Human dynamics

It is a privilege getting a chance to analyse the largest released mobile phone dataset for research community by that time. Data of Orange Telecom from Cote d’Ivoire are released for Data for Development challenge (D4D) in 2012.

So we asked: could such anonimized mobile communication (call timings, locations and person ids) serve as a socio-economic proxy indicator for the country? The answer is yes.

Mobility <–> communication frequency

For instance, from the averaged frequency and length of communication, one can well observe important events in the country, as well as correlate those with mobility of people.The black footprint shows mobility (calculated from calling locations). The red shows calling frequency and the blue, the duration.  We can see from the Figure 1 that the mobility and frequency correlate (we also calculated this to confirm). Interestingly, the duration of calls has a different pattern, and does not positively correlate with either frequency, nor mobility. Our conclusion is that people, when on the move,  communicate more in terms of number of calls, but when they want to make longer calls, they prefer to be in one place. Not that surprising when one thinks about it.Now, for all the 3 of the activities, one can identify easily New Year hours, Christmas, and Easter. Without previous knowledge, we could from those ananymized data find out that Cote d’Ivoire is a country in which religion is important (for a large part of its inhabitants).The graph (b) in the Figure 1 shows and averaged daily traveled distance and we have identified that the 3 peak periods match with December, 11, 2011 Parliamentary elections, then Africa Cup of Nations 2012 in football, where Cote d’Ivoire played in the final, and  Bouake Carnival and Fete du Dipri.

Fig1Figure1: Mobility vs. calling frequency and duration

 Economical status <–> radius of gyration

Another interesting finding is that a relatively simple quantification of human mobility, such as radius of gyration, can tell us a lot about different regions in the country.

This African country has its economic and development center in the city of Abijan, on the south east on the coast, while the northern and western parts are less developed, and on many indexes considered poor. Radius of gyration measures how far on average people do travel (very simplified interpretation, but serves our purpose). On the map (a) in Figure 2, we use the darker color for the regions in which people have a larger radius of gyration.

Now, it is apparent how the people in and near Abijan have relatively low radius of gyration, showing that they do not travel too far from their home location, and  people from the poor regions have considerably larger radius of their trajectories. That is because they do not have all the necessary services (hospitals, schools, ports) nearby, and they need to travel further and more frequently to fulfill their basic needs.  Moreover, it is rather clear how the whole country seems to gravitate towards the wealthy south-east coast and Abijan.

In the graph (b) on the right in Figure 2, we have averaged the radius of gyration for the 3 regions:

  1. Wealthy Abijan,
  2. Poor North-West,
  3. The whole Cote d’Ivoire.

Our aim is to show that this measure of human mobility, which is otherwise shown to be consistent over time for one country, differ in different regions, serving as an economic status fingerprint.

Fig5Figure2: User radius of gyration statistics

Administrative units and economic centers <–> commuting

Finally, to my own surprise,  using only the communication data, we were able to find the home and work locations for users, and based on those to calculate the the commuting network. Applying one of the common network partitioning algorithms (modularity detection), we were able to identify regions of commuting, that incredibly well match with with the administrative regions in the country. And for those areas where the obtained commuting regions do not match (mostly red, blue, green and cyan), we can easily identify the reasons: the borders are distorted by the economical centers (Abijan, Yamoussoukro, Gagnoa) that attract commuting.

On the left map in Figure 3, we show the important economic centers, that are identified after we have run a standard PageRank algorithm on the commuting network.


Figure3: Regions and centers of commuting importance


While this work shows a lot of ideas obtained based on what we already know about this particular country, there are at least a few points that amazed me-lover of data analysis and convinced me of its power:

  • When we call is different to how long we call.
  • How we move has to do with how wealthy we are.
  • We are free, but are we aware that we still move a lot under some invisible constraints?

For the rest and more details of our analysis, you can have a look at our PLoS ONE article.

Šćepanović, S., Mishkovski, I., Hui, P., Nurminen, J.K., Ylä-Jääski, A., “Mobile Phone Call Data as a Regional Socio-economic Proxy Indicator,” PLOS One, 2015.