Can Private School Growth Foster Universal Literacy? Panel Evidence from Indian Districts

Similar documents
BASIC EDUCATION IN GHANA IN THE POST-REFORM PERIOD

[For Admission Test to VI Class] Based on N.C.E.R.T. Pattern. By J. N. Sharma & T. S. Jain UPKAR PRAKASHAN, AGRA 2

According to the Census of India, rural

Research Update. Educational Migration and Non-return in Northern Ireland May 2008

Accessing Higher Education in Developing Countries: panel data analysis from India, Peru and Vietnam

A Comparison of Charter Schools and Traditional Public Schools in Idaho

Educational Attainment

ABILITY SORTING AND THE IMPORTANCE OF COLLEGE QUALITY TO STUDENT ACHIEVEMENT: EVIDENCE FROM COMMUNITY COLLEGES

Post-intervention multi-informant survey on knowledge, attitudes and practices (KAP) on disability and inclusive education

School Competition and Efficiency with Publicly Funded Catholic Schools David Card, Martin D. Dooley, and A. Abigail Payne

Effective Pre-school and Primary Education 3-11 Project (EPPE 3-11)

Australia s tertiary education sector

Production of Cognitive and Life Skills in Public, Private, and NGO Schools in Pakistan

AUTHORITATIVE SOURCES ADULT AND COMMUNITY LEARNING LEARNING PROGRAMMES

An Empirical Analysis of the Effects of Mexican American Studies Participation on Student Achievement within Tucson Unified School District

Educational system gaps in Romania. Roberta Mihaela Stanef *, Alina Magdalena Manole

MEASURING GENDER EQUALITY IN EDUCATION: LESSONS FROM 43 COUNTRIES

College Pricing. Ben Johnson. April 30, Abstract. Colleges in the United States price discriminate based on student characteristics

Literacy Level in Andhra Pradesh and Telangana States A Statistical Study

A Note on Structuring Employability Skills for Accounting Students

LANGUAGE DIVERSITY AND ECONOMIC DEVELOPMENT. Paul De Grauwe. University of Leuven

Peer Influence on Academic Achievement: Mean, Variance, and Network Effects under School Choice

Estimating the Cost of Meeting Student Performance Standards in the St. Louis Public Schools

NAVODAYA VIDYALAYA SAMITI PROSPECTUS FOR JAWAHAR NAVODAYA VIDYALAYA SELECTION TEST- 2014

The Comparative Study of Information & Communications Technology Strategies in education of India, Iran & Malaysia countries

IS FINANCIAL LITERACY IMPROVED BY PARTICIPATING IN A STOCK MARKET GAME?

Regional Bureau for Education in Africa (BREDA)

(ALMOST?) BREAKING THE GLASS CEILING: OPEN MERIT ADMISSIONS IN MEDICAL EDUCATION IN PAKISTAN

Is there a Causal Effect of High School Math on Labor Market Outcomes?

Management and monitoring of SSHE in Tamil Nadu, India P. Amudha, UNICEF-India

Schooling and Labour Market Impacts of Bolivia s Bono Juancito Pinto

Role Models, the Formation of Beliefs, and Girls Math. Ability: Evidence from Random Assignment of Students. in Chinese Middle Schools

The Impact of Group Contract and Governance Structure on Performance Evidence from College Classrooms

Teacher Quality and Value-added Measurement

What effect does science club have on pupil attitudes, engagement and attainment? Dr S.J. Nolan, The Perse School, June 2014

Annex 1: Millennium Development Goals Indicators

I set out below my response to the Report s individual recommendations.

1GOOD LEADERSHIP IS IMPORTANT. Principal Effectiveness and Leadership in an Era of Accountability: What Research Says

ROA Technical Report. Jaap Dronkers ROA-TR-2014/1. Research Centre for Education and the Labour Market ROA

EDUCATIONAL ATTAINMENT

Unequal Opportunity in Environmental Education: Environmental Education Programs and Funding at Contra Costa Secondary Schools.

Sl. No. Name of the Post Pay Band & Grade Pay No. of Post(s) Category

This Access Agreement is for only, to align with the WPSA and in light of the Browne Review.

A Diverse Student Body

Higher Education Six-Year Plans

EUROPEAN UNIVERSITIES LOOKING FORWARD WITH CONFIDENCE PRAGUE DECLARATION 2009

Class Size and Class Heterogeneity

UPPER SECONDARY CURRICULUM OPTIONS AND LABOR MARKET PERFORMANCE: EVIDENCE FROM A GRADUATES SURVEY IN GREECE

James H. Williams, Ed.D. CICE, Hiroshima University George Washington University August 2, 2012

CONFERENCE PAPER NCVER. What has been happening to vocational education and training diplomas and advanced diplomas? TOM KARMEL

Tamil Nadu RURAL. School enrollment and out of school children. Young children in pre-school and school

Massachusetts Department of Elementary and Secondary Education. Title I Comparability

Gender and socioeconomic differences in science achievement in Australia: From SISS to TIMSS

Teacher intelligence: What is it and why do we care?

Alpha provides an overall measure of the internal reliability of the test. The Coefficient Alphas for the STEP are:

GDP Falls as MBA Rises?

The number of involuntary part-time workers,

A STUDY ON AWARENESS ABOUT BUSINESS SCHOOLS AMONG RURAL GRADUATE STUDENTS WITH REFERENCE TO COIMBATORE REGION

American Journal of Business Education October 2009 Volume 2, Number 7

A comparative study on cost-sharing in higher education Using the case study approach to contribute to evidence-based policy

Sector Differences in Student Learning: Differences in Achievement Gains Across School Years and During the Summer

In reviewing progress since 2000, this regional

JOIN INDIAN COAST GUARD

Greek Teachers Attitudes toward the Inclusion of Students with Special Educational Needs

U VA THE CHANGING FACE OF UVA STUDENTS: SSESSMENT. About The Study

Systematic Assessment and Monitoring leading to Improving Quality of Education

Wisconsin 4 th Grade Reading Results on the 2015 National Assessment of Educational Progress (NAEP)

Entrepreneurial Discovery and the Demmert/Klein Experiment: Additional Evidence from Germany

PUPIL PREMIUM REVIEW

Education in Armenia. Mher Melik-Baxshian I. INTRODUCTION

Higher education is becoming a major driver of economic competitiveness

PROPOSED MERGER - RESPONSE TO PUBLIC CONSULTATION

About the College Board. College Board Advocacy & Policy Center

How and Why Has Teacher Quality Changed in Australia?

BENCHMARK TREND COMPARISON REPORT:

University of Toronto

Like much of the country, Detroit suffered significant job losses during the Great Recession.

GROUP COMPOSITION IN THE NAVIGATION SIMULATOR A PILOT STUDY Magnus Boström (Kalmar Maritime Academy, Sweden)

Rwanda. Out of School Children of the Population Ages Percent Out of School 10% Number Out of School 217,000

How to Judge the Quality of an Objective Classroom Test

Social and Economic Inequality in the Educational Career: Do the Effects of Social Background Characteristics Decline?

HIMACHAL PRADESH NATIONAL LAW UNIVERSITY, SHIMLA GHANDAL, P.O. SHAKRAH, SUB TEHSIL DHAMI, DISTRICT SHIMLA

Pupil Premium Impact Assessment

National rural Health mission Ministry of Health and Family Welfare government of India, new delhi

Dakar Framework for Action. Education for All: Meeting our Collective Commitments. World Education Forum Dakar, Senegal, April 2000

Listening and Speaking Skills of English Language of Adolescents of Government and Private Schools

PEER EFFECTS IN THE CLASSROOM: LEARNING FROM GENDER AND RACE VARIATION *

The Effects of Statewide Private School Choice on College Enrollment and Graduation

w o r k i n g p a p e r s

Conditional Cash Transfers in Education: Design Features, Peer and Sibling Effects Evidence from a Randomized Experiment in Colombia 1

The Good Judgment Project: A large scale test of different methods of combining expert predictions

IMPERIAL COLLEGE LONDON ACCESS AGREEMENT

The Future of Consortia among Indian Libraries - FORSA Consortium as Forerunner?

Asian Development Bank - International Initiative for Impact Evaluation. Video Lecture Series

Leprosy case detection using schoolchildren

Mathematics subject curriculum

School Size and the Quality of Teaching and Learning

FACTORS AFFECTING TRANSITION RATES FROM PRIMARY TO SECONDARY SCHOOLS: THE CASE OF KENYA

Engineers and Engineering Brand Monitor 2015

THE QUEEN S SCHOOL Whole School Pay Policy

Transcription:

Can Private School Growth Foster Universal Literacy? Panel Evidence from Indian Districts Sarmistha Pal * Brunel University, London and IZA (Germany) sarmistha.pal@brunel.ac.uk and Geeta Gandhi Kingdon Institute of Education, University of London and CSAE (Oxford) g.kingdon@ioe.ac.uk December 2011 Abstract: While the Education related Millennium Development Goals highlight the role of national governments and aid agencies, in view of the recent growth of private schools around the world, the present paper examines whether the private sector can foster universal literacy. Results using unique district-level panel data-set from India finds a significant positive association between the shares of recognised private schools and literacy, especially among 10-14 year old children; however, the potential role of private schools for reducing gender gap in literacy is rather weak in our sample. Further analysis suggests that the effect of private school growth is rather limited on literacy or gender gap among ST children as well those for the districts targeted by the District Primary Education Programme. Key words: Private school growth, Millennium Development Goals, Universal literacy, Gender gap in literacy, India, South Asia. JEL code: I21, I28, O15 * Corresponding author. Address for correspondence: CEDI, Brunel University, Uxbridge UB8 3PH, United Kingdom. Fax. 01895 269786. Sarmistha Pal wishes to thank Leverhulme Trust for funding this research, NCERT for access to AISES data and CSAE, Oxford University for their hospitality where much of this research was done. We would also like to thank Petia Topalova for poverty data and Mohammed Arif, Ana Maria Lugo, Yama Temouri for excellent research assistance, Paul Atherton, Erlend Berg, Sonia Bhalotra, Jean Drèze, William Evans, Irma Clots-Figures, Pramila Krishnan, Pushkar Maitra, Karthik Muralidharan, and also participants at the UKFIET Conference on Education and Development at Oxford, Economic Growth and Development Conference at the Indian Statistical Institute New Delhi, and Social Policy in India workshop at Warwick University for many constructive comments/suggestions. Any remaining errors are ours.

1 Can Private School Growth Foster Universal Literacy? Panel Evidence from Indian Districts 1. INTRODUCTION In April 2000 the World Education Forum s meeting in Senegal set the agenda for the attainment of universal primary education by 2015. It noted that the prime responsibility for achieving universal education lay with national governments, assisted by aid agencies. This prescription however ignored the role of the burgeoning private schools in different parts of the world including some emerging economies. While some aid agencies have recently shown some awareness of the importance of growing private sector (e.g., see World Bank 2009), this knowledge has not yet been reflected in policy. This is an important oversight which has led to a debate as to whether growth of private schools could be conducive to universalising literacy (Tooley, 2004; Watkins 2004). In an era of stagnating public budgets for education (as well as other accounts), private financing of education has gradually been gaining importance around the world. This is not an isolated phenomenon for the developing and emerging economies only; in fact, faced with binding government budget deficit, some European countries too (including the current coalition government in the UK) have been preaching for the enhanced role of the private sector for educational development (e.g., see Wilby, 2010). While there have been some assessments about the prospects of meeting the Millennium Development Goals (MDGs) by 2015 around the world (Bloom et al. 2006), we are not aware of any systematic analysis that assesses the effect of private provision of education on universal literacy. The present paper aims to bridge this gap of the literature. It has often been argued that greater market orientation makes private schools and teachers more accountable to parents, more sensitive to input costs and thereby more efficient. A fast pace of private school growth could however raise concerns for equity reasons, especially in the light of the MDGs. First, private schools are the fee paying schools and hence would naturally exclude children from poorer/disadvantaged background. Further, given the importance of son preference especially in some Asian countries, private school growth could widen the gender gap between boys and girls if this induces resource constrained parents to send only their boys to private schools. Given this trade-off, it is pertinent to explore the potential role of private sector for delivering universal literacy by 2015. India is an important case in point. While the state sector still dominates the schooling market in India, an important feature of the 1990s has been a significant growth of private schools in India (Public Report on

2 Basic Education; in short, PROBE 1999). 1 While about 16% of the villages surveyed in PROBE data had access to private schools, the corresponding figure rose to about 28% in 2003 (Muralidharan and Kremer 2008). There is a large and growing literature on child schooling in developing countries (e.g., see Glewwe, 2002; Hanushek and Woessman 2008) and a sizeable literature (e.g., see Bashir, 1994; Kingdon, 1996; Beegle and Newhouse, 2006) 2 on relative efficiency of private schools in imparting education; however, there is very little direct evidence, if at all, about the possible effects of recent growth of private schools on universal literacy. In this respect, two recent studies are worth citing. First, considering a nationally representative rural sample from the major Indian states, Muralidharan and Kremer (2008) suggested that the single most distinguishing feature of the private schools in rural India is that they pay much lower salaries to teachers than the government schools. This allows the private schools to hire more teachers, thus ensuring a lower pupil-teacher ratio than state schools. The same study found that private schools are more likely to be set up in poorer states and also in areas where state schools are failing. Using PROBE survey data from five north Indian states 3, Pal (2010) found that private schools are more likely to be present in villages with existing state schools as well as those with better off households and better access to public infrastructural facilities (e.g., transport, communications). Also year five pass rates are significantly higher in the villages with private schools while the presence of private schools is not associated with higher government school pass rates in the sample. Despite the absence of school fees, dismal state of the state schools have induced many households, including some poor, to take advantage of the newly emerging private unaided schools in India to meet their educational needs. To a large extent, the latter has been facilitated by the modest private school fees in India. While there is no systematic data available for private school fees across India, Tooley and Dixon (2003) found that the average school fees were only about 2 a month in Hyderabad while the median fee across rural India was estimated to be Rs. 63 per month in the 2003 survey by Muralidharan and Kremer (2008). While private schools generally tend to serve the better-off households, PROBE survey too found that poorer households were not totally excluded from access to private schools in the PROBE states (Dreze and Kingdon, 2001). Tooley and 1 See Pal (2010) for further details about PROBE survey. 2 There is however no general consensus that private schools are necessarily better than state schools. Bashir (1994) indicated that students in private schools had better Mathematics achievement, but less achievement in Tamil language, compared to government school students in Tamil Nadu. Kingdon (1996) found that, ceteris paribus, students in private schools performed significantly better than those in government schools in urban Lucknow district. In the context of Indonesia, Beegle and Newhouse (2006) contrast Bedi and Garg (2000) in that junior secondary level (grades 7-9) students in public schools in Indonesia outperform those in private schools, which they attribute to unobserved higher quality of inputs in public schools. 3 The survey covers households, schools and villages drawn from five Indian states including four of the country s worst performing states, namely, Bihar, MP, Rajasthan and UP; the fifth state is a much better-off state Himachal Pradesh (HP). See Pal (2010) for further details.

3 Dixon (2003) study further highlighted that fathers were largely daily-paid labourers and 30% of mothers were illiterate, but families were active in the school choice process in Hyderabad. In this context, we raise the question whether private school growth can foster universal literacy. Increase in private school share at a given level in a district could affect literacy through affecting demand and/or supply. On the one hand, private schools are directly accountable to parents and children and thus tend to resolve the incentive problems commonly present in the management of government schools (incentive effect). 4 On the other hand, the greater effectiveness of private schools may also be linked to the unobserved characteristics of parents/students attending private schools (selection effect). 5 The pertinent issue here is to explore whether an increased share of private unaided schools could boost literacy at a given level (upper primary, secondary) at the district level. If it did not, private school growth, in a sense, would be incompatible with the objective of attaining universal literacy by 2015, which is a corollary of the education related MDG. In principle, a higher private school share could either raise or lower literacy rates, or indeed have no effect on literacy. It may raise literacy if private schools impart higher learning than government schools, as has been found by some micro-level studies (e.g., see Kingdon 1996). It may also raise literacy if presence of private schools boosts the quality of local government schools (e.g., Hoxby, 1994), which is not supported by existing evidence from India (Pal 2010). It may lower literacy rates if the growth of private schools causes the closure of or deterioration in the quality of government schools; we are not yet aware of any such evidence. Private school growth may have no impact on literacy as such; other things remaining unchanged, if, for example, those who choose private schools are relatively better-off and/or more motivated towards schooling and would have become literate even in the absence of private schooling (e.g. via enrolment in government schools with/without private tuition), privatisation would not have a significant impact on literacy rates. A related issue is the implication of private school growth for gender differences in literacy rates. This is particularly important for a country like India where a pronounced gender difference in literacy persists, especially in the large north Indian states such as Bihar, UP, MP and Rajasthan. Private school growth could potentially exacerbate gender difference in literacy rates since it may enable parents characterised by pronounced son-preference to exercise that preference by sending sons to private and daughters to government 4 Another possible channel would be to assess the effect of school privatisation on performance of government schools. We however do not have any information on performance indicators of students attending private and public schools; hence we were unable to test this hypothesis. We also note that Pal (2010) argued that presence of local private schools fails to have a significant impact on government school pass rates in the PROBE villages. 5 With educational opportunities opened to all through the expanded government school system, there has been a search for ways of differentiating achievement. The latter has given rise to the demand for instructions in English that newly emerging private schools often offer, thus strengthening the selection effect.

4 schools. 6 If learning levels are better in private schools as has been documented by various micro-level studies (e.g., Kingdon, 1996), the gender gap in learning (and literacy) may increase with private school growth as girls, especially in the presence of son preference in school choice, are likely to be excluded from private schools. On the other hand, private schools may mitigate gender differences in educational outcomes. For example, if private schools fulfil differentiated demand for girls education, it may increase girls access to schooling and learning and thus reduce the gender gap in literacy. For example, private schools may encourage girls schooling in various ways: (i) provision of local schools so that girls do not have to travel far; (ii) provision of separate toilets for girls and boys, which may especially encourage schooling among adolescent girls; (iii) provision of English medium education, which may entail higher return for girls in the marriage market (Munshi and Rosenzweig, 2006). However, it remains possible that any apparently beneficial effect of private schools (in terms of a reduced gender gap in schooling/literacy) is in fact due to an aspect of sample selection. Parents who choose to send daughters to private schools (especially at the higher levels of schooling) are not a random draw from the population of all parents; they are likely to be the more enlightened in terms of attitudes to gender equality. Thus, if a higher share of private schooling in a district is associated with a lower gender gap in literacy, this could potentially be a spurious relationship, simply due to sample selection, i.e. in reality there may be no relationship between the two. On balance, whether the growth of private schooling has a negative, positive or neutral association with gender gap in literacy rates, ceteris paribus, remains an open empirical question that we explore here. Studies that analyse different aspects of private schooling growth in India (e.g., Kingdon 1996; Muralidharan and Kremer, 2006; Pal 2010) have primarily used single-year cross-section data. Consequently, existing estimates of the effects of private schools are likely to suffer from endogeneity bias primarily due to the unobserved heterogeneity among market participants (schools/parents/children). For example, in an achievement production function, a private school dummy variable is endogenous since it is likely to pick up the effect of child or family level unobserved factors (e.g. motivation, ambition etc.) that make it more likely that a child will attend private school and also raise achievement levels. Similarly, to extend to district level data, if we were to regress district literacy rate in a given year on the district s private school share in that year, the latter variable would suffer from omitted variable bias in such an OLS regression. There could also be reverse causality: just as private school presence affect achievement/literacy, the latter may also influence private school presence. 6 There is however no evidence that parental wealth is associated with greater gender difference in literacy in India (e.g., see Pal, 2004).

5 We use a unique two-period district-level data for 1992 and 2002 from 17 major Indian states compiled from various official sources. Access to this district-level panel data allows us to address the potential biases arising from unobserved heterogeneity as well as reverse causality better while examining the effect of private school growth on literacy rates and the associated gender gap in literacy, two of the key millennium development goals (MDGs). Our analysis focuses on literacy information available from the Indian Census where a person is defined as literate when s/he is able to read and write in any language. The extent of private schooling in a district is measured by the share of private unaided schools 7 in total schools at a given level, namely, upper primary 10-14 years, secondary 15-19 years; we also pool these two levels to consider children aged 10-19 years). 8 We start with conventional pooled OLS model of literacy and gender gap in literacy with control for various factors including private schools share at a given level and binary indicator for the year 2002. Clearly, however, a pooled OLS model would suffer from the omitted variable bias influencing the variable of interest. One alternative is to consider a first difference model of changes in literacy (and gender gap in literacy) in terms of initial values of private school share and other possible control variables; while this minimises the potential simultaneity bias from literacy to private school share, it still suffers from omitted variable bias. Hence our analysis focuses on the fixed effects OLS panel data estimates using this two year panel data information on sample districts. In other words, we exploit the variation across districts over these two years to identify the possible effect of private school share on literacy and gender gap in literacy in our sample. While fixed-effects estimates redress the potential bias arising from time-invariant omitted factors, concerns may still arise because of the potential bias arising from time-varying omitted factors. In an attempt to redress this bias, we experiment with some possible time-varying factor, namely, operation of district primary education programme (DPEP) in our sample, using difference-in-difference (DID) estimates (see further discussion in section 4.2). 9 There is evidence of a significant positive impact of private school growth on literacy though its effect on gender gap in literacy is rather weak in our sample. Compared to 15-19 year olds, literacy effect of private school share is more pronounced for 10-14 year olds, which is perhaps convincing: 10-14 year olds in 2002 7 See section 2 for a discussion of types of private schools in India. 8 Henceforth we use private school growth and growing private school share at a given level of schooling interchangeably for the rest of this paper. 9 Recent studies highlight the benefit of randomized experiments that may provide a good solution to the problem of reverse causality in some cases. For example, Angrist et al. (2002) attempted to randomize household access to private schools through vouchers awarded by lottery. However, randomization is unlikely to work in our analysis of the effect of district-level growth of private schools on literacy and gender gap in literacy as presence of private schools in a given location is a strategic choice of private investors (e.g., see Pal, 2010). Further despite collecting a rich array of information at the district-level, use of instrumental variable (IV) is unlikely to be convincing as the factors that influence private school growth are also likely to influence literacy or gender gap in literacy. Hence our analysis focuses on FE-OLS and DID estimates.

6 were the first group of children to enjoy the full benefits of pronounced private school growth as they started schools in the second half of the 1990s. Next we consider the DID estimates which highlighted the differential effect of DPEP on 10-14 literacy over 1992-2002 (which was also evident on 10-19 literacy); the effect was, however, insignificant for 15-19 year age group. There is also evidence of a significantly lower incidence of gender gap in literacy among 10-14 year olds, who turn out to be primary beneficiary of private school growth in our sample. However the differential effects of private school growth on literacy or gender gap in literacy in the DPEP districts remain insignificant. Further analysis suggests that private school growth exerted a significant influence only on SC children though failed to affect that among ST children. Taken together, these results generally highlight a potential role of the private sector in imparting learning, though its effect on marginalised ST population or DPEP districts remain much weaker in our sample. Thus there remains the option of introduction of school vouchers or subsidising the private sector to boost literacy among all sections of the population, which has not yet received sufficient attention either from the national authority or from international donor agencies operating in the country. The paper is developed as follows. Section 2 describes the data while section 3 explains the methodology. Results are discussed in section 4. The final section concludes. 2. DATA Data has been compiled from various sources: This includes the Sixth (1992-93) and Seventh (2002-03) All India School Education Survey (AISES) data and also the Census data (1991 and 2001). District-level AISES data cover information on the number of recognised schools (by management type, i.e., private/public, etc.), enrolment by gender and caste (scheduled castes, SC; scheduled tribes, ST), characteristics of teachers (gender/caste), and physical facilities at primary, upper primary and secondary levels of schooling in the district. District level census data from 1991 and 2001 provide information on population composition (by gender/caste); literacy rates for different age categories of the population (male/female and total); and access to various infrastructural facilities. In addition, we obtain district-level poverty head count rates information from the 50 th (1993-94) and 55 th (1999-00) rounds National Sample Survey (NSS) data. We merge 1991 Census data and 50 th round NSS data with 6 th AISES to generate district-level information for 1992. Similarly, we merge 2001 Census data and 55 th round NSS data with 7 th AISES data to generate the corresponding district-level information for 2002. 10 This 10 We need to do this because school census data collected every ten years are available only for 1992 and 2002.

7 allows us to build up a two-period panel data for the period 1992-2002. 11 There are three broad types of recognised schools in India, namely, government schools, private aided schools (PA) and private unaided schools (PUA) schools. 12 Government and aided schools are invariably government-recognised, i.e. they have the government stamp of approval. They are similar to each other in many respects since aided schools are almost entirely financed by the government and have little control over staffing (hiring/firing) and fee levels, despite being nominally privately managed. 13 PUA schools (whether recognised or not) are more autonomous than aided schools and are totally self-funded out of fee income. Thus PUA schools are the truly private schools in India. At the secondary school level, all schools including PUA schools have to be government-recognised. But at the primary and upper primary levels, many PUA schools remain unrecognised. 14 Non-recognised schools are not included in any government list of schools and are thus not included in the periodic school census (called the All India School Education Survey (AISES)). As a result, our analysis in this paper can only include the recognised PUA schools rather than all PUA schools. This is an unfortunate but an inevitable data limitation since there is no source that provides information on unrecognised PUA schools for all districts of India going back to early 1990s 15. However, in general, there is likely to be a positive correlation between the share of recognised PUA schools and the share of all PUA schools (recognised and unrecognised) since districts that have more recognised PUA schools are also likely to be the districts that have more unrecognised PUA schools. 16 As such our result would provide only a lower bound of the growth of private schools in the Indian districts. In the rest of the paper whenever we refer to PUA or private schools, we mean the recognised private unaided schools only. For the purposes of this paper, we 11 Two clarifications are in order. First, given that there have been changes in the number of districts between 1991 and 2001 Census, we consider only the districts present in both rounds of Census for the selected sixteen major states. These states are Andhra Pradesh, Assam, Bihar, Gujarat, Haryana, Jammu and Kashmir, Karanataka, Kerala, Madhya Pradesh (MP), Maharashtra, Orissa, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh (UP) and West Bengal. Second, between 1991 and 2001 Bihar, MP and UP were split to give rise to 3 new states namely Jharkhand, Chhatisgarh and Uttaranchal. Our sample districts include those districts which were parts of Jharkhand, Chhatisgarh and Uttaranchal in 2001 though were included under the parent states Bihar, MP and UP respectively. 12 In order to receive recognition, however PA and PUA schools must fulfil several requirements that are prohibitively expensive for many schools, especially those serving the poor (e.g., hold a sizeable cash bond with the government, provide sizeable playgrounds, etc.). 13 There is some inter-state variation in the management of PA schools. For example, PA schools in Uttar Pradesh have no control over hiring/firing of own teachers (who are appointed by the UP School Service Commission). In contrast, PA schools in Tamil Nadu have some autonomy to select and hire their own teachers. 14 However, in most states, there are no board examinations at the primary of upper primary levels so there is no strong incentive for private schools to seek government recognition except if the school wishes ultimately to become a secondary school and affiliate with an exam board. 15 Even the District Information System on Education (DISE) data collection exercise introduced in the late 1990s does not have the mandate/authority to collect information on all unrecognised PUA schools. Thus, even today there is no way of reliably knowing the number of unrecognised PUA schools in India, though see Kingdon (2008) for various estimates. 16 This can be supported by the PROBE data that distinguishes between recognised and unrecognised private schools in UP, Bihar, HP, Rajasthan and MP. The correlation coefficient is about 0.24 and it is statistically significant at 5% (our own calculation)

8 exclude PA schools and compare the case of pure government schools with the case recognised private unaided schools which we call simply private schools or PUA schools. Table 1 compares the quality of PUA and government schools at the district level over the decade 1992-2002, using some commonly used quality measures. In general, PUA schools tend to have better infrastructure (pucca building, access to drinking water and toilets) than government schools; however, government schools have significantly narrowed the gap in this respect over the decade. Compared to recognised PUA schools, Government schools have significantly higher pupil-teacher ratio (more than double at the primary level) and the situation does not change much over the decade under consideration. Furthermore, recognised PUA schools employ a higher proportion of female teachers. Although Table 1 does not show this, compared to government schools, PUA schools also have younger teachers, fewer teachers with (pre-service) training and fewer vacant teaching positions (see Pal 2010). Thus despite significant public interventions over the 1990s (for instance, the District Primary Education Programme since 1994 and Sarvashiksha Abhiyyan since 2001) to improve government schools, input differences between recognised PUA and government schools persisted by 2002. 2.1. Growth of private schools Using 6 th and 7 th AISES data, we first calculate the average share of recognised PUA schools in total schools at a given level (primary, upper primary and secondary) in a district, and also the corresponding district literacy rates, as summarised in Table 2. In each of the two years, the share of private schools at the secondary level (e.g., 15% in 1992) is significantly higher than at the primary level (e.g., 4.4% in 1992) 17. Over the course of the decade 1992-2002, the pace of private school growth gathered momentum at all levels, with private school share at secondary level reaching 28% in 2002. We also examine the nature of private school growth at primary, upper primary and secondary levels across the regions in our sample. This is shown in Table 3. As shown in Table 2, the share of recognised PUA schools is significantly higher at the secondary level (relative to primary and upper primary levels) over the period 1992-02. Table 3 highlights the pronounced inter-regional variation in the rate of private school growth. We classify all districts into five regions, namely, east (Assam, Bihar, Orissa, West Bengal (WB)), west (Gujarat and Maharashtra), north west (Punjab and Haryana), north (Madhya Pradesh (MP), Rajasthan and Uttar Pradesh (UP)) and south (Andhra Pradesh (AP), Karnataka, Kerala and Tamil Nadu (TN)). In general, the rate of private school growth is relatively lower in the eastern states, especially at the primary and the upper primary levels. At 17 It should be borne in mind that at the primary school level the share of private schools in total schools appears lower than the true private share because there are no data on the private unrecognised schools. Kingdon (2008) shows estimates suggesting that a high proportion of private schools at the primary level remain unrecognised. Thus, our estimate of the private share of total primary schools is an underestimate.

9 the secondary level, the highest share of PUA schools is found in the socially backward northern states, namely MP, Rajasthan and UP, which are generally known for failing government schools (see Dreze and Kingdon, 2001). 2.2. Literacy rates and gender gap Unfortunately AISES data do not provide information on any learning outcomes. Hence we combine 1992 and 2002 AISES data with age/gender specific literacy data available from the 1991 and 2001 Census data respectively. According to the Indian Census definition, a person is considered to be literate if s/he is able to read and write in any language. Additional information, e.g., knowledge of English (e.g., Census 1941-1971) are available in some Census. In the absence of any better alternative, we consider the ability to read and write in a given language as a learning outcome in our analysis. Our analysis focuses on children aged 10-19 years. This choice has been guided by the fact that we could not obtain literacy rates for primary school age children 5-9 years old. While 10-14 literacy rates correspond broadly to literacy rates for upper primary level of education, those for 15-19 correspond to that for the secondary level. We also analyse the rate of growth of literacy rate for 10-19 years old taken together, and we do so for both male and female children. As before, we classify our sample into five regions, namely, east, west, north1, north 2 and south; this allows us to consider private school share and literacy rates not only for the whole of India, but also for the sub-regions in our sample. Table 2 shows the literacy progress at primary, upper primary and secondary levels between 1992 and 2002, while Table 3 presents the male and female literacy rates for 10-14 and 15-19 age groups across the regions. Not surprisingly, literacy rates are lower for female children, in both the 10-14 and 15-19 age groups. The gender difference is significantly higher in the worse performing regions, e.g., see eastern (comprising of Assam, Bihar, WB and Orissa) and northern zones 2 (comprising of UP, MP and Rajasthan). Compared to the national average, age/gender specific literacy rates are lower in these two regions and higher in the west, south and north 1 (Punjab and Haryana) regions. 3. EMPIRICAL ANALYSIS As set out in the introduction, we empirically model literacy rates as well as the gender gap in literacy rates, among children aged 10-19 years old. We also split 10-19 years old children into upper-primary (10-14 years) and secondary (15-19 years) school age groups and repeat the analysis separately for each age group. This allows us to explore the difference, if any, in the estimates between upper-primary and secondary school age group

10 children. 3.1. Modelling universal literacy and gender gap Our central objective pertains to the determination of literacy rates (L ilt ) and gender gap in literacy rates (G ilt ) in the i-th district for a particular schooling level l in year t; l refers to upper primary (10-14), secondary (15-19) or the pooled category (10-19 years) while t refers to 1992 and 2002. We start with the conventional pooled OLS model to determine Y (indicating the variables of interest, literacy or gender gap in literacy) of the i-th district at level l, l=10-14, 15-19 or 10-19: Here P is the share of private school at level l while D is the year dummy that takes a value 1 for year 2002 and 0 for 1992. X refers to the set of other control variables (see discussion below and Appendix Table A1 for variable definitions). However the problem with the OLS estimates is that it cannot control for omitted factors and hence may produce biased effect of private school share on literacy and gender gap in literacy. Further there could be bias arising from reverse causality: just as private school share may affect literacy, existing literacy in a district may also affect private school share; also, a priori, it is not possible to indicate whether the estimation bias will be positive or negative. One option is to determine changes in literacy ( L) and gender gap ( G) in literacy over the period 1992-2002 at a given level (upper primary, secondary or both pooled) as functions of 1992 levels of private school share and other X variables as follows: (2) (3) Where P il is the level of private school share in the initial year 1992 at schooling level l (as defined above); also the set of all other explanatory variables X (see discussion below for a description of the control variables) refers to the initial period 1992. This approach may minimise the bias arising from reverse causality: in order for reverse causality to bias our estimates, private investors need to anticipate future literacy ten years in advance, which could be ruled out without much loss of generality. Since we have data for only two years, after differencing, this two-period information set is reduced to a single cross section one at the district-level; hence we omit the time subscript from equations (2) and (3). However the potential problem of omitted variable bias still remains. Also, in this case we only determine the changes in literacy and gender gap in literacy rather than level of literacy and gender gap in literacy as such.

Hence our analysis focuses on obtaining the fixed effects OLS estimates using this two-period panel data at our access. We start with the most general specification for determining literacy (L) and gender gap in literacy (G) as follows: µ (4) µ (5) Here P ilt is the share of private unaided schools (in total schools) at the l-th school level (l=10-14, 15-19 or 10-19 year olds) in district i at time t. In addition we control for other possible factors affecting literacy and gender gap in literacy rates in our sample. In particular, the set of control variables X includes adult (25-49 years) literacy rates, share of urban population, proportion of scheduled caste (SC) and scheduled tribe (ST) population, ratio of female to male child 0-6 year olds and also supply of schools per child at the given level with a view to minimise the omitted variable bias as far as possible. Given the close link between literacy and earnings, we consider adult (25-49 years) literacy rates to be a good proxy for income or wealth. Since scheduled caste and scheduled tribe population are more disadvantaged than the general population and are also over-represented in Indian poverty, these SC and ST variables would also proxy for poverty. Since urban literacy rates are often much higher than the rural literacy rates in the Indian context, we include share of urban to rural population as a proxy for urbanisation with a view to explore its effects on literacy and gender gap. 18 Son preference may also play an important role in parental allocation of resources for education and other accounts. In the absence of a better alternative, our measure of son-preference is the district ratio of surviving female to male children in the 0-6 age-range. Furthermore, it could be that in districts where there are more private schools, the overall supply of schooling is greater and that, access to schooling is greater. If so, the private school effect could capture an effect of schooling availability. In order to eliminate this possibility, we include number of total schools (at the relevant level, upper-primary, secondary or both pooled) per 100 children as an additional explanatory variable. Given the multi-level nature of our data, we include both district ( 1, 2) and year-specific (µ,µ fixed effects respectively in equations (1) and (2); the remaining errors are captured by (ε 1, ε 2 ), which are independently and identically distributed. Use of panel data fixed effects models allow us to obtain consistent estimates net of time-invariant unobserved heterogeneity. 19 11 In particular, both private school growth and literacy may be influenced by some unobserved time-invariant factors like district s culture, institutions, labour 18 Note that our results remain unchanged even when we exclude the rate of urbanisation variable. 19 Nevertheless there could remain some time-varying unobservables like culture, institutions, gender and caste relations, which may also influence literacy and gender gap in literacy. We however hope that a decade is not too long for these socio-cultural/institutional changes to affect literacy, accounting for which is a beyond the scope of the paper. Even if these changes are faster in some districts, we hope that on an average this variation will cancel out each other in aggregate in our sample.

12 market characteristics, gender and caste relations; the resultant estimates would be biased if these unobserved factors are correlated with the error term, thus justifying the use of panel data fixed effects models. Given that we have only two data points for each district, these fixed effects estimates are also equivalent to the underlying first difference estimates of changes in literacy in terms of changes in private school share and also changes in other X variables (that eliminates time-invariant unobserved heterogeneity). There is a further consideration. Clearly the fixed effects estimates ignore the effects of various timevarying factors that may also influence the variables of interest, namely, literacy and gender gap in literacy. While the effects of time-varying unobservable factors like culture, institutions, gender and caste relations may be ignored as these factors change rather slowly, there are various on-going policies, e.g., District Primary Education Programme (DPEP) initiated in 1994 in selected districts and Sarva Shiksha Abhiyaan (SSA) initiated in 2001 in all districts of India. DPEP was a collaborative effort between the Government of India and different international aid agencies, especially the World Bank while SSA has been a flagship programme of the government of India to secure education for all. While both programmes may influence literacy, SSA initiated in 2001 may not have a pronounced effect in our sample districts observed over 1992-2002; hence we focus on DPEP as a confounding factor in our analysis with a view to identify the pure private school effect (see section 4.2 for further discussion). 4. RESULTS We start with the simplest pooled OLS estimates literacy and gender gap in literacy before moving on to the alternative OLS estimates for changes in literacy and gender gap in literacy in terms of initial values of the private school shares among other covariates. However our preferred estimates are the FE-OLS estimates which controls for the observed time invariant omitted factors; we also check the robustness of these estimates by considering the potential effect of DPEP on literacy and gender gap in literacy not only in the full sample, but also among SC/ST children, who are generally poorer than the total population. 4.1. Pooled OLS estimates of literacy and gender gap in literacy Pooled OLS estimates of literacy and gender gap in literacy are summarised in Table 4. After controlling for all other factors, greater private school share is associated with significantly higher literacy for 10-14, 15-19 and also 10-19 year olds; private school share does not however have a significant association with gender gap in literacy. Underlying partial correlations are 0.13, 0.10 respectively for 10-14 and 15-19 year old children in our sample while it turns out to be 0.15 for 10-19 year olds when pooled together.

13 4.2. OLS Estimates of changes in literacy and gender gap in literacy Next we consider the robust ols estimates of models (2) and (3) determining changes in literacy and gender gap in literacy over 1992-02 in terms of lagged (1992) values of all explanatory variables, as summarised in Table 5; the underlying argument is that a decade-lagged value of private school share in 1992 can be treated as exogenously given for determining the changes in literacy (or gender gap in literacy) over 1992-2002. Clearly, greater private school share is associated with higher literacy and lower gender gap for all the age groups concerned. The underlying partial correlation is comparable to pooled OLS estimates for 10-14 year olds while it is about 4 percentage point less for 15-19 year olds in our sample. Taken together, there is suggestion that greater private school share in 1992 is associated with about 11 percentage point higher 10-19 literacy rate over the period. 4.3. FE-OLS estimates Given the potential bias of OLS estimates arising from the omitted factors, we shall now focus on the FE-OLS estimates of literacy and gender gap in literacy because these estimates control for the time-invariant omitted factors. These estimates for literacy and gender gap equations (4) and (5) are summarised in Table 6 (full set of estimates are shown in Appendix Table A2). All standard errors are robust to clustering at the district level. It follows that higher private school share is associated with significantly higher literacy for all age groups while it is associated with significantly lower gender gap in literacy only among 10-14 year old children. Clearly, the literacy effect of private school growth is most pronounced for the younger age-group, 10-14 year olds, who naturally benefitted more from the recent trend of private school growth around the country (this age group started school at a time when the private school growth picked up). In addition, adult literacy rates (25-49 years) tend to boost literacy and lower gender gap in literacy in all the relevant age groups (10-19 years old) that we consider. It also follows that districts with higher share of ST population experienced higher literacy during 1992-2002 while the effect of SC population has generally been insignificant in our sample. Rate of urbanisation also fails to have any significant effect on literacy. 20 It has often been argued that the necessity of being accountable to parents causes private schools and teachers to apply more effort. The notion that private management of schools leads to higher teacher effort is supported in some recent study on India. For instance, using data from 20 Indian states Muralidharan and 20 Note however that our result remains unchanged irrespective of whether we include the urbanisation variable or not.

14 Kremer (2008) find that within the same village, teacher absence rate in private schools is about 8 percentage points lower than in government schools. This is similar to the findings in Kingdon and Banerji (2009) for Uttar Pradesh and Bihar. More generally, our findings of a positive private school effect on literacy, which is a measure of cognitive skills, especially for 10-14 year olds are consistent with a growing body of literature that finds similarly, using data from different sources and using different methods (Desai, et al, 2008). A comparison of the first difference results with FE-OLS estimates is useful here. As with the first difference estimates, FE-OLS estimates highlight a favourable effect of private school share on changes in literacy over 1992-2002. However unlike the first difference estimates (Table 5), the private school effect on gender gap remains insignificant in FE-OLS estimates, which controls for unobserved time-invariant factors. Note also that compared to the level effects of private school share as shown in Table 6, the size of first order private school effects in Table 5 is somewhat smaller for both literacy and gender gap in literacy in literacy, which is to be expected. 4.3. DPEP effects As indicated in section 3, the fixed effects estimates may still suffer from bias generated by the omitted timevarying variables. In an attempt to address this bias, we now attempt to isolate the potential effect of the government of India s universal primary schooling programme, the DPEP, which operated between 1994 and 2000. DPEP operated in phases. The programme started with 120 districts in 1994; subsequently it included 52 more districts in 1996, 38 in 1997, 34 in 1998, 20 in 1999 and a further 108 in 2000. While the length of the operation varied across the districts, given that our sample includes two years 1992 (before the start of DPEP) and 2002 (after the introduction of DPEP), we pool all the districts together to generate a binary variable indicating whether a sample district was selected by DPEP. While there was no district under DPEP in 1992, about 45% of our sample districts came under DPEP operation by 2002 which includes 14 of our 16 states (excluding only J&K and Punjab). We also argue that there has been significant variation in private school growth in DPEP and non-dpep districts between 1992 and 2002, which we exploit to identify a causal effect of private school growth over 1992 and 2002.

15 We start by exploring the relationship between DPEP and private school growth in our sample and in doing so, we classify the sample districts in three possible ways: (i) districts with /without DPEP; (ii) districts with and without private unaided schools for 10-19 year olds and (iii) districts with and without access to both DPEP and private schools. Table 7 summarises the selected mean characteristics of these districts in our sample pooled for 1992 and 2002. Clearly DPEP districts tend to have higher share of private schools on average, but lower literacy and higher gender gap than those without DPEP. This is also reflected in a positive correlation between DPEP and private school share which varies somewhat with the level of schooling. These correlations are about 0.18 and 0.19 respectively at the upper-primary and secondary levels and aggregate to about 0.19 for both levels pooled together. Second, considering the districts with/without private school growth, clearly districts without access to private schools have lower literacy and higher gender gap compared to those with access to private schools. Finally, we consider the districts which are not only selected for DPEP, but also have access to PUA schools. These districts tend to have characteristics which mirror those of DPEP districts in that they have higher share of private schools, but lower literacy and higher gender gap. Given this positive correlation between access to DPEP and private school share, if the DPEP raised literacy rates then the private school share variable would pick it up and thus in the absence of a suitable DPEP control 21 would generate an upward bias on the coefficient of the private share variable in FE-OLS estimates. Since the available DPEP information is time invariant in our sample (see footnote 20), we now turn to difference-in-difference (DID) estimates using pooled data for 1992-2002. Following the literature on impact evaluation, one can define the DPEP districts as the treatment group and others as the control group. Accordingly, we construct a treatment group dummy DPEP, which takes a value 1 for the DPEP districts and 0 for others. A second treatment arises from the private school growth in our sample while a further treatment would pertain to the private school growth in DPEP districts (DPEP*PUA). Further, given the data for the two years 1992 and 2002, we generate a period dummy D that takes a value 1 for 2002 and 0 otherwise to account for any time trend indicating different unobserved changes influencing literacy over the decade 1992-2002. Thus in order to isolate the true effect of DPEP from private school growth, we include a number of interaction terms with the year dummy D: (i) DPEP*D, (ii) PUA*D and (iii) DPEP*PUA*D. Differential effect of the three treatments, namely, DPEP, PUA growth and both DPEP and PUA growth in our sample would in effect rely on the statistical significance of these three interaction terms (i)-(iii), which forms the basis of the DID 21 We are unable to include a DPEP control in the FE-OLS estimate as we only observe a time-invariant DPEP variable for our sample of 1992-2002.

16 estimates. As before, we retain the other control variables X (as in equation 3) 22 while estimating the literacy in the i-th district at a given age level l (where l=10-14, 15-19, 10-19) as follows: (5) We also estimate a second equation determining gender gap in literacy for a given age level l, as before. All standard errors are clustered at the district level. Finally, using the pooled sample of all districts for these two years 1992 and 2002, we obtain the robust OLS estimates of literacy and gender gap in literacy as summarised in Table 8. The coefficients of interest are,, and of the three interaction terms with the year dummy D, which forms the basis of the difference-in-difference (DID) estimates using a two-period pooled sample. The DID estimates shown in Table 8 helps us to clarify the FE-OLS estimates discussed earlier. While captures the differential effect of DPEP, shows that for PUA and finally, highlights the differential effect of private school growth DPEP districts in our sample. First, there are significant DPEP effects on literacy in that literacy significantly increases by about 2% in DPEP districts for all 3 age groups concerned; there is also evidence that DPEP significantly reduces gender gap for 10-14 year olds by about 2.2%. Second, we consider the differential effect of private school growth as highlighted by the estimate of. There is evidence that private school share is associated with higher literacy among both 10-14 and 15-19 and therefore 10-19 year olds in 2002. Estimated marginal effects are respectively 0.16, 0.21 and 0.09 respectively for 10-19, 10-14 and 15-19 year olds. In other words, as with FE-OLS estimates, the greatest return of private school share is obtained for 10-14 year olds, though the marginal returns tend to decrease as we control for DPEP. Private school growth also significantly reduces gender gap among 10-14 year olds in 2002. There is however no significant evidence that DPEP districts enjoy a significantly higher literacy or lower gender gap in literacy from private school growth, as the estimated coefficients remain insignificant for all cases. 4.4. Case of SC/ST children One may also argue that the effect of private school growth on overall literacy may blur what happens to the marginalised group of low caste population, who are far behind the general population in terms of literacy achievements. We use 2001 Census literacy data for SC/ST children aged 10-19 years old to examine if there is 22 In this case, we also include the state dummies to allow for variation in state-level policies and programmes these districts.