The Effects of Private and Public Schooling on Learning Outcomes in India

Evidence from Andhra Pradesh

Bachelor Thesis , 2019 , 45 Pages , Grade: 1,3

Pedagogy - School System, Educational and School Politics

This paper tries to estimate the causal effect of one year of private schooling in comparison to public schooling on standardized test score outcomes for Math and English in India, Andhra Pradesh. I also try to estimate the effect of fees on test score outcomes for private schools.

I use a 2-wave panel data set of the Young Lives School Survey of 2016 and 2017. I estimate the private schooling effect with a value-added model, using problem-solving and critical-thinking test scores as control for fixed cognitive ability of pupils. I use a similar model to evaluate the effects of fees. Effects are estimated separately for rural and urban environments. I find higher teacher wages, higher teacher absenteeism, less instructional time and better school infrastructure for public schools. The high cost associated with public schools seems to largely result from higher teacher wages.

I find statistically significant positive private schooling effects in the urban sample for both Math (0.314 SD) and English (0.228 SD). In the rural sample there is no private schooling effect for math and a positive effect for English test scores (0.393 SD). The positive effect for English test scores persists after controlling for the medium of instruction, in both environments but with a smaller effect size.

In absolute terms the performance for both private and public schools is underwhelming, and any positive private schooling effects only translate to a small absolute improvement for test scores. For the effect of fees, I find that higher fees have a positive effect on private school performance only in the urban sample with decreasing marginal returns. For the rural sample I find a negative effect for higher fees.

Content

Abstract

List of Tables

List of Figures

Abbreviations

1. Introduction

2. Private schools in India
2.1 Typology of schools
2.2 The share and growth of the Indian private school market
2.3 Fee and cost structure of private and public schools
2.4 The “Right to Education Act” and Public Private Partnership

3. Prior research of private schooling effects in India 8
3.1 Quasi-experimental estimate
3.2 Propensity score matching
3.3 Value-added model estimation

4. Theoretical Framework 12
4.1 Input factors and theoretical channels
4.2 Conceptual basis of the VAM model

5. Young Lives School Survey 15
5.1 Description of relevant variables
5.2 Summary Statistics
5.3 Attrition

6. Econometric Model

7. Results
7.1 Estimates for Math and English test scores
7.2 Results for effects of fees on private school performance
7.3 Limitations

8. Conclusion

9. References

10. Appendix

Abstract

This paper tries to estimate the causal effect of one year of private schooling in compari- son to public schooling on standardized test score outcomes for Math and English in In- dia, Andhra Pradesh. I also try to estimate the effect of fees on test score outcomes for private schools. I use a 2-wave panel data set of the Young Lives School Survey of 2016 and 2017. I estimate the private schooling effect with a value-added model, using prob- lem-solving and critical-thinking test scores as control for fixed cognitive ability of pu- pils. I use a similar model to evaluate the effects of fees. Effects are estimated separately for rural and urban environments. I find higher teacher wages, higher teacher absentee- ism, less instructional time and better school infrastructure for public schools. The high cost associated with public schools seem to largely result from higher teacher wages. I find statistically significant positive private schooling effects in the urban sample for both Math (0.314 SD) and English (0.228 SD). In the rural sample there is no private schooling effect for math and a positive effect for English test scores (0.393 SD). The positive effect for English test scores persists after controlling for the medium of instruction, in both environments but with a smaller effect size. In absolute terms the performance for both private and public schools is underwhelming, and any positive private schooling effects only translate to a small absolute improvement for test scores. For the effect of fees, I find that higher fees have a positive effect on private school performance only in the urban sample with decreasing marginal returns. For the rural sample I find a negative effect for higher fees.

List of Tables

Table 1 Percentage of children in private schools, by state and age, 2014-2015

Table 2 Summary statistics for the rural and urban sample

Table 3 Private schooling effect on normalized Math test scores

Table 4 Private schooling effect on normalized English test scores

Table 5 Private schooling effect on Math test scores

Table 6 Private schooling effect on English test scores

Table 7 Effect of fees on private school performance

List of Figures

Figure 1 Typology of private and public schools

Figure 2 Private school share in India and Andhra Pradesh

Figure 3 Histogram of yearly private school fees in urban Andhra Pradesh

Figure 4 Histogram of yearly private school fees in rural Andhra Pradesh

Figure 5 Sectioned wages of teachers by school type

Figure 6 Experimental design of the MS study

Figure 7 Young Lives Sites in Andhra Pradesh

Abbreviations

Abbildung in dieser Leseprobe nicht enthalten

1. Introduction

India has one of the largest student’s body in the world comprising about 260 million children. They also have the largest growing total youth population in the world (Gov- ernment of India, 2018). This represents great potential and a liability at the same time. Only if India can educate their large student body adequately will they be able to sustain- ably reduce poverty in the long term. Widespread poverty is still prevalent at 27,5%¹ of the total population of India remaining in poverty, according to the Global 2018 Multidi- mensional Poverty Index (Oxford Poverty and Human Development Initiative, 2018).

In recent years private schools have seen strong growth in urban and rural India, leading to a migration from public to private education (Kingdon, 2019). The parental abandon- ment of public schools calls into question their quality of teaching. Private schools show on average better test results than their public counterpart (ASER , 2019). A common critique of better private school performance is that it follows as a result of “cream-skim- ming”. This denotes that pupils in private school perform better because they have higher cognitive ability and a more positive socio-economic background. To convincingly esti- mate the causal private schooling effect (PSE), it is necessary to control for selection bias, because without random assignment, children might self-select into different school types based on observable and non-observable factors. Available literature on this topic is sur- prisingly spare considering the importance of the issue.

This paper tries to estimate the effect of one year of secondary private schooling on learn- ing outcomes including standardized test scores for Math and English in Andhra Pradesh (AP). I use a 2-wave panel data set of the Young Lives School Survey (YLSS) of 2016 and 2017. I also try to estimate the effect of fees on private school performance. To my knowledge this data set hasn’t been used for analysing the private school effect in India even though it provides a large sample size and a great number of covariates. There also isn’t much literature available on the effect of fees on private schools. To estimate the PSE and the effect of fees, I use a value-added model (VAM) with problem-solving (PST) and critical-thinking test scores (CTT) as control for the cognitive ability of pupils. I es- timate all effects separately for both urban and rural settings to account for the vast dif- ferences in these environments.

I find significant PSE for English test scores for both urban and rural environments. The effects persist after controlling for the medium of instruction with a smaller effect size. For Math test scores the results are more ambiguous, as I find no PSE effect for the rural setting but a significant effect for the urban setting. Private schools seem to on average provide better or at least the same learning outcomes with less resources, as such I find a positive PSE achieved with lower per pupil cost (PPC). Larger fees have a positive effect on test score outcomes for the urban sample and a negative effect in the rural sample.

In the first section I will provide an overview of the public and private education system in India, drawing from my own results and previous literature. In Section 2 I will assess the already established literature on the topic of private schooling effects in India with an emphasis on their statistical methods. In Section 3 I will establish the theoretical frame- work on which the VAM is based. Section 4 is a description of the Young Lives School Survey; I also define relevant variables and provide summary statistics. I describe my econometric model in Section 5 followed by presenting the results in Section 6. Lastly, I conclude my results in section 7.

2. Private schools in India

2.1 Typology of schools

There are four broad categories of schools that can be identified in India: public schools, private aided schools, private recognized schools and private unrecognized schools. The obvious difference being that private schools require fees in comparison to public schools². Private schools can be further differentiated between recognized and unrecog- nized. Recognized schools are officially acknowledged by the government, which allows them to emit certificates. These are required to allow for advancing in the official school system. Because fulfilling all criteria to be acknowledged is costly and takes much time, many private schools remain unofficial (Tooley & Pauline Dixon, 2005). The solution for children attending unrecognized schools is often double enrolment, which means enrol- ling in a public school to be able to pass exams and acquire certificates, as described by Kremer & Muralidharan (Kremer & Muralidharan, 2009). They also provide access to public school supplies. This solution is tolerated by all sides on the local level. It allows public schools to report higher enrolment rates. Parents can let their child attend a school they perceive as better and are still able to acquire certificates and additional school sup- plies. Unrecognized schools benefit by higher utility for their pupils and lower cost.

Part of the recognized schools are private aided schools. These schools are supported by either the state or central government or both. Their governance structure follows closely that of the public schools. Government influence on these school is strong, and their teachers are part of the public teacher unions and get payed by the government. As a result, those schools are de jure private but can most definitely be considered as de facto public schools, since they share a similar governance- and infrastructure. When evaluat- ing different surveys, it is important to consider how the distinction between private and public schools is drawn. Many official government statistics also don’t include unrecog- nized private schools even though they are a large part of the Indian education system (Kingdon, 2017).

To summarize the definition of private schools in this paper: private recognized schools and private unrecognized schools count towards the private school system, except for private aided schools. Private aided schools and government schools count towards the public-school system. See Figure 1 for a visualization of the definition.

Abbildung in dieser Leseprobe nicht enthalten

Source: Definition follows Kingdon (Kingdon, 2017); Author’s own illustration

Figure 1 Typology of private and public schools

2.2 The share and growth of the Indian private school market

In Table 1 you can see the share of private school pupils by age and state for the period of 2014-2015. The categorization follows the definition set out beforehand. When we look at the total share, independent of age in India, we see that in total about 30% are in private school. In rural India it is 20%. For urban India it is about double at 40%. A general trend for all states and ages is higher private school enrolment for urban settings. We also see that AP (and Telangana)³ have far higher private school shares in comparison to other states. Thus, the high share of private school enrolment in AP might be caused by a comparatively more dysfunctional school system. Another explanation could be that AP is more open to the establishing and operation of private schools than other states.

Looking at age structure in urban areas, there is a trend of declining private school shares with higher age. In rural India private school attendance is highest for the oldest cohort. This effect might be produced by general lower enrolment in rural areas for the 11-18 age bracket in public school, while private school pupils might stay enrolled for longer. What stands out is that AP goes against this trend. Their private enrolment rate is very high for the oldest cohort in urban and rural areas. Especially in rural areas their private enrolment rate is higher than the Indian average. In Figure 2 you can see the trend of private school shares in AP and India in total between 2010-11 and 2015-16. The trend is based on DISE raw data.

Abbildung in dieser Leseprobe nicht enthalten

Source: raw data from DISE, available at www.dise.in; Author’s own illustration

Figure 2 Private school share in India and Andhra Pradesh

Similarly, to the NHS data of 2014-15, AP has a larger private school share in comparison to India as a whole. But its growth of the private school sector is slower than for the rest of India. In AP it grew by 5 percentage points over 5 years and in India by 9 percentage points. It seems like the rest of India is catching up to private schooling levels of AP, which would speak against the notion that AP is an outlier and isn’t representative for the educational system of India as a whole.

2.3 Fee and cost structure of private and public schools

Previous literature consistently finds lower PPC for private schools. Estimates range be- tween ⅓ and ½ of operating cost in public schools (Kingdon, 2017) (Muralidharan & Sundararaman, 2015). They also find that public teachers are strongly unionized and re- ceive wages far above market rate. Teachers of private schools also are often younger, bare less credentials and have higher attendance and teaching rates. The head principle often exerts more control over his teachers by incentive schemes and more liberal hiring and firing. They have smaller teacher-pupil ratios and less multigrade teaching. They op- erate for more days in the year and for more hours per day (Muralidharan & Sundarara- man 2013). Reports show that in public schools, teacher absenteeism is a large problem, with public school teachers in AP being absent on average 27% of time (Kremer et al., 2015). This is likely due to their strong unionization, allowing them to shirk without get- ting fired.

Which complicates the question of operating cost further is double enrolment. Many pu- pils that are enrolled in private unrecognized schools also are enrolled in public school. This enables them to get access to public school supplies and enables them to get official certificates with which their pupils can apply for secondary or tertiary education (Kremer & Muralidharan, 2009). In most cases they mainly attend private school, for (perceived) better schooling, and attend public schools for exams and school supplies. This means that the cost of the private school system is partially carried by the public schools in the form school supplies like books and pencils. But even accounting for these estimation biases couldn’t close the large cost gap between low-fee private and public schools (Kingdon & Azam, 2015).

Fig. 3 and Fig 4 represents the fee structure present in my sample for the academic year 2016-2017 separated by school type and environment. As seen most schools certainly fall in the category of what is commonly referred to as low fee schools. This is expected as the schools of my sample are visited by Young Lives Survey children. Since the survey focuses on children of poor or average background it is unlikely that any elite private school would be part of the survey. We can also see that fees are in general lower in the rural sample opposed to the urban sample. This too is expected as rural residents on av- erage have lower incomes and possibly can’t afford similar fees.

Abbildung in dieser Leseprobe nicht enthalten

Note: *Andhra Pradesh also includes the state of Telangana

Source: Raw data from the Young Lives School Survey; Author’s own illustration

Figure 3 Histogram of yearly private school fees in urban Andhra Pradesh*

Abbildung in dieser Leseprobe nicht enthalten

Note: *Andhra Pradesh also includes the state of Telangana

Source: Raw data from the Young Lives School Survey; Author’s own illustration

Figure 4 Histogram of yearly private school fees in rural Andhra Pradesh*

As I don’t have data on the cost of public schools, I rely on estimates of Dongre & Kapur. They find PPC on average of about Rs. 14400 in a year for public schools in AP (Dongre & Kapur, 2016). For my whole sample I find average fees of about Rs. 9760 for private schools. It should be noted that Kingdon et al. suspects Dongre & Kapur’s estimate to be an underestimation, mainly because they don’t include pensions, free supplies provided by the central government and overreporting of enrolment (Kingdon et al., 2016). Re- gardless there is a large gap in costs between private and public schools. What drives the difference in cost? I have data on sectioned teacher wages. This doesn’t allow me to do a thorough comparison, but it at least enables a gauge. Looking at Fig. 5 you can see that in public school teachers in general receive higher wages than in private school, the graphs also resemble a normal distribution for both environments.

Abbildung in dieser Leseprobe nicht enthalten

Source: Raw data from the Young Lives School Survey; Author’s own illustration

Figure 5 Sectioned wages of teachers by school type

Public schools also seem to enjoy better infrastructure, at least in the Inputs observed. The differences are rather marginal though in comparison to teacher salaries. So, it can be concluded that the gap in cost between private and public schools is explained mainly by teacher salaries which is in line with prior research.

2.4 The “Right to Education Act” and Public Private Partnership

In 2009 the Indian government has enacted the “Right of Children to Free and Compul- sory Education Act” or “Right to Education Act” (RTE). The Act is supposed to ensure free and compulsory schooling to children between the ages of 6 and 14. The legislation also requires private schools to admit 25% of disadvantaged students. The admission of disadvantaged students is supposed to be reimbursed by the government at cost equivalent to public schools. Schools also aren’t allowed to interview the children or parents. The act also requires private schools to fulfil certain input criteria like teacher-pupil ratios, teacher pay, teacher qualifications and physical infrastructure. Not complying with these rules is supposed to result in closure of the school. According to the India School Closure Report 2018, about 2500 schools have been closed consequently and 5000 are threatened to be closed (Anand, et al., 2019). Some schools report that they aren’t reimbursed the full amount for disadvantaged pupils set by the state according to an Oxfam India report of 2014 (Sarangapani & Mehendale, 2014).

Considering the vast amount of bureaucratic work and time required to acquire an official schooling license and the corruption and bribes involved, this represents a new obstacle to private schools (Wadhwa, 2001). One the one hand starting a private school without proper license now involves the risk of shutdown. On the other hand, compiling with rules and regulations is associated with high cost and takes much time and possibly many bribes. This represents a clear competitive disadvantage for private school and might hin- der growth of the private school sector.

3. Prior research of private schooling effects in India

There already exists some literature that tries to estimate the effects of private schooling on learning outcomes in comparison to public schools in India. The general empirical problem that presents itself is that of selection bias. Simple OLS without any controls would only yield an upper bound for the PSE. In the following I present some recent literature on the topic of private schooling effects in India and emphasize their statistical method as much of the descriptive findings already have been discussed in the previous section.

3.1 Quasi-experimental estimate

When estimating the PSE on learning outcomes, the ideal setup would be to measure the same individual at the same time in either school. For obvious reasons this is not possible. The next best solution is to look at sample groups that are as similar as possible in all factors that influence learning outcomes. As discussed before this is most likely not the case if one simply looks at a sample created by self-selection. A possible solution to this problem is a setup which allocates students randomly to both type of schools. Granted the sample size is big enough to create an equal distribution of characteristics – simply com- paring the average outcomes should yield the PSE in the aggregate.

Muralidharan & Sundararaman (MS) use a quasi-experimental setup of a two-stage lot- tery-based allocation of vouchers to evaluate the impact of private schooling on learning outcomes for all subjects at the aggregate and individual level in AP for 2 and 4 years of private schooling (Muralidharan & Sundararaman, 2015). They also estimate several spillover effects to make the assessment of a large-scale voucher program possible. One of the big weaknesses of the analysis is that it is limited to rural areas. It is likely that both environments produce different PSEs as they face different competitive environments. In my analysis I will estimate two separate PSEs for both the urban and rural environment. Still their estimate provides a great measure for comparison to my PSE in rural areas, as their methodology used is one of the most robust against selection bias.

Their descriptive findings confirm that of other research: private schools have higher at- tendance by teachers, operate longer and on more days in the year. They also find that private school are more hygienic, teachers are younger, have less credentials, class sizes are smaller and there is less multigrade teaching. Time use and curricula also differs. Private schools spend 40% less instructional time on Telugu and 32% less time on math. Instead they spend more time on English, Hindi and EVS⁴. This runs contrary to my findings, as in my sample instructional time for Math and English is higher in private schools.

They conclude that private schooling doesn’t have significant effects on learning out-comes for Telugu and Math, but significant positive effects on Hindi⁵, English and EVS. On average across all subject’s private school students scored 0.13 SD higher, with the casual impact of private schooling being 0.23 SD. This is achieved even though instruc- tional time and resources spent is lower in private schools. This suggest that overall pri- vate school has slight positive effects on learning outcomes and is much more efficient. They also find heterogenic effects for private Telugu and English medium (EM) schools. EM schools unsurprisingly improve English scores but at the cost of worse performance in other subjects. This finding is also supported by other literature. Ramachandran shows that switching to the mother-tongue medium of instruction leads to significant increases in educational outcomes for Ethiopian children (Ramachandran, 2012). For this reason, I will also control for the medium of instruction in the results of the Math and English test scores. They find no significant spillover effects, which among other things, shows that the better test results of private school pupils don’t have adverse effects on other children.

In the following I describe their experimental setup more in depth. Their sample includes several villages in AP, of which some were assigned randomly to the treatment group, in form of partaking in the lottery, the rest is part of the control group. In villages of the treatment group, some children were randomly assigned a voucher. Using the voucher isn’t mandatory and they could choose to stay in public school. The survey followed the pupils for 4 years with surveys taken at the end of the 2nd and 4th school year. This setup not only allows to estimate individual effects but also spillovers. This allows identifica- tion of three different types of spillovers, between the groups of 2T and 2C, 1T and 1C & 4T and 4C, see Figure 7 for clarification. They estimate the effect of attending private school by using the offer of a voucher as an instrumental variable for the PSE. they also control for lagged test scores and fixed effects on the district level and a set of household and socioeconomic factors.

3.2 Propensity score matching

Chudgar & Quin use the propensity score matching technique of Rosenbaum & Rubin (Rosenbaum & Rubin, 1983). As Data they use the India Human Development Survey of 2005 (IHDS). They too estimate the PSE separately for urban and rural India. The IHDS includes children aged 8-11 from 41,554 households. They also try to identify low and high fee schools to make a fair comparison possible.

Contrary to research of others including my findings, they find no significant PSE. Their results only apply when excluding high-fee schools. They do this by setting a threshold.

This threshold simply is the comparison of fees of the local private school to the PPC of public school. If the private school charges less than the PPC of public schools in fees they are classified as low fee school. But it seems arbitrary to include schools that operate with much less PPC but exclude any schools that have a bit higher PPC. It doesn’t even the playing field, since the average PPC for private schools in their sample still remains lower in that case.

Their available data on test score outcomes are reading, math and writing tasks, these are measured in categorical criteria’s. This makes measuring marginal improvements very hard. The range for the reading test is 0 (cannot read at all) to 4 (can read a whole story). For math the scale only included 3 values (from “cannot recognize a 2-digit number” to “can divide a 3-digit number by a 1-digit number”). Writing was measured in a base cat- egory. They report none of these scores to follow a normal distribution. To make use of the full variation of pupil skills they estimate the PSE as a summation of all three tests called SCORE. They also use a 2nd outcome measure they call PROF to produce a nor- mally distributed score. PROF is made up by an ordinal evaluation of pupil skill: 1 - bottom third of distribution, 2 -medium, 3 -top third of distribution.

Their method of matching on covariates also greatly limits their initially great sample size (about 15% of the initial rural sample size and 30% of the initial urban sample size). It has the advantage of aligning the density of distribution. They acknowledge that this method still doesn’t reliably solve the selection bias. Even when matching observables variables, unobservable variables can still deviate. Their methodology is very elaborate, but their outcome variables are so broadly defined, that marginal improvements in edu- cation can hardly be measured. An important point they make is the heterogeneity of private schools based on the fee levels charged. They advise for future research to account for the heterogeneity of private schools when evaluating their performance. I incorporate this by identifying the effects of fees on private schools in an extra sample that only in- cludes private schools.

3.3 Value-added model estimation

Singh uses panel-data of the 2010 Young Lives Survey (YLS) and uses a VAM to analyse the PSE (Singh, 2015). He estimates the PSE separately for both rural an urban environ- ment and for two different age cohorts. His estimate for the older cohort, which were aged 16 at the end of the survey provide a useful comparison to my estimate, as I look at a similar age group and use a similar methodology.

In rural settings he finds positive effects for the older cohort especially for math and Tel- egu of about 0.2 SD. He finds no PSE for urban settings for both cohorts. His urban sam- ple size is small especially for public schools (67 pupils). This makes his result for the urban sample unreliable. For the younger cohort he finds only substantially better results in English. He also estimates PSE effects on psychosocial skills, though he finds no sig- nificant results.

In his model he makes use of a lagged variable test score as it is required for a VAM. He also includes a control for cognitive ability, which needs to be included to not bias the result upward. As proxy for pupil ability he uses Raven’s Test score. It isn’t clear whether my PST and CTT scores are better suited for controlling cognitive ability in comparison to Raven’s Test. He also doesn’t use equally weighed test scores, instead he uses Item Response Theory (IRT) to modify test results. The concept behind IRT is to assign weights to test questions based on their (estimated) difficulty. I will refrain from using IRT, because it opens the possibility to bias by pupils answering multiple choice (MC) questions right on a random basis with strong weights attached. It also adds further com- plexity to the model. His results for the rural sample are comparable to mine. For the urban sample I find very different results, as both test scores are positive and significant while he reports no effect.

4. Theoretical Framework

4.1 Input factors and theoretical channels

In this section I want to explore different input factors in the production of learning and emphasize their possible theoretical basis. Inputs on the household-level include house- hold income, number of siblings, parent’s education, family structure and whether it is conductive to study in the household. Income enables spending on nutrition and health as well as educational content, it might also induce a more conflict free household. The number of siblings influences the amount of money that can be spent per child and might also lead to certain children having to take up more work than others. The family struc- ture, for example age structure and who oversees household decision making could also have an immense impact on spending patterns relating to nutrition and human capital. For example, a very patriarchal household might value education for boys more than for girls. Parent’s education might affect how they value education and estimate wage returns on higher education.

Inputs on school-level include infrastructure like clean toilets, cooperation with parents, principal’s ability and effort, teacher’s ability and effort, teacher-pupil ratio, textbooks, utensils and location of school⁶. The effect of school infrastructure seems straightforward. A clean and pretty class environment induces higher productivity. Clean water sources and working toilets enable a hygienic environment and have an impact on child health, which in turn affects productivity and attendance. Cooperation and communication with parents can help align household and school tasks of children and enables parents to em- phasize relevant tasks relating to schooling. This effect might strongly depend on the in- volvement and importance the parents place on the education of their child and as such might be a very heterogenic effect. Teachers have a variety of possible effects on learning outcomes of children. First their ability and effort put into teaching, including their at- tendance which plays an especially important role regarding public education in India. They also can act as a role model especially in an environment of low educational achievement like a remote village.

Innate characteristics of the child include gender, health, age, genetic endowment and cultural and religious background. Gender itself might have an impact on many other channels like parent’s involvement and valuation of the child’s education. Health, age and genetic endowment could also directly affect learning capabilities and ability. Cul- tural and religious background and being of Scheduled Caste (SC) or of a Scheduled Tribe (ST) might also have significant effects on educational achievement.

4.2 Conceptual basis of the VAM model

One can think of human capital agglomeration being the result of a learning production function with several inputs. This production function is often thought of as a structural cumulative effects model (Singh, 2015). This means that educational achievement is a result of the history of a set of input variables. These inputs can be further categorized by relating to the student, household, school, village and so on. There is a large set of inputs that could be part of the learning production function as described in the previous chapter.

The Output of all these Inputs over time is human capital, which can be measured by wage returns in the long term. In the short-term human capital can be measured by cog- nitive abilities, more specifically through test scores. By measuring cognitive outcomes no exact effect on future wage returns can be made, but assuming a strict monotone rela- tionship between cognitive outcomes and human capital gains, a comparison about effi- ciencies between private and public school can still be made using test scores. Todd & Wolpin describe the learning production function as follows (Todd & Wolpin, 2007) (Todd & Wolpin, 2003):

Abbildung in dieser Leseprobe nicht enthalten

Yist is the achievement of pupil i at school s at period t as a function of a set of inputs. Todd and Wolpin categorize these Inputs in three ways. Xi describes Household-level inputs, an example would be household income. Si(t) describes the history of school-level inputs of the schools attended by the pupil at the respective year. An example would be the availability of computers in school. μis0 describes endowments of child i in school s, like gender or cognitive ability. Typically, it is not possibly to estimate the private school- ing effect by using Equation (1), because many inputs aren´t observed and/or their whole history isn’t known. If the functions of inputs are linear the production function can be written like Equation (2), following Andrabi et al. (Andrabi, et al., 2011):

Abbildung in dieser Leseprobe nicht enthalten

Yit is achievement of child i at time t. Xit represents a vector of inputs for child i at time t. summed μis are cumulative productivity shocks. Under the assumption that the coeffi- cients decline geometrically and adding and subtracting βyi,t−1 yields Equation (3). This is the lagged value-added model (Andrabi et al., 2011).

Abbildung in dieser Leseprobe nicht enthalten

By comparing Equation (2) and (3) you can see that under the assumptions made, the lagged variable controls for all previous inputs and for past endowments and shocks. This means that when at least two periods of test score data is available, no prior history of any relevant variables is needed, though it is still necessary to control for them in current estimation of test scores. This model is a vast improvement in comparison to a simple OLS only linking current test score and inputs, but the VAM still has many possible sources of bias. First are the assumptions already described beforehand, second introduc- ing a lagged test score enables measurement errors to bias the coefficient β but also could bias the input coefficients ά. For the lagged test score to capture the effect of cognitive ability it is necessary for it to enter through an endowment that has fixed effects. If, how- ever children with higher cognitive ability also learn faster, meaning fixed abilities have different effects at every age, then β and other input coefficient’s will be biased. For this reason, it is highly advised to control for cognitive ability, as correlation with the lagged variable test score and input factors will bias the results when omitted (Singh, 2015).

5. Young Lives School Survey

The Young Lives School Survey (YLSS) is a longitudinal study that was conducted for the first time in 2010-2011 in India, Peru and Ethiopia and was part of the Young Lives Survey (YLS). The Survey in India is restricted to schools of the Young Lives Sites in AP⁷. In 2016-2017 a second survey was conducted in two waves. The first wave was conducted in the period of July-August 2016 and the second wave in the period of Janu- ary-March 2017, respectively at the start and the end of the academic year. The subjects of the survey are pupils of secondary education in the 9th grade, English and Math teachers of the classes from which the pupils were interviewed and principals of the sample schools. The survey collects a wide array of data on the children including a repeated Math and English test in both waves. Another Test they took was either the critical-think- ing test (CTT) or problem-solving test (PST), though only in wave 2, the tests were split between them, meaning that they either took the CTT or the PST.

There are several school types recorded in this survey. They are differentiated by govern- ment schools, private schools (including both recognized and unrecognized) and Tribal/welfare schools (TWS). I will not use the Tribal/Welfare schools as part of my analysis since don’t make up a significant amount of the schools in India, as such they are not relevant to answering questions related to policy implementation regarding the broader school system.

The sampling was done at the site level to ensure that each site was represented in the survey. Fig. 7 shows the different Young Lives Sites in AP. The selection followed a simple draw at site level within each stratum. The number of schools sampled in each site is proportional to the number of schools in that site. To ensure a sufficient amount of every school type, a census-sample has been used to allow for all school types to be pre- sented more equally. In total, the sample covers 205 schools: 83 government schools; 54 private unaided schools; 30 private aided schools; and 38 tribal/social welfare schools. This adds up to 113 public schools and 54 private schools that are used in this analysis⁸.

5.1 Description of relevant variables

In the following I will more accurately describe the relevant variables. The relevant Var- iables measured include: English (ETS) and Math Test scores (MTS) in the 1st and 2nd round, CTT and PST scores in the 2nd wave, Gender, Age, Caste, Wealth Index, Father’s Education, Mother’s Education, Language Match, Absenteeism of Teachers, Household Index, Health Index, Time Use and Instruction Time. In the following I will explain these variables and their respective values in more depth.

All students had to complete a repeated English and Math standardized test at the begin- ning and end of the study. I mostly follow Moore et al. in my description of the design of these tests mostly follow. The 1st and 2nd test don’t differ, except for very small adjust- ments. Test scores are measured in percentage points of right answers, and as normalized scores.

The Math test is a MC test and consist of 40 questions. The test is designed to test curric- ulum knowledge of the 9th grade and the pupil’s ability to use these skills in unfamiliar context. The concepts covered are (Moore, et al., 2017):

- Basic number competency
- Integers, rational numbers, powers and bases
- Fractions, decimals, ratios and percentages
- Area, perimeter, volume and surface area
- Geometry and shapes
- Algebra
- Measurement, charts and graphs
- Reasoning, problem solving, and applications in daily life (Grønmo et al. 2015)

Those concepts are tested in three ways, knowing the concepts, applying them and rea- soning that goes beyond routine procedures.

The English test is functional, meaning that it represents the use of English in real life scenarios rather that testing curricular knowledge (OFQUAL, 2011). It also tests via MC and consists of 50 questions relating to reading & comprehension, and grammar. Because of testing via MC no assessment of writing skills is possible. The skill domains tested more specifically are:

- Vocabulary
- Identifying meaning of unfamiliar words through context
- Sentence completion
- Reading comprehension

Within this framework the tests focuses on the skills which 15-year old’s in AP currently use or may need in the future.

The just described Math and English Test scores are the relevant outcome variables of my analysis. ⁹ In the following I describe the PST and CTT, which serve as control for pupil’s cognitive ability. The tests were administered only in the 2nd round and no child took both tests but one or the other. Both tests are designed to test transferable skills in a cross-disciplinary manner. This serves the purpose as control variable for testing cogni- tive ability well, since subject specific knowledge isn’t required. The PST is supposed to test an individual’s ability to solve real, cross disciplinary problems without obvious so- lution paths (Greiff et al., 2013). The PST consists of 13 MC questions. The CTT is de- scribed is supposed to test an individual’s ability to use inference and evaluation to solve ill-structured problems for which no definitive solutions exist. (Kuhn 1999; Thomas and Lok 2015).¹⁰ Since the administration of those tests was random in assignment, but fixed on a 50:50 ratio, I assume their distribution across private and public school to be equiv- alent, which ensures that variation in difficulty across these tests doesn’t bias the results.

Gender, Age and Caste are straightforward. Age is measured in years; Caste is a dummy with the base category being of SC or ST. Gender is a dummy with being male as base category. The Wealth Index is a composite variable consisting of several dummy varia- bles relating to the ownership of certain items. The items are: bicycle, motorbike or scooter, television, electric fan, chair, table, mobile telephone, fridge and car or truck. Owning the Item represents the base category. The sum is divided by the total number of items. This means that the Wealth Index takes a value between 0 and 1, with 1 represent- ing the ownership of all items. Father’s and Mother’s Education is a categorical variable taking discrete values between 0 to 5 representing different levels of their highest educa- tion achieved. 0 represents no schooling. 1 represents Primary School (Class I-V), 2 Up- per Primary School (Class VI-VII), 3 High School (Class VIII-X), 4 Junior College (Class XI-XII), 5 Higher education (University, Diploma).

The English Medium variable is a dummy describing if the language used in the Section is English. If not, the used language is non-English, which is Telugu in most cases. Most schools fulfilling this are private schools teaching in English.

The Household Index consists of 5 variables. 4 of them relate to the frequency of interac- tion between parents and their child regarding education. They are answered using the following scale: 0 = never, 1 = less than once a month, 2 = 1-2 times a month, 3 = 1-2 times a week, 4 = every day / almost every day. The 1st variable is about how often the child is asked what they are currently learning in school. The 2nd variable is about how often the child speaks about his homework with anybody at home. The 3rd is about whether anybody at home makes sure the child sets aside time for homework. The 4th variable is about whether anybody at home checks if the child is doing his homework. The 5th variable is a dummy taking a value of 4 in case there is a quiet place at home for the child to study, and 0 otherwise. The Sum of these variables is divided by 20 to result in an Index that ranges from 0 to 1, with 1 representing the fulfilment of all items.

The Health Index consists of 6 dummy variables. The items relate to hearing, sight, stom- ach, headache, fever, and other. The Index ranges from 0 to 1. 1 represents a person that is plagued by all described health problems.

Time Use is a composite variable. It consists of variables describing work related task outside of education on a regular school day. This includes time spent on working on the farm or in the family, chores or caring for family members, work for pay. All these where answered using the following scale: 0=none, 1=less than one hour, 2= 1-3 hours, 3=more than 3 hours. Time Work is the sum of all variables. I will not use this variable for my main estimation for two reasons. First, it is sectioned data its interpretational power is limited. Second, time use might be causally related to time spent on educational tasks, which are one of the channels through which private schools might improve test results.

Teacher Absence in Math and English describes the reported absence of teachers in last week by their pupils when interviewed. The absence is reported on an ordinal scale rang- ing from 0 to 3, “0” represents no absence, “1” absence of 1-2 lessons, “2” absence of 3- 4 lessons and “3” absence in 5 or more lessons. Present Days measures school attendance since the start of the school year in days, self-reported by the pupil. Instruction Time describes how many hours of Math or English are supposed to be taught in a week. This survey question was answered by the head teacher. The School Infrastructure Index con- sists of 9 dummy variables related to different school inputs. The Items are: chalk or board marker, black or whiteboard, teacher’s desk, teacher’s chair, books, electric light, electric fan, windows with glass and windows without glass. A School Infrastructure Index of “1” represents all inputs being available.

5.2 Summary Statistics

In Table 2 you can see the summary statistics for the YLSS data of my rural and urban sample. The T-Statistics tests compare the control and treatment group. The Sample is differentiated by rural and urban environments as I will estimate the private schooling effect for both groups separately. Differentiating between both environments is important to account for their vast differences and the potentially heterogenic private and public school performance. In the following I will describe the summary statistics by starting with the rural sample and comparing it to the urban sample.

The treatment group has about 10% percentage points more boys than the control group. This is most likely, as explained beforehand, due to gender bias in the household which leads to the prioritization of boys in education decisions (Afridi, 2010). The difference is even bigger for the urban sample, the treatment group has about 15% percentage points more boys. In both cases the difference is statistically significant. In both schools the majority of pupils are part of a SC or ST. Though in public they are more prevalent. This could be explained by several reasons. First public schools must accept any student and no bias is possible here. Private schools might discriminate towards children from a SC & ST background. Considering that they make the majority of private school pupils, this explanation is unlikely. Households from a non-scheduled background might be more financially affluent, which allows them to afford the fees. They might also expect higher returns on education as their child might not face discrimination on the job market in contrast to SC & ST children. The Health Index reports rather low health issues overall. Unsurprisingly private school attendees report in less health issues for both samples. The difference is statistically significant, but rather weak in comparison to the other control variables.

As expected, the Wealth Index (WI) is higher for private school pupils in both samples. There also is a stark difference between rural and urban settings. Because the items of the WI aren’t weighed, any small differences between samples might entail large income differentials. For both Father’s and Mother’s Education we see a similar pattern as with the other variables. The urban sample has in general higher educational achievement than rural areas and the private school attendees also have parents with higher educational achievements than their public counterparts.

PST & CTT scores verify the expectation for both rural and urban environment. In both cases children in private school score about 10% percentage points higher than public school pupils. If the PST and CTT are reliable proxies for cognitive ability, we expect them to be higher for private schools. Interestingly the standard error for the urban sample is higher indicating a wider arrange of baseline ability among students.

The share of English medium schools shows stark differences within samples and be- tween them. Most private schools teach in English, while most public schools teach in Telugu. Also, in the urban sample English as medium is more common for both school types, especially for public schools.

Time Use the rural sample show that public-school pupils must spent more time on work related tasks in comparisons to private school students on a regular school day. In total they must work about 2 hours outside of school. In the urban sample public school pupils also must work longer but only 80 minutes in total and 20 minutes more than their private school peers.

Surprisingly instructional time for Math is higher for private school pupils in both sam- ples. This runs contrary to descriptive findings of prior research (Muralidharan & Sundararaman, 2013). In the rural sample private pupils receive about 20 more minutes of math lessons per week. In the urban sample they receive 24 minutes of teaching per week. The urban sample shows lower instructional time for Math than the rural sample. Both samples are statistically significant. Instructional time for English shows a very sim- ilar pattern. Overall instructional time is higher in the rural sample than the urban sample and higher for private schools. In the rural sample instructional time is about 10% higher for private schools and 15% respectively in the urban sample. Teacher Absence shows a clear pattern for both Math and English, for both subjects and samples, public school teachers are absent from class more often. It is not possible to specify the extend of the difference between both school types, because the data is sectioned. For Present Days of pupils, we see no major differences between school types and samples. This shows that higher teacher absenteeism might not contribute to higher absenteeism of pupils. It actu- ally is slightly lower for private school pupils. The School Infrastructure Index shows that public schools enjoy better infrastructure than private schools in both samples, although the difference for the rural sample is more significant, while in the urban sample there is no significant gap.

Table 2 Summary statistics for rural and urban sample

Abbildung in dieser Leseprobe nicht enthalten

Note: Standard Deviation is reported in parentheses; Observations deviate for PST- & CTT-scores.

Source: Raw data from YLSS; Author’s own illustration

5.3 Attrition

The attrition in my sample includes dropping all pupils that went to TSW schools. Since TSW schools only make up a small percentage of actual schools in India their relevance is negligible. In the dataset they are disproportionally represented on purpose of the Sur- vey creators to enable analysis on them. Their exclusion doesn’t represent a problem in estimating the private school effect. I remove any movers for several reasons. When they move between school types during the survey year no clear identification of private school effects is possible. Even moving within the same school type the estimation might be biased by pupils possibly having to adapt to a new medium, different teachers and the stress associated with changing school and home. The amount of attrition due to movers is negligible.

More relevant is the attrition accrued due to missing values. All control and outcome variables that are missing are dropped. This reduces the observations from 5734 to 5431. This represents about 6,3% of the sample. This create a certain risk of missing values biasing the result. Missing test scores for example could be driven by pupils being ill, which should be more prevalent in public school. This would bias the Private school ef- fect downward, since troubled pupils of public school aren’t considered, improving the relative scores.

The urban sample includes 2926 observations. 1299 of those are in private school and 1627 in public school, so the sample size is roughly equivalent. The urban sample also showed in general a much higher heterogeneity between private and public-school stu- dents for summary statistics than their rural counterpart. The rural sample includes 2505 observations. 406 of those are in private school and 2099 in public school. The rural sam- ple is more unbalanced, and their student body is more homogenic in their observable characteristics. The exact figures might slightly differ for different model specifications. The here stated numbers correspond to the first model specification.

6. Econometric Model

I will use a value-added model to estimate the effect of one year of secondary private schooling on test score outcomes for Math and English. Equation (1) describes a general value-added model for several periods. yit describes the test score for individual i at period t. β 0 describes the constant of the model. β 1 is the coefficient of yit− 1, which describes the test score of individual i in the previous period. β 2 is the coefficient of PSit which is the dummy variable indicating whether individual i visited a private school at period t. X represents a set of different control variables, which also include PST & CTT scores.

Abbildung in dieser Leseprobe nicht enthalten

Since my data set includes only two time periods and I only know about the school type in grade 9 the value-added model simplifies to Equation (2). yi 2 is the test score for indi- vidual i at period 2 (end of grade 9). β 0 describes the constant of the model. β 1 is the coefficient of yi 1, which describes the test score variable of individual i for period 1 (start of grade 9). β 2 is the coefficient of PSi which is the dummy variable indicating whether individual i visited a private school in the 9th grade. X represents a set of different control variables that could variate in both time periods, but the control variables I use don’t differ over time so Xi describes a set of time constant control variables for individual i. Although I don’t have knowledge about the history of control variables or previous school types, I can partially control for those by the lagged test score variable. This only applies if the assumptions laid out in section 4 are true. Because the assumption of fixed abilities having homogenic effects at every age. For this reason, a control for baseline ability is required. In my model I will use PST and CTT scores as proxy for baseline cognitive ability.

Abbildung in dieser Leseprobe nicht enthalten

I also specify two more models. In Equation (3) I further control for English Medium Schools. Even though the medium of instruction certainly is a channel through which schools affect learning outcomes, I am interested in their comparable performance when language is controlled for. EMi is a dummy representing schools in which English is the section language as base category.

Abbildung in dieser Leseprobe nicht enthalten

In Equation (4) I further dissect the PSE for Telugu & English Medium Schools by in- cluding EMi and the Interaction Term PSi * EMi. This model allows a comparison be- tween English medium public & private schools. The PSi captures non-English private school effects. PSi * EMi captures the effect of English medium private schools. EMi. captures the effect of public schools teaching in English.

Abbildung in dieser Leseprobe nicht enthalten

In Equation (5) you see the model for estimating the effect of fees on private school per- formance. The model is restricted to a private school only sample. It follows closely the model of Equation (3) with the difference that no private school dummy is included, but instead a term for yearly fees measured in thousand Rupees per unit. It’s necessary to use a similar model, because fees can create selection pressure similar to that of private and public schools.

Abbildung in dieser Leseprobe nicht enthalten

7. Results

I differentiate my analysis between two samples. The first being schools in an urban set- ting and secondly schools in a rural setting. I believe both samples will produce different PSE, for two major reasons. The differences between private- and public-school pupils vary strongly within and between urban and rural areas. Second the urban environment is much more competitive, especially for private schools that must compete against public but also other private school. When evaluating education policy, it is important to acknowledge the vast differences between these two environments, as for example pri- vatization might create natural monopolies in rural settings but leave competition intact in urban settings. I also estimate the effect of fees on private schools. For that I use a rural and urban sample and exclude public schools in both.

7.1 Estimates for Math and English test scores

In table 3, column 1 you can see the results of the VAM with no language specific controls for standardized Math test scores for both the urban and rural sample. There is a highly significant PSE for the urban sample (0.314) but no significant effect for the rural sample; In table 4, column 1 you see the same model for English test scores. I find significant and positive effects for both urban (0.288 SD) and rural (0.474 SD) settings. These effects correspond to exactly one academic year of private schooling. In absolute percentage points the general effect size is rather small. For absolute values of Math and English test scores see Table 5 and 6. In absolute terms the effects for the just reported model range between 3.8 and 6.2 percentage points. This is indicative of the general bad educational performance of India. In column 2 you can see the same model with the control for EM schools added. The resulting decrease in the PSE for ETSs isn’t surprising, as we expect teaching different subjects in English to improve ETSs. But even after controlling for the EM a private schooling effect for English is remaining for both urban and rural environ- ments (0.175 SD and 0,353 SD). There don’t seem to be major changes for MTSs when including the control, which might show that Math isn’t strongly affected by language.

In column 3 you can see the model expanded with the EM*PS interaction term. The In- teraction term captures the effect of private schools teaching in English. The Private School variable captures the effect of all non-English medium schools. English medium captures the effect of public schools teaching in English. This setup allows the compari- son of English medium private and public schools. EM public schools surprisingly have a more positive effect ETSs than EM private schools, but they fare worse in in MTSs.

Not surprisingly two of the largest predictors in all regressions is the lagged variable test score and the PST & CTT control. Those being significant and large doesn’t proof the legitimacy of this model, but it is a necessary requirement. They also seem to explain the gross of differences in predictive power between the different regressions. The strongest fit is the urban ETS model (0.75 adj. R²) followed by the rural ETS model (adj. R² 0.5). The urban MTS (adj. R² 0.5) and rural MTS (adj. R² 0.3) regression have in general weaker predictive power.

As expected, the Wealth Index is significant for all regressions, except for the MTS in the rural sample. Its effect in the urban sample is also larger than the rural sample. A possible explanation might be that in the urban sample the differences in the WI between house- holds is made up by more expensive items like a truck or car, thereby in fact representing larger income differences, which could explain the higher effect size.

Being male is supposed to positively influence test scores as prior research consistently finds them receiving parental priority in educational achievement (Afridi, 2010). Inter- estingly for MTSs there is no gender effect in urban settings but a relatively large negative effect in rural settings. This would speak for a gender bias persisting in rural environ- ments. Paradoxically I find the opposite for ETSs, i.e. a negative effect in the urban sam- ple and no effect in the rural sample. A possible explanation might be a combination of different levels of gender bias and different levels of prioritization regarding subjects, dependent on the environment.

The Health Index is highly significant only for the ETS of the rural sample and has a very large positive effect size of 0.46 SD. This might be explained by children being plagued by health issues not having to take up hard physical labour. The Household Index which is composed mostly of items relating to parental supervision is only significant for MTS in the rural sample (0.185 SD). This could be explained by parents prioritizing Math over English in educational supervision in rural environments. For parental education effect sizes are relatively small. It is notable though that in the urban sample the education of the father is comparable to that of the mother in effect size. In the rural sample the edu- cation of the father doesn’t seem to matter at all.

Table 3 Private schooling effect on normalized Math Test scores in rural and urban AP

Abbildung in dieser Leseprobe nicht enthalten

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Source: Raw data from YLSS; Author’s own illustration

Table 4 Private schooling effect on normalized English test scores in rural and urban

Abbildung in dieser Leseprobe nicht enthalten

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Source: Raw data from YLSS; Author’s own illustration

7.2 Results for effects of fees on private school performance

In table 7 you can see the results for the effect of fees on MTSs and ETS, differentiated by urban and rural samples. In urban environments the effect is positive for both Math and English. In rural environments the opposite applies, for ETSs. This pattern might emerge as higher fees could have ambiguous effects. On the one hand the additional fees could be used to raise investment in school inputs, like teachers or computers. On the other hand, additional fees reduce the disposable income of the household. We expect the negative effect of reduced income to be larger when household income is lower, which could explain why the overall effect of fees in rural environments is negative. This would mean that a voucher program that pays for fees, would improve learning out- comes for rural household more than for urban households. We also see that for the ur- ban sample the effect of increased fees has decreasing marginal returns, but the effect is rather low. For example, increasing yearly fees by 1000 Rupees from the current aver- age of private schools in the urban sample, would increase the average MTS by 0,15

Table 7 Effect of fees on private school performance in urban and rural AP

Abbildung in dieser Leseprobe nicht enthalten

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Source: Raw data from YLSS; Author’s own illustration SD.

Like the previous models, the lagged variable test score and PS & CT scores are all sta- tistically significant and of moderate effect size. The EM term is only of importance to the English test score. It is important to include this control to not bias the bias the effect of fees in favour to English medium schools. Parents see to be rather irrelevant.

7.3 Limitations

In the following I want to explore several problems relating to the estimation of the pri- vate schooling effect using my specified method and data. These problems in principle also apply to my estimation of fees on private schools, as I use the same methodology.

First the higher heterogeneity of the student body in the urban sample in comparison to the rural sample. Even though I control for all relevant observable variables that differ strongly for both groups, it is possible that there are many unobservable I can’t control for. The heterogeneity of the urban sample (regarding observables characteristics) indi- cates that there might be many other unobservable characteristics that differ strongly. The rural sample is much more heterogenic indicating that they may not differ as much in unobservable variables. These differences might further widen the gap if the assumptions of the Value-added-model of linearity of input factors doesn’t hold true. In this case the lagged variable test score doesn’t capture variation in past inputs. This means that the model couldn’t control for the selection bias in the urban environment and therefore pre- dicts a larger PSE.

Another possible source for bias could be the control for the fixed cognitive ability of pupils. I use the PST scores and CTT scores as control. If these tests are influenced by the school type, in the sense of private school having a causal positive effect on those scores over time, they don’t present a baseline control and capture part of the positive effect of private schools on test score outcomes. I can’t test this assumption since the PST & CTT were only administered in the 2nd round not allowing for observation of improve- ment related to school type.

Further there might be causal relationship between missing values and school type. For example, missing values in test scores are mostly the result of pupils missing on that day. This could be the result of low effort or pupils having to take up work outside of school. These factors should lead to lower scores compared to peers. Because of selection bias we would expect those students to be proportionately overrepresented in public school. This would result in the underestimation of the private schooling effect. Looking at the share of students of the whole sample for which test scores are missing by school type reveals that this is indeed the case. Though the overall number of missing students is still relatively low and thereby shouldn’t majorly influence effect size.

Another weakness of this study is the limit of scope. Since I report only results on Math and English, I can’t evaluate the whole spectrum of school performance. My analysis is also limited to the secondary level of education.

8. Conclusion

I find statistically significant modest positive effects for one year of private schooling on English test scores for students of the 9th grade in both urban and rural environments even after controlling for medium of instruction. For Math Test scores I only find significant effects in the urban sample. For the effect of fees on private schooling, I find significant positive effects on test scores with decreasing marginal returns in the urban environment while the effect is negative in rural environments, most likely due to income effects. On average private schools provide better learning outcomes. They likely achieve this through more instructional time and higher teacher attendance. The higher instructional time for both subjects runs contrary to previous findings and indicates the evolving nature of private schools in AP (Muralidharan & Sundararaman, 2015). Private schools of my sample on average also use much less funds in comparison to the conservatively estimated cost of public schools by Dongre & Kapur (Dongre & Kapur, 2016).

In absolute terms the private schooling effect is still lacklustre. Simply staying passive and relying on the growing private schooling sector to solve the educational misère won’t be enough. Still private schooling provides a great opportunity. The government shouldn’t see itself as antagonist to private schools but rather realize their potential in providing cost effective education. Channelling public funding from public schools into the private schooling system could significantly improve private school performance and create acceptable educational outcomes. This could for example be accomplished via a voucher program. Instead of closing unofficial schools and making strict input require- ments, the government should make acquiring licenses as easy, fast and cheap as possible and enable private schools to use their budgets freely.

Because the private schooling sector is still growing and evolving it is important to reval- uate its effects on a recurring basis. Since the scope of my analysis is limited to secondary education and the state of AP, further research encompassing larger parts of India and primary education would be helpful.

9. References

Note: The data used in this publication come from Young Lives, a 15-year study of the changing nature of childhood poverty in Ethiopia, India, Peru and Vietnam (www.younglives.org.uk). Young Lives is funded by UK aid from the Department for International Development (DFID). The views expressed here are those of the author. They are not necessarily those of Young Lives, the University of Oxford, DFID or other funders.

Afridi, F., 2010. Women's empowerment and the goal of parity between the sexes in schooling in India, Population Studies, 64:2, 131-145.

Anand, B., Dhingra, H., Sabharwal, V. & Shah, R., 2019. India Closure Report 2018, New Delhi: Centre for Civil Society.

Andrabi, et al., 2011. Do Value-Added Estimates Add Value? Accounting for Learning Dynamics. American Economic Journal: Applied Economics, 3 , pp. 29-54.

ASER , 2019. Annual Status of Education Report 2018, New Delhi: ASER Centre.

Chudgar, A. & Quin , E., 2012. Relationship between private schooling and achievement: Results from rural and urban India. Economics of Education Review, pp. 376-390.

Dongre, Ambrish and Avani Kapur, 2016. Trends in Public Expenditure on Elementary Education in India, Economic and Political Weekly. Vol. 51, Issue No. 39, 24 Sep, 2016.

Geeta Gandhi Kingdon. 2019. Trends in Private and Public Schooling. Growth, Disparities and Inclusive Development in India , pp. 343-370.

Government of India, Ministry of Human Resource Development, Department of School Education & Literacy, Statistics Divison, 2018. Educational Statistics at a Glance, New Delhi: Government of India

Grønmo, L.S., M. Lindquist, A. Arora and I.V.S Mullis, 2015. TIMSS 2015 Mathematics Framework. Boston, MA: TIMSS and PIRLS International Study Centre.

Kingdon, G., 2017. The Private Schooling Phenomenon in India: A Review, Bonn: IZA Institute of Labor Economics.

Kingdon, G., S. Sinha and V. Kaul, with G. Bhargava & K. Pental, 2016. Value for money from Public Education Expenditure on Elementary Education in India, Discussion Paper Series, Education Global Practice, South Asia Region, World Bank, New Delhi. April 2016.

Kingdon, G., 2005. Private and public schooling: The Indian experience, UK: University of Oxford. Available at:

Kingdon, G. & French, R., 2010. The relative effectiveness of private and government schools in Rural India: Evidence from ASER data, London: Institute of Education.

Kingdon, G. G. & Azam, M., 2015. Assessing teacher quality in India. Journal of Development Economics, Issue 117, pp. 74-83.

Moore, R. et al., 2017. Young Lives School Survey, 2016–17: Evidence from India, UK: Universtiy of Oxford.

Muralidharan, K. & Sundararaman, V., 2013. Contract Teachers: Experimental Evidence from India. Available at: http://www.nber.org/papers/w19440, last accessed 15 August 2019.

Muralidharan, K. & Sundararaman, V., 2015. The Aggregate Effect of School Choice: Evidence from a Two-Stage Experiment in India. The Quarterly Journal of Economics, August, pp. 1001- 1066.

Oxford Poverty and Human Development Initiative, 2018. Global Multidimensional Poverty Index 2018: The Most Detailed Picture To Date of the World’s Poorest People, UK: University of Oxford.

OFQUAL, 2011. Functional Skills Criteria for English. Entry 1, Entry 2, Entry 3, Level 1 and Level 2, Coventry: OFQUAL.

Ramachandran, R., 2012. Language use in education and primary schooling attainment: evidence from a natural experiment in Ethiopia, Barcelona: Institut d’Economia de Barcelona.

Rosenbaum, P. & Rubin, D., 1983. Assessing Sensitivity to an Unobserved Binary Covariate in an Observational Study with Binary Outcome, Journal of the Royal Statistical Society: Series B (Methodological) Volume 45, Issue 2, January.

Sarangapani, P. M. & Mehendale, A., 2014. Inclusion of Marginalised Children in Private Unaided Schools under the Right of Children to Free and Compulsory Education Act, 2009: An Exploratory Study, New Delhi: Oxfam India.

Singh, A., 2015. Private school effects in urban and rural India: Panel estimates at primary and secondary school ages. Journal of Development Economics, Volume 113, pp. 16-32.

Todd, P. E. & Wolpin, K. I., 2003. On the Specification and Estimation of the Production Function for Cognitive Achievement, The Economic Journal, Volume 1113, Issue 485, 13 February, pp. 3-33.

Todd, P. E. & Wolpin, K. I., 2007. The Production of Cognitive Achievement in Children: Home, School, and Racial Test Score Gaps. Journal of Human Capital 1, no. 1, Winter, pp. 91-136.

Tooley, J. & Pauline Dixon, 2005. An inspector calls: the regulation of ‘budget’ private schools in Hyderabad, Andhra Pradesh, India. International Journal of Educational Development, May, pp. 269-285.

Wadhwa, M., 2001. Licenses to Open a School: It’s All About Money. Research Internship Papers 2001 Centre for Civil Society, New Delhi: Centre for Civil Society.

10. Appendix

Table 1 Percentage of children in private schools, by state and age, 2014-2015

Abbildung in dieser Leseprobe nicht enthalten

Source: Raw data from the National Sample Survey 71st round, 2014-2015; illustration by Kingdon (Kingdon, 2017)

Note: *The average of the Northeast states; these are Arunachal Pradesh, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim and Tripura

Abbildung in dieser Leseprobe nicht enthalten

Source: illustration by Muralidharan & Sundararaman (Muralidharan & Sundararaman, 2015)

Figure 6 Experimental design of the MS study

Abbildung in dieser Leseprobe nicht enthalten

Source: Young Lives Survey, available at https://www.younglives-india.org/about-young-lives-india (last accessed 20.08.2019)

Figure 7 Young Lives Site in Andhra Pradesh

Table 5 Private schooling effect on Math test scores in rural and urban AP

Abbildung in dieser Leseprobe nicht enthalten

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Source: Raw data from YLSS; Author’s own illustration

Table 6 Private schooling effect on English test scores in rural and urban AP

Abbildung in dieser Leseprobe nicht enthalten

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Source: Raw data from YLSS; Author’s own illustration

[...]

---

¹ This translates to about 372.35 million people in poverty, visit http://hdr.undp.org/en/2018-MPI for more information.

² Some public and private aided schools technically require fees, the amount is very small though and therefore ignorable.

³ Andhra Pradesh and Telangana used to be a single state, they were broken up in 2014. From now on when I refer to Andhra Pradesh, I include Telangana for simplicity.

⁴ Social studies and science

⁵ Hindi is the most spoken language in India, but Telugu is the official language of AP

⁶ In this case not the exact location of the school is meant as a factor, but rather its relative distance to the different pupils.

⁷ Technically the Young Lives Sites are present in Telangana too, which was formerly just AP.

⁸ Numbers based on the definition of Kingdon (Kingdon, 2017)

⁹ Technically they also serve as a control in the form of a lagged variable

¹⁰ All tests, translated to English on a page by page basis, are available on the appended CD

• College: University of Heidelberg (Alfred-Weber-Institut für Wirtschaftswissenschaften)
• Grade: 1,3
• Author: Daniel Klein (Author)
• Publication Year: 2019
• Pages: 45
• Catalog Number: V504989
• ISBN (eBook): 9783346057907
• ISBN (Book): 9783346057914
• Language: English
• Tags: India, Development Economics, Human Capital, Private Schools