Protecting Learning During Drought: Experimental Evidence on Remote Education and Cash Transfers in Kenya

The CREST programme delivers a three-component drought-response package to households with school-age children (Grades 1--9) in Garissa and Tana River counties. All components are funded through a parametric insurance payout (ARC Ltd.), triggered automatically when satellite-measured vegetation indices (NDVI) and rainfall indicators (SPI) cross pre-agreed thresholds.

1. Phone-based Remote Learning (T1 and T2): Eight weeks of structured IVR (Interactive Voice Response) lessons delivered to learners via caregiver phones. Two lessons per week per learner (one English, one Maths), each split across two calls of 5--10 minutes. Content is aligned with the Kenyan national curriculum, co-developed with the Kenya Institute for Curriculum Development (KICD), and covers foundational literacy, numeracy, and safeguarding across three grade bands (1--3, 4--6, 7--9). SMS and WhatsApp are used for scheduling. Optional short quizzes and live weekend group sessions (20 minutes, teacher-led) complement the IVR content.

2. Phone-based One-on-One Tutoring (T1b subset only): Structured, adaptive one-on-one tutoring in foundational numeracy for learners in Grades 3--6, delivered via phone calls by trained tutors (approximately 20 minutes/session, 8 sessions total). Tutors are
recruited and trained by IRC with technical guidance from Youth Impact, using the ConnectEd model. Learner level is set using the baseline assessment; progression follows check-point questions at the end of each session. Tutoring runs in parallel with the remote learning programme.

3. Unconditional Cash Transfers (T2 only): A one-off, unconditional mobile money transfer to households with at least one learner registered in a participating school. Transfer values range from KES 5,000 to KES 10,000 (approx.\ USD 50--100) per household,
depending on the number of registered learners. Disbursed in the second week of February 2026. Receipt is not contingent on school attendance, survey participation, or use of any other programme component.

2026-02-08

2026-05-16

Child learning: Standardised numeracy test scores, assessed via phone-based instruments adapted from the ASER/Uwezo framework.

Learning scores will be standardised using the control group mean and standard deviation at each follow-up round.

(i) Food security: Child meals per day; dietary diversity; household food reserves.
(ii) Literacy: Standardised literacy test scores, assessed via phone-based instruments adapted from the ASER/Uwezo framework.
(iii) School attendance: Indicators of current enrolment, regular attendance (days attended in past week), and dropout status, measured through caregiver report.
(iv) School engagement: Time spent on educational activities; caregiver involvement in learning; engagement with IVR lessons (treatment arms only).
(v) Child labour: Hours on paid work, herding, domestic chores, or other economic activities.
(vi) Psychological wellbeing: Child and caregiver wellbeing indices (validated short-form instruments: WHO-5, GAD-2, and age-appropriate child items).
(vii) Migration: Household displacement; temporary migration for pastoralist activities

Secondary outcomes serve as mechanisms through which the primary intervention effects operate (per the theory of change).

Experimental Design
The evaluation uses a three-arm cluster-randomised controlled trial (RCT) at the school level, with a nested within-school household-level randomisation.

Between-school randomisation: 210 schools across Garissa and Tana River counties were randomly assigned to three arms:
C – Pure Control (70 schools, 2,300 learner-households)
T1 – Remote Learning (73 schools, 2,400 learner-households total in T1)
T2 – Remote Learning + Cash Transfer (67 schools, 2,200 learner-households)

The comparison T1a vs. C identifies the effect of remote learning; T2 vs. C identifies the joint effect of remote learning and cash; T2 vs. T1a identifies the marginal effect of cash conditional on remote learning.

Within-school randomisation (tutoring): Within T1 schools, households with at least one registered learner are randomly assigned at the household level to:
T1a – Remote Learning only (approx. 1,650 learner-households)
T1b – Remote Learning + Tutoring (approx. 750 learner-households)

Randomisation within T1 is conducted independently within each school, ensuring T1a and T1b households are drawn from the same school environment. The comparison T1b vs. T1a identifies the marginal effect of phone-based tutoring, holding remote learning constant.

Eligibility for all intervention arms is determined by pre-existing school registration data. Households are eligible if they have at least one learner enrolled in a participating school. Eligibility is not contingent on school attendance. Randomisation of schools was conducted prior to the start of the intervention using stratified computer randomisation, stratifying by county (Garissa / Tana River) and school size. Within T1 schools, household-level randomisation for tutoring was conducted by computer following completion of baseline data collection, using the list of households that completed the baseline survey.

Not available

Computer randomisation (stratified by county and school size for the between-school assignment; simple
computer randomisation within school for the tutoring assignment).

School (between-school experiment);
Household (within-school tutoring experiment).

Yes

210 schools (70 Control, 73 T1, 67 T2).

Approximately 6,900 learner-households surveyed at baseline and across up to three follow-up rounds (Follow-up 1: March 2026; Follow-up 2: April/May 2026; Follow-up 3: July 2026).

C (Pure Control): 2,300 learner-households, 70 schools.
T1a (Remote Learning only): approx. 1,650 learner-households, within 73 T1 schools.
T1b (Remote Learning + Tutoring): approx. 750 learner-households, within 73 T1 schools.
T2 (Remote Learning + Cash): 2,200 learner-households, 67 schools.

0.2 SD