Presentation on theme: "M SREENIVASA RAO AZIM PREMJI FOUNDATION 29 April, 2011 Achieving universal quality primary education in India Lessons from the Andhra Pradesh Randomized."— Presentation transcript:
M SREENIVASA RAO AZIM PREMJI FOUNDATION 29 April, 2011 Achieving universal quality primary education in India Lessons from the Andhra Pradesh Randomized Evaluation Studies (AP RESt)
2 Basic concepts, if not mastered early, may never be mastered Notes: Based on APRESt data for Control schools only, Oct = ? Less than half the students who don’t know single digit addition in 2 nd grade, learn it by the end of 5 th grade!
3 Higher spending in government schools may not be enough Source: “Teacher Absence in India”, Journal of the European Economic Association, 15-Sep-04 Motivation and effort-levels of government school teachers in India are a serious problem High levels of teacher absence (25%) ranging from 15% to 42% across states 90% of non-capital spending goes to teacher salaries Teacher that are paid more – older teachers, more educated teachers and head teachers – are more frequently absent Higher absence rates in poorer states (additional spending has highest leakage where it is needed the most)
4 Broad objectives of AP RESt (Andhra Pradesh Randomized Evaluation Studies) Move the focus of education policy from outlays to outcomes Focus systematically on institutional incentives for service delivery Improve evidence base for policy with rigorous (randomized) evaluations
5 How do you evaluate the impact of large social sector programs? Let’s use mid-day meals (a popular program in India) as our example: What has been the impact of the mid-day meal program? 3: Compare to appropriate control 1: Define outcomes 2: Measure outcomes The control and treatment groups are similar in all other ways except for the program The difference in the outcome measure between the two is a measure of the impact of the mid-day meals program Often, even this first step is not undertaken Let’s assume it is, and we define some outcomes, e.g. nutrition, attendance and learning Is this a valid measure of the impact of the program? No, because there are many other things that have changed at the same time Outcome Outcome TreatmentControl We use a randomised evaluation methodology: the “gold standard” in social science research
6 APRESt is a multi-stakeholder partnership Government of Andhra Pradesh (GoAP) -Main client – project initiated at request of Principal Secretary, Education -All relevant letters of permission and administrative support -Financial contribution (cost of contract teachers; direct contribution) Azim Premji Foundation -Main counterpart to MoU with GoAP -Fully responsible for all aspects of project implementation, school communications, test administration, and data collection Over 50 full time project staff and 750 part-time evaluators Continuous engagement with government Financial contribution as well World Bank -Technical support -Financial support (mainly through DFID) -Institutional continuity with government (6 secretaries in 6 years!)
7 We tested five specific interventions Contract teachers Block grants Performance pay Feedback + Monitoring Schools provided with additional teacher (on contract) Schools provided cash grants for student inputs Existing teachers provided with detailed feedback on students and subject to low- stakes monitoring Teachers eligible for bonuses based on improved student performance (either in own class or whole school) MOTIVATIONINTERVENTION One reason learning levels may be low is teachers don’t know how to help students Can better information help? Use of contract teachers is widespread, but highly controversial Are contract teachers effective? Significant amounts of money committed under RTE. What is the effectiveness of such spending? Teacher salaries are the largest component of education spending in India, but a poor predictor of outcomes Can linking pay to performance improve outcomes?
Sample Balance Panel A (Means of Baseline Variables)  School-level Variables Control Extra Para- teacher Block Grant Group Incentive Individual Incentive P-value (Equality of all groups) Total Enrollment (Baseline: Grades 1-5) Total Test-takers (Baseline: Grades 2-5) Number of Teachers Pupil-Teacher Ratio Infrastructure Index (0-6) Proximity to Facilities Index (8-24) Baseline Test Performance Math (Raw %) Telugu (Raw %)
9 Within two years we had tested 600 schools with five different interventions Input onlyIncentive only Feedback + Monitoring 100 schools Individual Incentive + Diagnostic Feedback 100 schools Group Incentive + Diagnostic Feedback 100 schools Extra Contract Teacher + Diagnostic Feedback 100 schools Block Grant + Diagnostic Feedback 100 schools Business as usual 100 schools
10 Timeline of experimental design and execution Jun-Jul ’05Conducted baseline tests in these schools Late Jul ’05Schools randomly assigned to various treatments Early Aug ’05 Provided diagnostic feedback on test performance to all schools Sep ‘05 – Feb ’06 Monitored process variables over the course of the year via unannounced monthly tracking surveys Mar-Apr ’06 Conducted endline tests to assess the impact of various interventions on learning outcomestests Jul ’06 Interviewed teachers in incentive schools, but before outcomes are communicated to them Next school yearRepeated process above
12 However, there was no difference in test scores between students in treatment and comparison schools Outcomes for treatment schools relative to comparison schools The lack of impact on test scores, despite enhanced teaching activity, suggests that teachers temporarily changed behavior when observed, but did not actively use the feedback reports in their teaching.
13 Impact of the program is lower after 2 years than after 1 year Household spending fell significantly when the grant was anticipated Student learning improved in the first year, but not the second Anticipated
14 We find that students perform better in schools given an extra CT CTs have lower rates of absence and higher rates of teaching activity Students in extra CT schools significantly outperform students in comparison schools
15 Potential concerns with such a program are addressed pro-actively in the study design Potential concernHow addressed Teaching to the test Test design is such that you cannot do well without deeper knowledge / understanding Less of a concern given extremely low levels of learning Research shows that the process of taking a test can enhance learning Threshold effects/ Neglecting weak kids Minimized by making bonus a function of average improvement of all students, so teachers are not incentivized to focus only on students near some target; Drop outs assigned low scores Cheating / paper leaks Testing done by independent teams from Azim Premji Foundation, with no connection to the school Reduction of intrinsic motivation Recognize that framing matters Program framed in terms of recognition and reward for outstanding teaching as opposed to accountability
16 Bonus schools perform better across the board Outcomes for bonus schools relative to control schools Students in bonus schools do better for all major subgroups, including: all five grades (1-5); both subjects; all five project districts; and levels of question difficulty No significant difference by most student demographic variables, including household literacy, caste, gender, and baseline score Lack of differential treatment effects is an indicator of broad-based gains Overall, no child in a bonus school was worse off relative to a comparable child in a control school
17 Individual incentives versus group incentives The theory on group- versus individual-level incentives is ambiguous −On the one hand, group incentives may induce less effort due to free-riding −On the other, if there are gains to cooperation, then it is possible that group incentives might yield better results Both group and individual incentive programs had significantly positive impacts on test scores in both years In the first year, they were equally effective, but in the second year, the individual incentives do significantly better Both were equally cost-effective In theory… Our findings…
18 Teacher opinion on performance pay is overwhelmingly positive It is easy to support a program when it only offers rewards and no penalties However, teachers also support performance pay under an overall wage-neutral expectation Strong teacher support for performance pay Significant positive correlation between teacher performance and the extent of performance pay desired beforehand −Suggests that effective teachers know who they are and there are likely to be sorting benefits from performance pay
19 The 5-year effects of individual incentives are even stronger PRELIMINARY In the long-run individual bonuses seem to work best It’s not that other programs are ineffective – just not as effective as individual bonuses Notes: Results reported on the cohort of students that entered class 1 in the first year of the study.
20 … And were significantly more cost effective Avg cost for 2 years (INR) Impact (SD) Cost per 0.1 SD impact (INR) Contract teacher20, ,184 Block grant20, ,816 Group bonus12, ,453 Individual bonus20, ,380 Overall, the incentive programs are 3× as cost effective as the input programs Performance pay was twice as cost effective as an extra contract teacher, and a contract teacher is five times more cost effective than a regular teacher Suggests that expanding a performance pay program would be 10 times more cost effective than hiring additional regular teachers
Towards a Just, Equitable, Humane and Sustainable Society Q&A Session Thank You 21