The assumption underpinning the extraction of mobility information from Call Detail Records (CDRs) or other mobile operator data is that, if a subscriber is recorded from cells that are in different locations, then the subscriber has moved.
This assumption is not always true however, as a subscriber’s consecutive calls can be routed by distant cells even if the subscriber has not moved. Call rerouting (or rerouting of SMS and data sessions) therefore creates spurious movements if a change in subscriber recorded location is interpreted as movement, which can make it appear as if subscribers travel further, and more often than they actually do.
Spurious movements can be detected when a change in recorded location occurs too quickly to be a plausible movement (e.g. 2 calls made with a 5 sec interval and routed by cells distant by 20km, which would be a speed of 14,400 km/h), which we call ‘teleports’. However, rerouting can also create spurious movements that are plausible (e.g. 2 calls made with a 5 hour interval and routed by cells distant by 20km) and not distinguishable from actual movements. Therefore we can only measure the tip of the iceberg (‘teleports’) of the larger ‘rerouting-without-movement’ problem.
In this abstract, we present the challenges in detecting 'teleports' and our preliminary approaches to fix this bias without losing spatial resolution
Read the full abstract
Flowminder Foundation, Eyre R. et al. (2023). The impact of apparent ‘teleports’ on the estimation of mobility from Call Detail Records (CDRs). https://doi.org/10.5281/zenodo.8414370