learning achievement in primary education in...

43
Learning Achievements in India: A Study of Primary Education in Rajasthan Sangeeta Goyal South Asia Human Development The World Bank Human Development Unit South Asia Region May 2007 Document of The World Bank

Upload: trankhanh

Post on 03-Feb-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

Learning Achievements in India: A Study of

Primary Education in Rajasthan

Sangeeta Goyal

South Asia Human Development

The World Bank

Human Development Unit South Asia Region May 2007 Document of The World Bank

Preliminary version: for comments

Abstract

This paper presents findings from a study of learning outcomes in grades IV and V of government, private aided and private unaided schools in Rajasthan. Approximately 6000 students were tested in 200 schools in three tests – two language tests (Reading Comprehension and Word Meaning) and one test in mathematics. The survey also collected information on student, family background and school characteristics. The survey results showed that overall learning levels were low absolutely and relatively in government schools. The average percentage correct scores in government schools ranged from 40-50 percentage points, a quarter to a fifth below the average scores in private schools.

The analysis of determinants of learning outcomes provided a number of

important insights. Firstly, the school attended by the child has the most substantive impact on the quality of learning. School fixed effects account for more than half the variation in test scores. Once we take school fixed effects into account, the type of school management and other school related characteristics lose all explanatory power. Secondly, private schools, whether aided or unaided, outperform public schools. Thirdly, there is large variation in the performance of public schools - a section of public schools has better test scores than the representative private school.

Findings from the study provide directions for policy interventions and for future

research for more evidence-based policy making. The variation in public school performance and the dominance of school specific factors in explaining test scores imply the importance of raising quality of schools in the public sector. Moreover, not only are the learning outcomes low, learning gains from one grade to another are flat with nearly constant and large dispersion of scores around the mean in both grades. Therefore, to achieve any given learning target, improving school quality would require increasing the amount of incremental learning that takes place in each grade. Government schools also do not use their resources effectively leading to inefficient allocation of public resources to provide education. Teachers in private schools get much lower salaries than teachers in government schools and private schools are 1.5-2 times more cost-effective than government schools in terms of learning gains per rupee. This indicates there is much room for improving the cost-effectiveness of public sector education provision. Among other determinants, we find that social group, household wealth and mother’s literacy have significant but small impacts on test scores. Note: The survey work for this study was funded by the EPDF Trust Fund (Project ID: P0554559-SPN-TF054642).

2

Preliminary version: for comments

1. Introduction

Countries seeking to increase the level and pace of economic growth, and to raise the

productivity and earnings of their citizens, have increasingly focused on increasing the

quantity and quality of their people’s educational attainment. Consequently, growth in

school enrollment has been phenomenal across the world in the last four to five decades.

However, even as the quantity of education has increased over time, the quality of

education, especially primary education, remains a cause for serious concern. The

experience of many developing countries including India is that children do not master

basic literacy and numeracy even after four and five years of schooling.1

In this paper we examine the determinants of learning achievements of students of

grades IV and V in language and math in government, private aided and private unaided

schools in Rajasthan. In India as in most developing countries, the public sector is the

dominant provider of primary education. Government managed and financed primary

schools are in principle ‘freely’ accessible by any child of school going age. According to

official data, nearly ninety percent of all primary school going children in India attend

government schools. Alongside free public education, there is a growing sector of fee-

charging private unaided schools. These schools are managed and financed privately,

often along profit-making principles. Children, even from poor families, are attending

these schools in large and increasing numbers. There also exists a hybrid variety of

schools in India known as government aided/private aided primary schools. These

schools are managed by the private sector but largely financed by the government.2

3

1 The Sarva Shiksha Abhiyan (SSA) or Universalization of Elementary Education Mission, which is the flagship mission of the Government of India in the education sector, was introduced in 2001 to ensure all eligible children go through eight years of schooling. In the first few years of the program, the focus was on improving access to schools, increasing enrollments and reducing drop-out of children from elementary grades. Increasingly, SSA is now focusing on quality of education in schools, including teacher presence and activity in class-rooms, teacher training and assessment systems. 2 All the schools that are a part of this study have ‘recognition’ from the government. There is also a fast growing sector of fee-charging private unaided unrecognized schools.

Preliminary version: for comments

According to the Census of India 2001, the western state of Rajasthan had a

population of 56.5 million with a literacy rate of 60.4% as compared to the national

average of 64.1%. Of the 9.96 million children in the age-group 6-14 years enrolled in

school in 2004-05, nearly 22% went to private schools. Historically, Rajasthan has had

lower than average economic growth and human development indicators.

This study is based on primary data collected in the state in early 2006. We use

percentage of correctly answered questions in language and mathematics tests as proxies

for cognitive competencies in literacy and numeracy, the true underlying learning

objectives. Because we use percentage scored in any particular test and not acquisition of

particular competencies to rank performance, the data is better suited for drawing

inferences about the relative effectiveness of different determinants of learning

achievements. In Section 2, we review the theoretical and empirical literature on the

determinants of learning outcomes pertinent to the Indian context. In Section 3, we

describe the sampling methodology, the data and the analytical framework used in this

study. In Section 4, we report the unadjusted average learning levels across school types,

genders, social groups and rural-urban locations. Section 5 provides the results from our

empirical analyses of the determinants of learning outcomes. In Section 6, we take a look

at the labor market for teachers in Rajasthan. Section 7 provides some policy implications

based on the findings of this study and concludes. We will use the terms government

school and public school interchangeably in this paper.

2. Background and Previous Literature

The question of how to improve the quality of educational attainment in schools has

become one of utmost importance to policy-makers. It is generating a large body of

research, previously in developed, but now also in developing countries. Most empirical

studies of determinants of learning achievement relate measurable school characteristics

and student and family background characteristics to learning outcomes.

4

Preliminary version: for comments

A number of studies show that school attended (school fixed effects) explains a large

amount of the variation in learning outcomes. Das et al (2005) in their study of primary

schools in Pakistan find that nearly 50% of all the variation in test scores in Pakistan can

be attributed to school fixed effects. A study similar to this one for the state of Orissa also

shows that between 50-60% of the variation in test scores is determined by school fixed

effects (World Bank, forthcoming)

School fixed effect plausibly captures (observable and unobservable) dimensions

of school quality. Standard proxies for school quality used in the literature are school

inputs such as pupil teacher ratio, the use of multi-grade classes, quantity and quality of

school infrastructure, teacher numbers and characteristics, provision of mid-day-meals

etc. The relation between observable schooling inputs and student outcomes however is

not consistent and in general weak in most studies. Empirical evidence from developed

countries generally does not find any effect of pupil-teacher ratio. Lavy and Angrist

(1999) for Israel and Urqiola (2006) for rural Bolivia, however find that smaller class-

sizes benefits students learning attainment. Regarding the use of multi-grade classrooms,

the general belief is that they are detrimental to learning though research from outside

India is non-conclusive (Miller, 1990). There are few studies that include the share of

graduate teachers and the share of non-regular teachers as controlling characteristics for

schools. It is difficult to predict the direction of the net effects of these characteristics.

Teachers with higher educational qualifications and more secure employment can be

expected to be more motivated to perform. There is also evidence that they are also more

prone to be more absent from schools (Chaudhury et al, 2004).

The type of school management, i.e. whether the school is a government, private

aided or private unaided school, has also been found to be a significant predictor of

educational outcomes in the Indian context. Access to schools is a necessary but not a

sufficient condition for ensuring the development of cognitive competencies. According

to the empirical evidence, private unaided schools in general outperform public schools

5

Preliminary version: for comments

(Kingdon, 1996; Smith et al, 2005; Tooley and Dixon, 2006). Few systematic studies

compare private aided schools quality with other types.

That individual student and family background characteristics influence school

outcomes even after controlling for school related factors is undisputed, even though the

research does not provide conclusive evidence regarding effects. Some studies find that

boys and children belonging to the upper castes perform better (Dreze and Kingdon,

2001; Aggarwal, 2000; Filmer et al, 1997). Household wealth and parents’ education

also have positive correlations with children’s educational outcomes (Pritchett and

Filmer, 1998).

3. Data Description and Empirical Methodology

Primary data was collected for the study in the three months from February to April

2006 by A C Nielsen ORG MARG on behalf the World Bank and in cooperation with the

Government of Rajasthan (GoR). Eight districts out of the 32 districts in Rajasthan were

chosen for the study, in discussion with government officials, to be representative of

accepted stratification of the state. Two blocks were randomly chosen from each district

and 25 schools were randomly chosen from each block – 12 schools from one block and

13 schools from the other. The 12-13 schools were distributed across the categories of

government, private aided and private unaided schools in the ratio 6:2:2. Where adequate

number of private aided and unaided schools was not available, the remaining schools

were replaced by government schools. Private unaided schools were restricted to those

that have government recognition. The schools were distributed such that there were 8

rural schools for every 2 urban schools.

A maximum of 30 students from grades IV and V were tested in each school, 15

students being randomly chosen from each grade. If more than one section of a grade was

available, then first a section was chosen randomly, and then students were randomly

selected from it. If fifteen or fewer children were present in a grade, then all of them were

included in the sample.

6

Preliminary version: for comments

Both grades IV and V students were administered the same tests in three subjects –

two language tests and one mathematics test. The two language tests were a reading

comprehension test and a word meaning test used by the State Council of Educational

Research and Training (SCERT) to test students in grade IV. The SCERT tests were the

same tests used by the National Council of Educational Research and Training (NCERT).

The mathematics test was a sub-sample of curriculum consistent questions selected from

the TIMMS 2003 Mathematics test for grade IV.

Tables 1 – 4 in the Annex provide descriptive statistics of the data used in this study

based on a number of other questionnaires that were also administered to collect

correlative information. These included:

(a) School Questionnaire: This questionnaire collected information on school and teacher

characteristics.

(b) Student Questionnaire: This questionnaire collected information on student and

family background characteristics.

Analytical Framework

We use a two-pronged empirical strategy to analyze the determinants of learning

achievement.

(1) In the first case, we use a ‘panel’ approach whereby we model the achievement of

student i in school j as a function of individual and family background

characteristics , a school fixed effect term and a random error term

ijY

ijX jz ijε . We

are able to do this because we have multiple observations from the same school.

ijjijij zXY εβα +++= ; where [A] ),0(~ 2σε Nij

This ‘panel’ strategy should in principle give us unbiased and consistent estimates

of individual and family characteristics.

7

Preliminary version: for comments

(1) In the second case, we model the achievement of student i in school j as a

function of individual and family background characteristics , a vector of

schooling resources which is constant across students from the same school,

and a random error term

ijY

ijX

jS

ijε such that

ijjijij SXY ελβα +++= ; where [B] ),0(~ 2σε Nij

We cannot have the school fixed effect and the observable school characteristics in

the same equation because in that case there is likely to be ‘perfect-collinearity’.

4. Learning Levels: Differences in educational attainment by School Type,

Gender, Social Group and Rural-Urban Location

In this section, we provide the unadjusted learning levels of students in grades IV and

V along a number of dimensions: school type, gender, social group and rural-urban

location of schools.

Table 4.1 below shows the means and standard deviations of the percentage scores for

the three tests for grades IV and V. Students in grade IV score below 50% in Reading

Comprehension and Math tests. Students in grade V have higher scores in all three tests

but the difference between the subject mean scores for the two grades is always less than

10 percentage points. The average gain in grade V is highest in reading and lowest in

word comprehension.3 The dispersion in test scores, measured by the standard

deviations, is very high – almost 20 percentage points – and is greatest for reading and

lowest for the word meaning test in both grades IV and V.

8

3 This gain is not true gain as the set of students whose scores constitute the average are different in the two grades.

Preliminary version: for comments

Table 4.1: Mean Scores in Tests for grades IV and V

Read Word Math Percentage (%)

Mean SD Mean SD Mean SD Grade IV 45.65 23.5 55.71 18.9 44.28 19.99 Grade V 53.31 23.28 61.76 18.53 51.12 19.97

School Type Differences: Table 4.2 shows the average percentage scores in reading

comprehension, word meaning and mathematics test by grade for each of the three school

types. Government school students have lower average scores than private unaided and

aided school students in both grades, ranging between 42-59 percentage points.

Government school students score between 2 and 6 percentage points less than private

unaided school students and between 9 and 16 percentage points lower than private

unaided school students in grade IV. In grade V, government school students score 8-9

percentage points lower than private aided schools in all three tests, and between 8-9

percentage points lower than unaided schools in the two language tests, and about 4

percentage points lower in mathematics. Private unaided schools catch up much more

with private aided schools in grade V compared to government schools.

The table also shows the standard deviations of subject scores (in parentheses).

Compared to the mean, the standard deviations are large across school types and across

grades. For government schools, large standard deviations combined with a small mean

imply that very little learning takes place for those children who are even one standard

deviation below the mean. For private schools, a large standard deviation combined with

a relatively large mean implies that students who score a standard deviation or more

above the mean score have very high scores. Later in the paper, we will see that school

specific differences account for most of the variation in test scores, and only a few

percentage score points remain unexplained due to unobserved within school differences,

after taking into account students background and school characteristics. This is true of

variation in scores for schools within a school type and across school types.

9

Preliminary version: for comments

Table 4.2: Mean Percentage Scores by School Type

Mean Percentage Score (S.D.)

Grade IV School Type Read Word Math

Government 42.8

(22.86)53.66

(23.84)42.34

(22.53)

Private Aided 58.34

(18.36)62.72

(19.87)54.72

(19.17)

Private Unaided 49.51

(19.91)59.43

(20.39)44.64

(17.44) Grade V

Government 50.63

(22.99)59.38

(22.12)49.25

(22.54)

Private Aided 59.71

(17.76)67.17

(17.92)58.17

(19.29)

Private Unaided 59.37

(19.81)67.25

(21.72)52.85

(17.10)

Gender Differences: There are virtually no gender differences in performance in all

tests in both grades as can be seen from table 4.3 below. In fact, the unadjusted scores of

girls are better than that of boys, albeit marginally. Figures 2 and 3 in the Annex also

show the unadjusted scores of boys and girls by school type. In math test in grade V, girls

in private aided and unaided schools score at least 5 percentage points more than boys.

What is also clear from the figures is that differences in scores between school types

outweigh differences in scores between boys and girls within any school type. This is true

of all tests in both grades.

Table 4.3: Mean Percentage Scores by Gender

Mean Percentage Score Grade IV

Read Word Math Boys 45.46 55.7 44.28 Girls 45.86 55.72 44.29 Grade V Boys 53.23 61.92 50.5Girls 53.41 61.58 51.84

Social Group Differences: There are relatively larger differences in performance

between children belonging to SC and ST on the one hand, and those belonging to

10

Preliminary version: for comments

General and OBC categories on the other. This is true for all three tests in both grades. In

grade IV, SC and ST score on the average 6-8 percentage points lower than

General/OBC. In grade V, the gaps narrow between SC and others, but widen for ST. The

score gaps between SC and others ranges from 3-6 percentage points and between 8-12

percentage points between ST and others. The mean scores for the different social groups

by test and grade is provided in table 4.4 below.

Table 4.4: Mean Percentage Scores by Social Group

Mean Percentage Score Grade IV Read Word Math SC 42.61 54.25 42.28ST 40.05 53.75 38.6OBC 48.14 56.46 46.24General 48.24 57.35 46.86 Grade V SC 51.68 60.95 49.46ST 46.91 59.5 43.49OBC 54.71 62.28 52.77General 57.51 63.44 55.72

Figures 4 and 5 in the annex also show the average percentage scores in the three tests

in the two grades by social group and school type. The figures confirm the following:

• OBC and General category children outperform SC and ST children in both

grades IV and V, and in all three tests. The performance of SC and ST children

are similar to each other and the performance of OBC and General children are

similar to each other.

• The gaps in test scores are narrower for SC and others in grade V.

• For social groups, differences between school types are larger than differences

between social groups in general, except in the case of mathematics scores in

grade V. All students, irrespective of their social group perform worse in

government schools and do better in private aided schools.

11

Preliminary version: for comments

Rural-Urban Differences: Average scores are lower in rural schools in all three tests

and in both grades, though the differences narrow in grade V. Rural-urban differences are

highest for reading comprehension test and lowest for math. Figures 6 and 7 in the annex

show the performance of schools by type and rural-urban location. From the figures, we

can observe that:

• The greater difference is between government and private unaided schools on the

one hand, and private aided schools in rural areas, on the other. Private aided

schools score 12-15 percentage points higher on the average than government

schools and private unaided schools in rural areas.

• The differences in the performance of the three types of schools are smaller in

urban areas.

5. Variations between Schools

In Section 4, we described the unadjusted learning achievement levels and gaps by

grade, gender, social group, rural-urban location and school type. We can use the method

of variance-decomposition of test scores to disaggregate the total explained variation by

source. The remaining which is unexplained variation can be attributed to omitted

variables and noise in the data. Using this method, we can also identify the adjusted

effects of particular characteristics such as school type, gender, social group etc. In a

multiple regression model, the adjusted effect is the coefficient on a particular attribute,

after taking into account all other characteristics.

Many studies find that cognitive achievement in schools can be predicted to a large

extent by the school attended. The effect of school attended can be measured by

including an indicator variable for the school, i.e. by accounting for school fixed effects,

using model A in Section 3 above. To determine the between and within school

variations, we regress test scores on an indicator variable for the school attended. The R-

square for the regression is the between variation, and the remaining is within school

variation. The between and within school variation for the two grades and the three tests

are shown in Figure 5.1 below. For Rajasthan, school fixed effects explain between 45%

12

Preliminary version: for comments

and 72% of the variation in test scores in the two grades. For both grades IV and V,

school fixed effects explains more than 70% of the test scores in mathematics and around

60% of the test scores in reading comprehension; for word meaning, school fixed effects

explain between 45% - 47% of the variation.

Figure 5.1: Between and Within Schools Variation by Test and Grade

Between and Within Schools Variation

0102030405060708090

100

Read Word Math Read Word Math

Class IV Class V

Per

cent

age

BetweenWithin

District as a source of variation in test scores by itself explains only 8-10%. Type

of school management – whether government, private aided or private-unaided – explains

between 3-4%; and the explanatory power of the grade the child attends – whether grade

IV or V – is only 2-3%. Also, district, school type and grade effects are stronger for grade

IV test scores compared to grade V. Once we take school fixed effects into account,

however, district, school type and grade lose any explanatory power.

13

We repeat the variance-decomposition exercise within each school type to compare

the variations in school quality across types and between each type and the overall

variation in test scores. We find that school quality within each type varies different

across the tests making the interpretation of results difficult. Nevertheless, we can draw

the following conclusions:

Preliminary version: for comments

• In general, government schools are the most variable, and private aided schools

are the least variable in quality. However, this does not hold for mathematics

where between 74 – 81% of the variation in test scores is explained by school

attended within the group of private aided schools, more than the other two school

types.

• The variation in school quality shrinks for grade V in private aided and unaided

schools but becomes only marginally lower for government schools.

Impact of School Characteristics

A critical issue in analyzing educational outcomes is how schools affect learning

attainment. School related factors that improve the quality of learning achievement can

provide education policy options. To unpack school quality, we replace school attended

in our OLS regressions by some standard measures of school quality used in empirical

research. This is the second of our two-pronged empirical strategy – model B in Section 3

above. We also include as a determinant the school type by management – i.e. whether

the school is a government, private aided or unaided school. The results are set out in

columns (3) and (6) of Regression tables I, II and III in the annex for grade IV and grade

V for the three tests results respectively. These results control for child and household

characteristics but not for school attended. We find:

• In all the regressions, the coefficients on school type are in the expected

direction. Private aided and unaided schools perform better than government

schools. Even though, the coefficient on private unaided school is significant

in only one regression – word meaning test for grade V – the robust t

statistics are small in all the regressions. The substantive impact of school

type on academic performance is not negligible, and the adjusted gap between

government school and private aided and unaided schools ranges between 5-7

percentage points.

• Anecdotal and more systematic empirical evidence generally document that

schools using multi-grade classrooms for teaching generally perform worse

14

Preliminary version: for comments

than schools that don’t. In the case of our data, we do not find any difference

between schools on the basis of this characteristic.

• Schools with a higher pupil teacher ratio (PTR) perform worse in all three

tests in grade IV, but no differently in tests in grade V. A 10 percentage point

increase in the PTR would reduce average percentage scores by 1-2

percentage points in reading comprehension, word meaning and mathematics

test scores respectively in grade IV.

• The average number of days of teachers absent from school in the previous

academic year has a small – a third to a quarter of a percentage point per day –

negative and significant impact, especially in grade V.

• A higher share of graduate teachers in the school has a small positive but

non-significant effect on test scores. A higher share of teachers who are non-

regular has a very small negative and non-significant effect on test scores.

Overall however, the standard measures of observable school characteristics used

in our analysis explain very little of the variation in test scores, and are unimportant once

we take school fixed effects into account.

Comparing the Distribution of Public and Private Unaided Schools Performance

So far we have compared mean scores of students in different school types. Even

though the typical government school performs poorly in comparison to the typical

private school, there is a lot of variation in performance within the category of any

particular school type. As we saw in section 5 above, most of the variation in test scores

is explained by school fixed effects. Figure 10 in the annex, shows the kernel density

distribution of the average school scores for reading comprehension and mathematics for

government and private unaided schools. We compare only these two school types

because they are ‘pure’ types. The average scores for the schools have been computed by

averaging individual student scores adjusted for child and family background

characteristics. Apart from the density distributions, the panels also show the location of

the median (the left vertical line) and the best (the right vertical line) private school. The

15

Preliminary version: for comments

kernel density distributions show that not all government schools perform badly, and that

there is also a substantial area of overlap between government and private unaided

schools. After adjusting for student and family background characteristics, 44.44% and

60.42% of public schools perform as well or better than the median private unaided

school in reading comprehension and mathematics in grade IV respectively; and 37.76%

and 44.06% in grade V respectively. This is especially true of distribution of schools in

grade V.

From the point of view of policy, the pertinent question to ask from a policy

perspective is how are good public schools different from bad public schools? For our

purposes, we are simply taking the good public schools to be those that have the same or

higher adjusted average scores than the median private unaided school. A comparison of

the mean characteristics of the two reveals that:

• In grade IV, for both language and mathematics, the better public schools

have a lower percentage of non-regular teachers.

• In grade V, for both language and mathematics, the better public schools

have a higher percentage of graduate teachers.

6. The Impact of Child, Family and Social Group Characteristics

As we have seen above, school attended has the maximum impact on test scores

on students. However, even after controlling for school attended, observable student,

family background and social group characteristics have significant, albeit relatively

small effect on test scores.

The regression results that identify child, family and social group characteristics

effects on test scores are provided in columns (2) and (5) of Regression Tables I, II and

III in the annex. In these regressions, apart from the school attended by the child, we

include as determinants the child’s age, gender, the child’s mother’s and father’s literacy,

whether the child lives in a rural or urban area, the social group of the child (general, SC,

ST, OBC), the number of days the child was absent in the week before the interview, and

16

Preliminary version: for comments

an asset index for the household the child belongs to (the construction of the asset index

is described in the annex).

The results of the regressions allow us to compare the unadjusted and adjusted

gaps in test scores for the relevant attribute under scrutiny of the child. To reiterate, the

unadjusted gap is simply the difference in average scores across an attribute such as

gender or caste, whereas the adjusted gap is the coefficient of that attribute in the

regression. For Rajasthan, the findings from these regressions are largely consistent with

expectations a priori and with findings from other studies. We find:

• Age of the child in general has no impact on test scores and has a small

negative impact in two of the regressions: for reading comprehension and for

mathematics in grade 5, age reduces test scores between ½ and a sixth of a

percentage point.

• Gender of the child has no or a very small and insignificant impact on test

scores. Unadjusted gaps are in favor of girls in general. The adjusted gaps for

girls are larger but still insignificant. This is shown in figure 7 in the annex.

• Social group does not seem to have a very strong impact on test scores in the

case of Rajasthan, though the differences across social groups are in the right

direction. Children belonging to the SC category in general score more than

those belonging to ST category and less than OBC and general category

children. SC children score on the average between 1 and 6 percentage points

more than ST children, and between 3 and 6 percentage points less than OBC

and general category children. Figures 8 and 9 in the annex show the

unadjusted and adjusted gaps between the test scores of SC students and other

social groups. Adjusted gaps between SC and ST children disappear and

become insignificant. The adjusted gaps between SC and OBC and general

categories also become much smaller and in most cases become insignificant.

Only in the case of the reading comprehension test in grade IV, the adjusted

gap is nearly half the adjusted gaps between test scores of SC and OBC and

general category students and is significant.

17

Preliminary version: for comments

• We do not find any consistent impact of mother’s and father’s literacy on test

scores. Mother’s literacy matters only for reading comprehension in grade 4.

Children of literate mothers on an average score 1.46 percentage points higher

in the reading comprehension test in grade IV than children of illiterate

mothers. Father’s literacy has an impact only on mathematics score for

children in grade IV. Children of literate fathers on an average score 1.33

percentage points higher in mathematics in grade 4 compared to children of

illiterate fathers.

• The asset index of the household to which the child belongs to has either a

very small, a quarter to a sixth of a percentage point, or no impact on test

scores.

7. The Labor Market for Teachers

Few studies analyze the characteristics of the labor market for teachers in India. In

this market, the government has a near monopoly in providing qualifications and is nearly

a monopsonist buyer since most teachers find employment in public sector schools.

Public sector (and private aided school) teacher salaries and other benefits are set by the

state – through pay commissions and other political processes – using considerations

other than qualifications or productivity.4 Salaries paid to teachers in private unaided

schools are a fraction of those paid to government school teachers, plausibly reflecting

local labor market conditions. Figure 7.1 shows the average salaries paid to regular

government school teachers and teachers in private aided and unaided schools. In

Rajasthan, private aided and unaided school teachers have similar average salaries, which

is approximately a third less than the average salary of a regular government school

teacher.

18

4 The variation in the salaries of regular government school teachers can largely be explained by seniority.

Preliminary version: for comments

Figure 7.1: Average Salary of Teachers by School Type

Average Salary of Regular Teachers

0100020003000400050006000700080009000

10000

Government Private Aided PrivateUnaided

School Type

Rupe

es

Salary (Rupees)

How do these salaries relate to differences in teacher characteristics across school

types? The data shows that the distribution of teacher educational qualifications is similar

between government and private unaided schools in Rajasthan. In both nearly 26% of

regular teachers are non-graduates and the remaining are graduates or above. On the other

hand, the share of graduates among teachers in private aided schools is on the average 7

percentage points greater.

Table 7.1: Distribution of Education: Regular Teachers (%)

Highest Education

Level

Government Private

Aided

Private

Unaided

Elementary 0.56 0 1.57

Secondary 2.66 2.56 3.15

Higher Secondary 16.50 7.69 14.17

Diploma/Certificate 5.87 6.41 6.30

Graduate 33.71 43.50 34.65

Post-Graduate 40.14 39.74 40.16

Other 0.56 0 0

Total 100 100 100

19

Preliminary version: for comments

Compared to government school teachers, private aided school teachers are

overwhelmingly not trained. More than 70% of regular teachers in private aided schools

have not received any pre-service training compared to 44% in private unaided schools

and only 30% in government schools. If we look at all the teachers (regular and non-

regular), then 38% in government schools, 77% in private aided schools and 57% in

private unaided schools have not received any pre-service training as non-regular

teachers are predominantly untrained

A small set of covariates in Mincerian type wage equations – years of experience, the

square of years of experience, age, gender, highest educational qualification, rank, status

(whether regular or otherwise) and rural-urban location explain nearly 60% of the

variation in teacher salaries in each of the school type. However, the set of predictors

vary across the school types. For teachers in government schools experience, status and

teacher rank are the significant predictors of salary. More experienced teachers earn

nearly Rupees 146 more per additional year of experience. Non-regular teachers earn

nearly Rupees 5534 less than regular teachers and teachers who are not head-masters or

their deputies earn Rupees 713 less than those who are. Thus experience, seniority and

status are strongly correlated with teacher pay in government schools. Rural-urban

location made no difference to teacher salary in government schools but in private aided

schools, teachers in rural schools earn Rupees 2050 less than their urban counterparts. In

private unaided schools, female teachers earn Rupees 1075 more than male teachers and

non-regular teachers earn Rupees 2960 less than regular teachers. Other observable

teacher characteristics were not significant predictors of salary variations in these two

school types.

Further analysis of teacher demographics in the various school types shows that while

only a third of teachers in government and private unaided schools are females, they

constitute nearly half the teaching force in private aided schools.

20

Preliminary version: for comments

Teacher Incentives

From the above it is clear that the representative regular teacher in a government

school is relatively older, more experienced and trained compared to his or her private

unaided school counterpart. Then what can account for the better performance of students

in private schools over and above personal and family characteristics, which by

themselves explain only very little? It is generally accepted that teacher incentives are

relatively weak in government schools that leads to poor teacher performance which in

turn results in poor student performance; better teacher performance in private schools,

even at much lower pay, is due to the stronger structure of incentives: private school

teachers can be penalized and even be fired for poor performance by school management

who are in turn accountable to fee-paying parents.

One aspect of teacher performance is teacher presence in schools. There is evidence

of wide-spread teacher absence in government schools in India (Chaudhury et al 2004).

We do not have data in our dataset that allows us to compute true absence rates for

teachers. Absence information in the data-set was reported by the school respondent on a

one time visit basis. The respondent was either the head teacher or a senior teacher in the

school, and the data refers to the number of days in the previous school year the teacher

was absent from school. On the basis of this information, teacher absence behavior was

worse in private aided and unaided schools, and better in government schools. In the

latter the average number of days absent (averaging by summing over all teachers) for

regular teachers was 12 days compared to 24 days for private aided schools and 16 for

private unaided schools. Contract teachers had similar absence behavior in government

schools but non-regular teachers in private aided and unaided schools had half the

number of days of absence as their regular counterparts.

8. Policy Perspective and Concluding Remarks

The analysis of determinants of learning achievement in grades IV and V in

Rajasthan provides important insights for the currently on-going debate on how to

improve the quality of public primary education. Firstly, the school attended by the child

21

Preliminary version: for comments

has the most substantive impact on the quality of learning. School fixed effects account

for more than half the variation in test scores. Once we take school fixed effects into

account, the type of school management loses all explanatory power. Secondly, private

schools, whether aided or unaided, outperform public schools. Thirdly, there is large

variation in the performance of public schools. Nearly a third of the public schools have

average performance better than the median private unaided school. From the point of

policy, this variation in public school performance provides the space for reforms that

will enable the public schools at the bottom of the distribution to perform. Future

research should explore the differences that separate the ‘good’ from the ‘bad’

government schools.

Learning Profiles and Learning Gains

What stands out from Table 4.1 above is that learning profiles are very flat: the

average gain in learning in terms of percentage points from grade IV to grade V for all

the students in the sample are: 7.66 in Reading Comprehension, 6.05 in Word Meaning

and 6.84 in Mathematics respectively. Even if we separate out the learning gains by

school type, gains are still very flat across all, particularly private aided schools where

scores increase on the average by 1-4 percentage points. Average gains in government

schools are between 6-8 percentage points and in private unaided schools are between 7-

10 percentage points. The standard deviations of achievement scores are very high in

both grades IV and V, relative to the mean. Given low mean scores, the implication is

that the students who are located even one standard deviation below the mean are

learning little. Moreover, there is little narrowing of the distribution of scores around the

respective means in the two grades implying that the incremental learning in the higher

grade is nearly constant over the entire distribution of scores.

The location and shape of the distribution of test scores has implications for

policy interventions aimed at improving the quality of education. Learning outcomes can

be improved in at least three ways: (a) better students, (b) more school years, (c) and

more learning in each grade.

22

Preliminary version: for comments

(a) Better students: We can expect three types of sorting taking place that can impact on

learning outcomes – sorting within schools where the better ability students progress

through grades, sorting across schools within a particular school type, and sorting across

school types. One way to deal with the issue of selection into schools is to offer school

choice by way of say school vouchers. Findings from this study show also that student

characteristics as we have seen above explain little of the variation in scores, most of

which is driven by school specific factors.

(b) More school years: The dispersion of scores in Table 4.2 above, in each grade and

relative to learning gains, is very high. If we assume a linear learning profile, then

everything else remaining constant, a child in the fourth grade of a government school

who is one standard deviation below the mean will take approximately six more years to

reach one standard deviation above the mean in each test (Read: 46/8 ≈ 6; Word: 38/6 ≈

6; Math: 40/7 ≈ 6).

(c) More learning in each grade: Currently, the amount of incremental learning taking

place in each grade is very low. If the ideal situation is one where students on reaching

grade V have mastered the intended curriculum for grade IV, then based on the findings

of this study, the average shortfall (=100-Mean Score) of 50 percentage points declines

only by 12-16% for the three tests in government school in the higher grade.

There is little education policy can do to improve the social background of

students – in the long run, economic development may be the best input into the

production of the quality of learning. Ensuring more school years is an untenable policy

intervention because with a given quality, it will take an infeasible number of years to

achieve any learning outcome target. The best option for policy makers is to steep-en the

currently flat learning profile – so that learning profiles in each grade approximate more

closely the shape of curves in Panel (B) below.

23

Preliminary version: for comments

Figure 8.1: Learning Gains in Government Schools Panel (A): Current Learning Profile Panel (B): Ideal Learning Profile

0102030405060708090

100

Grade IV Grade V

ReadWordMath

0102030405060708090

100

Grade IV Grade V

ReadWordMath

The objectives of education policy reform needs be to (a) improve the performance of

schools and (b) to keep costs down. Therefore, any education policy reform in the Indian

context will have to involve teacher quality. Teachers are the main input into the

teaching-learning process. Private schools perform better than public schools as is

evidenced by the better performance of their students. Salaries not only constitute the

largest share of the recurrent expenditures of public schools, but private school teachers

earn a fraction of the salary of public school teachers. It is not the personal

characteristics of the teachers but the incentives that are offered by the two school

systems that plausibly determine their behavior, which in turn determines teacher quality.

Most private school teachers do not have any training unlike regular public school

teachers, most of whom have at least pre-service training. Public school teachers also

have much higher experience in the profession on an average than private school

teachers. The higher education levels of private school teachers and their choice of low-

paying employment in private schools plausibly reflects the labor market conditions.

Government schools do not make use of their resources – mainly teachers, effectively,

and this is linked to technical or allocative inefficiency in the use of given resources. The

formal condition for the optimal allocation of resources is to equalize learning gains per

rupee for all inputs. In government schools, teacher salaries constitute the largest item of

expenditure on school resources. The table below shows the average learning gain from

grade IV to grade V by school type, divided by the average teacher salary for that school

24

Preliminary version: for comments

type. For ease of interpretation, the resultant ratios (shown in columns (1), (2) and (3))

were multiplied by 1000. As can be seen in the last column in the table (column (4)),

private unaided schools were nearly twice as more cost-effective than government

schools, implying that public school teachers earn considerable rents.

Table 8.1: Cost-Effectiveness of Education Delivery by School Type

Average Learning Gain Per Rupee

GovernmentPrivate Aided

Private Unaided (3)/(1)

(1) (2) (3) (4) Read 0.82 0.20 1.31 1.60 Word 0.60 0.64 1.04 1.73 Math 0.73 0.49 1.09 1.50

Other policy implications also emerge from the study, but by way of further research.

For example, in this study we find that schools with multi-grade classrooms record lower

test scores. Teachers in government schools are not trained to teach in a multi-grade

classroom context. This disconnection between the realities of the teaching environment

and the tools provided to teachers in government schools plausibly impacts negatively on

learning outcomes. A similar case can be made regarding teaching large class-sizes which

again is a reality for many government school teachers for which they may not be trained.

Educational quality determines individual earnings, income distribution and

economic-growth of countries (Hanushek and Woessman, 2006). Public schooling will

remain the dominant provider of schooling for the majority of the population. Policy-thus

makers need to find cost-effective ways to improve quality in public schools. Improving

the performance of public schools is made difficult by the fact that measurable school

characteristics have proven to be weak proxies for school quality in the standard

education production function approach. However, there are some desirable

characteristics that any reform agenda must have:

• Education policy reforms should be based on robust empirical evidence, given the

opportunity costs of scarce public resources. Policy makers should have a fair

idea about the returns to the marginal rupee across alternative interventions, and

25

Preliminary version: for comments

should choose those interventions where the returns are the largest. This requires

accurate assessment of the costs and benefits of any intervention.

• People respond to incentives. The success of any reform initiative will therefore

also depend on which outcomes are identified for monitoring and evaluation, for

establishing accountability and for judging success and failure of the reform. If

the objective of reforms is to improve learning outcomes, then education

providers – line department officials, school principals, teachers and other

stakeholder – will have to be made accountable for achieving this goal.

.

26

Preliminary version: for comments

References

Aggarwal, Yash (2000), “Primary Education in Delhi: How Much Do The Children

Learn?” NIEPA, New Delhi.

Chaudhury, Nazmul, Jeff Hammer, Michael Kremer, Karthik Muralidharan and F. Halsey

Rogers (2004), “Teacher Absence in India,” The World Bank, Washington D.C.

Das, Jishnu, Priyanka Pandey and Tristan Zajonc (2006), “Learning Levels and Gaps in

Pakistan,” World Bank Policy Research Working Paper #4067, The World Bank,

Washington D.C.

Dreze, Jean and Geeta G. Kingdon (2001), ‘Schooling Participation in Rural India’,

Review of Development Studies, 5(1), February, pp 1-24.

Filmer, Deon, King, Elizabeth M and Lant Pritchett (1997), ‘Gender Disparity in South

Asia: Comparison Between and Within States,’ World Bank Policy Research Working

Paper No. 1867, The World Bank, Washington D.C.

Filmer, Deon and Lant Pritchett (1998), ‘Education Enrollment and Attainment in India:

Household Wealth, Gender, Village and State Effects’, South Asia Region, IDP – 97, The

World Bank.

Fuller, Bruce (1986), Raising School Quality in Developing Countries: What Investments

Boost Learning, World Bank Discussion Paper No. 2, World Bank, Washington D.C.

Goldhaber, Dan, and Dominic Brewer (1997), “Why Don’t Schools and Teachers Seem

to Matter? Assessing the Impact of Unobservables on Educational Productivity.” Journal

of Human Resources, 32(3), pp. 505-523.

27

Preliminary version: for comments

Hanushek, Eric and L. Woessman (2006), “The Role of Education Quality in Economic

Growth,” xx.

Kingdon, G (1996), “The Quality and Efficiency of Public and Private Schools: A Case

Study of Urban India”, Oxford Bulletin of Economics and Statistics, 58(1), February, pp

55-80.

Lavy, Victor and Joshua Angrist (1999), “Using Maimonides’ Rule to Estimate the Effect

of Class Size on Scholastic Achievement,” Quarterly Journal of Economics, 114(2), pp

533-575.

Muralidharan, Karthik and Michael Kremer. “Public and Private Schools in Rural India.”

Working Paper, Department of Economics, Harvard University, March 22, 2006.

Smith, F., Hardman, F., and J. Tooley (2005), ‘Classroom interaction and discourse in

private schools serving low income families in Hyderabad, India’, International

Education Journal, 6(5), pp 607-618.

Tooley, James and Pauline Dixon (2006), ‘‘De facto’ privatization of education and the

poor: implications of a study from sub-Saharan Africa and India’, Compare, 36(4), pp

443-462.

Urqiola Miguel (2006), “Identifying Class Size Effects in Developing countries:

Evidence from Rural Bolivia,” The Review of Economics and Statistics, 88(1), pp 171-

177.

28

Preliminary version: for comments

Annex Table 1: Number Schools by Type and Location, Rajasthan

School Type Rural Urban Total Government 129 17 146

Private Aided 7 9 16

Private Unaided 25 13 38

Total 161 39 200

Table 2: Student Sample Size by Class and Gender, Rajasthan

Boys Girls Total

Class IV 1719 1508 3227

(0.53) (0.47) (1.00)

Class V 1702 1470 3172

(0.54) (0.46) (1.00)

Total 3421 2978 6399

(0.53) (0.46) (1.00)

Table 3: Mean Student Characteristics, Rajasthan

Scheduled Caste (%) 20.55

Scheduled Tribe (%) 22.71

OBC (%) 15.36

General/Other (%) 41.59

Father Literate (%) 71.95

Mother Literate (%) 31.52

Mean SD

Number of Days Absent 0.64 1.088

Household Asset Score 3.35 2.01

29

Preliminary version: for comments

Table 4: Descriptive Statistics of School Characteristics by School Type, Rajasthan

School Type Government Private Aided Private Unaided

Mean SD Mean SD Mean SD

Percentage of Graduate Teachers (%)

67.4

27.80

68.17

25.23

69.64

24.28

Percentage of Teachers with Pre-Service Training (%)

61.78

48.61

23.00

42.26

42.50

40.00

Average Teaching Experience (Years)

11.72

5.34

10.08

7.26

9.04

6.47

Average Age of Teachers (Years)

39.17 9.37 36.59 11.18 35.12 10.67

Pupil Teacher Ratio

44.33 37.51 33.68 10.32 49.58 66.59

SC/ST Share (%)

46 30 30 28 33 28

Percentage using Multi-grade Classrooms (%)

59.85

43.75

55.56

30

Preliminary version: for comments

Figure 1: Average Percentage Scores in Read, Word, Math in Grades IV and V by School Type,

Rajasthan

Average Percentage Scores

0

10

20

30

40

50

60

70

80

Read Word Math Read Word Math

Class IV Class V

Per

cent

age

Scor

e (%

)

GovtPvt AidedPvt Unaided

Figure 2: Average Percentage Scores in Read, Word, Math in Grade IV by Gender and School Type,

Rajasthan

Average Percentage Scores by Gender and School Type: Class IV

010203040506070

Read Word Math Read Word Math Read Word Math

Govt Pvt Aided Pvt Unaided

Per

cent

age

Scor

e(%

)

BoysGirls

31

Preliminary version: for comments

Figure 3: Average Percentage Scores in Read, Word, Math in Grade V by Gender and School Type,

Rajasthan

Average Percentage Scores by Gender and School Type: Class V

01020304050607080

Read Word Math Read Word Math Read Word Math

Govt Pvt Aided Pvt Unaided

Perc

enta

ge S

core

(%)

BoysGirls

Figure 4: Average Percentage Scores by Social Group and School Type, Rajasthan

Percentage Test Scores by Social Group

010203040506070

Read Word Math Read Word Math

Class IV Class V

Perc

enta

ge S

core

(%)

SCSTOBCGeneral

32

Preliminary version: for comments

Figure 5: Average Percentage Scores by Rural Urban Location for Grade IV, Rajasthan

Average Percentage Scores by Rural-Urban Location and School Type: Class IV

0

10

20

30

40

50

60

70

Read Word Math Read Word Math Read Word Math

Govt Pvt Aided Pvt Unaided

Per

cent

age

Scor

e (%

)

RuralUrban

Figure 6: Average Percentage Scores by Rural-Urban Location for Grade V, Rajasthan

Average Percentage Scores by Rural-Urban Location and School Type: Class V

01020304050607080

Read Word Math Read Word Math Read Word Math

Govt Pvt Aided Pvt Unaided

Perc

enta

ge S

core

(%)

RuralUrban

33

Preliminary version: for comments

Figure 7: Unadjusted and Adjusted Gaps between Boys and Girls Test Scores

-2.00-1.50-1.00-0.500.000.501.00

UnadjustedAdjusted

Unadjusted -0.40-0.02 -0.01-0.18 0.34 -1.34

Adjusted -1.10-0.87 -1.46-0.66 0.51 -1.50

ReadWordMathReadWordMath

Grade IV Grade V

Figure 8: Unadjusted and Adjusted Gaps between Test Scores of SC and other Social Groups, Grade IV

-8-6-4-20246

UnadjustedAdjusted

Unadjusted 2.56 -5.5 -5.6 0.5 -2.2 -3.1 3.68 -4 -4.6

Adjusted -0.5 1.1 1.21 0.59 1.05 -1.2 0.36 1.06 1.47

ST OBCGeneST OBCGeneST OBCGene

Read Word Math

34

Preliminary version: for comments

Figure 9: Unadjusted and Adjusted Gaps between Test Scores of SC and other Social Groups, Grade V

-10

-5

0

5

10

UnadjustedAdjusted

Unadjusted 4.8 -3.1 -5.8 1.5 -1.3 -2.5 6.1 -3.3 -6.3Adjusted 0.6 1.6 3.5 0.4 -0.3 -0.3 -0.2 1 0.7

ST OBCGen ST OBCGenST OBCGenRead Word Math

35

Preliminary version: for comments

Figure 10: Kernel density distribution of unadjusted and adjusted School Average Scores for

Government and Private Unaided Schools

Grade IV

0

.005

.01

.015

.02

Den

sity

-50 0 50Percentage Score

Govt Pvt Unaided

Adjusted Grade IV Reading Comprehension Scores

0

.005

.01

.015

.02

Den

sity

-40 -20 0 20 40 60Percentage Score

Govt Pvt Unaided

Adjusted Grade IV Mathematics Scores

Grade V

0

.005

.01

.015

Den

sity

-50 0 50Percentage Score

Govt Pvt Unaided

Adjusted Grade V Reading Comprehension Scores

0

.005

.01

.015

.02

Den

sity

-50 0 50Percentage Score

Govt Pvt Unaided

Adjusted Grade V Mathematics Scores

36

Preliminary version: for comments

OLS Regressions

Regression (I)

37

Dependent Variable: Percentage Score in Reading Comprehension Test

Grade IV Grade V

(1) (2) (3) (4) (5) (6)

School Yes Yes No Yes Yes No

Age -0.149 -0.772 -0.597 -0.999

-0.54 -1.2 (2.15)* -1.96

Age-Squared 0 0 0 0

-0.2 -0.08 -0.27 -0.79

Male -1.103 -0.359 -0.663 0.136

-1.57 -0.33 -0.81 -0.11

ST -0.465 -1.082 0.623 -4.839

-0.29 -0.39 -0.44 -1.74

OBC 1.099 3.18 1.611 0.811

-1.15 -1.67 (1.98)* -0.44

General 1.206 1.89 3.524 1.255

-1.05 -0.85 (3.31)** -0.57

Father Literate 0.289 2.044 0.315 1.443

-0.4 -1.63 -0.48 -1.23

Mother Literate 1.485 1.403 -0.112 -0.295

(2.12)* -0.97 -0.14 -0.2

Household Asset Score 0.36 1.464 0.605 1.313

-1.85 (3.22)** (3.29)** (2.93)**

Days Absent Last Week -0.186 -0.678 0.226 -0.363

-0.61 -1.38 -0.83 -0.76

Rural -31.744 -4.963 23.375 -2.533

(25.59)** -1.32 (15.63)** -0.65

Average Salary of Teachers in School 0 -0.001

-0.95 (2.33)*

Average Years of Teacher Experience in School -0.03 -0.017

-1.71 -0.92

Average Teacher Days Absent in the Last Academic Year -0.194 -0.388

-1.25 (2.68)**

Percentage Graduate Teachers 0.016 0.061

-0.29 -1.08

Percentage Non-regular Teachers -0.053 -0.089

-1.22 -1.88

Multi-Grade 0.296 -0.166

Preliminary version: for comments

-0.1 -0.06

Mid-Day Meals -6.336 -4.174

(2.09)* -1.4

Pupil Teacher Ratio -0.188 -0.127

-1.94 -1.26

Private Aided 5.042 1.661

-0.93 -0.35

Private Unaided 4.774 6.805

-1.34 -1.71

Constant 20.926 98.178 95.914 44.444 133.141 302.646

(9.96e+10)** -0.45 -0.24 (5.69e+11)** -0.49 -0.63

Observations 3255 3038 2386 3227 2994 2364

R-squared 0.62 0.62 0.14 0.6 0.59 0.13

Robust t statistics in paranthesis

* significant at 5%; ** significant at 1%

38

Preliminary version: for comments

Regression (II)

Dependent Variable: Percentage Score in Word Meaning

Grade IV Grade V

(1) (2) (3) (4) (5) (6)

School Yes Yes No Yes Yes No

Age -0.236 -0.247 -0.374 0.01

-0.96 -0.51 -1.42 -0.02

Age-Squared 0 0 0 0

(2.02)* -0.87 -0.87 -1.02

Male -0.868 -0.085 0.508 0.684

-1.29 -0.09 -0.7 -0.7

ST 0.586 -0.319 0.381 -0.7

-0.42 -0.18 -0.32 -0.38

OBC 1.047 -0.493 0.235 -0.588

-1.21 -0.34 -0.26 -0.39

General 0.674 0.151 1.023 -0.155

-0.66 -0.1 -0.9 -0.09

Father Literate 1.007 1.898 -0.356 0.533

-1.54 (1.98)* -0.56 -0.53

Mother Literate -0.016 -0.34 0.508 1.313

-0.02 -0.3 -0.75 -1.28

Household Asset Score 0.202 0.739 0.451 0.935

-1.01 (2.30)* (2.52)* (2.98)**

Dasy Absent Last Week 0.048 -0.546 -0.15 -0.506

-0.21 -1.5 -0.59 -1.23

Rural -11.95 -4.615 -7.47 0.868

(10.49)** -1.82 (5.18)** -0.29

Average Salary of Teachers in School -0.001 -0.001

(3.25)** -1.93

Average Years of Teacher Experience in School -0.011 -0.021

-1.23 -1.71

Average Teacher Days Absent in the Last Academic Year -0.257 -0.234

(2.53)* (2.28)*

Percentage Graduate Teachers -0.008 -0.013

-0.21 -0.31

Percentage Non-regular Teachers -0.089 -0.054

(2.82)** -1.65

Multi-Grade 2.19 -2.203

-0.95 -1.06

39

Preliminary version: for comments

Mid-Day Meals -3.495 -2.981

-1.54 -1.37

Pupil Teacher Ratio -0.157 -0.047

(2.52)* -0.78

Private Aided 1.587 3.786

-0.36 -0.99

Private Unaided 2.955 6.802

-1.27 (2.15)*

Constant 46.19 469.423 352.75 58.095 156.23 322.45

(5.31e+11)** (2.31)* -1.12 (9.90e+11)** -0.62 -0.84

Observations 3255 3038 2386 3227 2994 2364

R-squared 0.48 0.48 0.11 0.45 0.46 0.08

Robust t statistics in parentheses

* significant at 5%; ** significant at 1%

40

Preliminary version: for comments

Regression (III)

Dependent Variable: Percentage Score in Math Test

Grade IV Grade V

(1) (2) (3) (4) (5) (6)

School Yes Yes No Yes Yes No

Age -0.19 -0.099 -0.451 -0.418

-0.93 -0.16 (2.15)* -0.72

Age-Squared 0 0 0 0

-1.11 -0.14 -1.33 -0.66

Male -1.455 -1.206 -1.496 -0.63

(2.68)** -1.2 (2.55)* -0.52

ST 0.355 -3.617 -0.249 -7.727

-0.35 -1.42 -0.24 (2.73)*

*

OBC 1.057 0.767 1.042 1.207

-1.31 -0.38 -1.4 -0.53

General 1.47 2.523 0.72 3.69

-1.53 -0.99 -0.75 -1.36

Father Literate 1.326 1.396 -0.651 -0.678

(2.39)* -1.13 -1.1 -0.52

Mother Literate -0.222 1.019 0.135 1.192

-0.44 -0.79 -0.23 -0.99

Household Asset Score 0.223 1.021 -0.056 0.618

-1.38 (2.30)* -0.39 -1.55

Days Absent Last Week -0.317 -1.731 0.157 -0.72

-1.35 (3.62)*

*

-0.82 -1.94

Rural -24.964 1.626 -22.286 4.086

(24.16)** -0.47 (19.79)*

*

-1.32

Average Salary of Teachers in School 0 0

-1.04 -0.95

Average Years of Teacher Experience in School 0.001 -0.012

-0.05 -0.95

Average Teacher Days Absent in the Last Academic Year -0.174 -0.269

-1.49 (2.08)*

Percentage Graduate Teachers -0.01 -0.016

-0.2 -0.31

Percentage Non-regular Teachers -0.055 -0.057

41

Preliminary version: for comments

-1.33 -1.21

Multi-Grade -0.572 -0.75

-0.22 -0.28

Mid-Day Meals -0.971 -0.393

-0.37 -0.15

Pupil Teacher Ratio -0.209 -0.061

(2.47)* -0.62

Private Aided 7.479 6.81

-1.23 -1

Private Unaided 2.706 4.765

-0.89 -1.56

Constant 21.549 -157.441 -0.867 47.475 300.519 255.91

(1.18e+11)** -0.85 0 (9.57e+11)*

*

-1.68 -0.54

Observations 3255 3038 2386 3227 2994 2364

R-squared 0.72 0.72 0.1 0.7 0.7 0.09

Robust t statistics in parentheses

* significant at 5%; ** significant at 1%

42

Preliminary version: for comments

Construction of Household Asset Score: The household asset score has been constructed on a 12 point scale with each asset getting one point if available in the household of the student. The information was gathered through questions in the student background questionnaire.

43