ABSTRACT
We investigate how social identity, namely gender and caste, affects stream choice at the higher secondary level of schooling in India. The choice of science stream at this level is a crucial determinant of subsequent science, technology, engineering, and mathematics (STEM) education and labor market outcomes. Using nationally representative data spanning a decade, we show that females and individuals from historically disadvantaged castes are significantly less likely to study science. We analyze the role of various socio-economic and schooling-related factors in explaining these gaps. We also highlight the interplay between caste and gender using an intersectionality framework.
Acknowledgements
We are grateful to Colin Green and one anonymous referee for very constructive comments that have helped us to improve the paper. We also thank Ritwik Banerjee and seminar participants at IIM Bangalore for their helpful comments.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 The system of reserving quotas in educational institutes and public-sector jobs has existed since 1950 for the Scheduled Castes (SC) and Scheduled Tribes (ST) and since 1992 for the Other Backward Classes (OBC). This is one of the oldest such social identity based affirmative action programs in the world.
2 In 2018, 14% of seats have been added exclusively for females in the premier government institutes (including Indian Institutes of Technology and National Institutes of Technology) for engineering courses. These additional seats were increased to 17% in 2019 and to 20% in 2020 (JOSAA Citation2021).
3 Castes, in India, are a system of rigid hierarchical social groups, which affect the social and economic lives of its members (Deshpande Citation2011; Mosse Citation2018; Munshi Citation2019).
4 This policy, enacted in Jan 2019, reserves 10% quota in educational institutes and public-sector employment for individuals belonging to the economically weaker section (EWS), i.e. those with household income not exceeding INR 8,00,000 (approximately USD 10,000) per annum and not covered under the reservation scheme for disadvantaged castes.
5 Our finding is consistent with the literature showing the persistence of intra-household gender inequality in various educational outcomes (Kingdon Citation2002; Sahoo Citation2017; Sahoo and Klasen Citation2018).
6 On the gender gap in science choice, the study closest to our paper is Sahoo and Klasen (Citation2021) which documents the gender gap in science and commerce choice using the India Human Development Survey of 2011–2012. In this paper, we use larger and more updated surveys spanning the recent decade to analyze the trends in both gender and caste gaps.
7 The Right to Education (RTE) Act, 2009 mandates the government to provide free and compulsory education to all children in the age group of 6–14 years. It is the responsibility of the appropriate authority to ensure that all children in this age group are enrolled in school and attain elementary education (8 years of schooling). Figure A1 in the appendix indicates that enrollment rates have increased over time for the target age group of 6–14 years. Also, there is no sudden drop-out at the policy cut-off age of 14 years, indicating that most students continue education beyond compulsory schooling and gradually discontinue further education.
8 The details of the survey samples are provided in the appendix (Table A1).
9 Enrollment rates by gender and caste for this age group are provided in Table A2 in the appendix. While there is almost no gender gap in higher secondary enrollment rates, caste gaps are visible in every round. In a robustness analysis, we show that our regression analysis identifying the gender and caste gaps in science choice is not biased by any sample selection issue arising from non-random enrollment in higher secondary level.
10 Grade retention refers to instances where the grade a student is studying in current year is the same as that of previous year. NSSO collects data on this variable for children who are enrolled in school. Among students enrolled in grades 1–10 (i.e. from primary to secondary level), those repeating the same grade were 4.6, 3.5 and 2.4% in 2007–2008, 2014, and 2017–2018, respectively.
11 Considering the variation in age within this age-band due to late enrollment and/or grade repetition, we control for age in all our estimation models.
12 We also do our analysis on the sample restricted to those in the age group of 16–18 years and find that the results are almost identical, which is not surprising considering that 82.28% of our sample is from the age group of 16–18 years. These results are available on request.
13 We obtain these estimates from DISE data.
14 The district level measures are constructed for the years 2005, 2012 and 2015 which are lagged by two years from when the outcome on stream choice is collected by NSS surveys held in years 2007–2008, 2014, and 2017–2018 respectively.
15 We present linear probability model since we are interested in the marginal effects that are straightforward to interpret from linear models. However, we also check the results using nonlinear (probit and logit) models and find similar results (presented in the appendix). Another concern is that the sample is based on students who are enrolled at the higher secondary level, hence, there may be sample selection bias. In a robustness analysis presented later, we estimate a sample selection model to consider the enrollment decision. The main results are unchanged.
16 In the decomposition analysis, since our interest is to analyze how much of the gaps are explained by endowments, we use a specification that controls for observable individual and household specific characteristics along with survey-round fixed effects.
17 Equivalently, can also be estimated with the interaction of caste and gender dummies as in the following model:
.
18 We also report the full results of round-wise estimation in tables A4–A6 in the appendix.
19 Inclusion of household fixed effects implies that the estimation sample considers only those households where there are at least two individuals enrolled in higher secondary level. Therefore, the sample size reduces.
20 In a related study, using India Human Development Survey data collected in 2011–2012, Sahoo and Klasen (Citation2021) find an estimated gender gap in science choice (compared to humanities) to be around 10 percentage points. This magnitude is broadly similar to what we find in this study using a larger data from the National Sample Survey.
21 These effects are not significant when district fixed effects are included since there is not much variation in the number of HSS per 100,000 population and the percentage of HSS offering science stream over time within the same districts.
22 One such factor that could potentially affect enrollment in higher secondary level is grade retention. If students repeating grades are more likely to drop out, then gender- and caste-based differences in grade retention can influence stream choice through differential enrollment rates at higher secondary level by gender and caste. We find that gender and caste gaps in grade retention are either insignificant or negligible in magnitude (Table A9), hence this is unlikely to be a major concern in our analysis. Nevertheless, the sample selection model we discuss in Appendix B would address such issues of non-random enrollment at higher secondary level.
23 Differences in observables between the gender and caste-groups provided in Table A10.
24 The magnitude of the effect on boys is similar to the effect on girls when state fixed effects are considered, but the negative effect on boys has a larger magnitude than the positive effect on girls when district fixed effects are included, which makes the overall effect to be negative, as we found earlier while discussing the main results.
25 We also estimate the intersectionality hypothesis using interaction terms of female and caste dummies. The results, reported in Table A11 in the appendix, are equivalent to what we present in . We also include a specification with household fixed effects in column 4 of Table A11 and find broadly similar results although the precision of the estimates reduces with inclusion of household fixed effects.