasible. Her point is pragmatic: countries still benefit from combining methods, but mobile phone data offers a realistic alternative where surveys are costly or absent.

She says that for internet use, both mobile phone data and traditional sources can be used together, but also notes that surveys are very expensive and many countries are not collecting this information through traditional surveys . She therefore argues that a way must be found to produce the data, and that mobile phone data is the primary source ITU is investing in through tools and country support .

Paul defines a mobile phone data system not merely as data access, but as the full process that converts raw operator records into useful public-policy outputs. He emphasises that the whole chain between raw data and final anonymised statistics must be understood as one integrated production system.

He explains that a mobile phone data system is essentially a mechanism that transforms raw mobile phone data held by mobile network operators into anonymised statistics that provide actionable insights for public policy . He then adds that everything in between should be viewed as an integrated data production system .

Paul argues that sustainable implementation depends on governance architecture, not just analytical techniques. He presents a three-pillar framework to ensure systems are strategically directed, legally compliant and technically operational.

He says the focus is not on the technical algorithms alone but on the governance side needed to build sustainable mobile phone data systems . He describes the governance framework as structured across strategic, legal and technical pillars , and explains that these pillars are meant to make systems sustainable, legally compliant and operationally efficient .

Paul stresses that the governance framework is intended to be operational, not merely conceptual. He therefore highlights practical implementation tools that countries can use to structure agreements, assign roles and meet compliance requirements.

He says the project will deliver a framework report, MOU templates and operational checklists to support country teams in implementation . He illustrates this with a technical checklist covering responsibilities such as methodological design, quality assurance, coding, execution, documentation, data ingestion, secure transfer and privacy safeguards .

Paul argues that operational success depends on understanding what each stakeholder is willing and legally able to do. He sees formal agreements and shared protocols as the mechanisms that turn multi-actor interest into a functioning system.

In response to a question, he describes engagement processes that identify the regulatory framework and assess the willingness of mobile operators, regulators, national statistical offices and line ministries to share, process or use data . He then says these arrangements are formalised, primarily through MOUs and framework agreements that define the rules and protocols organising the system .

Paul treats privacy and security as built-in features of system design rather than as afterthoughts. His argument is that governance structures and agreements must embed legal and privacy protections from the outset.

He states that framework agreements organise the system and embed privacy by design within it . Earlier in his presentation, he also explains that mobile phone data governance consists of organisational and technical measures that create a sustainable, legally controlled, compliant and operational data production system , and that the legal pillar includes lawful processing, core principles and privacy and security obligations .

Paul argues that technical capacity for privacy protection is improving alongside the expansion of data use. This makes him optimistic that secure and appropriately governed access will become increasingly achievable.

He says that as data use expands, techniques for privacy preservation, privacy-enhancing technologies and cryptographic methods are also advancing significantly . He adds that this makes him hopeful that governance systems enabling secure data access will continue to improve as well .

Daniel argues for an operational model in which sensitive subscriber data remains under operator control while analysts work in a secure environment to produce aggregated outputs. This allows governments to benefit from the analysis without assuming the risks of direct access to personal-level records.

He describes a standard workflow in which a secure server is set up on the mobile network operator's premises, where call detail records are pseudonymised, geo-referenced and analysed by a limited number of security-cleared analysts before only aggregated outputs are exported for reports and maps . He later emphasises that across countries the individual-level data stays on operator premises and government actors do not need direct access to subscriber-level data .

Daniel makes a normative argument that privacy-preserving architectures should be preferred when they can still deliver strong public-policy outputs. In his view, if high-impact work is already possible without exposing individual-level data, more intrusive arrangements are hard to justify.

After reviewing long-term work in multiple countries, he says that high-impact analysis and subscriber privacy protection are both achievable with a similar solution in each country . He then directly asks why the work should be done any other way if that privacy-protective model already works .

Daniel argues that international organisations should not lower privacy standards when operating in jurisdictions with weaker protections. Instead, they should comply with both stricter external standards and local law, while promoting stronger practice in-country.

He states that Flowminder is governed by the GDPR and treats that as a minimum standard . He adds that the organisation must comply with both GDPR and local law, while building relationships with government actors and advocating best practice in data privacy locally .

Daniel argues that one of the strongest advantages of mobile phone data is operational readiness in emergencies. When systems and partnerships already exist, analysts can quickly generate movement insights that are directly relevant to crisis response.

He explains that during the Ebola outbreak in eastern DRC, Flowminder already had pipelines in place with information on connectivity and population movement in the area through its partnership with Vodacom Congo . Because of these pre-existing systems, they were able to begin cohort analysis immediately .

Daniel presents a concrete example of mobile phone data being used to anticipate disease spread. He argues that mobility-derived measures can act as a strong predictive tool in epidemics because human movement drives transmission.

He says they identified hundreds of thousands of subscribers in the outbreak region and calculated an 'intensity' metric combining time spent and the number of cohort members visiting other health zones . He reports that one week after the analysis was released, 10 more health zones reported Ebola cases and 8 of those 10 were among the top 16 zones identified in the earlier list, demonstrating strong predictive value .

Daniel argues that mobile phone data can correct weak official baseline data and directly influence service delivery. In the DRC example, improved population estimates strengthened local advocacy for health resources where official projections were outdated.

He notes that the last census in the DRC dates from 1984 and that population density was being projected uniformly nationwide despite major uneven growth, especially in the mining areas of the south-east . He explains that doctors in Haut-Katanga used the population data produced with Vodacom to advocate for sufficient vaccines, and when Flowminder visited in February doctors said they were finally able to vaccinate more children and secure other resources such as petrol for motorbikes as a result .

Daniel uses Haiti to show how long-term cooperation with an operator can sustain repeated humanitarian and public-policy applications over many years. He also indicates that such mature operational data production can increasingly be connected to formal statistical systems.

He says Flowminder has partnered with Digicel in Haiti for 16 years and that this was the first operator to enable its data to be used for humanitarian response work . He lists use cases including hurricanes, earthquakes, cholera spread and displacement from gang violence, and adds that with World Bank support they are now working towards an MOU with the Statistics Office so these long-produced data can feed into official models .

Daniel presents Ghana as a case where institutional maturity has moved beyond experimentation into routine governmental use. He also suggests that once the basic partnership is stable, more advanced analytical methods such as machine learning can be added.

He says Ghana has the strongest governmental buy-in, a long-term relationship with the Ghana Statistical Service and high capacity, including access to aggregated operator data through an API for use in official statistics . He further notes that discussions are under way with the Ghana Statistical Service and the disaster management agency NADMO to use artificial intelligence and machine learning to model population movement under flood shocks .

Daniel argues that investment in data should be protected because it improves the effectiveness of scarce development spending. He presents data systems as a way to ensure limited funds are directed to the right places despite broader cuts in funding.

He refers to recent reductions in funding following political changes in the United States and the United Kingdom, saying that funding for data was cut along with other funding . He then argues that investment in data must continue because it helps determine whether the few dollars available are being invested well .

Daniel argues that operator cooperation cannot be assumed; it depends on incentives and trust. For data sharing to be sustained and useful, operators must perceive a tangible benefit and be willing to provide data openly and consistently.

He states that the work depends on operators opening up their data in a way that allows high-quality analysis, and that this should not be done begrudgingly because that would undermine data quality and timeliness . He says what enables this is a value exchange with operators, and gives Vodacom as an example of a company that provides data without charge because the partnership supports its corporate social responsibility and public profile in the Congo .

Fredrik argues that indicator production should only begin after rigorous data checking. He warns that processing errors, timestamp inconsistencies and missing data can distort outputs, so quality assurance is a necessary first stage.

He notes that if a chart does not show the expected pattern there may be issues related to processing, different timestamps or missing data . He then says that only once good quality data are available can analysts move on to calculate indicators .

Fredrik presents home detection as the key step that converts raw anonymous event streams into usable statistics linked to place. Without it, the data remain just time-space traces; with it, subscribers can be associated with areas and combined with reference data.

He says home detection is probably the most important step in the workflow and is the bridge between mobile phone data and statistics . He explains that before home detection the data are only sequences of events in time and space, but after home locations are determined, subscribers can be mapped to areas and combined with population and other reference data .

Fredrik explains that because the data are anonymous, home locations must be inferred from behavioural patterns rather than directly observed. The ITU method gives greater weight to times when people are most likely to be at home, especially weekday evenings, nights and early mornings.

He states that anonymous mobile phone data do not explicitly show where people live, so home location must be inferred from user behaviour . He then says the notebooks use an algorithm that gives greater importance to weekday activity and specifically to events during the night, evening and early morning because these are the periods when people are typically at home .

Fredrik argues that home detection enables the actual production of ICT indicators from mobile phone data. Once subscribers are assigned to places, the data can be disaggregated into meaningful categories such as location, technology and demographic groups where available.

He says that from home detection, indicators of internet use can be generated . He specifies that these can be broken down by geographic areas, by technologies and by age groups and other categories depending on the available data, and that the notebooks include visualisations such as technology composition of internet use .

Fredrik stresses openness and practical capacity-building as central to adoption. He argues that countries need accessible tools, example data and hands-on training so they can experiment directly and build competence in using mobile phone data for official statistics.

He says that all the materials shown are openly available through the ITU MPD portal, including access to the notebooks, synthetic data, methodologies and country examples, while the GitHub repository provides the full source code . He also notes that ITU organises small-group hands-on trainings on the Jupyter notebooks because one of the best ways for countries to learn is by doing .

Fredrik acknowledges that mobile phone data are not bias-free and that additional methodological work is needed to improve representativeness. He identifies several distortions and presents current research efforts to correct them.

He lists biases including people holding multiple SIM cards or devices, differences in operator coverage or market share, and the fact that some people do not have mobile phones at all . He then describes research on subscriber deduplication, privacy-enhancing technologies, similarity scores, trajectory matching, and post-adjustments and re-rating measures to better infer results for the whole population before integrating validated methods into the notebooks .

Fredrik argues that institutional acceptance grows when stakeholders can see concrete value from the data. He suggests that persuasive, high-impact examples help overcome hesitation and make it easier to combine mobile phone data with other sources.

In response to the discussion on barriers, he says that the value of mobile phone data has become clearer through examples such as those from the Congo . He concludes that showcasing benefits like Daniel's examples is key to improving attitudes towards this data source and combining it with other sources .

Fredrik responds to the broader concern about underused data by arguing that the core problem is not lack of data itself. Instead, he sees the main bottlenecks as combining datasets, analysing them and overcoming cultural barriers to integration.

He says that after many years working in data, he does not necessarily believe there is a scarcity of data . He then states that there is a scarcity of combining data and analysing large amounts of data, and that cultural aspects also play a role .

The audience member raises a challenge rather than a policy proposal, asking how mobile phone data systems can manage consent and varying legal protections as they scale up. The point underscores that public trust and weak data protection frameworks remain unresolved barriers to normalising these practices.

Francesca Chacano asks how countries should handle consent and different legal frameworks as they move from experimentation to sustainable data gathering . She specifically asks about experiences in countries that do not have strong data protection frameworks, signalling concern about weak legal safeguards and trust .

The audience member argues that the underuse of data is not a new problem created by digital technologies. He suggests that institutions and professionals have long resisted sharing and combining data because of security concerns, public backlash and fear of losing control.

He refers to longstanding dormant reservoirs of data in administrations and gives examples such as controversy around census data in Switzerland and the non-use of cross-institutional health data for epidemiology because of security concerns . He then suggests that learned professions are often afraid of sharing data and being bypassed, which makes implementation of progress difficult .