In order to make robust inferences about subscribers’ mobility, we need a suitable amount of sufficient quality location data.
However, many subscribers, especially in Low- and Middle-Income Countries (LMICs), use their mobile devices irregularly (i.e. their patterns of activity change) and/or infrequently (i.e. they only use their mobile devices occasionally). Both of the characteristics make it difficult to make inferences about their mobility, particularly identifying “places of interest” (PoIs) such as subscribers’ home locations.
We therefore only include subscribers who are active enough (i.e. use their mobile devices frequently enough) when calculating mobility aggregates. We term these “active subscribers”.
The threshold for a subscriber to be described as “active” depends on the application and the required temporal resolution.
In general, we discard subscribers who were only observed for a short period of time (e.g. for one week or less). We also discard subscribers who do not use their mobile device often enough to provide the necessary temporal resolution. For example, to calculate a subscriber’s weekly home location we require at least one network event each week. If we were instead to calculate daily home location using a three-day rolling window we would require at least 1 call in any given three-day period.
It is therefore important to consider what your use case requires. Increasing the temporal resolution of the aggregates may result in substantially fewer subscribers being included in the analysis. This may affect the representativeness of the data if higher usage of mobile devices is associated with socio-economic status or age, for example.