Dial 999 to Receive Fast Cash: How Data Science and Machine Learning Can Identify and Aid People Living in Poverty

Share on linkedin
Share on twitter
Share on facebook

Instead of dialing 911 for emergency assistance, what if you could dial 999 for the cash you need to feed your family and buy supplies during a crisis? Imagine a world in which you, along with thousands of others, could rapidly and accurately be identified as someone needing assistance, and this assistance could be sent to you via text message almost immediately after you’d applied for it.

This service, called MobileAid, is now a reality in Togo. And it will soon be scaling internationally thanks to new funding from the data.org Inclusive Growth and Recovery Challenge supported by founding partners the Mastercard Center for Inclusive Growth and The Rockefeller Foundation.

MobileAid is just one example of data-driven global transformation among many being considered by more than 1,500 people from 52 countries at Good Tech Fest 2021. The conference attendees attended numerous workshops to engage with compelling case studies that illustrate what data and data science for social good can do.

GiveDirectly, one of the fastest growing NGOs focused on international issues, and its academic partner, the Center for Effective Global Action (CEGA) at UC Berkeley, have used data science and machine learning to identify and send cash with unprecedented speed to 95,000 people living in poverty in Togo during the COVID-19 pandemic. Governmental advisors in Togo have described the project as “foundational” in terms of setting up the country’s social protection system. It’s one of eight projects chosen to receive a combined $10 million in funding and technical assistance in the Challenge. The Challenge drew ideas and proposals from 108 countries around the world to tackle social problems with data and technology.

On the road to recovery from COVID-19, many countries lack reliable and up-to date information about economic conditions on the ground and have no way of collecting it. Traditional aid modalities relying on in-person enrollment and delivery are no longer safe or scalable, with governments and NGOs lacking personnel and relief taking weeks to arrive.

GiveDirectly and CEGA proposed addressing the challenge by developing and testing a new model for humanitarian support that enables cash transfers to be deployed effectively, accurately, and at scale to those who need them most. The project incorporates new data and computational technologies to identify people and places in economic distress and integrates data from mobile phones, satellite imagery, and traditional surveys. The plan piloted in Togo during the pandemic and aims to develop a transparent framework to scale globally.

Han Sheng Chia, Vice President, Innovation at GiveDirectly, and Emily Aiken, Data Scientist and UC Berkeley PhD candidate working at CEGA, gave a talk titled “Dial 999 to Receive Emergency Cash” during the 2021 Good Tech Fest.

Han Sheng and Emily explained in detail how they ran the project.

Direct Cash Payments

“GiveDirectly is a non-profit that focuses on just one thing, which is providing unconditional cash to families living in extreme poverty around the world,” said Han Sheng.

The organization has so far delivered direct payments to 695,000 recipients in 10 countries. GiveDirectly manages end-to-end processes for its programs, designing each program, determining its location, targeting criteria and payment amounts, and then executing on it with targeting to identify eligible recipients. GiveDirectly then enrolls and audits people for the programs, paying them and monitoring their spending, evaluating each project through surveys and controlled trials.

Han Sheng said that a common concern raised when he talks about his work, is that people might spend the money on so-called temptation goods such as alcohol or tobacco. But there are more than 165 studies that have shown the efficacy of cash transfers in helping people in poverty, including 15 randomized controlled trials of GiveDirectly’s programs. These studies have shown that cash transfers consistently lead to increases in positive outcomes such as school attendance, use of health facilities, and crop production. Cash transfers also lead to decreases in negative outcomes including child labor and intimate partner violence. And a review of 19 studies found that cash payments led to no increase in spending on temptation goods such as alcohol or tobacco.

Algorithmic Targeting of Aid

What was new about GiveDirectly’s partnership with CEGA on this project was the rapid identification of recipients using machine learning.

CEGA “thinks about how to implement these programs at a massive scale, entirely remotely,” said Emily.

Good income data doesn’t exist in many developing countries. In Togo, for example, the last census didn’t contain any information on poverty. Traditionally, poverty information could be ascertained only through a door-to-door survey, which takes time and resources. Emily and the team at CEGA and their partners have focused on two nontraditional “big data” sources to identify poverty in developing countries. The first source, satellite imagery, is used to produce hyper-granular poverty maps by identifying infrastructure and building materials.

“The intuition here is that wealthy places and poor places look different from above,” said Han Sheng. “By and large, wealthy places have more organized urban planning. Roads are made of more long-lasting material, homes may have roofs that are more durable. And these are patterns you can pick out from satellite data.”

The second source, mobile phone data, is used to identify individual-level poverty by looking at the number of international calls, data purchases, and size of contact networks, Emily said. The organizations work with cellphone companies to create a list of predicted poor cellphone subscribers. If one makes a lot of international phone calls, buys a lot of data, then the intuition is that one is classified as wealthier and not eligible for the programs.

“We start with mobile phone metadata, which basically details every transaction occurring on the cell network. We don’t get anything like the content of text messages, but who is calling who, at what time,” said Emily. “We calculate a bunch of statistics about mobile phone use, things like how many days are they using their phone out of the year, how many transactions are they making in each part of the country based on cell tower locations, how many unique contacts do they have, and so on.”

The two organizations then combine that information with survey data, matching actual data about wealth outcomes with the data from the cellphones, and using that, can predict poverty estimates for each phone user.

The team at CEGA has been researching improved targeting using nontraditional big data sources for more than a decade.

COVID-19 Relief in Togo

“[COVID-19] has led to an increase in poverty for the first time in 20 years, and an increase at tremendous scale,” said Han Sheng. “The prediction for the increase in extreme poverty in 2020 was 120 million new people in poverty. As we stared at this giant problem, we needed a new way to respond at tremendous speed and scale and to distribute cash.”

The tenets of a successful response to this unprecedented crisis include the ability to scale globally, to give help now, having a secure model that is replicable at scale, and to be able to precisely identify and prioritize those who need aid most, and because of the nature of COVID-19 and the need to avoid in-person contact, a fully remote approach.

GiveDirectly and the team at CEGA worked in partnership with the Ministry of Digital Economy and Innovation in Togo to choose the beneficiaries for cash payments. They looked at the 400 cantons, or administrative regions, in the country and used satellite imagery combined with algorithmic learning to identify the 100 most poverty-stricken cantons. They then used cellphone data to identify potential program recipients.

To reach those recipients, they used catchy radio ads combined with in-person outreach to community leaders in two enrollment phases. The gender split for recipients was 48% male and 52% female, and the groups found that the machine learning targeting approach was more accurate compared to available alternatives such as distribution by income or mere geographic targeting by canton alone.

Measurement also showed more beneficiaries are living in poverty, and that a better proportion of people living in poverty were paid versus other identification methods.

Scaling in Togo and Around the World 

“What we’re super excited about is the fact that we have tremendous support, at the policy level and at foundations with really strong partners, to transform how the sector thinks about not just social protection but about emergency response as well,” said Han Sheng.

The new approach can augment regular safety net programs. The use of machine learning and data science can increase the amount going to regular beneficiaries and identify new beneficiaries. It also frees up in-person capacity to identify more hard-to-find beneficiaries for benefits.

Increased enrollment can come from combining the approach with ecosystem partner screening, government social protection lists, and in-person office-based and home-based screening, Han Sheng said. He also pointed out that while mobile phone use in Togo exceeds 80 percent, there are other ways to reach recipients including radio ads, community leader meetings, and loud trucks in neighborhoods announcing the opportunity. Supplementing text-based service with measures like these along with voice-based support and in-person visits is the likeliest way forward based on the experience in Togo.

The approach has drawn partner support and press coverage around the world in addition to the support of the data.org Inclusive Growth and Recovery Challenge.

As a next step, GiveDirectly and CEGA will be looking to improve rates of uptake by improving the pathways to getting people money using their technology. In Togo, for example, not everybody has a phone, and keeping phones charged is difficult. Literacy can also pose a challenge, making it more difficult for people to navigate the questions to get accepted for benefits. There are also issues with consistent cell coverage. Han Shen and Emily said that while countries with more established infrastructure make it easier to swiftly scale a program like theirs, it is vital to figure out how to serve the most vulnerable people in the world. And that is where they plan to focus their efforts, using data science to deliver life-changing social impact.

Find out more about GiveDirectly, CEGA and all the data.org Challenge awardees here.

Building a better future with data

data.org brings people and organizations together to effect positive social change and build the field of data science for social impact.

Data.org uses cookies to monitor usage of the site to better understand user interests and help us plan for future iterations of the site such as regional specific pages. By using this website you agree to our Privacy Policy. If you do not want to have cookies track your movements on the website we recommend incognito mode.