Does the Medium of Education Matter
Does Medium of Instruction matter? A propensity score analysis
of the effect of English Medium Education on students learning
outcomes in India
by
Sandeep Shankar Murugesan
MPP Essay
Submitted to Oregon State University
in partial fulfillment of the requirements for the degree of Master of Public
Policy
Presented on March 27th, 2015
1
Master of Public Policy essay of Sandeep Shankar Murugesan presented March 27th, 2015
Approved:
______________________________________________________________
Dr. Victor Tremblay, Economics
______________________________________________________________
Dr. Alison Johston, Political Science
______________________________________________________________
Dr.Todd Pugatch, Economics
______________________________________________________________
Sandeep Shankar Murugesan, Author
2
Abstract:
Over the past two decades there has been sharp increase in the demand for Englishlanguage skills in India. This led to a rapid expansion of English Medium Schools, more so in
the private sector than the public sector. Recent studies have estimated the economic benefits of
English- language skills but few studies have compared the educational performance of students
from the English Medium Schools (EMS) versus Regional Medium Schools (RMS). This study
seeks to address this gap in the literature. Specifically, this paper uses propensity score matching
methods to examine whether students between 8 and 11 years old have higher reading,
mathematics and writing scores when they attend EMS instead of RMS. The study makes use of
the data from the Indian Human Development Survey (IHDS) that was collected by the National
Council of Applied Economic Research (NCAER) in collaboration with University of Maryland
in 2005. The estimates from the propensity score matching analyses suggest that there is no
significant difference in learning outcomes between the two groups. Based on the findings, I
further discuss the policy implications, limitations of the study and the potential areas for future
research.
3
Acknowledgements:
I am immensely thankful to Professor Victor Tremblay for his constant encouragement and
guidance throughout this project. I have gained a lot both personally and professionally from
working under him.
My understanding of the use of statistics in social science research has evolved a lot over these
years. I owe it to the lectures of Professor Alison Johnston and Professor Todd Pugatch that I
have attended. I consider them as my Guru’s and I am greatly indebted to them, both for their
sharing their wisdom and offering me valuable feedback on this project.
I would like to thank Professor Brent Steel and Professor Denise Lach, who decided to give me
an opportunity to pursue my graduate education. I would be failing in my duty, if I do not
mention their names.
I would like to thank my friends from the MPP cohort for making me feel at home here in
Corvallis. I have learned a lot from interacting with them and their friendship has been a great
source of joy and comfort.
Last but not least, I would like to thank all my family members and friends back in India. Their
emotional and financial support has helped me a great deal throughout my stay here in the United
States.
4
Table of Contents
1.
Introduction ............................................................................................................................. 8
2.
Background ............................................................................................................................ 10
3.
2.1.
Colonial History and Post Independence Language Policy ........................................... 10
2.2.
Post-Liberalization shift in priorities.............................................................................. 12
2.3.
Private Vs Public Schools Debate .................................................................................. 13
Literature Review .................................................................................................................. 14
3.1.
The link between language of instruction and learning outcomes ................................. 15
3.2.
Gender Discrimination in Educational Choice............................................................... 16
3.3.
Economic Class and differential access to Primary Education ...................................... 17
3.4.
Government Programs and Primary School Enrollment ................................................ 18
3.5.
The influence of Caste on Education ............................................................................. 18
3.6.
Schooling and the Urban Rural Divide in access to education ...................................... 19
3.7.
Household Composition and Choice of Education ........................................................ 20
3.8.
Parental Education and Choice of Education ................................................................. 20
4.
Empirical Framework ............................................................................................................ 21
5.
Data ........................................................................................................................................ 25
6.
Data Analysis ......................................................................................................................... 31
7.
6.1.
Model Validation............................................................................................................ 32
6.2.
Matching procedures ...................................................................................................... 33
Discussion .............................................................................................................................. 34
7.1.
Characteristics of English Medium School Students ..................................................... 34
7.2.
Propensity Score Model Results Discussion .................................................................. 38
8.
Limitations ............................................................................................................................. 40
9.
Policy Recommendations ...................................................................................................... 42
References .............................................................................................................................. 44
5
List of Tables
1.
2.
3.
4.
5.
6.
Descriptive Statistics for Regional Language Schools Vs English Medium Schools ...... 27
Basic Regression Model ................................................................................................... 31
Characteristics of English Medium School Students Logit Model ................................... 35
Net Impact of EMS on Students computed using different Matching Techniques .......... 38
Descriptive Statistics of differences in student inputs between EMS and RMS .............. 39
Net Impact of EMS on student’s time allocation for Home Work ................................... 40
List of Figures-.
Student Reading Skills ...................................................................................................... 50
Student Mathematics Skills............................................................................................... 50
Student Writing Skills ....................................................................................................... 51
Student Enrollment Status................................................................................................. 51
Comparison of Percentage of Students Enrolled in English Medium Schools between
Public and Private Schools................................................................................................ 52
Student Enrollment in Different Schools .......................................................................... 52
Comparison of Student Enrollment in Private and Public Schools .................................. 53
Balancing Property Test for Model 1................................................................................ 53
Balancing Property Test for Model II ............................................................................... 53
Overlap Test for Model 1 .................................................................................................. 54
Overlap Test for Model II ................................................................................................. 55
Distribution of propensity score of Matched Samples ...................................................... 55
6
List of Abbreviations
IHDS
Indian Human Development Survey
MI
Medium of Instruction
EMS
English Medium School
RMS
Regional Medium School
PSM
Propensity Score Matching
NN
Nearest Neighbor
EGS
Education Guarantee Scheme
VEC
Village Education Committee
SC
Scheduled Caste
ST
Scheduled Tribe
OBCs
Other Backward Classes
ASER
Annual Survey of Education Report
CBSE
Central Board of Secondary Education
NGO
Non-Governmental Organization
NCAER
National Council of Applied Economic Research
NKC
National Knowledge Commission
UPA
United Progressive Alliance
TLF
Three Language Formula
7
1. Introduction
For more than two decades, scholars have debated the impacts of globalization on the
economy, culture, politics and society. One of the significant outcomes of globalization is the
emergence of English as the lingua franca of the entire world. In recent years, scholars have
concentrated on the impact of globalization on the language policy in the developing world
(Chang, 2006; Her, 2007; Hornberger & Vaish, 2009; Kirkgöz, 2009). Undoubtedly, the
emergence of English as the ‘global language’ poses both social and economic challenges in
many parts of the world and policy makers around the world are beginning to grapple with this
change.
In a global market there are significant economic benefits of learning English. In fact, many
studies indicate that there is a high “wage premium” for English skills in the labor market
(Azam, Chin, & Prakash, 2013; Kapur & Chakraborty, 2008; Munshi & Rosenzweig, 2003). The
growing demand for learning English medium schools1 is partly driven by the expectation of
earning a “wage premium” in the labor market.2 From a public policy perspective, improving
the access to English education is important for two main reasons. First, English skills enhances
the capability of individuals to participate in the global economy. Second, it can address the
growing wage inequality which arises due to differential access to quality English training.
However, the question still remains whether the government should expand public English
medium schools. The argument in favor of expanding English medium schools is that it can
improve the access to learning English. On the other hand, there are genuine concerns over
1
By definition an “English Medium School” is one in which ‘English’ is the primary medium of pedagogy. The
same logic applies to the Regional Medium Schools as well.
2
Data collected by the National University for Education Planning and Administration indicates that those opting
for English Medium Education have increased by 150% between the years 2003 to 2008.
8
shifting language policy in favor of promoting English medium schools. For instance, education
psychology literature shows that for children may be better-off if they are taught in their native
language (Abadzi, 2006). Kosonen (2005) argues that when children are offered opportunities to
learn in their native language, they are more likely to attend and succeed in school. Also, some
studies show that when children learn in their native language the parents are more likely to
participate in their children‘s learning (Benson, 2002).Therefore if English medium education
has shown to have adverse effects on learning outcomes, then students from disadvantaged
backgrounds would be put at a “double disadvantage”. Unfortunately, few studies have estimated
the impact of English Medium Education on the student learning outcomes in India. This paper
seeks to address this gap in the literature.
In this paper I use the nationally representative student-level dataset (Desai, Vanneman, &
National Council of Applied Economic Research, 2013), to estimate the impact of English
Medium Schools on student learning outcomes. Since the expansion of EMS has been
predominantly in the private sector, only students between the age 8 and 11 years, who are
currently enrolled in private schools,3 were included in the study sample. The challenge in
estimating the impact of Medium of Instruction on learning outcomes is to overcome the
endogenous factors such as student motivation and teacher skills which can affect learning
outcomes. Propensity score matching analysis allows us to partly overcome this problem,
although it is no “magic bullet” to solve all problems concerning endogeneity. In short, the
propensity score matching method ex post tries to recreate the conditions of a randomized control
trial. This estimates obtained by using this method are shown to be an improvement over those
that are derived from traditional regression techniques (Dehejia & Wahba, 2002).
3
“Private Schools” include Private recognized and unrecognized Schools, Government Aided Schools, Convents,
Madrassas and junior college.
9
The results from the propensity score analysis suggest that there is no significant difference
in the learning outcomes between the students who attend EMS and RMS. This shows that the
medium of instruction does not appear to influence learning outcomes. The results of this study
can have important implications on the debate on language policy. In the later sections I discuss
the limitations of the study and then, based on the findings of this paper, I discuss why a more
pragmatic policy is required in dealing with the question of language and point out the possible
areas for future research.
2. Background
2.1. Colonial History and Post Independence Language Policy
India’s tryst with English dates back to the early nineteenth century, when the British
formally introduced English education. Viswanathan (2014) argues that the British education
policy was “set out to create a middle class serving as an agency of imperialist economy and
administration and, through it, to initiate social change through a process of differentiation”.
During the British Raj, English evolved as the language of the Indian elite and it came to be
associated with power and privilege. Therefore, historically speaking, English has played a huge
role in shaping India’s political, social and economic landscape.
The “British Raj” ended in 1947. Post Independence, the Constitution Assembly debated the
role of English in free India. There was a general feeling that an Indic language should be made
the official language. There were calls for Hindi, which was spoken by the majority, to be made
the national language4 but it was not accepted by the non-Hindi speaking members of the
4
Many members of the Constituent Assembly, especially those from the Southern States opposed the imposition of
Hindi as the national language. T.A.Ramalingam Chettiar, a member from the State of Madras, told the Assembly
“you cannot use the word national language, because Hindi is no more national to us than English or any other
language. We have got our own national languages.”(Volume IX, Constituent Assembly Debates)
10
Constituent Assembly. The members reached a compromise and it was decided that both English
and Hindi shall be the official languages of the Central Government, each state shall have its
own official language and the communication between the Centre and States shall be in English
and Hindi. The implication of this decision was that by default all bureaucrats were expected to
have a working knowledge of English Moreover, in practice, English continues to be the link
language between the Hindi speaking and non-Hindi speaking population in India.
As far as the language policy with respect to schools was concerned, the government adopted
a “three language formula”. And these three languages were to be taught as school subjects,
regardless of the Medium of Instruction, namely: 1) mother tongue or regional language; 2)
Hindi or English; 3) one modern Indian language or foreign language not covered under 1 and 2.
By definition, an English medium school is one in which the core curriculum is taught in
English. In addition to that, the students are taught a secondary language which is either the
regional language or Hindi. Since education was included as a State subject, the “three language
formula” (TLF) was not uniformly implemented across all the states. For instance, only two
languages are being taught in the public schools in the State of Tamil Nadu —namely Tamil and
English. According to the Seventh All India Education Survey (2002), 90.61% and 84.86%
schools followed the TLF in the upper primary and secondary stage respectively. The survey also
reveals that 55.05% and 54.1% schools taught English as the second language in the upper
primary and secondary stage respectively. 5
5
In most of the states Classes I-IV/I-V constitute ‘primary stage’; Classes V-VII/VI-VII/VI-VIII constitute ‘upper
primary stage’; Classes VIII-X/IX-X constitute ‘secondary stage’;and Classes XI-XII as ‘higher secondary stage’.
Source: 7th All India Education Survey
11
2.2. Post-Liberalization shift in priorities
India dismantled the “License Raj”6 and shifted to a more open economy in the early 1990’s.
Over the past two decades, there has been a tremendous growth in India’s Information
technology sector. It has grown from contributing around 2% of the GDP in the late nineties to
4.8% at the end of 2006 and now stands nearly around 7.5%. This growth was fueled by an
English-speaking educated middle class, which is currently estimated to be around 300 million.
Since the service sector has been the main driver of India’s economic growth and generates
relatively high paying jobs, the demand for English education has increased tremendously. For
instance, a recent study by Azam et al (2013) reveals that the hourly wages for men who speak
fluent English 34% higher than those who did not know English. This has forced the
Government to rethink its language policy. In 2006, the United Progressive Alliance (UPA)
appointed the National Knowledge Commission (NKC) to study the problem and make a set of
recommendations. The NKC report recommended making English as a compulsory language and
advised the government to expand EMS in the public sector (India, 2007). According to 7th All
India Education Survey (2002), 12.98%, 18.25%, 25.84% and 33.59% schools used English as
the medium of instruction at the primary, upper primary, secondary and higher secondary stage
respectively. The corresponding figures from 6th Survey (1993) were 4.99%, 15.91%, 18.37%
and 28.09% respectively.
The post-liberalization phase of India’s economic policy was also marked by the failure of
the government to provide universal quality public education. This fueled a rapid expansion in
private provision of primary schools throughout India. The state of public schools was so
6
Post Independence, India adopted Socialistic Economic Policies. The State curbed the growth of private sector by
instituting severe licensing requirements for producing goods and services within the country.
12
abysmal that the PROBE Team (1999) study reported that "even among poor families and
disadvantaged communities, one finds parents who make great sacrifices to send some or all of
their children to private schools, so disillusioned are they with government schools”. This lead to
a situation when parents started to send their children to fee charging private schools rather than
to the free of cost public schools. In 1993 approximately 10% of children aged 6–14 were
enrolled in a private school. By 2008 an around 22.5% of young school children were enrolled in
a private school (Cheney, Ruzzi, & Muralidharan, 2005).
2.3. Private Vs Public Schools Debate
In the recent past, many studies have compared the performance of private and public
schools in India (Muralidharan, 2006; Muralidharan & Kremer, 2006; Muralidharan &
Sundararaman, 2013b) These studies reveal that despite the fact private schools in India use
fewer resources, students who attend private schools outperform their counterparts in public
schools on almost all parameters. A recent study revealed that the rate of teacher absenteeism in
public schools is higher than that of private schools (Kremer, Chaudhury, Rogers, Muralidharan,
& Hammer, 2005). Therefore “teacher absenteeism” and poor instructional quality could be
among the most important factors that explain the performance gap between private and public
schools in India. A randomized control trial study to evaluate the effect of teacher performance
pay incentive seems to indicate that those students who are assigned to the incentive classroom
performed much better than those students in the control group (Muralidharan & Sundararaman,
2009) On the other hand, private school teachers are usually less qualified and paid less than one
third the salaries compared to teachers in public schools. The same study reveals that 28% of the
population in rural India has access to fee-charging private primary schools in the same village,
and that 16.4% of children aged 6 – 14 in rural India attend fee-charging private schools.
13
Moreover, private schools are more likely to have more contract teachers7 hired within local
areas than public schools (Muralidharan & Sundararaman, 2013a). A more recent study has also
made the case that there is little evidence of “cream skimming”8 in private schools in India
(Tabarrok, 2013).
This study excludes the students who attend public schools for three main reasons 1) There is
a huge performance gap between private and public schools in India due to the reasons which we
have discussed in the previous section. 2) The expansion of EMS has been mainly in the private
sector. With our overall sample we can observe that among those students who have enrolled in
public schools only 2.66% students attend public EMS (see Figure 5) 3) The PSM method is a
data intensive technique and it can only match on observable differences. Including only private
school students helps to match the unobservable variables that might lead parents to choose
private instead of public schools. There are fewer differences between private English Medium
Schools and private Regional Medium Schools. Therefore such a comparison can produce more
accurate results in the propensity score analysis.
3. Literature Review
In this section, I will first discuss the theoretical and empirical link between the student
learning outcomes and the medium of instruction. After that I will discuss the variables which
affect parental school choice. These variables include gender of the student, economic class of
the household, caste, government incentives, the neighborhood of the household (urban or rural),
parental education and household composition.
7
“Contract teachers” broadly refers to those teachers recruited by the community (though not always) on a
contractual basis and on a fixed honorarium to overcome the problem of teacher shortage and teacher absenteeism in
rural and remote areas
8
“Cream Skimming” refers to pattern of school enrollment wherein the best students from the public schools are
absorbed into private schools.
14
3.1. The link between language of instruction and learning outcomes
The study examining the link between the language of instruction and learning outcomes has
several antecedents. The education psychology literature shows that for children may be betteroff if they are taught in their native language (Abadzi, 2006). The evidence from the literature
suggests that mother tongue instruction in the early years of childhood is can avoid cognitive
disadvantages (James Cummins, 1978; Jim Cummins, 1979). This significance is more for
children who come from disadvantaged backgrounds. Kosonen (2005) argues that when children
are offered opportunities to learn in their native language, they are more likely to attend and
succeed in school. Also, some studies show that when children learn in their native language the
parents are more likely to participate in their children‘s learning (Benson, 2002).
On the other hand, there is also evidence from the psychobiological literature which shows
that younger children learn languages more easily than adolescents and adults. This is referred to
as the “critical period hypothesis” by cognitive scientists (Hakuta, Bialystok, & Wiley, 2003;
Johnson & Newport, 1989). The application of this theory would suggest that if children are
exposed to the new language at a very young age they will acquire the language skills more
easily and therefore this will not have any negative impacts on the learning outcomes of students.
According to Human Capital Theory, improvement in “knowledge stock and learning
capabilities” of the population would have a significantly positive impact on the overall economy
(Foray & Lundvall, 1998). Using this framework, within the economics of education literature, it
there is evidence to suggest that the “medium of instruction policy in education” does have an
effect on human capital formation. R. Ramachandran (2012) provides evidence from Ethiopia
showing that a switch to mother tongue instruction for primary school led to a significant
increase in student educational attainment. Likewise, evidence from Yoruba indicates that
15
“higher repetition rates, dropout rates and overall lower achievement” can be partly explained by
difference in the medium of instruction (Bamgbose, 2005). In the United States, Thomas &
Collier (1997) analyzed the impact of Bilingual Schools versus English Schools on language
minority students over the period-. The study revealed that language minority students
in English schools performed poorer in English tests and had higher dropout rates and lower
educational attainment compared their counterparts in Bilingual Schools.
In the Indian context there are not many studies which have empirically examined the link
between the medium of education and student learning outcomes except for Muralidharan &
Sundararaman (2013b) .Although this study cannot test the accuracy of these theories in general,
it can throw some light on the state of English medium schools in comparison to the regional
medium schools.
RQ: Does English Medium Education have a negative impact on Student Learning
Outcomes?
3.2. Gender Discrimination in Educational Choice
The phenomenon of gender discrimination in intra-household allocation of resources in India
has been documented extensively (Alderman & King, 1998; Duraisamy & others, 1992). A more
recent study has pointed out that “girls experience gender discrimination especially from age 10
onwards, with almost universal disadvantage in the amount of education expenditures in the
group of 15-19 year olds.” (Zimmermann, 2012) Moreover, gender bias tends to be more acute
in rural areas when compared to urban areas (Azam & Kingdon, 2013). In general, parents tend
to spend allocate more resources for the education of the male child compared to the female
child. The differential treatment of the female child can be due to social and economic reasons.
16
(Dreze & Sen, 2003) argue that “entrenched belief of gender division of labor” is an important
factor for gender discrimination. Results from Gandhi Kingdon (2002) reveal that a plethora of
factors influence girls educational attainment—namely “parental background, wealth, opinions,
individual ability, age-at-marriage and quality of primary education”. Based on the findings of
these studies, we can expect that girls will be less likely to attend EMS due to gender
discrimination in Indian households.
3.3. Economic Class and differential access to Primary Education
The causal link between economic status and access to quality education is quite robust and
there is copious literature to support the hypothesis. Geeta G. Kingdon (1996) and Filmer &
Pritchett (2001) reveal that on average a “rich” child is 31 percentage points more likely to be
enrolled than a “poor” child. The study goes on to add that this gap is not even in all States. For
instance, the State of Kerala the gap is just 4.6% whereas Bihar is 42.6%. However, recently
there has been a sharp rise in the growth of “low cost” private schools in India. Even though free
government schools are available, parents tend to send their children to fee-levying private
schools (Muralidharan & Kremer, 2006). In India, there is a gulf between regulation “on paper”
and regulation “in practice”. Dixon & Tooley (2005) show how actual regulation on paper can
stifle innovation and entrepreneurship in the education sector and that in practice the growth of
low cost private schools remains “extra legal”. Muralidharan & Kremer (2006) argue that nearly
53% of private schools are “unrecognized” by the government. But the question whether even
the low cost private schools is accessible to the economically and socially disadvantaged
children still remains relevant. Härmä (2011) shows that low cost schools are still unaffordable
to the bottom two wealth quintiles families and argue that “increased reliance on a market in
education will not help to achieve equitable access to primary schooling for all”.
17
3.4. Government Programs and Primary School Enrollment
The Constitution of India guarantees every child, aged between 6 and 14, the right to free and
compulsory education. In order to achieve universal primary education, in the past, different
governments have designed various government programs aimed at achieving this goal. Besides
the Central government, various State governments provide financial and material incentives for
students. As per the Fifth All India Education Survey (1986), throughout India, out of the 5.29
lakh primary schools, 1.47 lakh schools provide mid-day meals, 2.48 schools provide free
uniforms and 3.13 lakh schools provide free textbooks. These incentives are intended to enhance
primary school enrollment.
Although there are not many empirical studies examining the
effectiveness of the free uniform and free text book programs in India, there are some studies
which have analyzed the impact of mid-day meals (MDM) program. Afridi (2011) points out
that, in Rural Areas, the mid-day meals program has been successful in “improving participation
rates of girls thereby reducing gender disparity in schooling.” Likewise, a study based on
Karnataka’s experience shows that the MDM program has a positive effect on enrollment,
attendance, dropout rate and retention rates, and a marginal impact on student’s scholastic
performance (Laxmaiah et al., 1999). Besides these educational parameters, the MDM scheme is
shown to have positive impact on the nutritional factors and the general health of the students
(Singh, Park, & Dercon, 2014). Based on the results of these studies and given the high demand
for EMS, we would expect government incentives to have a positive impact on enrollment into
EMS.
3.5. The influence of Caste on Education
The caste system plays a huge role in determining social expectations in India. Although in
modern times the influence of caste is waning primarily due to globalization and urbanization, it
18
still places an important role in Indian society. Borooah & Iyer (2005) show that, when the
parents are illiterate, the “community effects” are more pronounced on the educational outcomes
of the student. In recent times, the analyses of Munshi & Rosenzweig (2003) is a seminal
contribution to the understanding of influence of caste in modern Indian society. The study
makes the following conclusion:
Caste continues to play a particular (gender-specific) role in shaping schooling
choices in the new economy of the 1990s. But the overall increase in English
schooling in recent years, and the growing mismatch in education choices and hence
occupational outcomes between boys and girls in the same caste, suggest that the
remarkably resilient caste system might finally be starting to disintegrate.
In a globalized world English education is being viewed as an instrument of social mobility. This
seems to be the case across all caste based communities. A recent newspaper reported that the
Dalits in Uttar Pradesh have constructed a temple dedicated to the “Goddess of the English
Language”. Dalit activist and intellectual, Chandra Bhan Prasad remarked.
She (Goddess of English) holds a pen in her right hand which shows she is literate.
She is dressed well and sports a huge hat - it's a symbol of defiance that she is
rejecting the old traditional dress code. In her left hand, she holds a book which is
the constitution of India which gave Dalits equal rights. She stands on top of a
computer which means we will use English to rise up the ladder and become free
forever. (P, News, village, & Pradesh, n.d.)
Although it is just a piece of anecdotal evidence, it is symbolic of how the socially disadvantaged
communities perceive the value of English education.
3.6. Schooling and the Urban Rural Divide in access to education
Although there is a clear divide between rural and urban private school enrollment rates,
Tilak (2001) argues that “the relative size of both the government and the government-aided
sectors seem to be shrinking and that of the private unaided sector is increasing”. Muralidharan
19
& Kremer (2006) argue that approximately 28% of the rural population have access to fee
charging private schools and around 16.4% of children aged 6 – 14 in rural India attend feecharging private schools. Drèze & Kingdon (2001) show that parental education and motivation,
the distance of the school, the quality of the school, work opportunities, village development,
teacher postings, teacher regularity and mid-day meals, as the basic factors which determines a
child’s participation in school in rural India. Kochar (2004) argues that the gap between rural and
urban schooling can be explained by not just conditions in the local village economy, but also the
functioning and size of the relevant labor market. The study shows that among at least the
landless laborers, the schooling choice reflects the possibility of employment in urban areas.
Based on the results of these studies, we expect students from urban neighborhood to be more
likely to attend EMS.
3.7. Household Composition and Choice of Education
India is going through a demographic transition. It has witnessed a secular decline in the
fertility rates and the size of the family (Drèze & Murthi, 2001). Intra-household resource
allocation depends on the composition of the household. With respect to educational choice,
studies show that household size and the number of children in the household are negatively
correlated with the choice of schools (Desai, Dubey, Vanneman, & Banerji, 2009). On the other
hand, ceteris paribus, female headed household (especially from backward castes) are more
likely to live in poverty, more so in rural areas and studies have shown that children from female
headed household in rural areas are less likely to attend school (Ray, 2000).
3.8. Parental Education and Choice of Education
The causal link between parental education and child schooling has been widely recognized
in the literature (Desai et al., 2009; Dreze & Kingdon, 2001; Duraisamy & others, 1992;
20
Kambhampati & Pal, 2001). There is also evidence of “same sex effects”. In other words,
maternal education level has a more significant effect on the girl’s school participation and the
paternal education level on the boy’s (Dreze & Kingdon, 2001; Kambhampati & Pal, 2001). The
link between the demand for English Medium Education and parental education is also
recognized in the literature. (Kapur & Chakraborty, 2008)
4. Empirical Framework
Ideally, a randomized experimental design is the best approach to calculate the average
treatment effect of EMS. However, conducting such large scale social experiments is not always
practicable. The standard non-experimental regression techniques are likely to be inaccurate
since data for all relevant variables are not easy to obtain. Alternatively, a simple comparison
between the average outcomes of students from EMS with those from RMS will obviously suffer
from “selection bias” because the students are not randomly assigned to the schools. To
overcome the problem of selection bias there are various empirical strategies that can be used to
adjust the systematic differences between the treatment and control groups. In this paper I make
use of “propensity score matching” (PSM) method. The advantage of using the PSM method is
that it allows us to make causal inferences even in a non-experimental setting, conditional on the
basic assumptions being met. The estimates obtained by this method are shown to be an
improvement over other non-experimental estimates and are much closer to the estimates
obtained from experimental studies (Dehejia & Wahba, 2002). However, more recent literature
on this subject points out that this method can be effective only if it satisfies certain basic
conditions. (Smith & Todd, 2005)
According to Rosenbaum and Rubin (1983), “the propensity score is the estimated
conditional probability of assignment to a particular treatment given a vector of observed
21
covariates”. Based on the propensity scores of each sample, the statistical model generates a
comparison group that has similar characteristics to those in the treatment group, except for the
fact that they do not get the treatment. In other words, this comparison group is similar to the
“control group” in an experimental design. The probability of being a part of the “treatment
group” or “control group” is predicted as a function of a set of observed covariates that influence
the program participation (in this case it is student enrollment in English Medium School) and
outcomes. A PSM estimator pairs each program participant with a non-participant and the
difference in the mean of the outcome variables of interest between both groups can be
interpreted as the “average treatment on the treated” of the particular program.
The underlying basic assumptions in the PSM method are:
a) The assignment to a treatment only depends on the observable pre-intervention variables.
In other words, after controlling for the covariates, the treatment is as “good as random”.
This is known as the ‘unconfoundedness’ or ‘selection on observables’ assumption.
Where ‘
’ and ‘ ’ is the potential outcome of untreated and treated individuals, ‘D’ is
the treatment and ‘X’ is the set of observed covariates.
b) The probability of assignment to a treatment is bounded away from zero and one,
otherwise known as “overlap” assumption.
If the two assumptions are not met, then the PSM estimators are likely to be biased. In
practice, the “unconfoundedness’” assumption can never really be tested. Therefore the choice of
22
variables to be included in the logit model is very important. As a rule, only those variables
which affect both the program participation and the outcome must be included in the model.
Therefore the choice of variables should either be based past research or should have strong
theoretical backing. More recent studies stress the importance of avoiding variables which can be
affected by the “anticipation of participation in the treatment”, and makes a strong case to avoid
“over-parameterization” (Caliendo & Kopeinig, 2008).
This study exploits the data at the student level i.e. the unit of analysis is the individual. The
first step of the PSM method is the calculation of the propensity scores of every sample. The
propensity score is derived from a logit regression in which the outcome variable of interest is
whether a student is enrolled in an EMS. The determinants of EMS enrollment can be broadly
classified into four categories, namely 1) Student Characteristics 2) Household Characteristics 3)
School Characteristics and 4) Socio-Religious Characteristics. Within each category there are a
set of variables, which in detail will be covered in the “Data” section of this paper. The baseline
logit model is depicted in the following equation:
Where ‘ ’ represents the dependent variable and it is a dummy in which ‘1’ stands for EMS and
‘0’ stands for RMS. ‘ ’ represents the error term.
In this study we have three outcome variables namely mathematics, reading and writing
scores. The baseline logit model will remain the same for all the three outcome variables. After
23
estimating the propensity scores based on the logit model, the next step is to choose a matching
method. There are various types of matching methods. Depending on the algorithm of the
matching method, a individual from the treatment group is compared to one or more individuals
from the control group.
The “Nearest Neighbor” method every individual from the treatment group will be
compulsorily matched with the nearest individual from control group, and in the next step the
difference between the outcomes will be computed. And finally, the “Average Treatment on the
Treated” is obtained by computing the average of these differences. The drawback in this method
is that the routine matches the nearest neighbor even if the propensity score is significantly
different from one another and thereby could result in poor matches. This problem can arise if,
for example, there are too many individuals with high propensity scores in the treatment group
and few individuals in the control group with high propensity scores; as a result the NN matches
individuals with relatively low propensity scores with high propensity score individuals. In such
a scenario, the NN method will not produce the most reliable estimates and therefore the
researcher should choose alternative algorithms. (Caliendo & Kopeinig, 2008)
The stratification matching algorithm partitions the common support region of the propensity
score into a set of intervals and computes the mean of difference in outcomes between treated
and control observations to estimate the ATT of a particular program (Rosenbaum & Rubin,
1983). An advantage of this method is that the outcome of all individuals in the treatment and
control group are factored in estimation of the treatment effect. The disadvantage of this method
is that some strata may contain a relatively small number of individuals in the treatment group
compared to control group members and vice versa.
24
Kernel matching algorithm compares the outcome of each individual in the treatment group
to a weighted average of the outcomes of all individuals in the control group. The highest weight
is attached to those with scores nearest to the treated individual. Since more information is used,
one advantage of this method is that the variance is very low. This method has the same
drawback as the NN matching method i.e. some observations may have poor matches (Smith &
Todd, 2005).
Each matching algorithm has its own merits and demerits. Therefore selecting any one
method can be a problem if different methods produce significantly different results. To
overcome this problem, in this study I make use of three matching methods namely Nearest
Neighbor, Kernel and Stratified Matching.
5. Data
The data used for this study was collected by the National Council of Applied Economic
Research in collaboration with University of Maryland in 2005 (Desai et al., 2013). The Indian
Human Development Survey (IHDS) is a multi-topic, nationally representative survey (with the
exception of Lakshadweep and Andaman and Nicobar islands) covering 41,554 households from
1503 villages and 971 urban neighborhoods across India. The data was collected using stratified
random sampling procedure to ensure linguistic, religious and caste based subpopulations are
adequately represented. The topics covered in this survey include health, education, employment,
economic status, marriage, fertility, gender relations, and social capital.
As a part of the survey exercise, children aged between 8 and 11 in the household were
administered reading, writing, and arithmetic knowledge tests. The test questions were designed
in collaboration with researchers from PRATHAM, one of India’s leading non-governmental
organizations (NGO) working in the field of primary education. Prior to administrating the test,
25
the surveyors were trained by Pratham. The tests were developed in 13 Indian languages and
standardized to ensure that the results can be compared across different languages. The students
were allowed to take the test in the language they were most comfortable in. and based on the
test results, IHDS classified student’s capability into different categories.
The student’s reading capability was classified into five categories: (1) Cannot read at all (2)
Can read letters (3) Can read words (4) Can read paragraph (5) Can read a one-page short story.
Similarly, a student’s arithmetic skills was classified into four categories: (1) Cannot read
numbers above 10, (2) Can read numbers between 10 and 99 but cannot do more complex
number manipulation, (3) Can subtract two-digit numbers, and (4) Can divide a number between
100 and 999 by another number between 1 and 9. And finally, the writing test classified students
into two categories: (1) Cannot write, (2) Writes with two or fewer mistakes. Figures 1, 2 and 3
show the students reading, arithmetic and writing skills in each category level. The reading,
arithmetic and writing skills of the students are the dependent variables in this study.
As discussed earlier, post 1991 India has witnessed a rapid expansion privately run schools
and that the trend was not just restricted to urban areas. The IHDS survey uses the 2001 census
to make the distinction between rural and urban areas. Figure 7 shows the private school
enrollment rate in rural and urban areas within our study sample. We can observe that about
57% in urban students and 22% in rural students attend private school. This should not come as a
surprise since many studies have confirmed this phenomenon earlier (Drèze & Kingdon, 2001;
Geeta Gandhi Kingdon, 2007). Since this study is concerned with only private schools
investigating the impact of private English medium schools on students learning outcome, only
the test results of children between the age of 8 and 11 who are currently enrolled in a school
will be included in the study sample.
26
The main dependent variable is a dichotomous variable which indicates whether the student
has enrolled in an English medium school (EMS) or a regional medium school (RMS). By
definition an “English Medium School” is one in which ‘English’ is the primary medium of
pedagogy. The same logic applies to the Regional Medium Schools as well. Post Independence,
Indian most states were re-organized on linguistic basis. Therefore each state has a dominant
regional language. To simplify our analysis, all RMS are coded as ‘0’ and EMS are coded as ‘1’.
In our overall data, out of 11,060 “currently enrolled” students only 1339 are enrolled English
medium schools (See Figure 4) and out of the 1339 students who have enrolled in English
medium schools only 214 are in government run English medium schools. In our study sample
we consider only those students who are currently enrolled in private schools, out of a total of
3731 (69%) students 1144 (30.66%) are enrolled in EMS (see figure 5).
The independent variables are split into four broad categories: Student characteristics,
Household characteristics, School characteristics and Socio-Religious characteristics. The
descriptive statistics of independent variables for Regional and English Medium Schools is given
below in Table 1. From the results of the means comparison test between the students who attend
RLS and EMS, we can observe that, on an average, EMS students spend more time for
homework and they are more likely to attend private tuition. This is consistent with the evidence
from the more recent literature on the difference between student study effort in private and
public schools. (Muralidharan & Sundararaman, 2013b)
Table 1: Descriptive Statistics for Regional Language Schools Vs English Medium Schools
Factor
N
Student Characteristics
Male
EMS = 0
2587
EMS= 1
1144
p-value
1451 (56.1%)
682 (59.6%)
-
Age 9
Age 10
Age 11
Standard (in years)
School Characteristics
Mid Day Meals Scheme
Free Uniform
Free Books
School Fee Paid By Government
Household Characteristics
Household Size
Number of Children
Household Monthly Consumption (log)
Household Income (log)
Fathers Education (in years)
Mothers Education (in years)
Female head
Either Parent Knows English
Urban Neighborhood
Socio Religious Characteristics
Brahmins
Other Backward Caste
Schedule Caste
Schedule Tribe
Muslims
Sikh & Jains
Christians
556 (21.5%)
856 (33.1%)
520 (20.1%)
3.651 (1.665)
306 (26.7%)
322 (28.1%)
199 (17.4%)
3.802 (1.464)
-
340 (13.5%)
91 (3.6%)
408 (16.0%)
190 (7.6%)
37 (3.4%)
3 (0.3%)
30 (2.7%)
27 (2.4%)