Old Delhi is dense and its lanes are narrow. Government and private schools sit within walking distance of each other, sometimes on the same street, so families have choices. We wanted to know what shapes those choices when the child is a girl.
In 2010 we designed what became India’s first randomised controlled trial of school vouchers for girls’ education. The design was simple. Randomly selected families received a voucher covering the fees at a private school of their choice, and we measured whether enrolment changed.
The hypothesis was that cost was the barrier. Private schools in Old Delhi were already trusted, for reasons that included the language of instruction, perceived teacher quality and social aspiration. If that trust was in place, a voucher that removed the fee should increase girls’ enrolment in private schools.
The design
Randomisation was at the household level. Families were identified through door-to-door surveys in particular Old Delhi wards. Treatment families received a voucher and control families did not, and both groups were followed over the school year.
The trial came early. In 2010, randomised evaluations of education policy in India were uncommon. We had to build the supporting systems alongside the study: survey teams trained in RCT protocols, consent processes suited to families with low literacy, and data systems that could follow individual students across school types.
What we learned
Cost turned out to be one barrier among several.
Cost did matter. Families who received the voucher were more likely to enrol their daughters in private schools. When the fee disappeared, behaviour changed.
But cost was tangled up with other factors that a voucher could not address. Some families used the voucher and then withdrew the girl in the middle of the year, because the private school’s timetable clashed with her work at home. Some enrolled the girl and kept her brother in a government school, so the voucher created a decision about how to divide resources within the household that the design had not anticipated.
The most interesting finding concerned information. Some families in the control group, who received no voucher, changed their choice of school during the study anyway. The survey had prompted conversations about schooling that these families had not had before. Being asked which school the daughter attends, and why, worked as an intervention in its own right.
What this means for RCTs
The study taught us to be honest about what an RCT can and cannot tell you.
An RCT can tell you whether a specific intervention, here a voucher, causes a specific outcome, here enrolment in a private school, on average across the treatment group. That is valuable, and observational studies cannot provide evidence of that kind.
The averages hide individual households. One family used the voucher for its daughter and took her out in October. Another enrolled its daughter and saw her grades improve but could not afford the uniform. A third never used the voucher, because the mother-in-law decided the government school was good enough.
These cases are part of how the intervention worked in people’s lives, and they should not be dismissed as statistical noise. A programme designer needs to see them in order to make the voucher work better next time.
The gender lens
The trial was about girls by design, and that choice shaped everything from the sampling frame to the household survey to the analysis plan.
Girls’ schooling in Old Delhi faces constraints that boys’ schooling does not. Families worry about safety on the walk to school. Girls are withdrawn early to prepare for marriage. Household work competes with school hours. Families also decide which child is “worth” the cost of a private school fee.
A voucher addresses the fee. It does nothing about safety, household work or marriage. The trial showed that lower cost is necessary for girls’ enrolment in private school and is not enough by itself. The other constraints need their own interventions, and they need to be studied directly. A programme that assumes a voucher solves the problem ignores them.
This connects to all of our measurement work, because the indicator you choose shapes what you see. An enrolment figure tells you the girl is in school. It does not tell you whether she is staying, whether she is learning, or whether she is free to choose.