Is all mobile positioning data GPS data?

No. Network-derived data often identifies cells or service areas. Device-based satellite positioning is a different collection mechanism.

Does it cover everyone?

No. Phone ownership, usage, operator membership and multiple subscriptions all affect representation.

Can two recorded cells prove that someone traveled between them?

Not by themselves. Network changes and location uncertainty must be considered before inferring a journey.

AI Glossary / Data engineering and architecture / Beginner

Mobile positioning data

Definition

Mobile positioning data describes where a mobile device was estimated to be at particular times. In official statistics, the term commonly refers to location-related records generated through mobile network activity. More broadly, mobile location datasets can also come from device positioning technologies such as satellite navigation. The collection method determines what a record actually tells an analyst.

What it means

A typical observation links a device or subscription identifier, a time and some location information. For network-derived records, that location may be an associated cell or service area rather than an exact coordinate for the person carrying the phone.

Aggregated observations can help study changing population presence and movement. However, the observations concern devices and network events. Turning them into statements about people, journeys or visits requires assumptions and validation. A large number of records does not remove the difference between what was observed and what the analyst hopes to measure.

How it works

Mobile networks generate operational records when devices use services or interact with network infrastructure. Depending on the source, records may be associated with calls, messages, data sessions or signaling activity. They differ in frequency, coverage and the information retained.

An authorized analysis process prepares the records, accounts for unsuitable or duplicate observations and relates network locations to geographic areas. It may estimate stays or transitions using defined rules. Published results should normally communicate aggregation, uncertainty and coverage rather than suggest exact continuous tracking.

Network association is not a perfect position measurement. Service areas overlap, and a device can change its serving cell without a corresponding physical journey. Analysts must distinguish changes in observed network state from evidence of actual movement.

Three practical examples

1. Transport planning: Aggregated transitions between broad areas can help investigate commuting patterns. Planners compare the results with other evidence because device observations do not directly identify travel purpose or transport mode.

2. Tourism analysis: Statistical teams study patterns consistent with temporary presence in a region. Definitions of visitor, resident and observation period must be explicit; one recorded connection does not establish a meaningful tourism visit.

3. Infrastructure planning: Authorities examine how estimated daytime presence differs from residential population. The results can inform questions about service demand while remaining complementary to surveys and other population sources.

Think of it this way

Think of intermittent entries in a visitor book at several buildings. They show that an identifier was recorded at certain times, but they do not describe everything that happened between entries.

The analogy is limited because network records are generated by technical systems rather than intentional sign-ins. A serving antenna can cover an area beyond its own site, and one person may carry multiple devices. Neither the location nor the person count can be read mechanically from the log.

Worked example

Consider a fictional dataset with 1,200 observed subscriptions in a district during an hour. A tempting shortcut is to call this “1,200 people present.” That ignores multiple subscriptions, people without phones, operator coverage and the rules used to assign location.

Suppose a simplified teaching adjustment first identifies 200 observations as additional subscriptions associated with already counted users, leaving 1,000 estimated users. If a separately justified model assumes these represent half of the relevant population, its estimate is 1,000 / 0.5 = 2,000 people.

The answer depends on both assumptions. If coverage is actually 40%, the same calculation gives 2,500. These invented numbers show why population estimation needs evidence for its adjustments and uncertainty. They are not a recommended universal correction procedure or a claim that duplicate users can always be identified reliably.

Illustrative case study

Fictional case study: MovementMeadow compares activity around two proposed community centers. An initial chart shows many more observed subscriptions near the first site, and the planning team almost interprets the difference as a direct ranking of community need.

Further review finds that the available operator's subscribers are unevenly distributed across the neighborhoods. The team also discovers that the time window emphasizes workers present at lunchtime rather than residents who might use evening services.

MovementMeadow revises the question, compares additional sources and reports the dataset's coverage limits. The analysis becomes more useful because its interpretation is narrower and better supported, not because the original observations have become a complete census.

Common misconceptions and mistakes

A common mistake is treating a cell tower's coordinates as the device's exact position. Another is assuming the nearest tower must serve the device. Network operation and overlapping coverage complicate both assumptions.

Do not equate a subscription with one person, or absence of a record with absence of a person. Observation frequency and phone use vary. Removing names also does not automatically remove privacy concerns from detailed movement histories.

Finally, an inferred home, workplace or journey is an analytical classification. Preserve its method and uncertainty rather than presenting it as a directly observed personal fact.

Business use

Organizations can use responsibly governed, appropriately aggregated positioning information to investigate broad demand and mobility patterns. Before acting, they should understand the source, observation interval, geographic resolution, population coverage and permitted purpose.

For AI analysis, these details are part of the model's input limitations. A model trained on one operator or region can learn patterns that do not transfer elsewhere. Useful validation compares estimates with independent counts or surveys and checks whether conclusions change under plausible alternative assumptions. The business value lies in answering a well-defined question, not accumulating the most detailed individual trajectories.

Check your understanding

A district has twice as many observed subscriptions as another. Does it necessarily have twice as many people present?

Answer: No. Differences in coverage, device ownership, usage and location assignment can affect the comparison. Population conclusions need a justified method and independent validation.

Related terms

Sampling bias
Geolocation
call detail record
mobility analysis
population estimation
spatial uncertainty
location privacy

Learn next

Compare a raw device observation, an inferred stay and an aggregated population estimate. For each, list what was measured directly and which assumptions were added during analysis.

Sources

UN-CEBD, mobile phone data guide, limitations: https://mdgs.un.org/wiki/spaces/MPDDPM/pages/143098186/3.4.%2BMPD%2Blimitations

UN-CEBD, representativity: https://mdgs.un.org/wiki/spaces/MPDDPM/pages/143098221/6.2.4.2.%2BRepresentativity

Ogulenko and colleagues, primary research abstract on probabilistic positioning: https://arxiv.org/abs/2007.12464

ITU, mobile phone data training material, indexed definitions consulted: https://www.itu.int/itu-d/sites/projectumc/wp-content/uploads/sites/42/2024/07/ITU-Big-Data.pdf

Examples are fictional and do not describe an operational tracking service.

Previous: Mobile cloud computing Next: Modal logic