In this twenty-third installment of our RECOVR Roundup series, we are sharing new findings and analysis from the RECOVR Research Hub and from our partner organizations, as well as links on what is happening in the Social Protection landscape in response to COVID-19. Read the previous installment if you missed it, and sign up for our mailing list if you'd like to receive this roundup series directly to your inbox. 

As always, we encourage you to write to our team with ideas for features.

New Findings & Analysis

Togo: Designing and Implementing a Fully Digital Social Assistance Program during COVID-19

A new case study highlights digital transfer implementation lessons for policymakers.

In April 2020, Togo’s Ministry of Digital Economy and Digital Transformation (MENTD) launched the Novissi cash transfer scheme. An unconditional cash transfer (UCT) to assist informal workers whose livelihoods have been upended by the coronavirus pandemic, Novissi is a fully digital social assistance program. As of March 2021, Novissi has reached 819,972 beneficiaries and disbursed approximately US$23.9 million (13,308,224,040 FCFA). This new case study details the program design process and its implementation during the coronavirus pandemic, highlighting lessons learned on the use of mobile money to support fully digital social assistance during times of crisis. Read more here.

Togo: How Can Mobile Phone Data and Machine Learning Help Find the People Who Need Assistance?

Targeting with machine learning outperformed other available options, but is best used as a supplemental tool to conventional approaches.

Following the launch of Novissi, the government partnered with GiveDirectly, CEGA, and IPA to further scale the UCT program through remote identification, enrollment, and disbursements to over 138,500 Togolese recipients. A new working paper by Emily Aiken, Suzanne Bellue, Dean Karlan, Chris Udry, and Joshua Blumenstock analyzes how “big” data from satellites and mobile phones can improve anti-poverty program targeting. The team used traditional survey-based measures of consumption and wealth to train machine learning algorithms to recognize patterns of poverty in non-traditional data. These algorithms were then used to prioritize aid to the poorest regions and mobile subscribers. Researchers compared the machine learning method to the geographic targeting which was used for the Novissi program and found that machine learning reduced exclusion errors by 4-21 percent. They also simulated other targeting methods using existing household-level data, and found that machine learning worked approximately as well as an asset-based wealth index, but less well than a poverty probability index or a proxy means test. Read more here.

What We're Reading & Watching

• The UN has released a brief outlining key lessons for social protection based on policy responses throughout the pandemic. Among the topics: expanding social protection coverage to informal workers, migrants, and specific vulnerable populations; adopting gender-responsive measures; introducing digital innovation in delivery mechanisms; mobilizing new financing; and enacting legislative reforms to support employees and the self-employed.
   
• Another working paper highlights the potential for high-resolution satellite imagery and deep learning methods to complement the survey data traditionally used for program evaluations. Based on a recent anti-poverty program in rural Kenya, the team infers changes in household wealth based on satellite data about roofing types—with metal roofs indicating greater wealth—and finds that these results are consistent with survey-based data.
   
• Cash transfers for disaster response often arrive after a humanitarian emergency or natural disaster has occurred. However, new research from Bangladesh suggests that the impacts of these transfers can be strong if they arrive before the disaster. Researchers from Oxford University and the Centre for Disaster Response studied a World Food Program intervention that provided cash transfers to Bangladeshi households predicted to be affected by floods in July 2020. They found that households used the transfers to buy food and water in order to prepare for the impact of the flood and were less likely to skip meals or have to sell assets afterward.
   
• US social assistance programs often include work requirements with the stated aim of fostering “self-sufficiency.” However, a new working paper from the National Bureau of Economic Research questions this logic. The authors look at participation in the SNAP food stamp program in Virginia, where work requirements were suspended in 2009 and reinstated in 2013. They find that the reinstatement of the work requirements reduced program participation by eligible working-age adults by 53 percent without increasing employment rates. Unhoused adults were particularly likely to lose their benefits due to the new requirements.
   
• A new report from the UN Economic Commission for Latin America and the Caribbean offers case studies from 15 countries in the region that have used social registries to distribute welfare benefits during the pandemic.