Experimental Design
Eight government-run primary schools were selected based on proximity to the city center and availability of public transportation nearby. In seven schools all first, second, and third grade students participated in the program. In one school, first-grade children were excluded due to administrative difficulties in obtaining these students’ addresses. The experiment consisted of a pretest, announcement of the child’s incentive scheme, and a post-test approximately two months later.
Children were initially tested for reading ability during school time to determine baseline learning levels. The test used an instrument developed by Pratham and used in national assessments of child reading ability. Each child was evaluated on a five-point scale: 0) the child could not recognize letters, 1) the child could recognize letters, 2) the child could read simple words, 3) the child could read a simple paragraph, and 4) the child could read and understand a multi-paragraph story. Each child scoring below the highest level on the test was given a goal competency based on his pretest score and was administered one of six randomly-assigned incentive schemes.
The treatments were assigned at the individual level. In order to increase power to detect heterogeneity by pretest score, the randomization was stratified by pretest score within each school, grade, and classroom. Award of the incentive depended on the performance of the children on a post-test, conducted two months after the program announcement. The prize value was set at 100 rupees (about $2.50 at the prevailing exchange rate) for all treatments. At the time of the study, 100 rupees was the approximate daily wage for an unskilled laborer in these areas.
The experiment consisted of six treatment groups. Four treatment groups assigned the household a reward that varied along two dimensions: the direct recipient of the reward (either the parent or child) and the form of the reward (either money or a toy). The two remaining groups offered the parent a choice between money for herself and a toy, either upon program announcement or conditional on reaching the goal. Regardless of treatment category, all children were invited to attend free after-school classes run as part of the program.
The classes were led by teachers trained to assist the children in achieving their literacy goals. The profile and training of the teachers followed the para-teacher model of Pratham, a large India-wide NGO specializing in literacy and numeracy. In each school, enough teachers were provided so that there was at least one teacher for every 20 to 30 students who attended the classes. Classes ran for three hours every afternoon that school was in session. Children were free to attend on a drop-in basis, and teachers were given flexibility to customize lessons based on the reading levels of the children who attended.
A sample of 330 children per treatment group had initially been planned, but the sample was ultimately limited due to budget constraints. Out of 1466 children who took the pretest, 331 were excluded from the study because they achieved the highest possible test score, and 49 others were excluded because they lived too far from the schools, making surveying impractical. 1086 children were thus available for the randomization. Eighty-five percent of children out of the randomized group of 1086 were reached for the baseline survey and program announcement. The attrition between the randomization and program announcement was primarily due to difficulty in locating the children’s homes and in reaching the parents at home. Of the 925 children offered the program, 900 (97 percent) took the post-test after two months. Most of the 25 students who were not available for the post-test had moved away since the program announcement. The final analysis sample contains approximately 150 children in each of the six treatment groups.