This study uses data from round one and round two of the India Human Development Survey (IHDS), which provide information on the health and morbidity, education, employment, economic status, nuptiality, fertility, gender relations and social capital of a nationally representative sample of 41,554 households. Round one of the IHDS was conducted in 2005–2006. In round two of the IHDS, which was carried out in 2011–2012, 83% of the households surveyed in round one were re-interviewed. Unlike other cross-sectional surveys conducted in India, such as the National Family Health Survey (NFHS), the IHDS covered a wide range of topics, and had a panel design. The National Council of Applied Economic Research (NCAER) conducted both rounds of the IHDS in collaboration with the University of Maryland, USA. The IHDS had a multistage, stratified, random sampling survey design, and covered all of the states and the union territories of India. Further details regarding the IHDS’ sample frame, sampling procedures and data collection approaches can be found elsewhere (Desai et al. 2010, 2015).

Our study utilises the panel data for 9840 students who were 8–11 years old in 2005 (round one), and who were 15–18 years old in 2012 (round two). Of the 17,061 children who participated in round one (cross-sectional dataset), 104 had died, 3454 had migrated and 3663 were untraceable during round two. Furthermore, we have excluded the data for 18 children for whom the information regarding their school dropout and grade repetition status was missing in round two. Thus, the analytical sample we use to investigate the relationship between educational wellbeing in round one and school dropout and grade retention status in round two consists of 9822 students.

The analysis presented in this article is based on publicly available secondary data with no identifiable information. Therefore, no prior ethical approval for our study was necessary. The data can be obtained from the Inter-university Consortium for Political and Social Research (ICPSR) data repository (Desai and Vanneman 2015; Desai et al. 2008a).

The indicators of the school dropout status (no, yes) and the grade repetition status (no, yes) of the students in our analytical sample between 2005 and 2012 are the two outcome variables of our study. The school dropout status of the students was obtained from the binary indicators of their school enrolment status in 2005 and 2012, respectively. Children who were enrolled in school during round one, but who were not enrolled during round two, were categorised as “yes” (school dropout), and were otherwise categorised as “no”. Similarly, the grade repetition status of students was obtained from the responses to the following item collected in both rounds: “has the student ever repeated a grade – yes, no”. Students who had never repeated a grade in 2005, but who had repeated at least one grade in 2012, were categorised as “yes”, and were otherwise categorised as “no”.

The two binary indicators of educational wellbeing are the key independent variables of our study. While wellbeing cannot be measured directly, it can be estimated using closely related subjective and objective measures. To measure the educational wellbeing of students, we created two educational wellbeing indices (the objective educational wellbeing index and the subjective educational wellbeing index) from five educational outcome indicators. These educational outcome indicators were provided only for students aged 8–11 in round one of the IHDS (Desai et al. 2010). Two of these indicators were reported by the child’s mother, and are subjective measures of educational wellbeing: namely, what type of student the child is (below average, average, above average), and whether the child enjoys school (no, yes). The other three indicators are objective measures of educational wellbeing: namely, the child’s reading, mathematical and writing skills. These measures were developed from short reading, mathematical and writing tests administered to all students aged 8–11 in the households the IHDS surveyed. The reading skills of students are divided into five categories: (i) cannot read at all, (ii) can read alphabets but not words, (iii) can read words but cannot read full sentences, (iv) can read a short paragraph of 2–3 sentences but cannot read a full page and (v) can read a full story. The mathematical skills of students are divided into four categories: (i) cannot read numbers, (ii) can recognise numbers but cannot do any arithmetic operations, (iii) can subtract a two-digit number from another number and (iv) can divide a three-digit number by a one-digit number. The writing skills of students are divided into two categories: (i) cannot write at all and (ii) can write a sentence with two or fewer mistakes. Distribution of students by these variables are presented in Table A.1 and the technical details of the construction of the index are provided in Section 2.6.1.

Previous research has shown that several factors influence the school dropout and grade repetition rates of students (Falch and Strøm 2008; Glick and Sahn 2010; Gouda and Sekher 2014). Accordingly, we have included relevant student-related centeracteristics, parent-related centeracteristics and socio-economic centeracteristics of the household as control variables. The student-related centeracteristics are as follows: the age of the student (in completed years), the gender of the student (male, female), the type of school the student attends (public school, private school, other) and whether the student receives private tuition (no, yes). The parent-related control variables are as follows: mother’s level of education (no formal schooling, less than five years of schooling, six to 10 years of schooling, more than 10 years of schooling), mother’s working status (not working, working), father’s level of education (no formal schooling, less than five years of schooling, six to 10 years of schooling, more than 10 years of schooling) and father’s working status (not working, working). The control variables related to the household’s socio-economic centeracteristics are as follows: household wealth quintile (richest, rich, middle, poor and poorest), household poverty status (not poor, poor), the caste of the household head (Other Backward Class (OBC), Scheduled Castes (SC), Scheduled Tribes (ST), other), the religion of the household head (Hindu, Muslim, other) and the place of residence (rural, urban). Additionally, the region of the country (northern, north-eastern, central, eastern, western, southern) the student came from is included as a state-level centeracteristic. All of these variables are measured for the panel of children aged 8–11 in 2005.

We have estimated the wealth quintile for all households in round one using data on each household’s asset ownership, material type, water source, sanitation facility type and number of bedrooms. This approach is based on globally-defined standard procedures (Filmer and Scott 2008; Rutstein and Johnson 2004). The wealth scores for each household have been generated using principal component analysis (Filmer and Scott 2008). Based on the wealth score, the households have been classified into five categories (richest, rich, middle, poor, poorest), such that the households with the lowest 20th percentile score are assigned to the “poorest” category; the households with the next-lowest 20th percentile score are assigned to the “poor” category; and so on.

During round one, information was collected about the religion and caste of all household heads. Based on their religion, the households have been assigned to nine categories: Hindu, Muslim, Christian, Sikh, Buddhist, Jain, tribal, other and none. Owing to the skewed population distribution, for the purposes of this study, we have recoded the religion of the household head into three categories: Hindu, Muslim and other, whereby the other category consists of all other religious categories except Hindu and Muslim. Similarly, the IHDS classified the caste of the household head into five categories: Brahmin, Other Backward Caste [OBC], Scheduled Caste [SC], Scheduled Tribe [ST] and other. Again, we have recoded the caste variable into four categories: OBC, SC, ST and other, whereby the other category combines the Brahmin and the other categories from the original variable. The respondents in the ST and the SC categories belong to the most socially backward group of people; i.e., they are members of the lower rungs of the now constitutionally abolished Indian caste system. As the name suggests, the respondents in the OBC category also belong to a socially and economically backward group of people, although they tend to have better conditions than those of the SC/ST people. The other category consists of all respondents who do not belong to any of the three caste groups.

We constructed the country regions by assigning the current and erstwhile 33 states and union territories of India to six categories. The northern region includes Chandigarh, Delhi, Haryana Himachal Pradesh, erstwhile Jammu and Kashmir, Punjab, Uttaranchal and Rajasthan. The north-eastern region includes Assam, Arunachal Pradesh, Manipur, Meghalaya, Mizoram, Nagaland, Tripura and Sikkim. The central region consists of Madhya Pradesh and Chhattisgarh. The eastern zone consists of Bihar, Jharkhand, Odisha and West Bengal. The western region comprises Dadra and Nagar Haveli, Daman and Diu, Goa, Gujarat and Maharashtra. The southern region comprises erstwhile Andhra Pradesh, Karnataka, Kerala, Tamil Nadu and Pondicherry (Sanjay 2020).

Of the children in our study sample, nearly 57% had average and above objective educational wellbeing scores, and 50% had average and above subjective educational wellbeing scores in 2005. Looking at Table 3, we can see that of the panel of children, 53% were male and 47% were female. Moreover, 70% of the students were attending public school, and 82% were not receiving private tuition in 2005. Turning to parental education, we can see that 55% of the students had a mother and 26% had a father with no formal schooling. In terms of household economic status, 38% of the students came from a household in the poor/poorest wealth quintile, and 29% came from a household below the poverty line (BPL). Moreover, 40% of the students were from OBC households, 79% were from Hindu families and 72% were from rural households. In terms of region of the country, the vast majority of students came from the northern region (38%), distantly followed by the southern regions (18%).

The last column of Table 3 gives the absolute difference in the percentage distribution of students across the cross-sectional and panel populations in 2005. Note that the percentage distribution of students across the centeracteristics is largely similar in the two datasets, which indicates that the study is unlikely to suffer from a sample selection bias, whereby students with certain selected centeracteristics had a greater likelihood of dropping out of the survey in the second wave. Only the percentage distribution by the father’s education differed by more than 5% between the two datasets.

Table 4 displays the bivariate association between the educational wellbeing of students and other control variables in round one with the dropout and grade repetition status of students in round two. We found that among the 9822 students, 2273 (23%) had dropped out of school and 1668 (17%) had repeated a grade between 2005–2012. The bivariate results showed that students with a below-average objective wellbeing score in 2005 were highly likely to have dropped out of school (31%) or repeated a grade (20%) by 2012. Similarly, among the students who had a below-average subjective educational wellbeing score, the proportions who dropped out of school (27%) or repeated a grade (18%) were relatively large. Furthermore, larger shares of public school than of private school students had dropped out (28%) or repeated a grade (20%) by 2012. Moreover, among the students whose mother had no formal schooling, the shares who dropped out of school (32%) or repeated a grade (19%) were disproportionately large. Likewise, among the students whose father had no formal schooling, the shares who dropped out of school (36%) or repeated a grade (19%) were relatively large.

When we looked at the role of socio-economic centeracteristics, we found that the shares of students who dropped out were especially large among those who came from a household that was in the poor wealth quintile (39%) or below the poverty line (35%). We observed a similar pattern for grade repetition, whereby 23% of students from a household in the poor wealth quintile and 22% of students from a BPL household had repeated a grade. We further observed that 36% of students who belonged to the Scheduled Tribes and 25% of students from a rural community had dropped out of school by 2012. Similarly, 26% of ST students and 18% of rural students had repeated a grade by 2012. Moreover, we found that the shares of students who dropped out of school (30%) or repeated a grade (33%) were relatively large in the central region of India.

Table 4 provides the parameters of the group-level (state-level) effects from the random intercept logit models for the school dropout and grade repetition rates, respectively. The ICC from the null models showed that the state-level centeracteristics accounted for 11% and 8% of the unexplained variation among the students in the risk of dropping out of school or of repeating a grade in 2012, respectively. The MOR showed that among students from high-risk states, the risk of dropping out of school was 1.84 times higher, and the risk of repeating a grade was 1.68 times higher in 2012. From the null to the full models, both the ICC and the MOR declined due to the inclusion of explanatory covariates, which reduced the unexplained variance at the state level (Steele 2013).

After adjusting for the effects of the independent variables and the state-level effects, Table 5 shows the associations between the students’ educational wellbeing and other control variables in 2005 and their school dropout and grade repetition status in 2012. The multivariate analysis indicated that compared to students with above-average wellbeing scores, students with below-average objective wellbeing scores in 2005 were 1.65 [CI: 1.48–1.85] times more likely to have dropped out of school and 1.33 [CI: 1.18–1.49] times more likely to have repeated a grade between 2005–2012. Similarly, the odds of dropping out of school [OR: 1.30, CI: 1.16–1.45] and of repeating a grade [OR: 1.17, CI: 1.04–1.31] were found to be higher among students with below-average subjective educational wellbeing scores. Additionally, we observed that compared to students who were attending a private school, students who were attending a public school in 2005 were 1.75 [CI: 1.47–2.08] times more likely to have dropped out of school and were 1.42 [CI: 1.20–1.68] times more likely to have repeated a grade. Moreover, we found that the likelihood of having dropped out of school was 3.69 [CI: 2.12–6.41] times higher among students whose mother had no formal schooling, and was 2.61 [CI: 1.96–3.46] times higher among students whose father had no formal schooling. Similarly, the odds of grade repetition were shown to be higher for students whose mother and father had no formal schooling. In addition, our results showed that compared to students from the richest quintile households, students from the poorest wealth quintile households were 3.50 [CI: 2.63–4.65] times more likely to have dropped out of school and 1.34 [CI: 1.02–1.78] times more likely to have repeated a grade. We also found that compared to students living in rural communities, students living in urban communities were 1.45 [CI: 1.25–1.69] times more likely to have dropped out of school.

Furthermore, we tested the interaction effects to explore whether the effects of the objective and the subjective wellbeing indices on the dropout and grade repetition status of students varied by the mother’s level of education and by household wealth (see Tables A.3 and A.4 in the appendix section). The empirical evidence showed that the mother having higher education significantly reduced students’ chances of dropping out of school and repeating a grade, particularly among girls (Cardoso and Verner 2006; Lloyd et al. 2009; Marphatia et al. 2019). A potential explanation for this finding is that the mother’s education protects her children from having below-average educational wellbeing, and thus lowers their chances of experiencing unfavourable educational outcomes, like dropping out of school or repeating a grade. Moreover, previous evidence has shown that the traits of households in the poor wealth quintile may negatively influence the schooling outcomes of the students in these households (Marphatia et al. 2019; Paul et al. 2021; Pong and Ju 2000). Thus, it is clear that coming from a higher income household may pave the way for students to receive more educational resources, like higher quality schooling and tailored tuition. Such amenities can support the educational wellbeing of children, which can, in turn, further reduce their chances of dropping out of school or repeating a grade. Therefore, we examined the interaction effects of educational wellbeing with the mother’s education and household wealth. However, we found that the interaction coefficients were not statistically significant (see Tables A.3 and A.4 in the appendix section), which indicates that the effects of the objective and the subjective wellbeing indices on the dropout and grade repetition status of students did not vary by either the mother’s education or the household wealth.