A subscriber’s home location is an estimate of the location in which they are residing during a given period of time.
This may be a permanent home or temporary accommodation. This location is never the location of a subscriber’s place of residence itself, only the location of a cell tower estimated to be associated with a subscriber's place of residence.
We can use a range of different methods for estimating the home locations of subscribers, ranging from simple to more complex.
Most simply, we can assume that the area in which a subscriber is most often observed (i.e. has a network event, such as a call, recorded) is their home location in a given period of time (e.g. over the course of a week). However, this method is likely to catch other places of interest (PoIs) such as place of work or education.
We can develop this approach by using the area in which a subscriber is most often observed overnight (e.g. between 18:00 and 08:00) and during the weekend in a given period of time. This approach is more likely to capture a subscriber’s accommodation but does make assumptions about working patterns which may not apply to certain roles or industries.
Flowminder’s methodology uses the most frequent location at which a subscriber is last observed each day. For example, a subscriber who makes their last call of the day in Area A 4 times, Area B one time, and Area C 2 times over the course of one week will be assigned Area A as their home location. Any other network events during that week are not used to infer the home location. This methodology attempts to capture the area where a subscriber ends their day, and therefore where they stay overnight.
Each of these methods can be calculated for different time periods and with different frequencies, depending on the requirements of the analysis.
For example, we can assign subscribers’ home locations each month using a one month time window. Similarly, we could assign subscribers’ home location weekly using a one week time window.
For shorter time windows, such as daily home locations, we can also use rolling time windows. For example, we could assign subscribers’ home locations each day using the previous seven days data.
In general, using longer time windows to estimate home location will produce more robust estimates of home location.
For example, when using the Flowminder methodology, a three-day window would assign a subscriber’s home location to an incorrect location if the subscriber had their last network of day on two of those days away from home; extending the window to seven days reduces this risk.