prying the private school effect: an empirical analysis of - peri

28
Prying the Private School Effect: An Empirical Analysis of Learning Outcomes of Public and Private Schools in Pakistan Kiran Javaid, Tareena Musaddiq and Atiyab Sultan 1 As the preference for private school education becomes more widespread in Pakistan, the debate on the relative merits of public and private education has gained increasing relevance and importance. To assess the differences in the educational outcomes of the students in the two streams, it is necessary to isolate the pure effect of school choice (private versus public). Using ASER Pakistan data for the years 2010 and 2011 we employ various techniques to analyse the effect of private schooling. In particular, the Oaxaca decomposition is applied to assess achievement differences between public and private school students, while Fixed Effects estimation is used to study province, district, village and household level differences. An in- depth study of the learning outcomes of the private school students as opposed to those enrolled in government schools is enabled by a pooled cross-sectional analysis using data from both years, as the level of fixed effects is made increasingly strict (from province to village and then household level). Private school advantage is significant at each of these levels, whereas gender discrimination present at only the village and household level. Oaxaca decomposition shows that only five percent of the achievement differential can be attributed to the endowment differences between the two groups. JEL Classifications: I20, I21 Keywords: Learning outcome differentials, Private vs. Public Schooling, Pakistan, 1 Kiran Javaid, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan; [email protected]; Tareena Mussadiq, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan; [email protected]; Atiyab Sultan, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan; [email protected]. Kiran Javaid, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan

Upload: others

Post on 11-Feb-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Prying the Private School Effect: An Empirical Analysis of - PERI

Prying the Private School Effect:

An Empirical Analysis of Learning Outcomes of Public and Private Schools in

Pakistan

Kiran Javaid, Tareena Musaddiq and Atiyab Sultan1

As the preference for private school education becomes more widespread in Pakistan, the

debate on the relative merits of public and private education has gained increasing relevance and

importance. To assess the differences in the educational outcomes of the students in the two

streams, it is necessary to isolate the pure effect of school choice (private versus public). Using

ASER Pakistan data for the years 2010 and 2011 we employ various techniques to analyse the

effect of private schooling. In particular, the Oaxaca decomposition is applied to assess

achievement differences between public and private school students, while Fixed Effects

estimation is used to study province, district, village and household level differences. An in-

depth study of the learning outcomes of the private school students as opposed to those enrolled

in government schools is enabled by a pooled cross-sectional analysis using data from both

years, as the level of fixed effects is made increasingly strict (from province to village and then

household level). Private school advantage is significant at each of these levels, whereas gender

discrimination present at only the village and household level. Oaxaca decomposition shows that

only five percent of the achievement differential can be attributed to the endowment differences

between the two groups.

JEL Classifications: I20, I21

Keywords: Learning outcome differentials, Private vs. Public Schooling, Pakistan,

1 Kiran Javaid, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS),

Pakistan; [email protected]; Tareena Mussadiq, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan; [email protected]; Atiyab Sultan, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan; [email protected]. Kiran Javaid, Teaching Fellow, Department of Economics, Lahore University of Management Sciences (LUMS), Pakistan

Page 2: Prying the Private School Effect: An Empirical Analysis of - PERI

1

1. Introduction

The debate on public versus private education has gained increasing importance in recent

years. The issue is of special importance for developing countries, many of which are home to

widespread networks of private schools. The existence of such networks questions the role and

capacity of the government in providing education for all. It also underlines the need to build

state capacity in providing education, more so since many individuals are unable to afford

private schools. In some countries, the acknowledgment of dysfunctional government schools

has given rise to public-private partnerships in an attempt to increase literacy and access to

schools. However, a key question remains regarding the quality of education in private versus

public schools. To evaluate their performance, many researchers have used tests that have been

specially designed to gauge the quality of public versus private education. This paper adds to this

body of knowledge by comparing public and private schools in 85 districts of Pakistan using the

results of a test that evaluated the linguistic and mathematical skills of the students.

Education in Pakistan has assumed increasing importance recently with the country still

far from attaining literacy for all. The country is lagging behind in the pursuit of the Millennium

Development Goals, especially in education. This is perhaps both a result of a resource constraint

as well as a lack of political will to devote a larger budget share towards education. Over the

years, the state-run system of schools has become an inferior or second-grade mode of education

in popular perception, with the concomitant rise of a vast network of private schools. Private

schools in Pakistan in 1983 numbered 3,343 which rose astronomically to 36,096 by

20002.Interestingly, not all private schools are expensive or elite schools and there is a wide

variation in the fees and quality of education being provided by these schools. The situation is

2FBS(2001)

Page 3: Prying the Private School Effect: An Empirical Analysis of - PERI

2

also complicated by the different streams of education in the country: the government

examination boards conduct the Matriculation and Intermediate /F Sc. exams, whereas many

private schools instead enable students to take the British GCE O and A Levels exams. A few

schools are also conducting Baccalaureate examinations and religious seminaries (madressahs)

also have their own exams. The dominant tracks are the Matric/FSc and the O/A Levels schemes.

A couple of studies have analyzed learning outcomes in Pakistani schools by comparing

public and private schools.3 However, since previous datasets mostly studied primary schools,

these studies only analyze the differential in outcomes at this level. This paper uses the latest

data (for the years 2010 and 2011) from ASER (Annual Status of Education Report) which spans

over the entire country and collects data on all levels of schooling including primary, middle and

secondary schools. The paper uses this data to evaluate the differential in learning outcomes of

public and private schools, thereby enabling an informed and insightful comparison of the

quality of education being provided by these schools.

The paper is structured as follows: Section 2 contains a literature review, Section 3

describes the dataset in greater detail, while Section 4 discusses the methodology employed.

Section 5 contains descriptive statistics and also discusses the results of the study and finally

Section 6 concludes.

2. Literature Review

A wealth of theoretical and empirical research exists on the effectiveness of private

versus public schools. The results of these studies present a varied and granulated picture in

which private schools can sometimes outperform public schools, and in other locations lag

3 See Alderman, Orazem,andPaterno 2001; Arif and Saqib 2003; Das, Pandey, and Zajonc 2006 ).

Page 4: Prying the Private School Effect: An Empirical Analysis of - PERI

3

behind them. In Pakistan, private schooling is generally considered to be of superior quality as

opposed to state-run government schools. In a paper exploring the extent and limitations of

private schooling in Pakistan, Andrabi, Das and Khawaja (2006) provide a comprehensive

treatment of the subject concluding that a key reason for the proliferation of private schools is

their low fees. The limits to their growth are interestingly defined in terms of horizontal and

vertical constraints with the former referring to the fact that private schools can only exist in a

community where there is already a pool of potential teachers (often women). The vertical

constraint identified by the authors refers to the fact that since most teachers are only educated

up to the secondary school level, providing education beyond that level is not possible for the

schools. Overall, private schools are found to have been quite instrumental in increasing

education at the primary school level and also in reducing gender discrimination.

Apart from these resource constraints however, a richer understanding of the private

school phenomena involves analyzing the service they provide and comparing it concretely with

the alternative(s), especially public schools. Academic research has focussed on disentangling

the private school ‘premium’ that might explain the difference in outcomes from the two types of

schools in two distinct ways: comparing the wages accruing to the graduates of public versus

private schools, or comparing the learning outcomes from the two types of schools.

Some of the research in this area focuses on why this differential exists i.e. can it be

explained in terms of a higher quality of education imparted in private schools or due to non-

academic gains that result from the environment of private schools: such as networks, resources

and other opportunities that have a payoff in the labour market.4 Other considerations in studying

4Brown and Belfield (2001) and Asadullah (2005)

Page 5: Prying the Private School Effect: An Empirical Analysis of - PERI

4

the demand for private schools relate to the level of fees charged e.g. Alderman et al (2001) find

that the demand for private schooling in Pakistan increases as the school fees charged by these

schools fall. The increased demand comes from both students switching from public schools to

private schools and from students that were not previously enrolled. They also find that the

increased demand is consistent with superior results on both mathematics and language tests by

private schools. While their analysis is conducted by simulating elasticities and changes in the

probability of enrollment in a school against different variables, Andrabi et al.(2002) provide a

nuanced account of the clientele of private schools in their paper ‘The Rise of Private Schooling

in Pakistan: Catering to the Urban Elite or Educating the Rural Poor?’ It is seen that even low

income households opt for private schooling in urban as well as rural areas. Private schools have

lower student-teacher ratios, encourage co-education and employ locally resident teachers, all of

which contribute to lowering the costs of the provision of education.

A more cynical view of private school education is provided by Brunello and Rocco (2008) in

their discussion of educational standards in public and private schools using data from Italy and

the United States. By modeling pre-set educational standards and the expectations from public

and private schools, they show that multiple equlibria are possible. Dispelling with the simplistic

notion that private schools are always of better quality, they consider the possibility that private

schools may be charging higher fees for other services e.g. leisure, access to certain networks or

religious groups, etc. They show that distinct equilibria are possible, one with better quality

education provided by private versus public schools as in the US, and one where private schools

are worse off as in Italy. This contextual awareness of the effectiveness of private schools is

pertinent in our study and we investigate the issue in great vertical depth by looking at district,

household, village, etc. effects.

Page 6: Prying the Private School Effect: An Empirical Analysis of - PERI

5

Wage differentials between public and private schools

In general, research has focussed on two different measures of estimating this premium.

The first relates to the wage differentials between students from public and private schools. In

this vein of research, the overwhelming thrust has been in the favour of private school graduates

with these earning higher salaries than their state-schooled counterparts e.g. A detailed study by

Bedi et al (2000) using data from Indonesia presents evidence that students who were enrolled in

private schools at the secondary school level ended up with higher salaries than those educated in

public schools.

Learning Outcomes in public and private schools

The other measure of estimating the private school premium has been through a

comparison of learning outcomes in private and public schools. A wealth of scholarship has

emerged in this regard, and it is progressively getting more sophisticated in the estimation and

explanation of this premium. The key point here is to disentangle the ‘private school effect’ from

other factors that may be influencing learning outcomes. Since often the results of these studies

are based on a comparison of test scores, the performance of students on these tests needs to be

evaluated in closer detail to disengage those factors that affect academic achievement. These

include family background, the age of the child, family income, parents’ education and so on. In

addition, other variables like the number of siblings or birth order can also have an effect.

Isolating the contribution of a private school education to the test score thus requires careful

analysis with many control variables to prevent the conflation of effects of different factors.

Studies focussing on the disentanglement of the true private school effect have mostly

taken one of two forms: either an econometric analysis of pooled data from private and public

Page 7: Prying the Private School Effect: An Empirical Analysis of - PERI

6

schools with a dummy variable representing the respective school type, or separate estimation for

private and public schools using an equation for each type. The results are mixed however with

some studies finding that private school students perform better than public school students and

vice versa in other cases.

In many studies, private school students have been seen to outperform students enrolled

in public schools e.g. Jimenez, Lockheed and Paqueo (1991) use data from Colombia, the

Dominican Republic, the Philippines, Tanzania and Thailand and find that at the secondary

school level, private school students obtained higher scores on standardized mathematics and

language tests even after controlling for the fact that on average, private school students in these

countries hail from more advantaged backgrounds. The study also finds that the unit costs of

private schools are lower than public schools. In another study, Jimenez, Lockeed and

Wattanawaha (1988) use a mathematics test for students in eighth grade in Thailand in public

and private schools. Again, private school students are seen to do better and the difference in

outcomes is explained in terms of the smaller size of private schools and their location in

wealthier neighbourhoods (leading to access to better resources and peers) even though there

were fewer certified teachers in private schools than in public schools.

Some studies have employed a more experimental set-up e.g. one discussed by Rouse

(1998) on the Milwaukee Parental Choice Programme. By randomly assigning students as

treatment (private schooling) or control groups through a voucher programme, the study finds

that private school students performed on a mathematics test better but there was mixed evidence

on the reading test. Similarly, Kim, Alderman et al (1999) use the Quetta Urban Fellowship

Program in Pakistan to conduct a random experiment by looking at enrollment in private schools

in a poor neighbourhood of the city. Enrollment for both boys and girls witnessed an increase

Page 8: Prying the Private School Effect: An Empirical Analysis of - PERI

7

with the setting up of a private school suggesting that subsidizing private school education can

increase demand for it even while holding other features of the school constant.

However, other studies have detected no clear advantage for private schooling. Using a

randomized lottery, Cullen, Jacob and Levitt (2006) study the impact of school choice on

academic attainment and finds that there is no systematic gain found in academic measures of

measuring performance. However, students are seen to benefit in a number of non-traditional

measures e.g. discipline. Other papers have refined the comparison still further e.g. by only

focusing on the teaching of Economics at the secondary school level, Grimes (1994) shows that

students in public schools outperform their counterparts in private schools in the United States.

Other papers have also reported better performance by public school students e.g. Newhouse and

Beegle (2006) use data from Indonesia to show that at the junior secondary school level, public

school students obtain better results than madressahs or private schools.5

A number of studies have also focused on Pakistan. Das, Pandey and Zajonc (2006) use a

test at the end of third grade and find that a large differential exists between schools as opposed

to differentials between students from varying backgrounds; e.g. the gap in the scores for the

English language test between government and private schools was found to be twelve times the

gap between children from rich and poor families. By conducting a village level analysis, they

also find that good quality and bad quality schools co-exist in every village and therefore the

differences are a result of the quality of schools, not differences across villages.

Monazza Aslam (2009) also studies public and private schools in Pakistan and finds that

boys are more likely to be enrolled in private schools than girls within the household. She also

5However they find that secular private schools perform on the same level as public schools.

Page 9: Prying the Private School Effect: An Empirical Analysis of - PERI

8

finds that private schools are of better quality and more effective in imparting quantitative and

linguistic skills. However she finds that gender significantly determines the learning

opportunities that accrue to a child with girls getting less educational expenditure within the

household and also attaining education of poorer quality.

We explore the same issue in greater detail and by using the latest survey data to obtain a

fine-grained understanding of private versus public school learning outcomes at different levels.

The next sections detail the data and methodology we employ.

3. Data

Data for Annual Status of Education Report (ASER) for the years 2010 and 2011 for

Pakistan is used for the purpose of this study. For 2010, the data was collected from 33 districts

of Pakistan and the sample size employed for the analysis is of 24,018 students. For the year

2011, the data collection was expanded to 85 districts and the sample size analyzed consists of

80,310 observations. Therefore, for the pooled analysis the sample size totaled up to 104,328

observations. The data has been divided into different groups for the various FE estimations. The

divisions at different levels comprise of eight provinces/administrative areas, 85 districts, 2692

villages and 49,244 households.

In terms of sample selection ASER uses the Probability Proportional to Size (PPS) technique

so that villages with higher population have a higher chance of being selected into the sample.

Within each village, for the purpose of selecting households the village is divided into four

hamlets such that the population of the village is divided into four equal parts. Next, a household

is picked from the centre of each of these households and interviewed. Every fifth household

Page 10: Prying the Private School Effect: An Empirical Analysis of - PERI

9

from the left of the first one selected is chosen until five households have been selected from

each hamlet to yield an overall sample of 20 households from each village.

The head of the household answers the information concerning the characteristic of the

household such as the asset base of the house etc. The children belonging to the household

available at the time of the survey are asked to sit for the learning assessments which have three

components namely English, Mathematics and Urdu (the national language).

4. Methodology

The aim of the study is to determine whether there are significant differences in the

outcomes of private vs. public schools. Traditionally, this can be done in two ways. Since the

primary benefit of schooling is thought to be the wage premium derived from higher or better

quality education, one way to assess the difference between public and private schooling is to

examine the premium of private schooling in labor market earnings accruing to graduates of

various schools, both private and public.6 The other method is to look at the differences in

students learning achievements at the school level. This approach treats learning outcomes as

output from an education production function with various types of inputs contributing to the

outcome. The inputs considered can be categorized at the individual level (students’ age,

gender), household level (parenting choices, parents education, number of siblings) and school

level (facilities at school, option of tuition etc). The output is determined by a score of students’

performance on a test.

6See Nasir (1999) and Asadullah (2005)

Page 11: Prying the Private School Effect: An Empirical Analysis of - PERI

10

Within the second approach, the analysis can be conducted by either pooling the data for

students belonging to both public and private schools and adding aprivate enrollment dummy

variable to ascertain the effect of private schooling on the educational attainment of an

individual, or separate education production functions for public and private students can be set

up.

This study employs the first approach and estimates the education function using the

ordinary least squares approach as:

Where Y is the test score and X is a vector encompassing the child, household and village

level control variables as discussed earlier while D is a dummy variable taking the value of one if

the student is from a private school and zero otherwise. Table 1 below lists the independent

variables and provides a brief description of each.

Table 1: Descriptive Statistics

Variable Description

Private Dummy variable: Takes a value of one if the student goes to a private

school and zero otherwise

Female Dummy variable: Takes a value of one if the student is a female and zero

otherwise Child Age Age of the student in years

Tuition Dummy variable: Takes a value of one if the student takes outside tuition

zero otherwise Number of Siblings Number of siblings of the student

Father’s Age Age of the students father in years

Father’s Age Squared Age squared of the students father in years

Father attended school Dummy variable: Takes a value of one if the students father attended

school and zero otherwise Mother’s age Age of the students mother in years

Mother’s age squared Age squared of the students mother in years

Mother’s attended school Dummy variable: Takes a value of one if the students mother attended

school and zero otherwise Birth order of child Number of elder siblings in the house

Wealth Wealth index

Page 12: Prying the Private School Effect: An Empirical Analysis of - PERI

11

At the individual level the gender and age of the student are expected to impact the

learning outcome. Age of the child should have a positive relationship with the test score, since

higher age should most likely correspond to a higher grade and a better performance on the test.

The gender dummy will help assess any gender differences in learning outcomes that might be

there. At the household level father and mothers age and age squared are added as explanatory

variables. Apriori it is expected that the relation of parents’ ages with child’s education should be

positive initially and should eventually become negative. The coefficients on both mothers and

fathers age are hence expected to be positive while that of the age squared should be negative.

Also if the parents of the student are educated it is more likely that s/he would perform better at

school. The reasoning is twofold; not only will the parents be able to better motivate the child but

would also be able to help the child with any course work. Likewise the birth order and number

of siblings are also expected to impact the child’s performance. On the one hand the lower down

the child is in the order of birth in the house and greater the number of siblings that the parents

need to support, less will be spent on the child’s education. On the other hand however theory

also suggests that having elder siblings could also mean additional support in learning and might

motivate the children to learn/study more effectively. Additional support may also come from

tuition that the child may be taking outside of school which is expected to impact the learning

outcome positively and increase the score.

Wealth of the family is also an important determinant when it comes to educational

attainment. It is indicative of a households financial position and hence the ability to afford

quality education. For the purpose of this study a wealth index is constructed which takes into

account various assets that would directly and indirectly contribute in the well being of the child

and ease access to different resources which might result in better learning achievement. The

Page 13: Prying the Private School Effect: An Empirical Analysis of - PERI

12

indicators used in construction of the index include conditions of the house the family lives in

(kind of dwelling and toilet facility), ownership of assets pertaining to information and

communication technology (mobile phones, electricity) and assets that ease transportation (car,

cycle, motorbike).

The outcome or the dependent variable is a test score which in the case of this study

consist of an overall score based on assessment of student’s learning/skills in English, Urdu and

Mathematics. For each of the three subjects scores are assigned depending on the students’

ability e.g. whether a student can recognize alphabets only, form a word, has counting skills or

whether s/he can add or subtract. A higher score indicates a better performance. The score of

each child is standardized by converting them into Z-scores i.e. subtracting the mean and

dividing by the standard deviation. This enables an easier interpretation of the results.

The analysis, using the OLS technique as described above, is performed at various levels

namely fixed effects at the household, village, district and provincial levels. Conducting the

analysis in this manner allows us to control for differences between the units at the various

levels. The fixed effects technique helps in mitigating the endogeneity bias inherent in a simple

OLS estimation.

Oaxaca Decomposition:

Once the estimates of the regression analysis are available it is worthwhile to apply the

Oaxaca decomposition7 to the results to determine the proportions of the differential in z-scores

explained by the endowment differences between the students in the two types of schools and the

quality of education provided. Although the Oaxaca decomposition has traditionally been

7 See Oxaca (1973)

Page 14: Prying the Private School Effect: An Empirical Analysis of - PERI

13

employed to explain wage differentials in the labor markets and evaluate differentials due to

discrimination, the approach has also been applied for differentials in education.8

This approach decomposes the mean difference in learning outcomes, based on the OLS

estimates, into the characteristic (explained portion) and coefficient effect (the unexplained part).

The unexplained portion in this case will possibly indicate the better educational services

provided by the private schools, amongst other unobservables.

The procedure involves the use of the OLS coefficients in conjunction with mean of the

variables in the two sectors to estimate the endowment and discrimination effects. The difference

in the z scores is given by

Zprivate

– Zpublic

= Xprivate

βprivate

– Xpublic

βpublic

where βprivate

represents OLS estimates for the private school children which is considered the

advantaged group and βpublic

represents the OLS estimates for public sector (disadvantaged

group). Xprivate is

a vector representing the means of the independent variables for the private

while Xpublic is

for the children enrolled in public school.

The difference between the coefficients of the two sectors is given by:

Δβ = βprivate

– βpublic

implying that βpublic

= βprivate

- Δβ

If this second relationship is substituted in the first one, the following equation results:

Zprivate

– Zpublic

= βprivate

(Xprivate

– Xpublic

) + Xpublic

Δβ

This equation will be used to determine the proportion explained by endowment differences and

the unexplained portion.

8 See Thapa (2011) and Desai et al (2008)

Page 15: Prying the Private School Effect: An Empirical Analysis of - PERI

14

5. Results

Descriptive Statistics

Before a discussion of the regression results, it is pertinent to look at some summary

statistics from the dataset. Table 2 shows the mean values for each variable used in the

estimation for both public and private schools. The last column shows the result of a t-test

conducted to determine whether or not the difference between the values for public and private

schools is significant. The average z-score of a public school student is significantly less than

that for a private school student. However, it is important to note that other variables which

might also have an effect on the score are also significantly different between the two categories,

hinting towards the existence of a selection bias.

Of particular interest are variables that explain the differing characteristics of students in

the two types of schools. On average 24.5% of private school students take tuitions compared to

only 7.8% of students enrolled in public schools. This indicates that taking tuitions might be an

important determinant of the differential in learning outcomes between public and private

schools. Similarly, a higher proportion of parents with privately enrolled children attended

school compared to parents of children enrolled in public schools. While 71.6% of fathers and

41.6% of mothers of private schools students were educated, the figures are much lower for

public schools (52.1% and 20.3% respectively.) These figures can imply that educated parents

might prefer private schooling and their children might also have the advantage of more help and

guidance from their parents in their schoolwork. Lastly, there is a significant difference in the

wealth index, indicating that richer parents can afford private schooling for their children and

prefer private schools to public schools.

Page 16: Prying the Private School Effect: An Empirical Analysis of - PERI

15

Table 2: Raw differences between public and private school students’ z-score

Public Private T value

Z-score 0.127 0.323 -30.05

Female 0.364 0.39 -7.9

Child age 9.501 9.04 21.04

Tuition 0.078 0.245 -78.98

Number of siblings 1.908 1.794 12.9

Father age 42.31 40.906 23.24

Father school 0.521 0.716 -59.65

Mother age 36.99 35.734 18.97

Mother school 0.203 0.416 -74.36

Birth order 1.934 1.96 -3.49

Wealth 0.236 0.401 -88.07

Note: The t-values indicate that all the variables are significant at the 5% significance level

Figure 1: Z scores for Public and Private School Students

Figure 1 above shows that the z-scores of public and private school students for the years

2010 and 2011. The difference between the achievement levels of private and government school

students is falling over the year, and the gradient for the students enrolled in public schools is

steeper than the one for private schools. This means that public schools students are catching up

with the private school students. However, this is the raw difference in z-scores and the data is

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

2010 2011

private public

Page 17: Prying the Private School Effect: An Empirical Analysis of - PERI

16

only for two years therefore this result might not hold when a longer time series is taken into

account.9

Pooled Regression:

The results for the estimation using OLS regression with fixed effects at the provincial,

district, village and household levels for the pooled data for the years 2010 and 2011 are

presented in Table 3. Fixed effects estimation at these four levels helps in controlling for those

characteristics that have an invariant effect on the test scores at the provincial, district, village

and household levels respectively. A dummy for the year 2010 has been added to analyze the

achievement differences over the two years. Significantly negative coefficients for the entire set

of regressions shows that the students on average performed better in 2011 than in 2010.10

Fixed effects at the provincial level help control for observable and unobservable

differences between provinces/regions. The Eighteenth Constitutional Amendment devolved

education to the provinces, which makes clustering at this level all the more important. There

exist cultural differences between provinces that can also be translated into different attitudes

towards education as well as different levels of political stability, for instance the military

operation on the tribal regions of FATA. However, even after using province fixed effects the

private school advantage of 0.103 standard deviations remains.

9 Analysis of a longer time series not possible in this paper as ASER Pakistan was formed only recently, and

therefore only reports the data for the past couple of years.

10 To see the regression results for the latest ASER data (year 2011) refer to appendix 1

Page 18: Prying the Private School Effect: An Empirical Analysis of - PERI

17

Table 3: Learning Assessment Pooled Regression using Fixed Effect Estimates

No FE Province FE District FE Village FE HH FE

Private 0.123***

0.103***

0.111***

0.0958***

0.0378***

(21.17) (5.97) (8.13) (12.04) (3.79)

Female -0.00436 -0.00848 -0.00904 -0.0100**

-0.0145***

(-1.11) (-0.55) (-1.12) (-2.20) (-3.16)

Child age 0.176***

0.178***

0.178***

0.177***

0.153***

(176.81) (24.06) (58.21) (122.61) (74.29)

Tuition 0.157***

0.146***

0.145***

0.141***

0.111***

(21.51) (3.81) (9.24) (14.77) (8.15)

Number of

siblings

0.0621***

0.0590***

0.0539***

0.0500***

(23.50) (8.21) (12.39) (14.90)

Father’s age 0.00882***

0.00404 0.00725***

0.0115***

(5.77) (1.35) (2.96) (6.50)

Father’s age sq. -0.0000655***

-0.0000325 -0.0000610***

-0.0000911***

(-4.29) (-1.44) (-2.69) (-5.71)

Father attended

school

0.0692***

0.0436* 0.0381

*** 0.0271

***

(11.76) (1.95) (2.97) (3.48)

Mother’s age 0.0159***

0.0169**

0.00546* 0.00348

*

(15.83) (2.96) (1.79) (1.68)

Mother’s age sq -0.000179***

-0.000187**

-0.0000648* -0.0000441

**

(-14.41) (-2.92) (-1.88) (-1.99)

Mother attended

school

0.124***

0.102***

0.0887***

0.0654***

(18.94) (10.00) (7.01) (9.48)

Birth-order -0.0542***

-0.0507***

-0.0501***

-0.0521***

-0.104***

(-17.00) (-9.04) (-8.54) (-12.62) (-20.06)

Wealth 0.00145***

0.00120***

0.00111***

0.000816***

(34.87) (9.63) (12.52) (14.11)

Year 2010 -0.205***

-0.220**

-0.198***

-0.185***

(-30.72) (-2.94) (-5.22) (-9.10)

Constant -2.512***

-2.323***

-2.141***

-2.119***

-1.125***

(-70.37) (-44.42) (-27.68) (-49.03) (-38.32)

Observations 104328 104328 104328 104328 104328

R2 0.4686 0.4685 0.5047 0.5629

t statistics in parentheses

*p< 0.10,

**p< 0.05,

***p< 0.01

Moving the analysis to the district level, some district are largely urban as opposed to

others, which means that they do not represent a typical rural area that one would observe in a

Page 19: Prying the Private School Effect: An Empirical Analysis of - PERI

18

predominantly rural district. Therefore there is an inter-district variation in the extent of urban

influence on the rural area that needs to be controlled for. Furthermore, since the district is the

smallest administrative unit in our study, fixed effects at this level also take into account the

differences in resources allocation to the education sector. We find that district level fixed effects

regressions still show a private school advantage of 0.11 standard deviations.

At the village level we then control for difference between villages in terms of school

quality and the number of educational opportunities available to the children. For example,

residents of villages closer to towns or cities have relatively better developed educational

facilities as well as the possibility of attending the schools in the nearby urban centers, as

opposed to residents of villages farther away from urban areas. The private school effects falls to

0.096 standard deviations when we use village level fixed effects, which means that differences

at the village level were critical in determining achievement levels.

At the household level, we control for household level characteristics that are uniform for

all children in a household e.g. age of the parents, the wealth level of the household and whether

or not the parents attended school. It is important to note however that the private school dummy

remains highly significant at all levels even after the strongest fixed level estimation at the

household level.

Estimation at the household level enables us to study the private school affect assuming

that all the children in a household face an identical environment. Therefore the assumption is

that unobservable characteristics like the attention and care that the parents give to the children

are uniform and that siblings in a household have the same ability. This implies that when

household fixed effects are employed, we are observing the same child attending different

Page 20: Prying the Private School Effect: An Empirical Analysis of - PERI

19

schools. Although it is a very strong assumption, the basic premise of siblings being more like

each other than any other individual still holds. This also suggests that this narrowest

specification gives the tightest upper bound on the private school effect that is still turning out to

be a significant 0.038 standard deviations.

Private school students perform 0.12 standard deviations better than public school

students in simple OLS. The coefficients on the private school dummy have a decreasing trend as

the fixed effects estimation becomes more stringent. When we move to household level fixed

effects, this difference decreases to 0.038 standard deviations. These coefficient estimates were

significantly below the OLS estimates indicating that using fixed effects eliminates a large

fraction of the bias that remained in the simple OLS. However, from the least (provincial level)

to the strictest fixed effects (from provincial level to the household level) the private school

effect remains significantly positive, implying that though there are various factors determining

achievement levels, the effect remains consequential.

Figure 2 below shows the unadjusted (raw) difference between the achievement levels of

private and public school students versus the adjusted (after controlling for other variables –

regression estimates from Table 3) differences. The unadjusted difference is almost 0.2 standard

deviations and falls to 0.123 using simple OLS. It continues to fall as fixed effects are used at

various levels and is lowest for the household level. This means that at least 0.138 standard

deviations difference in achievement levels is explained by factors other than private school

enrollment; however private enrollment remains a significant factor.

Page 21: Prying the Private School Effect: An Empirical Analysis of - PERI

20

Figure 2: Adjusted-Unadjusted gaps between public and private school students

Overall, factors that were intuitively expected to impact the test score positively were

seen to do so. These include the age of the child, ages of parents, wealth, father and mother’s

school attendance, and private tuitions outside school. A factor that can have an ambivalent

effect: the number of siblings of a child is seen to impact the test score positively. This suggests

that the educational benefit from having additional siblings outweighs the potential negative

effects of the same. However, the effect of birth order has a negative effect on the score: so a

child lower down in the birth order performs worse than an earlier-born child. This possibly

indicates that less investment or attention is given to younger children compared to their older

siblings.

The ages of the parents have a non-linear effect on the test score i.e. for both mothers and

fathers, test score increases positively up to an optimal age suggesting younger parents devote

more energy and resources into child rearing but the effect tapers off as the parents grow older.

0

0.05

0.1

0.15

0.2

0.25

Raw OLS Province District Village HH

Unadjusted difference Adjusted difference

Page 22: Prying the Private School Effect: An Empirical Analysis of - PERI

21

A dummy variable captures the effect of the gender of the child on the test score11

. The negative

coefficient on this variable in all the pooled regressions suggests that if the child is female, the

test score is lower suggesting some level of gender discrimination. It is important to note that

gender discrimination is significant at the village and household fixed effects level implying that

it is a within-village and within-household phenomenon. For the household fixed effects level

boys perform 0.015 standard deviations better than girls. This suggests that the evil of gender

discrimination has to be eradicated from the household level to have a nationwide impact.

Oaxaca Decomposition

Finally the Oaxaca decomposition is applied to the regression results to disaggregate the

effects of differences in endowments and private schooling/ unobservables. Table 4 below lists

the decomposition details.

Table 4: Oaxaca decomposition

endowment

differences

Private

school

effect

sum

Female 0.000589 0.013814 0.014402 Child Age -0.08335 0.066019 -0.01733

Tuition 0.021354 -0.00707 0.014285

Number of siblings -0.00591 -0.01885 -0.02476

Father’s Age -0.0183 0.386714 0.368413

Father’s Age squared 0.013128 -0.16139 -0.14826

Father’s School

Attendance

0.015538 0.019941 0.035479

Mother’s Age -0.01077 -0.30981 -0.32058

Mother’s Age Squared 0.011749 0.113048 0.124797

Mother’s School

Attendance

0.018789 -0.0021 0.016685

Birth Order -0.00025 0.033396 0.033146

Wealth 0.041556 -0.05347 -0.01192

0.004112 0.08024 0.084352

Proportion 0.0487 0.9513

11

This assumed a value of 1 if the child is female.

Page 23: Prying the Private School Effect: An Empirical Analysis of - PERI

22

Observing the decomposition at a disaggregated level we see that, on average, parents of

children attending private school are younger and have fewer children. Table 4 shows that

around 5% of the variation in the scores between public and private schools is explained by

endowment differences between the two while 95% is owing to the private school effect or some

unobservables like innate abilities. This implies that the quality of education, school facilities, etc

offered in the private schools contribute proportionately much more to raising the score of

children as compared to the differential in endowments.

6. Conclusion

The main objective of this study was to critically analyse the differential in learning

outcomes of students enrolled in public and private schools using an econometric technique that

controlled for biases at the provincial, district, village and household levels to disaggregate the

true ‘private school effect.’ The existence of such an effect would imply that children enrolled in

private schools are able to outperform their peers in public schools despite any differences in

endowments. It also highlights the existence of a private school premium, perhaps caused by

better provision of education or school facilities that are lacking in public schools. The existence

of such a differential therefore feeds into the debate on the level and quality of education being

provided by the state.

The results of the study suggest that the private school impact is significant firstly

because the dummy for private school in the pooled regression is significant at all levels with a

positive sign. Additionally the Oaxaca decomposition shows that even when endowment

differences between children enrolled in the two kinds of schools are controlled for, a substantial

proportion of the mean difference still remain unexplained which hints towards better learning in

Page 24: Prying the Private School Effect: An Empirical Analysis of - PERI

23

private schools. However the study does not control for endogeneity completely so it is unclear

whether the positive differential is entirely due to better quality of education provided in private

schools or some unobservables such as innate abilities of children, difference in motivation and

performance of teachers etc. Furthermore, school level factors like quality of teachers or school

facilities have not been controlled for so the differences in outcomes are not explained fully.

These suggest important directions for future research in this area, with this preliminary study

providing some interesting results that pave the way for more rigorous investigation of this issue.

Page 25: Prying the Private School Effect: An Empirical Analysis of - PERI

24

References

Alderman, Harold, Peter F. Orazem and Elizabeth M.Paterno. “School Quality, School Cost and

the Public/Private School Choices of Low-Income Households in Pakistan.” The Journal

of Human Resources 36.2 (2001): 304-326.

Alderman, Harold, Jooseop Kim and Peter F. Orazem. “Can Private School Subsidies Increase

Enrollment for the Poor? The Quetta Urban FellowshipProgram.” The World Bank

Economic Review 13.3 (1999): 443-465.

Andrabi, Tahir, Jishnu Das ad Asim I. Khwaja. “The Rise of Private Schooling in Pakistan:

Catering to the Urban Elite or Educating the Rural Poor?”

---. “Private Schooling: Limits and Possibilities” 2005

---. “A Dime a Day: The Possibilities and Limits of Private Schooling in Pakistan.” (2006)

Arif, G.M., and N. Saqib. “Production of cognitive life skills in public, private, and NGO

schools in Pakistan.” Pakistan Development Review 42.1(2003): 1–28.

Asadullah, M.N. “The effectiveness of private and public schools in South Asia.” Mimeo.

Centre for the Study of African Economies, Department of Economics, University of

Oxford, UK.(2005)

Aslam, Monazza. “The Relative Effectiveness of Government and Private Schools in Pakistan:

Are Girls Worse off?” Education Economics 17.3 (2009): 329-354.

Page 26: Prying the Private School Effect: An Empirical Analysis of - PERI

25

Bedi, Arjun S. and Garg, Ashish. “The Effectiveness of Private versus Public Schools: the Case

of Indonesia.” Journal of Development Economics 61 (2000): 463-494.

Beegle, Kathleen, and David Newhouse. “The Effect of School Type on Academic

Achievement: Evidence from Indonesia.” The Journal of Human Resources 41.3 (2006):

529-557.

Brunello, Giorgio, and Lorenzo Rocco. “Educational Standards in Private and Public Schools.”

The Economic Journal 118 (2008): 1866-1887.

Cullen, Julie Berry, Brian A. Jacob and Steven Levitt. “The Effect of School Choice on

Participants: Evidence from Randomized Lotteries.” Econometrica 74.5 (2006): 1191-

1230.

Das, Jishnu, PriyankaPandey and Tristan Zajonc. “Learning Levels and Gaps in Pakistan.”

World Bank Policy Research Working Paper WPS4067 (2006).

Federal Bureau of Statistics. 2001. Census of private educational institutions in Pakistan,

Federal Bureau of Statistics, Statistics Division, Government of Pakistan.

Grimes, Paul W. “Public versus Private Secondary Schools and the Production of Economic

Education.” The Journal of Economic Education 25.1 (1994): 17-30.

Jimenez, Emmanuel, Marlaine E. Lockheed and Vicente Paqueo. “ The Relative Efficiency of

Private and Public Schools in Developing Countries.” The World Bank Research

Observer 6.2 (1991): 205-218.

Page 27: Prying the Private School Effect: An Empirical Analysis of - PERI

26

Jimenez, Emmanuel, Marlene E. Lockheed and Nongnuch Wattanawaha. “The Relative

Efficiency of Private and Public Schools: The Case of Thailand.” The World Bank

Economic Review 2.2 (1988):139-164.

Oaxaca, R. “Male–female wage differentials in urban labor markets.”International

Economic Review 14 (1973): 693–709.

Rouse, C.E. “Private School Vouchers and Student Achievement: An Evaluation of the

Milwaukee Parental Choice Program.” The Quarterly Journal of Economics 113.2

(1998): 553-602.

Thapa, A. “Does private school competition improve public school performance? The case of

Nepal.” Unpublished thesis, Columbia University (2011)

Page 28: Prying the Private School Effect: An Empirical Analysis of - PERI

27

Appendix 1

2011 Learning Assessment Cross-sectional Regression using Fixed Effect Estimates

No FE Province FE District FE Village FE HH FE

Private 0.121***

0.0952***

0.104***

0.0938***

0.0582***

(18.18) (4.03) (6.55) (11.00) (5.04)

Female -0.00834* -0.0134 -0.0163

* -0.0188

*** -0.0175

***

(-1.85) (-0.81) (-1.94) (-3.85) (-3.32)

Child age 0.176***

0.177***

0.179***

0.177***

0.154***

(159.16) (28.01) (55.97) (115.41) (67.22)

Tuition 0.162***

0.143***

0.133***

0.133***

0.0942***

(18.94) (4.14) (7.58) (13.26) (5.84)

Number of siblings 0.0593***

0.0562***

0.0498***

0.0488***

(20.13) (9.00) (10.18) (15.32)

Father’s age 0.00807***

0.00415 0.00647***

0.00984***

(5.05) (1.72) (3.14) (6.77)

Father’s age sq. -0.0000546***

-0.0000312 -0.0000508**

-0.0000738***

(-3.48) (-1.67) (-2.56) (-5.27)

Father attended school 0.0563***

0.0710***

0.0578***

0.0346***

(8.56) (6.97) (4.57) (4.70)

Mother’s age 0.0159***

0.0162**

0.00581***

0.00434***

(15.59) (2.38) (3.13) (3.75)

Mother’s age sq. -0.000180***

-0.000169* -0.0000642

*** -0.0000461

***

(-14.34) (-2.23) (-2.93) (-3.39)

Mother attended school 0.106***

0.0852***

0.0723***

0.0486***

(14.24) (4.45) (5.18) (6.71)

Birth-order -0.0524***

-0.0495***

-0.0468***

-0.0500***

-0.100***

(-14.79) (-7.40) (-7.03) (-11.79) (-17.55)

Wealth 0.00127***

0.000987***

0.000967***

0.000636***

(27.41) (5.15) (9.31) (12.34)

Constant -2.443***

-2.294***

-2.117***

-2.069***

-1.103***

(-64.11) (-33.21) (-27.46) (-53.20) (-33.97)

Observations 80310 80310 80310 80310 80310

R2 0.4698 0.4697 0.5171 0.5599

t statistics in parentheses

*p< 0.10,

**p< 0.05,

***p< 0.01