Early childhood circumstances and educational wellbeing inequality among tribal and non-tribal children in India: evidence from a panel study

Despite efforts towards bridging the education gap between tribal (Scheduled Tribe) and non-tribal (Non-Scheduled Tribe) children, contrasting poor-quality education questioned the tribal children’s educational wellbeing in India. Early childhood circumstances render a remarkable impact on the educational wellbeing of children in later years. This study examined the influence of early childhood circumstances (child, household and community characteristics) during 2005 on the educational wellbeing inequality (among India’s tribal and non-tribal children) during 2012 using the India Human Development Survey panel dataset of 8611 children. The Educational wellbeing score was obtained from reading, mathematical and writing test scores using Principal Component Factor Analysis. We performed the Blinder-Oaxaca decomposition of the educational wellbeing inequality among India’s tribal and non-tribal children. The ST children’s average educational wellbeing score (−0.41) was much lower than the Non-ST children (0.04). Findings from the Blinder-Oaxaca decomposition show that the household economic condition in children’s early ages contributed to 24% of educational wellbeing inequality among tribal and non-tribal children. Further, the education status of males and female adults and the sanitation condition of families considerably impacted educational wellbeing. The present study concludes that caste antagonism has not reduced with time. The missing focus on the minority groups resulted in a deteriorated educational wellbeing.

IHDS round-I was a nationally representative multi-topic survey of 215,754 people from 41,554 households 21 . Round-II, conducted in 2012, was a multi-topic panel survey of 204,569 people from 42,152 households in India 22 . The University of Maryland, USA and the National Council of Applied Economics Research (NCAER), India, conducted the two IHDS rounds in India's states and union territories (except Andaman and Nicobar Islands and Lakshadweep). Round-II of the IHDS re-interviewed 83% of the original families from round-I living in the same village. IHDS used a stratified random sampling design to choose samples. More information on the sampling design, survey timeframe, and data collection methods used in rounds I and II can be found elsewhere 23,24 . The analytical sample is the panel of 8611 children aged 1-4 years in round-I who became 8-11 years old during round-II, after excluding the missing observations (see Fig. 1). Of the 8,611 children, 7850 (91%) and 761 (9%) belonged to the Non-ST and ST caste groups.
Constructing the educational wellbeing index. The continuous indicator of educational wellbeing during round-II is the outcome variable of this study. The mathematical, reading and writing test scores of children aged 8-11 years were used to prepare the educational wellbeing score.
The reading skill of students has five categories: (1) cannot read at all (score 0), (2) can read alphabets but not words (score 1), (3) can read words but cannot read entire sentences (score 2), (4) can read a short paragraph but cannot read a whole page (score 3), www.nature.com/scientificreports/ Equivalently, the mathematical skill of students has four categories -(1) cannot read numbers (score 0), (2) can recognize numbers but cannot do any arithmetic operations (score 1), (3) can subtract a two-digit number from another number (score 2), (4) can divide a three-digit number by a one-digit number (score 3).
The writing skill of students is categorized: (1) cannot write at all (score 0), (2) can write a sentence with two or fewer mistakes (score 1).
The educational wellbeing variable was constructed using Principal Component Factor Analysis (PCFA) on the reading, mathematical, and writing skill variables and the detailed procedure is described elsewhere 25 . Notably, the PCFA for educational wellbeing indicators resulted in a one-factor solution. From supplementary Table S1, we observed that the first factor had an eigenvalue of more than one and explained 73.4% of the total variability of all three educational wellbeing indicators. All three indicators had factor loading values of more than 0.80. Further, all indicators had Kaiser-Meier-Olkin (KMO) values greater than 0.70, thereby justifying our use of PCFA (KMO values greater than 0.50 are necessary for conducting PCFA). Finally, we generated the standardized educational wellbeing score based on the first factor.
Group variable. The binary caste group variable, whether an individual belongs to the Scheduled Tribes (ST) or Non-ST, is the group variable. The caste system is a form of social hierarchy native to India. Notably, the Indian constitution recognizes three distinct social groups-Scheduled Tribes, Scheduled Castes and Other Backward Classes. People in the ST (predominantly tribal population) and SC categories are the most socially backward. They traditionally belonged to the lowest rung of India's now-defunct caste system. People of the OBC category, as the name implies, are also members of a socially and economically backward community. However, their circumstances are better than those of the SC/ST population. The "Others" category consists of all people  Explanatory variables. Taking a cue from extant research, we included the child-, household-and community-related independent variables which explained educational wellbeing in Indian children 16,[26][27][28] . All the variables were obtained for children aged 1-4 years in round-I unless mentioned otherwise. The child-related explanatory characteristics are: (1) Age of children in years (One, Two, Three, Four).
(3) Stunting status of children (Stunted, Not stunted). Stunting indicates chronic undernutrition in children and is denoted by low height-for-age z-scores (HAZ) 29 . Although IHDS does not provide readymade HAZ scores, we obtained the HAZ scores of children aged 1-4 years in 2005 from their anthropometric data and the WHO Anthro software 30 . Children with HAZ scores of less than −2 standard deviations (SD) and more than − 6SD were coded as "Stunted", and those having HAZ scores of more than − 2SD and less than + 6SD were coded as "Not stunted". (4) Type of school attended in round-II (Public school, Private school).    www.nature.com/scientificreports/ (7) Type of cooking fuel (Solid fuel, Clean fuel). (8) Household sanitation condition (Poor, Average, Good). Based on extant studies, the sanitation condition of the household was prepared from the information on the type of drinking water, type of toilet facility and the number of members per room during 2005 16,31 . Households having "improved" drinking water and toilet facilities were scored as "1", and households in the "unimproved" counterpart were scored as "0". Equivalently, households with less than three members per room were scored as "1" and "0" otherwise. We added scores of the three variables to obtain household sanitation scores ranging from 0 to 3. Households with a score of 3, 2, or less than 2 were classified as "good, " "average, " or "poor" sanitation households, respectively. (9) Water purification in the household (No, Yes). (10) Household wealth quintile (Poorest, Poor, Medium, Rich, Richest). The household wealth quintile variable was constructed using standard procedures documented elsewhere 32,33 . We generated wealth scores by applying principal component factor analysis (PCFA) on the household asset ownership, livestock ownership, and type of building material information during 2005. The households were then classified into five wealth quintiles from "poorest" to "richest" based on the wealth scores. The community-related explanatory characteristics are: (19) Solving community problems (Each family individually, All families together). (20) Domestic violence in the community (Yes, No). IHDS 2005, collected information from a single woman (aged 15-49 years) from each household on whether husbands in the community assaulted their wives if-"her natal family does not provide money, jewelry and other items", "she does not cook food properly", "she goes out without telling him", "she neglects the house or the children" and "is suspected of having a relationship with other men". If a woman responded positively to any of the five questions, we classified domestic violence status in the community as "Yes" or a "No" otherwise. Statistical methods. At the start, we showed the absolute and percentage distribution of children by the background characteristics in round-I. The caste difference in average educational wellbeing score was assessed using the two-sample t-test. The caste difference in educational wellbeing across the explanatory variables was assessed using the chi-square test for independence. Next, we estimated multivariable linear regression models to examine the association between child-, household-and community-related variables in 2005 with the educational wellbeing of children in 2012. The coefficient in the multivariable models gives the adjusted change in the educational wellbeing score in round-II of children belonging to a particular category of an explanatory variable in round-I after adjusting for the effect of all the other explanatory variables 35 . The above analyses were performed separately for ST and Non-ST children. Further, we used the Blinder-Oaxaca twofold decomposition technique to identify the contribution of explanatory covariates in 2005 behind the caste differential in the educational wellbeing of children in 2012 36 . We show the overall and detailed decomposition of the caste differential in educational wellbeing. In the overall decomposition, the caste gap in educational wellbeing is decomposed into an explained (E) component and an unexplained (C) component 36 . The detailed decomposition shows the relative contribution of each child-, household-and community-related early childhood characteristics to caste inequality in educational wellbeing during 2012.
Note that the Non-ST group is heterogeneous (as it comprises SC, OBC and Others groups), and the results of the decomposition estimates might vary if we compare the ST children's educational wellbeing with that of SC, OBC and Others children individually. Therefore, we decomposed the educational wellbeing gap between ST and SC children, ST and OBC children, and ST and Others children, respectively. We performed this sensitivity analysis to check the sensitivity of the decomposition estimates (shown in Table 4) to the categorization of comparison groups. In our tests, none of the multivariable models violated the assumption of multicollinearity 37

Results
Descriptive statistics. Table 1 revealed the distribution of the panel of children aged 1-4 years by sociodemographic, health, household and community-related characteristics. Approximately one in every ten children were from the Scheduled Tribes group, and 47% were females. ST children's mean educational wellbeing score (−0.41) was significantly lower than that of Non-ST children (0.04). Nearly 53% and 46% of ST and Non-ST kids were stunted. The level of private schooling was higher in non-tribal children (38%) than in their tribal counterparts (15%). Moreover, only 10% of tribal kids took tuition compared to 24% of non-tribal kids. Further, most ST children come from families who cooked using solid fuel (95%), and nearly 71% of ST children belonged to households having poor sanitation conditions and no means of water purification. Three-fourths of the ST children were from the poor-poorest wealth quintile households, and six in ten ST kids were from households below the poverty line (BPL). In the scheduled tribe population, most of the household males were uneducated and unfortunately, the figure doubles in the case of females. The male-headed household was prominent in the ST group (94%), and the presence of violence was almost 80% in the community of the ST group. Nearly 90% of the tribal children resided in a rural community, and 32% lived in the central regions of India.
We checked the percentage distribution of children by relevant demographic and socioeconomic characteristics in the cross-sectional and panel survey during the baseline period for possible attrition bias. From the results in Table S2, we found that the percentage distributions of children across the selected characteristics were similar in the cross-sectional and panel surveys. Only the percentage distribution of children by age and type of community differed by greater than 3% between the two surveys. Table 2  Decomposing the caste inequality in educational wellbeing score. Table 3 (Table 4). Among 79.1% explained share, household-level variables of wealth quintile (24.3%), highest educational level of adult males (13.7%) and females (13.1%) in the families has a prominent contribution to ST and Non-ST children's educational wellbeing inequality. Among the child-related characteristics, taking private tuitions (6.9%), school attending status (6.2%), and stunting status (2.5%) had significantly higher contributions to caste inequality. Additionally, household sanitation conditions and poverty status contribute to children's educational wellbeing inequality. Domestic violence in the community (−0.8%) shows a significantly negative influence on the inequality of educational wellbeing scores. This negative value indicates that ST children experience higher educational wellbeing scores in a community with no domestic violence, and if we eliminate this advantage, it would further deteriorate their children's educational wellbeing scores. Table 5 shows the decomposition of the educational wellbeing gap among children in ST and Non-ST, ST and SC, ST and OBC, and ST and Others groups, respectively. We find that the direction of contribution is the same across all the statically significant contributors in the four decomposition estimates. The explained educational wellbeing difference between ST and Non-ST groups is similar for the ST and OBC, and ST and Others groups. The magnitude of the percentage contribution of each statistically significant contributor varies across the four decomposition estimates. However, the difference in the percentage contribution is not more than 5% in the contributors across the four decomposition estimates.

Sensitivity analysis of decomposition estimates to the categorization of comparison groups.
Ethics declarations. The present study utilized a publicly available secondary dataset with no information that would lead to the identification of the respondents. IHDS obtained the consent of respondents before data collection. Therefore, no ethical approval was necessary. All survey methods were performed following the relevant guidelines and regulations.

Discussion
Understanding the power of education in changing the lives of individuals, families and communities, the government of India has made a constant effort to bring children to schools and provide primary education. However, before celebrating the success of bridging the schooling gap in tribal (ST) and non-tribal (Non-ST) children, it is essential to determine the quality of education these children received in the past few years. Sadly, the present study shows a challenging face of the education system, where the educational wellbeing score of tribal children is significantly lower than their non-tribal counterparts. The salient findings of the study and their explanations are as follows: First, although the government has tried to eradicate caste-based discrimination in the education system, it is still prevalent with the tribal population at the receiving end. This finding is supported by the multivariable regression and decomposition analysis results. Such a situation may arise due to the unavailability of good schools in the community and qualified teachers. Past evidence has shown that their reading, writing, and mathematical competence was shallow even if the tribal children were attending schools. Curriculum and communication play an essential role in preventing children's educational wellbeing. It has been observed that the inclusion of local culture, folklore and history, and the local dialect in the curriculum builds confidence in tribal children. Further, interpreting through paintings, music, and storytelling can improve their educational wellbeing as they are common in their culture. Table 1. Socio-demographic, health-related, household and community characteristics profile of the panel of children aged 1-4 years during IHDS round-I. (a) N sample size, SD standard deviation, %: column percentage; (b) difference in educational wellbeing score by caste group was tested using T-test while the caste difference of explanatory variables was tested using the chi-square test for independence.  www.nature.com/scientificreports/ Second, early life stunting status can hinder the educational wellbeing of children. Consistent with the present study, an Indian study showed that child nutritional status affects their physical, cognitive and language development 16 . Moreover, the present study confirms that the child's educational wellbeing depends on their type of school and private tuition. Besides, these child characteristics-stunting status in early life, private schooling, and private tuitions largely contribute to the educational wellbeing inequality between tribal (ST) and non-tribal Table 2. Linear regression models showing the multivariate association between educational wellbeing score and the individual, household and community characteristics of the panel of children aged 1-4 years during IHDS round-I. (a) Coef coefficient, CI confidence interval, (Ref) reference category; (b) statistical significance denoted by asterisks where *p-value < 0.1, **p-value < 0.05 and ***p-value < 0.01.

Characteristics
Educational wellbeing score of children aged 8-11 years in round-II   Table 3. Overall Blinder-Oaxaca decomposition of the caste differential in educational wellbeing score of the panel of children during IHDS round-II. (a) CI confidence interval; (b) statistical significance denoted by asterisks where *p-value < 0.1, **p-value < 0.05 and ***p-value < 0.01.  39 . Most tribal households cannot fulfill their basic living needs, so education becomes their secondary priority. Even if the government had introduced free education and mid-day meal schemes for bringing the tribal children to schools, the financial constraints of households would restrict them from completing their education. Studies have shown that absenteeism and dropout are higher among tribal children, especially during crop cultivation. This situation can leave the children behind in the classroom compared to other regularly attending schools. Fourth, parents' education or the education of household members can also affect the educational wellbeing of children. Since uneducated elders in the household cannot help the children efficiently, there is past evidence that parent involvement has a commendable role in a child's educational achievement 40 .

95% CI Percent
To the best of our knowledge, the current study is among the few studies examining inequality in India's educational wellbeing scores of tribal and non-tribal children. Further, using the decomposition analysis, the study shows the contribution of early life circumstances to such inequality. We know that the early childhood period represents the development pedestal for the later years. Children's exposure to physiological and socioeconomic stress during this period gets manifested as reduced educational wellbeing in the long run 16,20,41 . Therefore, crosssectional studies examining the relationship between educational wellbeing determinants will misestimate the effect. The panel nature of this study helps us point out the role of individual, household and community factors of children aged 1-4 years (early childhood period) behind the differential in educational wellbeing in tribal and non-tribal children when they become 8-11 years.
Moreover, the study's findings did not suffer from attrition bias as the demographic and socioeconomic distribution of children in the cross-sectional and panel surveys during the baseline period were similar. This finding is similar to other studies that have used the IHDS panel dataset 16,25,40 . Additionally, the decomposition estimates were not sensitive to the heterogeneity in the non-tribal group. The sensitivity analysis revealed that the decomposition estimates were robust to the categorization of comparison groups. However, the study has its shortcomings. This study did not provide any causal inference. Further, due to the requirement of including a nationally representative panel dataset to show the contribution of early life circumstances, we have to use data from 2005 and 2012. Therefore, readers need to be cautious of the survey date while interpreting this study's findings. Table 4. Detailed Blinder-Oaxaca decomposition of the caste differential in educational wellbeing score of the panel of children during IHDS round-II. (a) CI confidence interval; (b) statistical significance denoted by asterisks where *p-value < 0.1, **p-value < 0.05 and ***p-value < 0.01.

Characteristics
Caste differential in educational wellbeing among children in round-II

Conclusion
The missing focus on the minority groups excluded these communities from education participation. Historically, tribal children faced rejection and discrimination in terms of their backwardness. Such discrimination can be seen in inequality in their educational wellbeing due to their early life circumstances. Commendable progress has brought tribal children to schools in the past few years. Still, efforts should also be made towards reducing their discontinuation and improving their quality of education which can improve their educational wellbeing. Quality education refers to both qualities of the school infrastructure, teacher and the learning process. Inclusion of an interactive curriculum based on their culture with proper communication at basic levels can help improve children's educational wellbeing. Besides these factors, policies should also focus on providing targeted interventions during the early childhood period of tribal children by improving their household conditions, sensitization of parents and the community about educational opportunities and advantages during their initial years, and creating a peaceful and healthy community. Notably, early childhood conditions can be improved by providing targeted benefits to ST children through existing nutrition-security and wellbeing programs (Integrated Child Development Services, Antyodaya Anna Yojana and Poshan Abhiyan) of the Indian government.

Data availability
The study utilizes a secondary source of data that is freely available in the public domain from the Inter-University Consortium for Political and Social Research (ICPSR) data repository (https:// www. icpsr. umich. edu/ web/ DSDR/ series/ 507).