3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

Economics of Education Review

Volume 97, December 2023, 102465

Improving public school productivity: Evidence from

model schools in India ☆
G. Naveen Kumar

Show more

Share Cite

https://doi.org/10.1016/j.econedurev.2023.102465
Get rights and content

Abstract

This paper studies the impact of India’s “model” school program which aimed to provide high quality
education to economically disadvantaged students. Model schools combine better infrastructure with more
accountability, contract teachers, and lower per-pupil spending than regular public schools. Using a fuzzy
Regression Discontinuity Design based on entrance exam cutoffs, I find attending a model school for five
years increases test scores in math by 0.38 standard deviations, in science by 0.26 sd, and in social science
by 0.26 sd on average. Furthermore, model schools increase the probability of joining pre-university by 11.5
percentage points. The results suggest it is possible to deliver substantial improvement of outcomes in
public schools at a slightly lower level of school spending through a package of reforms.

Introduction

A growing body of high-quality evidence has documented the poor translation of increased education
spending into improved learning outcomes in developing countries (Mbiti, 2016). Reviews of evidence
suggest that low productivity of resources in traditional public schools reflects weakness in governance and
pedagogy (Duflo et al., 2015, Glewwe and Muralidharan, 2016, Mbiti et al., 2019). This suggests it may be
possible to deliver substantial improvement of outcomes in public schools even without increasing school
spending, if productivity of the school resources is improved. However, there is limited evidence on whether
such efficiency gains are possible.

This paper presents evidence that public schools can substantially improve learning outcomes at the same
or lower cost through a package of reforms. Specifically, I study India’s model schools program targeted at
grades 6 to 10. The program’s objective was to start one high-quality public school in each of the
educationally backward blocks (EBB) to promote equitable access to quality education.1 The model schools
program bundled superior infrastructure, improved accountability, and better value for the money (through

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 1/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

the use of contract teachers with lower salaries). Importantly, the spending per child in model schools was
21%–31% lower than regular public schools.

A key challenge for quality middle school education in developing countries such as India is the large gap in
classroom instruction and student learning levels (Muralidharan et al., 2019). Model schools reduced
pedagogical mismatch between student preparation and curriculum standards by selecting students
through an entrance exam. I use fuzzy Regression Discontinuity Design (RDD) at the student level based on
entrance exam cutoffs to study the impact of model schools in Karnataka, a southern state in India. Fuzzy
RDD is best suited for the setting as not all students that receive admission end up attending model schools.

The entrance exam is conducted at the block level; hence, students residing in a particular block compete
for the model school in that block. Karnataka has a total of 74 model schools (one school per block) and the
first cohort of model schools was admitted in 2009. Additionally, students apply to attend a model school
under eight caste categories, and admission is based on their within-category performance in a given block.

A model school admits students up to the school’s capacity of 80 students. Using the admission lists
prepared by the examination authority, each model school will admit students in descending order, based
on their entrance exam score and caste category. The nature of the selection process creates a cutoff for each
category within each model school, meaning that each model school can have up to eight school-by-
category cutoffs.2 This cutoff score for admission into a model school is not known to the school or to the
potential students beforehand. Thus, whether students near the cutoff fall to the left or the right of the
cutoff is as good as randomly assigned.

I assemble three restricted student-level administrative data sets to track the students who appear for the
model school entrance exam in fifth grade at two future points: 10th grade and pre-university (12th grade).
With a data set of over 63,000 students that applied to 74 model schools across three cohorts, I am able to
investigate two dimensions of schooling outcomes: (i) academic achievement as measured by test scores
and final grades; and (ii) educational attainment indicators using years of schooling and choice of stream.

My econometric strategy combines all 1,513 cutoffs under one framework to identify the local average
treatment effect of model schools. Adopting the Fuzzy RDD, I compare the outcomes of students who scored
barely above and barely below the admission cutoff score within their block and caste category. Using the
indicator for whether the entrance exam score is above the relevant school-by-category cutoff as an
instrument for the model school attendance indicator, I find that attending a model school raises academic
achievement and educational attainment significantly.

For academic achievement, attending a model school increases math test scores by 0.38 standard deviations
(sd), science test scores by 0.26 sd, and social science test scores by 0.26 sd on average ( ).3 To put
these estimates into perspective, the median effect size of education interventions on learning outcomes in
developing countries is 0.10 sds and the 75th percentile effect size is 0.23 sd (Evans & Yuan, 2022).
Attending a model school also increases the probability of obtaining an A or A+ grade in 10th grade by 20
percentage points ( ). For educational attainment indicators, attending a model school increases the
probability of passing 10th grade by a statistically insignificant 5.3 percentage points and increases the
probability of joining pre-university college by 11.9 percentage points ( ). However, model schools
have no statistically significant effect on the probability of choosing either science, arts, or commerce as a
stream in pre-university education.

The primary identification concern is the possibility of differential follow-up rate for students above and
below the cutoff.4 I find students with an entrance exam score just above their respective school-by-
category cutoff are about 4.5 percentage points more likely to be found in the 10th-grade exam data set

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 2/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

than those who scored just below the cutoff. As a robustness check, I re-sample the 74 model schools in
several ways such that the differential follow up rate is not statistically significant and the sample is
different each time.5 The estimates are very similar to the main estimates even after altering the differential
follow-up rate threshold and altering the percent of model schools sampled.6 Further, I adopt two intuitive
trimming procedures to construct lower and upper bound estimates: worst-case scenario bounds (Lee,
2009) and bounds using covariates (a variation of Lee, 2009). In both trimming techniques, the lower-bound
estimate of the math, science, and social science test scores is positive and not statistically significant.

I consider the direction of the bias to help interpret the main estimates in light of the lower bound
insignificant positive effects. With entrance exam scores as the dependent variable, I determine the
discontinuity within the 10-point bandwidth separately for the sample I can and cannot track. The
magnitude of the entrance exam scores discontinuity in the sample that I am not able to track is higher than
the discontinuity in the sample that I am able to track.7 The bigger discontinuity among the non-trackable
sample is either coming from those just below the cutoff having lower entrance exam scores or from those
just above the cutoff having higher entrance exam scores. If entrance exam scores are correlated with 10th
grade test scores, then having 10th grade test scores for the additional 4.5 percent of the non-trackable
students below the cutoff would likely increase the magnitude of the main estimates. Hence, the direction
of the bias is likely downward.

On average, model school attendance improves educational outcomes; but an important issue is whether
the effects vary by caste, gender, or other dimensions especially given the explicit concerns about the
inequality in access to quality schooling. I find no evidence of heterogeneity in test score gains by caste, or
gender suggesting broad-based learning gains from the model schools. I also examine heterogeneity based
on initial student learning levels as measured by students’ entrance exam scores, and cannot reject that
there are similar positive effects at all parts of the learning distribution. These results of broad-based gains
are consistent with those of Bagde et al. (2016) who study the impact of affirmative action in the Indian
higher education setting.

This paper makes two main contributions. The first is the literature on improving school productivity in
developing countries. In recent years, several randomized evaluations have studied the impacts of
improving individual components of school productivity including hiring contract teachers (Duflo et al.,
2015, Muralidharan and Sundararaman, 2013), improving teacher incentives (Duflo et al., 2011, Mbiti et al.,
2019, Muralidharan and Sundararaman, 2011) and reducing the mismatch between student preparation and
curricular standards (Banerjee et al., 2007, Muralidharan et al., 2019). However, there is much less evidence
on whether it is possible to improve school productivity through a package of these reforms. My results
suggest that such a package can be effective within public school systems and at a lower per-child cost than
the status quo.

Second, this paper contributes to the literature on improving equity in access to quality education. Prior
work has examined whether giving vouchers to attend private schools can be a solution to providing quality
education for the economically deprived (Angrist et al., 2002, Hsieh and Urquiola, 2006, Muralidharan and
Sundararaman, 2015, Romero and Singh, 2022). My results show that a combination of selective entrance
(to reduce the mismatch between student preparation and curricular standards), and improved teacher
accountability, can deliver positive impacts on learning for low-income students. From a policy perspective,
this may provide a scalable template for other countries to provide quality education to academically high-
achieving but economically-disadvantaged students.

Section snippets

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 3/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

Policy experiment and student selection process

In this section, I describe a policy that created a high-quality public school in each of the Educationally
Backward Blocks (EBB) in India, thus giving low-income high-achieving students an opportunity to attend a
high-quality public school. I further explain the key features of the selection process for admitting students
from all castes. In particular, students are selected based on their performance on an entrance exam within
their caste and block.

Model schools program. With the intention of …

Data

In this section, I describe the three sources of administrative data that allow me to track those who
appeared for the model schools entrance exams at two future points: the end of high school (10th grade)
and the end of senior secondary school (pre-university). In particular, I exploit rich restricted data which
include students’ names, parents’ names, and date of birth to match across data sets and overcome the
challenge of the non-existence of a unique identifier in the Indian education…

Empirical strategy

In this section, I first discuss the approach taken to construct the cutoffs. I then discuss the strategy to
determine the effects of model schools by combining all the cutoffs. In particular, I combine all cutoffs
under one framework by assuming homogeneity in effects across all cutoffs to determine the treatment
effects.…

The first stage & threats to identification

The identification strategy discussed above relies on the validity of the instrument and the identification
assumption. First, the empirical strategy relies on entrance exam scores’ ability to predict model school
attendance. I find a statistically significant jump in the probability of attending a model school at the cutoff,
validating the instrument. Second, the key identifying assumption is that individuals on either side of the
cutoff are similar. The internal validity of the estimates…

Results

In this section, I present the estimates of the effect of attending model schools on various short and longer-
term schooling outcomes. In general, schools can affect several outcomes ranging from learning to social
behavior. With the data available, I am able to investigate two dimensions: academic achievement as
measured by test scores and final grades, educational attainment as measured by years of schooling, and
choice of stream in pre-university.…

Heterogeneity analysis

The main findings discussed in the above section are important when determining whether model schools
work or not. In this section, I explore heterogeneity in effects given the explicit concerns about the
inequality in access to quality schooling.

I begin by exploring whether the model schools benefit those who may not necessarily be at the bottom of
their class or those who join the model schools with high-prior learning levels. The main idea is to estimate
https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 4/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

multiple local average treatment…

Discussion

In this section, I explore the potential change mechanisms by pointing out the differences between model
schools and other types of schools. Using administrative data on school characteristics, interviews, personal
visits to schools, and anecdotal evidence, I attribute the effect of model schools primarily to teacher contract
structure, school accountability and governance, and student effort/motivation, but peer effects also appear
to be a contributing factor.…

Conclusion

In this paper, I exploit an education policy in India to determine whether it is possible to improve the
productivity of public schools without increasing school spending. The model schools program was
implemented to create public schools that consist of a package of reforms in each of the educationally
backward blocks in India. Using three restricted administrative data sets, I examine the effect of attending a
model school in Karnataka (a southern state in India) on two dimensions: academic…

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that
could have appeared to influence the work reported in this paper.…

Special issue articles Recommended articles

References (39)

ChakrabortyT. et al.
English language premium: Evidence from a policy experiment in India
Economics of Education Review (2016)

DufloE. et al.
School governance, teacher incentives, and pupil–teacher ratios: Experimental evidence from
Kenyan primary schools
Journal of Public Economics (2015)

GlewweP. et al.
Improving education outcomes in developing countries: Evidence, knowledge gaps, and policy
implications

HsiehC.-T. et al.
The effects of generalized school choice on achievement and stratification: Evidence from
Chile’s voucher program
Journal of public Economics (2006)

McCraryJ.
Manipulation of the running variable in the regression discontinuity design: A density test

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 5/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

Journal of Econometrics (2008)

ParkA. et al.
Magnet high schools and academic performance in China: A regression discontinuity design
Journal of Comparative Economics (2015)

AngristJ. et al.
Vouchers for private schooling in Colombia: Evidence from a randomized natural experiment
American Economic Review (2002)

AzamM. et al.
The returns to English-language skills in India
Economic Development and Cultural Change (2013)

BagdeS. et al.
Does affirmative action work? Caste, gender, college quality, and academic success in India
American Economic Review (2016)

BanerjeeA.V. et al.
Remedying education: Evidence from two randomized experiments in India
Quarterly Journal of Economics (2007)

View more references

Cited by (0)

☆ I thank the editor Priya Mukherjee and the first referee for thoughtful comments and suggestions. I am grateful to my
supervisors, Steven Rivkin, Ben Feignberg, Karthik Muralidharan, Ben Ost, Javaeria Qureshi, Abhijeet Singh for their support
and guidance. I also thank Katherine McElroy, Prashant Bharadwaj, Gordon Dahl, Gaurav Khanna, Erik Hembre, Jaselyn
Taubel, Shogher Ohannessian, Sabareesh Ramachandran, Frances Lu, Radhika Goyal, Kaspar Wuthrich, and participants at
various seminars for helpful comments. I thank the Department of Public Instruction, Secondary Education Exam board, and
the Department of Pre-University Education of Karnataka for providing me with the data. I am grateful to the Indian
Administrative Service officials, Dr. P C Jaffer, Dr. Rajkumar Khatri, Umashankar S R, Akash S for their invaluable support. The
findings, interpretations, and conclusions expressed in this paper are my own and do not necessarily represent the views of the
Government of Karnataka, India. All errors are my own.

View full text

© 2023 Elsevier Ltd. All rights reserved.

All content on this site: Copyright © 2024 Elsevier B.V., its licensors, and contributors. All rights are reserved, including those for text and data mining, AI training, and similar
technologies. For all open access content, the Creative Commons licensing terms apply.

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 6/7
3/29/24, 12:06 AM Improving public school productivity: Evidence from model schools in India - ScienceDirect

https://www.sciencedirect.com/science/article/abs/pii/S0272775723001127 7/7