January 12, 2022

Micro-estimates of wealth for all low- and middle-income countries

By: Guanghua Chi, Han Fang, Sourav Chatterjee, Joshua E. Blumenstock

What the research is:

The World Bank estimates that there were 120 million people in extreme poverty in 2020 due to the impacts of the COVID-19 pandemic [1]. At the same time, in low- and middle-income countries (LMIC), nonprofits and governments face difficult decisions about which populations to prioritize for assistance due to the lack of information about where poor people live. To meet this need, Meta’s Data for Good team developed the public data set of Relative Wealth Index (RWI) in collaboration with researchers at the University of California, Berkeley, which provides micro-estimates of wealth and poverty for low- and middle-income countries at 2.4 km resolution.

The estimates are built by applying machine learning (ML) algorithms to vast and heterogeneous data from satellites, mobile phone networks, and topographic maps, as well as aggregated and deidentified connectivity data from Facebook. We train and calibrate the estimates using nationally representative household survey data from 56 LMICs, then validate their accuracy using four independent sources of household survey data from 18 countries. We also provide confidence intervals for each micro-estimate to facilitate responsible downstream use.

These estimates are provided free for public use in the hope that they enable targeted policy response to the COVID-19 pandemic, provide the foundation for new insights into the causes and consequences of economic development and growth, and promote responsible policymaking in support of the Sustainable Development Goals.

How it works:

Our approach, outlined in Figure 1, relies on ground truth measurements of household wealth collected through traditional face-to-face surveys with 1,457,315 unique households living in 66,819 villages in 56 different LMICs around the world. These Demographic and Health Surveys (DHS), which are independently funded by the U.S. Agency for International Development, contain detailed questions about the economic circumstances of each household, and make it possible to compute a standardized indicator of the average asset-based wealth of each village.

We then use spatial markers in the survey data to link each village to a vast array of nontraditional digital data. This includes high-resolution satellite imagery, data from mobile phone networks, and topographic maps, as well as aggregated and deidentified connectivity data from Facebook, as previously mentioned. The data is processed using deep learning and other computational algorithms, which convert the raw data to a set of quantitative features of each village. We use these features to train a supervised ML model that predicts the relative wealth of each of the populated 2.4 km grid cells in LMICs.

Figure 1: Overview of approach. a) Nationally representative household survey data is obtained from 56 countries around the world. b) In Nigeria, for example, 40,680 households are surveyed in 899 unique survey locations (“villages”). Nontraditional data from satellites and other existing sensors is also sourced from each location. c) This data is used to train an ML algorithm that predicts micro-regional poverty from nontraditional data, even in regions where no ground truth data exists.

The estimates of wealth and poverty are quite accurate. Depending on the method used to evaluate performance, the model explains 56-70 percent of the actual variation in household-level wealth in LMICs (Figure 2a). This performance compares favorably to state-of-the-art methods that focus on single countries or continents.

To validate the accuracy of these estimates, and to eliminate the possibility that the machine learning model is “overfit” on the DHS surveys, we compare the model’s estimates to four independent sources of ground truth data: (1) 15 LMICs that have collected and published census data since 2001, (2) a nationally representative sample of 6,172 households collected by the government of the Togolese Republic (Togo) in 2018 and 2019, (3) a nationally representative sample of 22,104 households collected by the government of Nigeria in 2019, and (4) 5,703 households in two counties in Kenya surveyed by GiveDirectly. Importantly, these data sets are independently collected and are never used to train the ML model. We find that the ML model explains 50-86 percent of the variation in household wealth in the ground truth data.

Figure 2: Model performance. a) Distribution of model performance, across 56 countries with ground truth data, using three different approaches to cross-validation. b) Much of the model’s predictive power comes from being able to differentiate between rural and urban locations, but the model also detects wealth differentials within urban and rural locations. c) The ML model explains 72 percent of the variation in wealth, as measured with independent census data from 15 LMICs. Population-weighted regression lines are shown in blue; 95 percent confidence intervals are shown in dashes.

Why it matters:

How might these estimates be used to guide real-world policy decisions? One key application is in the targeting of social assistance and humanitarian aid. In the months following the onset of the COVID-19 pandemic, hundreds of new social protection programs were launched in LMICs, and in each case, program administrators faced difficult decisions about who to prioritize for assistance [2]. This is because in many LMICs, governments, and nonprofits do not have comprehensive data on the income or consumption of individual households [3]. The new estimates provide one potential solution.

Based on the strength of these results, several countries in the developing world have already benefited from insights derived from the RWI. Through a collaboration with the Government of Nigeria, the World Bank, and UC Berkeley, the RWI has proven useful for identifying pockets of poverty and potential beneficiaries for the Rapid Response Register for the COVID-19 Cash Transfer Project, which aims to benefit one million Nigerians living in poverty and has been distributing approximately $25 million per month to households in need across the country. Likewise, the Government of Togo has been using these estimates to target mobile money transfers to hundreds of thousands of the country’s poorest mobile subscribers in collaboration with GiveDirectly. To date, the Novissi program has distributed over $23 million to over 800,000 people living in poverty in Togo [7].

These examples highlight how the RWI can improve targeting performance even in countries with robust national statistical offices and work in tandem with administrative data to reach the poorest of the poor with critical financial assistance. In the large number of LMICs that have not conducted a recent nationally representative household survey, these micro-estimates create an option for geographic targeting that would otherwise not exist.

Read the full paper:

Microestimates of wealth for all low- and middle-income countries

Learn more:

Learn more about Meta’s Data for Good program on their website. Download the data here. View the interactive map here.


[1] K Atanda, A Cojocaru. Shocks and vulnerability to poverty in middle-income countries. World Bank blog (2021). available at https://blogs.worldbank.org/developmenttalk/shocks-and-vulnerability-poverty-middle-income-countries

[2] U Gentilini, M Almenfi, I Orton, P Dale, Social protection and jobs responses to covid-19: A 928 real-time review of country measures (2020).

[3] K Lindert, T Karippacheril, I Caillava, K Chávez, Sourcebook on the Foundations of Social 930 Protection Delivery Systems. (World Bank Publications), (2020).

[4] 1m Nigerians to benefit from Covid-19 cash transfer, Osinbajo says. The Guard. (2021) available at https://guardian.ng/news/1m-nigerians-to-benefit-from-covid-19-cash-transfer-osinbajo-says/.

[5] J Blumenstock, J Lain, I Smythe, and T Vishnaswath, “Using Big Data and Machine Learning to Locate the Poor in Nigeria.” World Bank Blog. Available at: https://blogs.worldbank.org/opendata/using-big-data-and-machine-learning-locate-poor-nigeria

[6] T Simonite, A clever strategy to distribute covid aid—with satellite data. Wired (2020) available at https://www.wired.com/story/clever-strategy-distribute-covid-aid-satellite-data/.

[7] Government of Togo, Novissi Program Overview. Available at: https://novissi.gouv.tg/en/home-new-en/