Using Big Data to diagnose poverty in Senegal

Clémence Cluzel

Journalist in Dakar

In Senegal, more than seven out of ten people are considered poor. This estimate is based on data collected through household income and consumption surveys, coupled with the population census. The problem is that such an approach is costly and requires significant human resources. In fact, many developing countries conduct such surveys only at rare intervals, which limits the monitoring of poverty. Another pitfall is that the traditional approach also provides an inaccurate picture of the situation. 

“In Senegal, the method used to measure poverty is the monetary approach based on income or consumption. It does not account for the deprivation suffered by individuals in such areas as health care, education, etc. But the tools for measuring poverty should allow public authorities to understand the phenomenon in its different facets”, explains Mamadou Amouzou, a demographer at National Agency of Statistics and Demography of Senegal (ANSD, Agence Nationale de Statistique et de la Démographie).

The mathematical approach based on Big Data is proving to be a good way to fill in these gaps. The study “Combining disparate data sources to improve poverty prediction and mapping” by Neeti Pokhriyal and Damien Christophe Jacques (2017) is an example. It combines traditional data such as census, income and consumption statistics with unconventional data, such as those provided by cell phones. 

Call data records provide information on the habits of users over a large part of the country. Generated with each call or message sent, they can tell when, where and with whom users communicate. These digital footprints provide information on literacy levels, mobility and access to the electricity grid, which correlates with regional wealth distribution. 

In the Senegal survey, the data provided by the telephone operator Sonatel included eleven billion calls and texts from nine million cell phone users. To this information, the researchers added satellite images that indicate parameters such as night lighting, paved roads, density of infrastructure and the type of roofing on houses.

These elements, combined with census data, provide a more complete and accurate picture of the inhabitants’ standard of living. “Artificial intelligence makes it possible to use traditional survey data to build machine learning algorithms to recognize patterns of poverty. The goal is to learn how to make data talk about poverty also when this was not their original purpose. Using such enriched data is much cheaper, more accurate and easier to update,” explains Christiane Rousseau, a professor in the Department of Mathematics and Statistics at the University of Montréal.

These enriched data have made it possible to develop poverty maps that provide a more dynamic view of the phenomenon. These maps, produced at the commune level, reflect a spatial and temporal distribution of socio-economic deprivation. For example, they show that the communes in the interior of Senegal have higher levels of poverty than the capital, Dakar, and the coastal communes.

These maps are a tool for policy-makers to direct aid to the poorest. “Mapping has a role to play in the organization of the country. By using telephone data, we can define who is eligible for humanitarian aid. This is what the organization Give Directly, which promotes direct money transfers, has done in Togo,” notes Damien Christophe Jacques, a doctoral researcher in geomatics and co-author of the survey on Senegal. “Mathematics allows us to optimize limited resources,” adds Christiane Rousseau. 

The opportunities offered by Big Data are promising, but they do raise certain questions. Indeed, telephone data belong to cell phone operators who have no real interest in communicating these data and are not always inclined to share them. Moreover, the data, which most often come from a single operator, do not reflect the situation of the entire population. “Some people have several SIM cards, while others, such as the ultra-poor, the elderly or children, do not own a phone," says Jacques, adding that “even if the information provided is very rich, these biases can raise questions in the context of rigorous studies.”

Another drawback, the use of these personal data, even if they are aggregated by area, meaning they are studied by group and not individually, raises ethical questions. “This sensitive information offers a gold mine for development projects, but it also presents a danger to privacy. We must therefore find a balance between protection of individual interests and benefits for the public interest,” says Jacques. 

“Non-traditional data represent a complement of information. They are a substitution model, in case of absence of census data, or an information augmentation model, when data are already available. The method gives real signals, but it is not a miracle solution,” emphasizes Jacques. This view is shared by Emmanuel Letouzé, director of the NGO Data-Pop Alliance and founder of OPAL (Open Algorithms) – a programme that aims to facilitate access to cell phone data and their use for social purposes in low-income countries.

“Globally, the areas and causes of poverty are already known. The real question is what to do with this information,” he says. For now, the results of the study in Senegal are not yet translated into policy changes on the ground. “But it isn’t useless, because it helps to raise awareness among policy-makers," he says. “It takes time to change habits and perceptions.”