Test score gaps between private and government sector students at school entry age in India

Abstract

Various studies have noted that students enrolled in private schools in India perform better on average than students in government schools. In this paper, I show that large gaps in the test scores of children in private and public sector education are evident even at the point of initial enrolment in formal schooling and are associated with previous attendance in private and public preschools. Children in the sample were aged between 4.5 and 6 years at the time of the survey and were tested on receptive vocabulary and quantitative skills. Most (albeit not all) children in the sample had attended private and/or government preschools and at the time of the survey, about 44% had made a transition into formal schooling. Attending private preschools is associated with substantially, and significantly, higher test scores relative to attending public preschools. A considerable portion of this difference can be removed by controls for parental background and selected child characteristics but the gap remains significant. Possible implications of these results are discussed.

Keywords:

• school
• preschool
• learning
• inequality
• private education

Acknowledgements

I am particularly grateful to Francis Teal for helpful discussion and guidance over several years. I am also grateful for detailed and useful feedback to the Editors of the special issue, Angela Little and Caine Rolleston, and two anonymous referees for detailed comments which helped improve this manuscript greatly. Helpful comments were also received at early stages from Stefan Dercon, Caroline Knowles, Geeta Kingdon, Pieter Serneels and Martin Woodhead. Young Lives is core-funded by UK aid from the Department for International Development (DFID) and co-funded from 2010 to 2014 by the Netherlands Ministry of Foreign Affairs. www.younglives.org.uk

Notes

1. For example, the SchoolTELLS survey carried out by Geeta Kingdon and co-authors and the repeated surveys carried out by the Andhra Pradesh Randomized Evaluation Studies have been very useful in facilitating detailed analyses of factors affecting the rate of learning in different schools.

2. See, however, the influential Lancet Studies on the importance of preschool factors, especially Engle et al. (2011), who pay particular attention to the potential effects of preschool education to the development potential of children and Walker et al. (2011), who discuss the risk factors and protective factors affecting early childhood development in developing countries.

3. In many cases, private preschools refer simply to the kindergarten sections of the private schools and therefore the close correspondence in characteristics between these institutions is not surprising.

4. This is also a point that has been stressed by James Tooley and co-authors in a range of different work focused on urban, peri-urban and rural areas in different parts of the country (see, e.g., Tooley & Dixon 2006; Tooley, Dixon, & Gomathi, 2007).

5. Notably the India Human Development Survey 2005 only collected test data for children between 8 and 11 years of age.

6. In Argentina, for example, Berlinski, Galiani, & Gertler (2009) report that a one year increase in pre-primary school increases average third grade test scores by 23% of a standard deviation. If similar relationships exist in India, then inequalities in preschool access probably contribute to inequalities in educational attainment observed later in the educational trajectory.

7. The first round of the data collection was meant to cover children between 6 and 18 months of age. Combined with a four-month period of data collection, and a few cases of initial mis-recording of date of birth in the absence of initial documentation, this leads to the 18-month age range observed in the dataset. There are few children born in the months at the tails of this distribution: 97% of children resurveyed in the second round were born between April 2001 and May 2002.

8. In the older cohort, born in 1994/5, about 70% of children started schooling at the age of five; 98% of the children were observed to be enrolled at the age of eight in 2002.

9. Andhra Pradesh had a population of about 84 million in 2011, larger than the population of Germany, and an area of 276,754 sq. km., which is larger than the United Kingdom. The data are relatively clustered and interview children in 100 communities (villages or urban wards) in 20 sub-districts (mandals); Kumra (2008) compares the Young Lives sample to representative household data for Andhra Pradesh and finds that the sample does contain similar variation to that found in representative surveys.

10. This restricts the sample used for the analysis of PPVT scores to 1829 children. The exclusion of children who took the test in other languages is necessitated by possible differences in the complexity of the same words across languages (thus leading to Differential Item Functioning) and has been recommended by the original adaptors of the test in Telugu for Young Lives (Cueto et al., 2009).

11. This subscale requires children to indicate which one of a set of pictures fits the description provided by the examiner. Notions such as a few, most, half, many, equal, a pair, etc. are assessed with statements such as: ‘Point to the plate that has a few cupcakes’. See the detailed Young Lives technical note by Cueto et al. (2009) for further elaboration on the content and characteristics of the two achievement tests.

12. For a more detailed explanation of Item Response Theory (IRT) models, please see Das & Zajonc (2010) and Van der Linden & Hambleton (1997). It is worthwhile to note that key results of the analysis are not driven by the choice of score aggregation method; results are similar when using raw scores instead of IRT scores.

13. Site fixed effects, included here by including a vector of site dummies, allow for removing any levels differences between different sites (mandals). This is particularly important in this case since the take-up of different institution types varies much across the sites in the sample. In effect, these regressions compare children who are in different types of education but living in the same cluster.

14. The wealth index is entered with dummy variables for the terciles of the wealth index (with the poorest third being the base category). This procedure is adopted to allow for non-linearities in the relationship between wealth and test score, and to recognise explicitly that wealth indices based on asset ownership are not inherently cardinal.

15. This pattern should not, of course, be interpreted causally: it is, for instance, possible that this pattern is caused by the transition into formal schooling being brought forward by parents for children with higher academic achievement or potential.

16. The number of children who cross over between sectors when going from preschool to school is very limited in the sample, thus precluding sufficiently precise statistical inference.

17. Note that all the aspects of enrolment as observed in the data, including whether a child is enrolled, the type of institution enrolled in and the age of enrolment into both preschool and school are choices made by parents, conditional on endowments. It is plausible that the determinants of these choices (e.g. parental preferences over education), some of which might be unobserved in the data, also independently affect learning and/or they influence not just choices around enrolment but also other investments (such as time spent reading to children) which then affect test scores.

18. In addition to the studies cited in section 2, see also the article by James and Woodhead (this issue) and the recent special issue of this journal on private and public education (Vol. 39, No. 4).

19. Singh (2013) uses the third round of the Young Lives survey in 2009/10, along with data collected from schools in 2011, to document that gaps between students enrolled in private and government schools in mathematics and Telugu, which are large cross-sectionally, are significantly reduced upon controlling for the gaps in PPVT and CDA that are documented in this paper along with controls for the differing socio-economic background of children; controlling further for differences in the time-use patterns across children in private and government schools, the gap is closed entirely. Clearly, the gaps documented in this paper are central in explaining the differences in human capital acquisition of children from different backgrounds educated in the private and state sectors.