On this page
Should a school spend money on extra helpers for children who are behind, or on computers? A pair of experiments in two Indian cities tested both ideas side by side. The answer is more interesting than "both work", because the results faded, and the costs were very different. This post explains the study, its numbers and its limits.
What question did the study ask?
The paper is Remedying Education: Evidence from Two Randomized Experiments in India by Abhijit Banerjee, Shawn Cole, Esther Duflo and Leigh Linden, published in the Quarterly Journal of Economics in 2007. It asked whether low-cost additions to government schools in poor urban areas could raise learning, especially for children at the bottom of the class.
The study was carried out in municipal schools in Mumbai and Vadodara with Pratham, a large Indian NGO. Even though teachers and children were coming to school, learning was very low. According to the working-paper version, only 19.5% of third-grade children in Vadodara could pass grade one maths competencies.
What were the two programmes?
Balsakhi (remedial education). A young woman from the community, called a balsakhi, took children in grades 3 or 4 who had not mastered basic skills out of class for two hours a day (the school day was about four hours) and worked on basic reading and arithmetic. Balsakhis were paid 500 to 750 rupees a month and trained for two weeks.
Computer-assisted learning (CAL). In Vadodara, each child in grade 4 got two hours of shared computer time a week, playing maths games pitched at different levels of difficulty. The games used computers the Gujarat government had already given to schools, many of which were lying unused.
Schools were randomly assigned to get each programme, so programme schools could be compared with similar schools that did not get it.
What did it find?
The published abstract gives the main results:
The remedial programme raised average test scores of all children in treatment schools by 0.28 standard deviations, mostly through large gains for children at the bottom of the class.
The computer programme raised maths scores by 0.47 standard deviations.
One year after the programmes ended, gains remained significant for the children who had been targeted, but they faded to about 0.10 standard deviations.
The working paper adds detail. The remedial programme's effect was 0.14 standard deviations in the first year and 0.28 in the second; the computer programme's maths effect was 0.35 and then 0.47. Children in the bottom third gained the most from the balsakhi programme.
Which was more cost-effective?
The working paper's cost section is revealing. The balsakhi programme cost about 107 rupees per child per year. The computer programme's recurring cost was 367 rupees, rising to 722 rupees once computers and software were included over five years. That is about 6.7 times as much. On the authors' estimates, the balsakhi programme was about five times as cost-effective for maths.
What does the study not show?
Fade-out is real. A large gain at the end of a programme can shrink once it stops. Short bursts of help may need follow-up.
The software was simple. The computer games of the early 2000s were not adaptive tutors. The results do not tell us how modern AI tutors would do in the same schools.
It covers grades 3 and 4 in two cities. Older students and rural schools were not part of this trial.
The comparison is not a head-to-head contest. The two programmes reached different children: the weakest children in grades 3 and 4 for balsakhis, and all grade 4 children for the computer games.
What does this mean for Indian students and parents?
Two practical points stand out. First, help aimed at the exact gap a child has, such as reading or basic arithmetic, can be powerful even when the helper is not a trained teacher. Second, gains need upkeep. A crash course before an exam may lift marks for a while, but without regular practice afterwards, much of it can fade.
The study also started a long line of research. Pratham's later approach, Teaching at the Right Level, grew from this work; see what the Teaching at the Right Level trials found. For a modern adaptive-software trial in Delhi, see the Mindspark study.
How to read a study like this
Look for a follow-up measurement taken after the programme ended. Many studies report only the end-of-programme effect, which can overstate long-term benefit. This one measured a year later and reported the fade honestly.
If you are rebuilding maths basics, the Fermi Problem Library has worked problems with a step-by-step solver that explains the reasoning, which is useful for regular practice after a topic is "done".
Set a small weekly review slot for last term's maths, so this term's progress does not fade.
