We routinely think of traffic patterns as representative of where carriers should invest, where investments are making the best return, or simply as a regulatory responsibility.
But there is increasing evidence that, in some cases, they carry far more information than we imagine. As the cellphone becomes pervasive in developing countries, and as poorer people buy airtime on an ad hoc basis (and even, in some markets, trade that airtime as a sort of alternative currency), it is realistic to think that the spending patterns on airtime might correlate with the income of the people who are using the handsets.
This is not a new thought: Joshua Blumenstock and Nathan Eagle found in 2010 that spending on airtime in Rwanda was a possible proxy for income. This follows a path literally trodden by development economists for more than 50 years: before cellphones, economists would visit houses in remote areas of developing countries and record how much families spent on clothing, medicine, or different types of food, to find goods that would act as useful indicators of poverty.
Cellphone-use research has advantages compared to this method: top-up data can be provided in a data file, so it is more precise, and in developing countries few people have landlines or contracts. Also, working people do not come home to find a researcher sitting at their dinner table asking how much they paid for rice three weeks ago.
Recent papers have found a correlation between socio-economic class and the number of cell towers that the owner's handset connected to each week (poorer people travel less often), or the greatest distance between cell towers used in a week (poorer people do not travel far). This is interesting, but even the authors would concede that it is unlikely to be a basis for policy.
The latest research* – by Thoralf Gutierrez and Vincent Blondel of Université catholique de Louvain and Gautier Krings of Real Impact Analytics – is potentially much more useful. Their hypothesis is that, in developing countries, the poorest phone users buy small amounts of airtime. Wealthier users will buy airtime in bigger chunks, because they can and because it is more convenient.
It is similar to our behaviour at the ATM. If we earn $200 per week, we do not take it all out of the bank in one go. It we earn $2,000 a week, we take out at least $200 because it is going to save time later.
For their seven-month sample of anonymised Call Detail Records and top-up history from a cellular network company, individual users tended to buy roughly the same amount of airtime in every transaction. We do not know with confidence that wealth determines this behaviour, but there is some evidence from previous research that poorer users make smaller, more frequent purchases.
Their sample was a network's subscribers in Côte D’Ivoire. About the size of Germany but with a quarter the population, it is certainly a poor country: its nominal GDP per capita, according to the IMF, is about $3 per day.
Government census information is undoubtedly a poor guide to poverty in the country for several reasons. The first is that this type of research is not a priority in poor countries, and often the results are unreliable guides to reality – for example, not everyone has an address, and villages are inaccessible. The second is that many people work in the informal economy, and so they have income that they are unlikely to reveal in a survey. Finally, Côte D’Ivoire experienced a civil war that made tens of thousands of its citizens into refugees in Liberia and Ghana after its November 2010 election, and the last census was in 2008. So, even if the census data was accurate then, it isn't today.
The paper (it is short, with heat maps of the results) has fascinating insights. The average purchase is higher in urban areas, but also on the coast, the border with Liberia – the centre of smuggling and other cash-generative activity – and the roads to Mali and Burkina Faso. Using the data to create a Gini index of purchase averages, which measures inequality, again we clearly see that some urban areas contain both rich and poor. Cities such as Korhogo are large, but surprisingly uniform in purchasing patterns, which the authors attribute to a larger middle class.
Finally, the research uses the data to examine user networks – communities of people who call each other a lot. Again, it is not surprising to find that wealthy individuals tend to communicate with other wealthy individuals, and the same for the poor. This replicates an effect found in similar data in the UK.
On the other hand, some cities, such as the capital Abidjan and the second-largest city Bouaké, have diverse communities in which the amount that users spend on their top-ups is not a good guide to who is in their social network – the poor appear to mix more with people who have more money. Some cities also combine high inequality with low diversity: suggesting there are both rich and poor, but they do not mix much.
With no accurate census data to compare this to, this is still speculation. Until the model is tested in a developing country with better census data, we cannot know for sure whether top-up size is a useful proxy for income or wealth.
But if average top-up size is a good proxy, the data is better than a census in several ways: it uncovers informal income, it measures inequality and diversity, and it provides information in real time. In an evolving economy, this could rapidly show the effect of policy. Mobile networks could be creating some of the most important data in the whole of development economics, as an afterthought.
* Gutierrez T, Krings G, Blondel V (2013) "Evaluating socio-economic state of a country analysing airtime credit and mobile phone datasets" Download: arxiv.org/pdf/1309.4496
