Acting early under uncertainty: Anticipatory cash transfers in the context of flood disasters
with Stefan Dercon, Rohini Kamal, Prabhmeet Kaur Matta, Ashley Pople, Munshi Sulaiman and Hannah Timmis
Grants: J-PAL King Climate Action Initiative & Weiss Asset Management Foundation
[BL survey instrument]
Details The project evaluates a targeted risk-informed early action pilot in response to floods in Bangladesh, testing efficacy of early warning messaging, timing of cash transfers, and data-driven innovations in targeting approaches. Through a randomized evaluation, we will target approximately 20,000 households, with some households receiving unconditional cash transfers ahead of or after a flood event. We will address two critical knowledge gaps that impede adopting early actions at scale. First, they will explore the optimal timing for delivering assistance: they will evaluate when best to act by examining how households use assistance before, during, or after a disaster. Second, we will evaluate the accuracy of data-driven approaches in targeting the most vulnerable households and the trade-offs thus incurred vis-a-vis timing.


Learning Traps? Experimental Evidence on Threshold Dynamics in Education
with Noam Angrist and Claire Cullen
Details Learning gains from educational interventions in low-income countries often fade after the intervention ends. This project asks whether fade-out reflects intrinsic differences in ability, or whether there are thresholds in foundational skills beyond which learning becomes self-sustaining — a learning analogue to the poverty trap. Working with the ConnectEd phone-tutoring programme in the Philippines, we randomise 3,500 grade 3–5 students into a control group and four treatment arms tutored to progressively higher numeracy operations (addition through division). Because randomisation varies the target skill level a child is tutored to — not merely whether they are tutored — the design generates exogenous variation in the level of mastery a child reaches while holding ability constant. Weekly assessments track progression through operation levels, and follow-up assessments measure whether gains persist, providing an experimental test of whether the persistence of learning gains is continuous in the level attained or exhibits the discontinuities that a threshold model predicts.


The Labour Market Returns to Schooling: Evidence from Compulsory Schooling Reforms in Africa
with Noam Angrist, Sharnic Djaker and Julio Rodriguez
Details This project estimates the labour market returns to schooling across Africa, using the staggered introduction of compulsory schooling reforms as a source of identifying variation. Reforms in fifteen African countries changed the minimum years of schooling or the school-leaving age, shifting educational attainment for cohorts exposed to the law relative to those who narrowly preceded it. Combining harmonised labour force and household survey data with Demographic and Health Surveys, we use the first birth cohort affected by each reform to instrument for years of schooling, and estimate the effect of additional schooling on employment, earnings, and occupational structure. The design recovers the returns to a marginal year of compulsory education for populations at the lower end of the attainment distribution, where the policy relevance of schooling investments is greatest.


Climate Resilient Education Systems: Cash transfers and remote learning during drought
with Noam Angrist, Stefan Dercon and Nithya Srinivasan
Grants: Strategic Impact Evaluation and Learning, IPA
[BL survey instrument]
Details This project evaluates a climate risk–triggered package of interventions designed to protect children’s learning during droughts in northern Kenya. Leveraging a parametric insurance payout based on vegetation and rainfall indices, the intervention delivers unconditional cash transfers and remote education support to households with school-age children at risk of drought-induced learning disruptions. Using a randomized controlled trial across 210 primary schools in Garissa and Tana River counties, the study tests the effects of remote learning alone and in combination with cash transfers, alongside a nested household-level experiment evaluating one-on-one phone-based tutoring in numeracy. The evaluation examines whether timely, shock-responsive support can mitigate learning losses during climate shocks, and how combining income support with targeted educational interventions affects learning outcomes and household well-being.


The Role of Multifaceted Social Protection Programmes and Microcredit in Fostering Adaptation to Climate Change
with Prabhmeet Kaur Matta and Anindita Bhattacharjee
[OSF Registration]
Details In cyclone-prone Bangladesh, we investigate how beneficiaries and non-beneficiaries of microcredit and multifaceted social protection programmes differ in their experiences with climate events, adaptation strategies, and livelihood decisions. Combining rich qualitative data collected through semi-structured interviews with quantitative survey data collected through a structured household survey, we seek to investigate whether asset transfers function as potential safety nets during climate shocks, and the market conditions that affect households’ ability to leverage these assets during crises. We investigate how households prepare for and recover from climate shocks, their perceptions of future climate risks, and the role of migration and insurance in their adaptation portfolios. While qualitative data collected through semi-structured interviews offers richer and more nuanced perspectives than structured survey data, the analysis of such data is often subject to cherry picking and narrative fallacies due to researcher bias. Natural language processing (NLP) methods may help overcome these issues but come at the cost of losing some of the narrative richness of qualitative data. This paper develops a method which aims to balance these concerns: We pre-specify how we use NLP methods to identify key themes and conduct sentiment analysis within these themes, and structure our qualitative analysis of the open-ended text data collected through the semi-structured interviews accordingly. This approach — both the act of pre-specification and the use of NLP to draw out key themes and conduct sentiment analysis — allows us to overcome core concerns with researcher bias, while at the same time retaining the richness of a qualitative analysis.


Fiscal Savings from Anticipatory Action: Under What Conditions Does Early Action Pay for Itself?
with Prabhmeet Kaur Matta and Jasper Andrée
Prepared for the Inter-American Development Bank and UN-OCHA
Details Governments in Latin America and the Caribbean spend billions each year responding to climate disasters after they occur. This report asks whether anticipatory action — pre-arranged financing released automatically when a forecast trigger fires — can reduce that fiscal burden by enough to pay for itself. We develop a model of expected government fiscal savings from anticipatory action as a function of the probability that a disaster occurs, the forecast hit rate, the share of emergency expenditure that early action avoids, and the government’s baseline fiscal outlay per event. Calibrating the model across eleven country–hazard frameworks in four Latin American and Caribbean sub-regions, we find that anticipatory action pays for itself when it reduces government emergency expenditure by at least roughly twelve per cent in triggered disaster years; the best-documented case, Honduras drought, generates $1.69 in fiscal savings per dollar of programme cost under the preferred specification. The framework — published as an open-access repository — identifies the conditions under which the fiscal case for anticipatory action holds, and the data investments needed to evaluate it with greater precision.