One weekly call helped children regain math skills across five countries

When a pandemic, climate disaster or conflict closes schools, online teaching assumes a connection many families do not have. Noam Angrist and colleagues tested a simpler alternative: for eight weeks, a tutor called once a week for about 20 minutes to work on math at the level each child could already follow. Five randomized trials conducted from December 2020 to July 2022 in India, Kenya, Nepal, the Philippines and Uganda indicate that the combination of calls and messages improved learning across different settings.

The researchers randomly assigned families or school groups to control, text messages and, at most sites, messages plus phone tutoring. The final assessment reached 9,148 children: 708 in India, 1,985 in Kenya, 2,678 in Nepal, 2,469 in the Philippines and 1,308 in Uganda. The main scale ran from zero to four according to the most advanced operation a child could solve—none, addition, subtraction, multiplication or division; in Kenya, where students were younger, the ceiling was two-digit subtraction.

Across the five countries, calls plus messages raised performance by 0.321 standard deviations relative to control groups, with p below 0.001. Messages alone produced an average gain of 0.078 standard deviations, with p equal to 0.006, but a statistically detectable effect appeared only in Uganda and the Philippines. The contrast does not show that messages can never help; it shows that, in these trials, individual conversation and instruction adjusted to the child's level were the more consistent parts of the intervention.

The gain also appears in a concrete measure. The share of participants able to solve division increased by 13.5 percentage points over the 14.6% rate in the control group, a 92% relative increase. In Uganda, where schools were closed for nearly two years, the estimated effect reached 1.193 levels on the operations scale; in the Philippines, it reached 0.56 levels. Those two trials occurred later and where educational alternatives were scarcer, so the data cannot isolate which condition explains the larger effects.

Nepal and the Philippines also allowed the team to compare who made the calls. Across the two samples, government teachers increased scores by 0.399 operation levels and instructors from non-governmental organizations by 0.296; both effects had p below 0.001. The difference between delivery models had p equal to 0.110, so the study did not detect sufficient evidence that either was superior. That matters because programs that work in small trials often lose strength when transferred to public systems.

The calculated average cost was about US$11 per child. By combining cost and effect in a metric designed to compare education policies, the authors estimated 3.6 years of high-quality instruction per US$100 spent. This equivalence does not mean eight calls literally added 3.6 school years to each participant; it converts test gains and costs for comparison. The study shows that common infrastructure can sustain learning during disruptions, but the persistence of gains and performance at national scale still need to be measured.