By Luis Tejerina and Juan Miguel Villa
According to the US government, the 2020 census could cost as much as $15.6 billion, or $49 per inhabitant. While in developing countries, according to a study of 77 countries carried out by Development Initiatives, the average cost of conducting a household survey ranges from $1 million to $1.6 million,
These initiatives to calculate a country’s population or to measure poverty levels are important for guiding public policy decision-making. But gathering data is often done haphazardly and is expensive– especially for developing countries.
Traditional methods used by statistical institutes to collect information are therefore being challenged by the (relatively) low cost of using what is known as “big data.”
These data are generated continually in different forms, ranging from spatial information produced by hundreds of satellites orbiting the planet to geographic information and information from calls by people using cell phones.
An emblematic example is the nocturnal satellite image of the Korean peninsula: North Korea is totally dark, while its neighbor, South Korea, is completely illuminated. Without access to public data extracted from surveys and social accounting, it could be inferred which of the two countries has a higher level of economic development and therefore higher levels of well-being.
Access to this type of information brings into question the very future of statistical information.
According to World Bank data suggest that in 2016 there were 102 cell phone lines per 100 inhabitants on the planet. Joshua Blumenstock, Gabriel Cadamuro, and Roberto On used this cell phone data in a study to map poverty in Rwanda. For their part, Jessica Steele and her co-authors combined household survey data with phone call records to produce similar maps in Bangladesh. Both studies produced similar results, and with better resolution than traditional maps (see Figure 1).
In fact, big data can do much more than map poverty. It can go as far as measuring a wide range of health, education and migration indicators.
Figure 1. Poverty and Income Maps based on (a) Satellite images, (b) Cellphone records, and (c) Household surveys
Satellite data on lighting provide a unique opportunity to measure poverty, given that the data are in the public domain and cover the entire planet. This coverage is better at low latitudes, where global poverty is concentrated..
Maxim Pinkovskiy and Xavier Sala-i-Martin used data on lighting to validate measurements of well-being based on conventional household survey methods and national accounts. Their analysis details that data on lighting represent a good approximation for the measurement of poverty.
Neal Jean and his co-authors went further and combined information from satellite data on nighttime lighting and daytime characteristics with information from household surveys. Using a machine learning algorithm, they succeeded in developing a neural network to predict household living conditions with 75 percent precision in African countries where a large portion of income depends on subsistence production.
Measuring poverty is transitioning toward the use of big data. It costs less and has great informative value.
Governments of developing countries have not yet officially adopted these steps to measure the well-being of their populations. But the United Nations has decided to tap into the advantages of using big data by creating a Global Pulse laboratory whose mandate is to explore big data to monitor compliance with its Sustainable Development Goals, at a fraction of the cost of traditional methods.
Hopefully, little by little, this approach will go beyond a novelty and become a reality for measuring and monitoring poverty indicators at the local level while providing government authorities with timely information.
About the authors: