Incentives to Learn

Publication Details

Review of Economics and Statistics, August 2009, v. 91, iss. 3, pp. 437-56. Available From:

Link to Source
Michael Kremer, Edward Miguel and Rebecca Thornton
Sub-Saharan Africa (includes East and West Africa)
Girl's Education
Equity Focus
Evaluation design
Randomised Control Trials (RCT)
Journal Article


This study evaluates the Girls’ Scholarship Program, a merit scholarship programme administered in Kenyan primary schools by the Dutch NGO ICS Africa. Merit scholarships reduce the costs of education for recipients and may improve academic performance by inducing greater effort by eligible students. Scholarships may also increase teacher effort and generate positive externalities for non-eligible students. The programme awarded scholarships to grade 6 girls with end-of-year standardized exam scores in the top 15 percent in rural Busia and Teso districts. The scholarships covered school fees and supplies for the next two academic years. This study examines this programme’s impact on academic performance and teacher attendance.
The basic methodology is a randomised controlled trial (RCT). From a sample of 127 primary schools, 64 were randomly selected to receive treatment. The programme was implemented over two academic years beginning in January 2001, covering two cohorts of grade 6 students. At baseline, 11,728 students in these cohorts were enrolled in schools in the sample. However, after the first year, six schools in Teso withdrew from the programme after a lightning strike killed several students in a Teso school. The study therefore estimates treatment effects separately for students in the entire baseline sample and for the restricted sample of students in schools that did not withdraw. The study also estimates separate treatment effects using longitudinal data for the subsample of first cohort students who took grade 5 end-of-year tests in 2000. In all cases, the authors disaggregate effects by district, and impact is estimated using ordinary least squares (OLS) models including an indicator for school treatment status. As a robustness check, nonparametric bounds on programme effects are also estimated. The authors use data from end-of-year tests, unannounced visits to monitor student and teacher attendance and in-school student questionnaires on study habits administered after the programme’s first year.

Main findings

This study finds that the Girls’ Scholarship Program generated improvements in girls’ academic performance. According to ordinary least squares (OLS) estimates, the scores for girls in treatment schools improved between 0.18 and 0.19 standard deviations. The programme’s effects, however, were concentrated largely in Busia schools. For all samples, test score improvements were greater and statistically significant for girls in Busia schools, ranging from 0.19 to 0.27 standard deviations; in Teso schools, programme effects ranged from –0.01 to 0.09 standard deviations and were insignificant for all samples. The difference in effects by district may be due to greater attrition of scholarship programme schools and students in Teso.
The programme’s overall positive impacts seem to be driven by increased teacher and student attendance. On average, the programme was associated with a 4.8 percentage point increase in teacher attendance. Parent interviews suggest that increased monitoring of teachers by parents may account for this channel of impact. The program was associated with an approximately one-quarter reduction in mean school absenteeism among girls in Busia schools. The programme’s positive effects may also be driven by increased investments in girls’ school supplies by households. The study reports a positive, though insignificant, programme effect on number of textbooks at home and number of new books purchased by households.
The programme also generated positive externalities for boys, who were ineligible for scholarships, and for girls with poor baseline academic performance. The programme was associated with significant test score improvements of between 0.05 and 0.09 standard deviations for boys. Additionally, the programme generated improvements of 0.12 and 0.13 standard deviations for girls who scored in the bottom two quartiles on grade 5 end-of-year tests, respectively. This finding suggests that scholarship eligibility generated positive effects even for students who were ex ante unlikely to win the scholarship competition.

Scroll to Top