3.1 The Tests

The tests were based on the National Council of Educational Research and Training (NCERT) tests for grade four in language and mathematics. The language tested is Hindi, the language in use in both states. All the tests were in the multiple choice format. Both grade 4 and 5 students took the same test.

Reading Comprehension Test: This consisted of 34 items aimed at assessing the student’s ability to comprehend paragraphs.

Word Meaning Test: The test consisted of 35 items aimed at assessing the student’s ability to identify synonyms and antonyms.

Mathematics Test: The test consisted of 33 items aimed at assessing the students’ ability to do simple additions, subtractions, multiplications, fractions, and area and weight analysis.

The test items correspond to competencies that children are expected to have mastered by end of grade 4. Each child’s score on a test is the number of questions he or she answered correctly converted into a percentage.

4.1 Unadjusted Test Scores

Mean test scores, overall and by type of school management are presented in Tables 9 and 10. Scores are low in absolute terms and certainly much lower than 60% which is the government’s own indicator for a child’s acceptable level of competency on a test.

Table 9: Mean Scores, Uttar Pradesh

Read Word Math

Private aided 37 (24) 50 (20) 33 (22)

Private unaided recognised 37 (18) 50 (18) 26 (13)

Private unaided unrecognised 40 (21) 52 (18) 31 (16)

Grade 5 Overall 34 (19) 48 (19) 26 (14)

Government 30 (14) 43 (18) 22 (12)

Private aided 44 (25) 59 (22) 42 (29)

Private unaided recognised 43 (20) 56 (18) 31 (13)

Private unaided unrecognised 44 (24) 57 (21) 32 (17)

Standard deviation in parentheses.

Table 10: Mean Scores, Madhya Pradesh

Read Word Math

Grade 4 Overall 30 (21) 42 (22) 25 (17)

Government 24 (18) 38 (22) 21 (15)

Private aided 25 (8) 49 (12) 18 (6)

Private unaided 40 (21) 49 (22) 32 (17)

Grade 5 Overall 36 (22) 47 (23) 29 (17)

Government 30 (18) 45 (22) 25 (15)

Private aided 40 (14) 56 (17) 27 (15)

Private unaided 48 (24) 54 (23) 37 (18)

Standard deviation in parentheses.

June 2, 2012 vol xlviI no 22

Within Within

100 80

100 80 60 40 20 0 Between Grade 4 Grade 5 Read Word Math Read Word Math Percentage (%) 58 70 58 54 70 57 42 30 42 46 30 43

Grade 5 Word 47 6553 35 55 53 34 56

45

47

66

44

Percentage (%)

60 40 20 0

Read

Word

Math

Read

Math

Grade 4

Each question had four or five options to choose the answer from. Therefore, even if a child was randomly guessing the answer, he or she could score an average of 20-25%. Accounting for guessing would imply an even smaller level of learning.

The standard deviations of the test scores are high. High variation implies that there are few students who do exceedingly well but the majority performs poorly. If we were to take the difference in the mean scores between grades 5 and 4 as an indicator of average gain in learning between the two grades, the mean gain is only about 3-4 percentage points in each area. The variation in scores also increases which implies a further pulling away of the top scorers with little improvement in the test scores of the majority.

Tables 11 and 12 (p 72) give the distribution of scores by percentile. The distribution of scores of government schools is to the left of private schools. But learning is poor in all school types. If we take scoring 50-60% on a test as a benchmark of acceptable levels of learning (NCERT uses 60%), government schools in both states achieve this standard somewhere between the 90th and 95th percentile, whereas private schools do so between the 75th and 90th percentile.

4.2 Varied Test Scores Within and Across Schools

The total variation in test scores is the sum of variation arising due to differences between schools and within schools. What share each source of variation contributes can be computed using OLS regression analysis with test scores as the dependent variable and the school attended as the only independent variable. The amount of variation “explained” in this case is the share of the variation coming from differences between schools. The remaining is that due to sources of differences within schools (i e, what happens if all the schools were identical).

Variation between schools accounts for 30-56% of the total variation in scores (Figures 1-2). The remaining variation in scores is within schools. We repeated the above analysis separately by school type. The results (available with the authors) are not very different from those of the overall sample. This implies there are good and bad schools within all school types. School quality differences matter; however differences across students within a school also matter considerably for test scores. From the point of view of policy, there is opportunity for improving education outcomes both by pursuing polices that improve school quality and also policies directed towards students.

4.3 Is There a Private School Effect?

We analyse the private school effect before and after controlling for differences in observed student and school characteristics. Tables 13 (Uttar Pradesh) and 14 (p 73) (Madhya Pradesh) show the unadjusted and adjusted difference in mean scores between private and government schools. The adjusted differences in mean test scores are the remainder differences after

Economic & Political Weekly

EPW

June 2, 2012 vol xlviI no 22

controlling for a set of characteristics. Scores have been adjusted using two models: (a) the model uses as controls child and family background characteristics, school characteristics, district of location dummies and rural location dummy;5 and (b) same as (a) except the district and rural location dummies are replaced with village dummies.

The remaining effect of school type is not totally unbiased because there is a likelihood of systematic selection into various types of schools correlated with unobserved children and family characteristics. If more able or motivated students select into private schools then any private school advantage over government schools in the test score, after controlling for observed student and school characteristics, cannot be attributed to school-type. In fact, as we see later, there are reasons to believe that “better” students attend private schools and this may be partly responsible for the higher average private school test scores. To obtain an unbiased private school

estimate when selection is going on, one needs a way to correct for selection bias. In the commonly used approaches to correct for selection bias, one needs a valid instrument which belongs in the school choice equation but not in the test score equation. Since we do not have a convincing instrument, we do not correct for selection bias.

Without adjusting there is a significant private school effect in every test and grade. Results change once controls are included. The advantage varies by state, type of private school and grade. In UP, the private schools have an advantage in grade five. Private unrecognised schools outperform private recognised schools in having a greater number of signifi cant differences from government schools. In MP, there is no robust private school advantage in either grade.

5.1 Socio-economic Characteristics of Students in Government and Private Schools

For both states, most factors of disadvantage are less represented in the private school, and all the differences across government and private schools are significant at the 1% level (Tables 15-16, p 74). Private school students are less likely to be from the scheduled castes (SCs) and scheduled tribes (STs) households, are more likely to be male and have parents educated above primary school. They also are more likely to have fathers who are in occupations other than agricultural labour, and come from households that own more than the median landholding in the sample.

The adjusted mean differences in these characteristics between school types become larger in magnitude, and are larger for private unaided unrecognised schools. These results suggest considerable sorting of students across school types. It is likely that sorting also goes on along unobserved family and student characteristics such as attitude and motivation.

5.2 School Inputs

There are few consistent differences in infrastructure between private and government schools except private schools have significantly lower pupil-teacher ratios (Tables 17-20, p 74). Mean teacher attendance and activity at the school level do not differ between private and government schools, except for private unrecognised schools in UP that have higher activity.

5.3 Do Teachers Differ across School Types?

Demographics: Teachers in private schools are likely to be younger than teachers in government schools (Table 21, p 74). In MP, teachers are also more likely to be from the local area than teachers in government schools (Table 22, p 74).

Professional Credentials and Salary: Overall, teachers in government schools are more likely to be trained, have greater experience and a higher salary than teachers in private schools. Teacher salary in government schools is eight times that in private schools. The differences in these characteristics are bigger between regular teachers in government schools and teachers in private schools.

5.3.1 Mean Teacher Effort

Government and private schools are similar in rates of teacher attendance, while differences in rates of teacher activity vary by state and by the type of school.

June 2, 2012 vol xlviI no 22

Figure 3: Within and between School Variations in Teacher Effort, Uttar Pradesh and Madhya Pradesh

Between

Percentage (%)

100 80 60 40 20 0

Attendance Activity Attendance Activity Uttar Pradesh Madhya Pradesh Within 84 16 20 80 85 15 40 60

Private recognised and unrecognised schools have similar rates of attendance and higher rates of teacher activity compared to government schools in UP. Activity rates are 11-18% points higher in private schools (Table 23, p 75). However after controlling for teacher characteristics and district or village fixed effects, teachers in private and government schools are similar in mean attendance and activity rates, except for private unrecognised schools that have higher teaching activity. In MP, private schools are similar to government schools in rates of teacher attendance and activity, before as well as after controlling for teacher characteristics and district or village fixed effects (Table 24, p 75).

5.3.2 Variation in Teacher Effort between and within Schools

Differences between schools explain 40% or less of the variation in teacher effort. This implies more than 60% of the variation in rates of teacher attendance and engagement in teaching is within schools. Only a small fraction of the variation in effort within schools is explained by observed teacher characteristics.

The r-square from a regression of teacher attendance (and activity) on school fixed effects gives the percentage of

Table 13: Difference between Private and Government Schools (Private-Government), Uttar Pradesh

Private Unaided Recognised	Private Unaided Unrecognised
Grade 4
Read	Unadjusted	13**	16**
	Adjusteda	3.38	9.47
	Adjustedb	9.68	16.17*
Word	Unadjusted	11**	12**
	Adjusteda	0.95	2.90
	Adjustedb	22.06**	27.04**
Math	Unadjusted	7**	11**
	Adjusteda	1.36	7.24
	Adjustedb	3.64	6.67
Grade 5
Read	Unadjusted	15**	16**
	Adjusteda	13.16*	17.49*
	Adjustedb	26.64**	33.40**
Word	Unadjusted	13**	13**
	Adjusteda	14.07**	16*
	Adjustedb	31.64**	36.57**
Math	Unadjusted	9**	11**
	Adjusteda	7.20	11.1*
	Adjustedb	24.34**	28.69**

*Significant at 5% level; **significant at 1% level. a Controls + District FE + rural dummy, b Controls + Village FE.

Economic & Political Weekly

EPW

June 2, 2012 vol xlviI no 22

variation in teacher effort that is due to differences across schools and villages. The remaining variation would be attributable to variation in within school variables such as observed and unobserved teacher characteristics, classroom characteristics etc.

We then add observed teacher characteristics to the school fixed effect regression to see how much of the within school variation can be explained by these. The vector of teacher characteristics includes age, gender, caste, education, whether teacher has pre-service training, number of years of service, number of days of in service training in last school year, whether teacher’s appointment is on a contract basis and whether teacher is a resident of the village.

There are two main points here. First, most of the variation in teacher effort is within schools. Variation in teacher attendance that is explained by differences between schools and villages is 15% in MP and 16% in UP (Figure 3). Variation in teacher activity that is explained by differences between schools and villages is 40% in MP and 20% in UP. In particular, whether the school is government or private recognised or private unrecognised contributes less than 2% of the variation in teacher effort. Second, observed teacher characteristics explain very little of the variation in teacher effort within schools. These observations are consistent with the fi ndings of other studies, although mainly from developed countries, that find: (a) substantial variation in teacher quality within schools, and (b) observed teacher characteristics explain very little of the variation in teacher quality within schools (Rockoff 2004; Rivkin, Hanushek and Kain 2005).

Madhya Pradesh

grade to the next are low

% Unless Stated Otherwise DistrictFE+ Unaided Rural Dummy

for both school types.

Recognised Unaided Recognised

guessing every answer, Graduate degree -.08** -.09**

he or she could have Experience (years) -7.72** -8.34**

scored an average of Salary (rupees) -5,675** -5,732** **Significant at 1% level

20-25%. Accounting for guessing will imply even lower actual learning.

There is a great degree of variability in test scores within and between schools in government as well as private schools. Observable school and teacher characteristics are weakly correlated with test scores. Most of the variation in teacher effort is within schools and has weak links with observed teacher characteristics that are commonly used by school administrators as indicators of teacher quality such as training, experience and education. This suggests rewarding teachers on the basis of their credentials may not be effective in raising effort. The existing salary structure is related to effort neither

June 2, 2012 vol xlviI no 22

Table 23: Difference in Teacher Effort between Private and Government Schools, Uttar Pradesh

Unadjusted Adjusted Adjusteda Adjusteda District FE District FE+ Village + Rural Rural FE + Dummy Dummy Controls

+ Controls

(1) (2) (3) (4)

*Significant at 5% level; **Significant at 1% level. a Controls are a full set of teacher characteristics.

Table 24: Difference in Teacher Effort between Private and Government Schools, Madhya Pradesh

Unadjusted Adjusted Adjusteda Adjusteda District FE + District FE + Village Rural Rural FE + Dummy Dummy Controls

+ Controls

(1) (2) (3) (4)

Attendance Private unaided recognised .02 -.01 .06 .09

Activity Private unaided recognised -.06 -.08 -.08 -.02

*Significant at 5% level; **Significant at 1% level. a Controls are a full set of teacher characteristics.

*Significant at 5% level; **Significant at 1% level. aControls are a full set of teacher characteristics.

in government nor in private schools, except for contract teachers in government schools. It fails to reward those more present and active in the classroom.

After controlling for student and school characteristics, the private school advantage in scores varies by state, type of private school, grade and subject. In UP, private unrecognised schools outperform private recognised schools in having a greater number of significant differences from government schools. In MP, there is no robust private school advantage. This is unlike the findings from other studies in India where after adjusting for student and school characteristics the private school advantage usually remains significant. One reason for the difference in results can be that our sample is largely rural. The urban areas in the sample are small towns located within the same block as the villages and are more likely to resemble the villages than the larger cities where private schools are found to perform better after controlling for sample characteristics as in studies such as Tooley and Dixon (2006) and Kingdon (1996a). Another reason could be that our data are from two states lagging in most development outcomes, where just as the public sector has low accountability, the private sector may be functioning in a largely unregulated

Economic & Political Weekly

EPW

environment and parents who largely have little education in the sample may be making school choices based on perceived school quality.

The sources of private school advantage lie in the types of students choosing these schools, lower pupil-teacher ratios and much lower teacher salaries. Private schools differ considerably in the types of students who attend even within the same district or village. Students in these schools are less likely to belong to low caste groups. They are likely to have more educated and wealthier parents. It is likely that sorting is also going on along unobserved family characteristics such as attitude and motivation. Private and government schools do not differ in physical facilities but private schools have a lower pupil-teacher ratio which implies greater teacher-time per student. Teacher salary in private schools is between one-seventh and one-eighth of the government schools, although teacher effort is largely similar except for private unrecognised schools in UP.

Since data indicate considerable sorting among students into school types, it is not surprising that the private school effect is less systematic after controlling for observed student and school characteristics. In the cases where the private school effect remains, we cannot be sure this effect is attributable to school type as there may be sorting in unobserved characteristics. Nevertheless, as teacher salaries in private schools are one-seventh or one-eighth of government schools and assuming salaries form a large fraction of the operating cost as is the case for government schools, private schools would unambiguously be more cost effective even in the case of no absolute advantage in test scores.

Our results may suggest at first that government regulations are redundant, and it is the market in schooling that is more effective in determining quality: private unrecognised schools, which account for half of all private schools in the sample in UP, do better than private recognised schools. But this is not so clear. In both states, learning standards are strictly enforced in neither government nor private recognised schools. The government schools have a minimum level of learning framework, but no functioning mechanism that ensures this standard. The private recognised schools can pay bribes to get recognition without meeting the required criterion for obtaining formal registration. The market does not ensure good quality education either since the unregulated schools are also way off the mark in basic competencies; moreover, we cannot disentangle the sorting effect from the school quality effect with respect to learning outcomes. Private schools may be choosing to locate above but close to government schools along the quality spectrum because it is rational for them to do so given supply-side (incomplete government regulations, poor enforcement) and demand-side (poverty and illiteracy of parents) characteristics. It is costly for schools to adhere to enforced standards of quality. We speculate that if the government were to enforce learning standards on all schools, there would be a change in the composition of supply of private education with low cost (and low quality) private schools likely leaving the market.

Notes

1 A gram panchayat is the lowest administrative unit in rural areas consisting of two to three revenue villages on average. The lowest administrative unit in urban areas is an urban ward. A block is an administrative unit between a district and a gram panchayat/urban ward.

2 In UP, a recognised school must be a registered society, have an owned rather than a rented building, employ only trained teachers, pay salaries to staff according to government prescribed norms, have classrooms of a specifi ed minimum size and charge only government-set fee rates (Kingdon 1994).

3 On further probing it came out that in MP, all schools which have up to grade 5 have to have a registration number at least. This registration number is not unique to schools but unique to an umbrella organisation like an NGO/trust/ society. Because of this, schools up to grade 5 are perceived as recognised by the villagers but may not necessarily be so. The distinction between recognised and unrecognised schools in MP is therefore blurred or ambiguous and this should be kept in mind when interpreting the results.

4 We did a check by doing the analyses both with and without private aided schools. The results stay similar in both cases.

5 Child and family background characteristics included as controls are the child’s age, gender and caste, sibling size, whether the child takes private tuition or not, mother’s and father’s education levels, father’s occupation, and landownership. School characteristics included as controls are infrastructure, mid-day meal provision, free textbook provision, and average teacher characteristics at the school level – female, education level, training and experience.

References

Pandey, Priyanka, Sangeeta Goyal and Venkatesh Sundararaman (2010): “Public Participation, Teacher Accountability, and School Outcomes in Three States”, Economic & Political Weekly, 12 June.

Mehta, Arun (2006): “Elementary Education in India, Annual Report 2004-05”, National Institute of Educational Planning and Administration and Department of Elementary Education and Literacy, Ministry of Human Resources Development, Government of India.

PROBE Team (1999): Government Report on Basic Education in India (New Delhi: Oxford University Press).

Rockoff, Jonah (2004): “The Impact of Individual Teachers on Student Achievement: Evidence from Panel Data”, American Economic Review, May, Vol 94(2).

Rivkin, Steven, Hanushek, Eric and John Kain (2005): “Teachers, Schools and Academic Achievement”, Econometrica, Vol 73 (2).

Social and Rural Research Institute (SRI) (2005): “Survey on Assessing the Number of Out-of-School Children in the 6-13 Years Age Group”, New Delhi.

Tooley, James and Paula Dixon (2003): Private Schools for the Poor: A Case Study from India, CfBT Research and Development.

– (2006): De facto’ Privatisation of Education and the Poor: Implications of a Study from Sub-Saharan Africa and India, Compare, Vol 36, No 4, December, pp 443-62.