The impact of life skills education on socio-emotional development and school-related outcomes among adolescents in India

Abstract

Background

The Magic Bus India Foundation (MBIF) Life Skills program, known as the Childhood to Livelihood (C2L) program, aims to build life skills among adolescents aged 11 to 15 years. This three-year program focuses on education, gender equality, and socio-emotional skills development. The study evaluates both short-term and long-term impacts of the program on school-related outcomes and socio-emotional skills, providing insights into its effectiveness across multiple sites in India.

Methods

Data were collated from five study sites with a pre-post cohort research design. Baseline data were collected at the start, midline data at 18 months, and endline data at three years. The study used harmonized data across projects, with a sample of 1898 children for short-term analysis and a larger sample of 5582 children for long-term analysis. The empirical strategy involved panel data analysis using multiple linear regression models to evaluate program effects on outcomes such as school attendance, self-efficacy, resilience, and gender attitudes.

Results

The short-term analysis showed significant improvements in school attendance, aspirations, and socio-emotional skills among participants. On average, the C2L improved egalitarian gender attitudes by 0.6% points and perceived self-efficacy by 4% points in the short-run. Over the long-term, there was an average increase of 2.5 points in perceived self-efficacy, 1.1 points on resilience, and 0.45 points in egalitarian gender attitudes. The odds of regular school attendance increased by 66.5% points.

Conclusions

The C2L program has a positive impact on both educational and socio-emotional outcomes among adolescents. The findings suggest that life skills interventions can effectively enhance school attendance, self-efficacy, resilience, and gender attitudes. These results have important implications for policy, indicating that integrating socio-emotional learning (SEL) programs in school curricula can contribute to better educational and developmental outcomes for adolescents in resource-constrained settings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-024-21195-0.

Introduction and Background

Socio-emotional skills can equip children with the knowledge and abilities to maintain healthy friendships, resolve disputes effectively, appropriately deal with stress and anxiety and boost confidence [1, 2]. Many studies explain the relevance of socio-emotional learning (SEL) and how it should be viewed as complementary to intellectual abilities to achieve overall success in academic and professional life [3]. This aligns with wide-scale systematic reviews that explore the interplay between personality traits and economics [4, 5]. Past literature suggests that building such socio-emotional skills can be important for human capital outcomes [6], including educational attainment [7, 8] and health-related behaviour [9]. Finally, in light of the COVID-19 pandemic and the shift to online and hybrid learning modes, SEL has important implications for educational outcomes. Competence-based curricula and those promoting creativity and problem-solving can also be important in tackling the challenges and learning losses due to COVID-19 [10]. Yorke et al. [11] highlight the relevance of SEL during COVID-19 in the face of large-scale school closures, which puts children from marginalized groups at greater risk of adverse educational outcomes. This is especially relevant in the case of India, which has among the largest share of adolescents to the total population in the world, and has reported tremendous levels of learning loss following the pandemic [12].

In this paper, we examine impacts of an SEL intervention (Childhood to Livelihood, or C2L) implemented by the Magic Bus India Foundation (MBIF) on school-related outcomes in India. The design of the C2L program, although not a cleanly designed randomized control trial (RCT), allows us to examine program impacts on adolescent school-related outcomes and development of socio-emotional skills such as egalitarian gender attitudes, resilience, and perceived self-efficacy. To the best of our knowledge, this is the first formal test of an intervention across multiple sites in India, allowing for an understanding of whether SEL interventions can have impact at scale. In order to identify the short- and long-term impacts of the program, we use a pre-post cohort research design and panel data to analyze the short-term and long-term program impacts, respectively. In the absence of meaningful ways to test whether the parallel trends assumption holds (as there is no data available on a comparable cohort who did not receive the program), we advise caution on interpreting causality from our estimates.

Our analysis reveals a robust and statistically significant positive impact of the program across multiple dimensions over the short and long term. The program led to a 0.6-point improvement in egalitarian gender attitudes and a 4-point increase in perceived self-efficacy on average, with the strongest gains in gender attitudes observed among older children (> 13 years) and improvements in self-efficacy concentrated among the youngest participants in the short term. Notably, girls experienced a 1.5-point rise in gender attitudes and a 55% increase in the odds of regular school attendance. In the long-term data, school-related outcomes improved significantly, with a 66.5% rise in the odds of regular attendance and a 27.1% increase in aspirations to pursue higher education. Additionally, participants showed a threefold increase in the odds of studying until graduation, alongside heightened awareness of the Right to Education (RTE) program.

SEL interventions have been tested in a variety of contexts around the world. Ashraf et al. [13] find that teaching girls interpersonal and non-cognitive skills (especially negotiation skills) improved educational outcomes in the subsequent three years, and dropout rates fell by 10% points in Zambia. In rural China (where dropout rates are similar to those in developing countries), an SEL intervention aided in reducing dropout by 1.6% points at the midline, and by 6.1% points among older students nearing 16 [14]. In South Africa, Hofmeyr [15] finds that grit (similar to resilience) interacts with school characteristics (and quality of schooling) to impact school outcomes and cognitive scores. Given the context-specificity of such interventions, it becomes important to tailor interventions instead of approaching SEL via the one-size-fits-all approach: Culturally-adapted research can raise participant involvement from diverse racial and ethnic backgrounds since, historically, it has been the case that African-Americans’ engagement in SEL interventions has been consistently low [16]. Elsewhere, using survey data from the Young Lives survey in Peru, Arapa et al. [17] find that agency and pride were positively associated with school attendance, and further with cognitive scores on math and vocabulary. However, they find a negative association between self-efficacy and school attendance among older children.

In India, the focus is on developing SEL is on girls and shifting gendered norms around education. Edmonds et al. [7] test whether group mentoring and life skills sessions can enhance women’s agency in females, bridge perceptions of gender inequality, and improve females’ schooling attainment. They find improvements in perceived gender equality among girls and boys, as well as a 30% fall in the dropout rate at the endline. Bhadwal and Panda [18] find that socio-emotional learning lowered test anxiety among fifth class students. Research also suggests that students involved in SEL programmes outperform other students in indicators of academic achievement by 11 percentile points [19]. In the eastern state of Jharkhand, it was inferred that a one-standard-deviation increment in self-efficacy translated into a 0.73-unit rise in girls’ aspirations to study for a greater number of years [20]. In similarly resource-constrained Bihar, a psychosocial intervention by Leventhal et al. [9] enabled equality in fostered gender attitudes, and improved health awareness and menstrual hygiene. This shows that an adolescent health curriculum supplemented by a curriculum aimed at resilience yields more remarkable results than implementing just a strictly curriculum-based intervention.1

The MBIF Program

The C2L program is aimed at building life skills among adolescents and starts between ages 11 to 15. Running for a duration of three years, the program has the following key components which are monitored and on which data is available: (a) Education (school regularity, 5 or more days a week attendance, Right to Education, and making children aware of their rights and facilities, class participation and benefits of education, staying in school); (b) Gender (across domains, equality, equity perceptions, challenging [cultural] stereotypes), and (c) Socio-emotional skills (self-efficacy/resilience - problem solving, perceptions in the community).

The program is mainly implemented by Community Youth Leaders (CYLs) who are volunteers from the local community where the program is being implemented. They serve as mentors for adolescents, conducting weekly sessions using activity-based curricula, covering aspects of schooling, gender, and socio-emotional skills. Specific details on the timing and duration of these sessions are unavailable, but CYLs are specifically instructed to conduct sessions either during break times or after school instructional hours. There are additional components of the C2L program that we do not study here; these are related to workplace readiness and building employability skills. This is primarily because currently data is still being collated on employment-related outcomes of adolescents who participate in the program. The details of the programs from which data for this study is drawn can be found in Table A.1 in the appendix. In what follows, both the short-term and the long-term program are similar in all aspects except the points at which data is available for participants, as well as whether there is data on a control group of children who did not receive the intervention.

The remainder of the paper is organized as follows. In Sect. 2, we present the methods and the data available. In Sect. 3, we present the key findings of our analysis. The paper concludes in Sect. 4 with a discussion of the implications of the study for policy and recommendations for future research.

Methods

Sampling

As part of the program, data are collected at three intervals: baseline (at the start of the program), midline (approximately 18 months into the program), and endline (approximately three years into the program, upon completion). Questionnaires were administered using computer-assisted personal interviewing (CAPI) methods in the local language.2 The details of the questionnaire and administration can be found in Mehta and Kapoor [24]. Some programs run for a shorter duration and data collection intervals change accordingly but are not considered in this analysis.

Baseline-Midline (Short-term) data

We use harmonized data from five (of eight) separate projects that all had a similar research and sampling design and contained survey data on all variables of interest (Projects 2, 4, 6, 7, and 9 in Table A.1). Specifically, only those studies that had a “pre-post” sampling design, which made the projects amenable to comparisons between two time periods, were considered for this paper. These are studies where measurements were taken from eligible children before and 18 months after the life-skills interventions were completed. In any given study site, all adolescents (11 to 15 years) were eligible to participate in the study. Of these eligible children, a random sample was drawn and assigned to the intervention groups. To ensure representativeness of the sample, they were drawn proportionate to their characteristics (age, gender, religion, and caste) to their overall population in each program location. Typically, surveys3 with adolescents were conducted in a staggered manner across 18 months.

Baseline-Endline (Long-term) data

In contrast to the simple pre-post cohort evaluation design used in the short-term impact studies, the data for the longer-term impact come from a large-scale project conducted between 2015 and 2018 (Project 1 in Table A.1). This study randomized at the child-level, resulting in separate treatment and control groups, the latter of which consisted of adolescents that were not exposed to the program but on whom data is available. This means that there is first baseline data on children who were part of both an intervention as well as a control group, and subsequently endline data on intervention and control was also collected. This is the largest study under consideration in this paper (N = 5582 at baseline; and N = 10,413 combined baseline-endline), and allows us to overcome some of the limitations of the pre-post approach in the short-term studies by comparing the impacts of the program on the treated children with a control group of children that did not receive the program. Details of this long-term project is included in the first row (Project ID 1) in Table A.1 in the appendix.

The longer-term impact program has a slightly wider age range covered in the sample, and is implemented with children and adolescents between 10 and 17 years of age. This represents the age range from the baseline to endline (i.e., all children who were between 10 and 15 years were eligible for participation). The intervention was mainly focused on nutrition, but also had intervention components that dealt with staff/volunteers, life skills education (similar to other factors mentioned here); healthy eating, engagement with parents, the community, and other local institutions. Thus, an intent-to-treat estimate can be derived from this evaluation design.

Data and empirical framework

Data are available in two groups: one sample where data collection is complete at the midline (18 months following program rollout) and another where data collection is complete at the endline (3 years after the program was implemented). Our empirical strategy therefore relies on evaluating program effects by forming longitudinal (panel) data between baseline and midline, and baseline and endline (where endline data is available). To retain comparability with the baseline analyses, we harmonized data across studies to retain key variables of interest on school-related outcomes, perceived self-efficacy, resilience, and gender attitudes. Additional child and household characteristics were also retained, where data was consistently available. Only those projects where a panel data formation was possible were retained for these analyses (i.e., a cohort approach, where children are tracked from the baseline to the midline or endline as described above). The final data come from five out of eight projects, details of which are presented in Table A.1 in the appendix.4

Across studies, data were harmonized to maintain consistency in measurement. All items on the resilience, self-efficacy, and gender attitudes scale were coded and cumulative scores were constructed as in the baseline analysis. Variable definitions can be found in Table 1.5

Table 1.

Variable definitions

Variable	Definition	Sample item	Internal consistency (Cronbach’s α)
Resilience	Total score cumulatively from 12 items rated along a 3-point Likert scale (1 = no, 2 = sometimes, 3 = yes). Higher total scores imply higher resilience. Maximum score possible is 36.	A sample item for the scale is “I try to finish activities that I start”.	0.94
Perceived self-efficacy (Schwarzer & Jerusalem, 1995)	Cumulative score from 10-item scale rated on a 4-point Likert scale (1 = strongly agree to 4 = strongly disagree). This index was reverse scored. High cumulative scores indicate higher perceived self-efficacy. Maximum score possible is 40.	A sample item for the scale is “I can always manage to solve difficult problems if I try hard enough”.	0.89
Gender attitudes	Cumulatively scored from 7 items scored on a 4-point Likert scale (1 = fully agree to 4 = fully disagree). Higher scores indicate more liberal gender attitudes and lower scores mean conservative gender attitudes. Maximum score possible is 28.	A sample item for the scale is “Teachers should encourage boys to take more classes in science and mathematics as compared to girls”.	0.76
Problem-solving	Cumulative score from 8 items along a 4-point Likert scale (1 = strongly agree to 4 = strongly disagree). This index was reverse scored to reflect higher scores indicating better problem-solving ability.	A sample item of the scale is “I easily identify my problems”. The data was not available consistently and hence not used in all analyses.	0.78
School attendance	Binary variable that takes a value of 1 if the child reported attending school at least 5 times a week, and zero otherwise.		NA
Liked attending school	Binary variable that took a value of 1 if the child responded yes or sometimes, and zero otherwise		NA
Awareness of RTE	A binary variable that took a value of 1 if the child reported that they had heard of the Right to Education (RTE) Act and zero otherwise.		NA
Aspirations	Ordered variable, where a value of 0 meant dropping out; 1 meant aspired to attending school till grade 8; 2 meant till class 10; 3 = class 12; 4 = graduation; and 5 = postgraduation		NA

The summary statistics are presented in Tables 2 and 3 separately for the samples that form the short-term impact evaluation and the longer-term impact study. The average age of the sample in the short-term and long-term sample is around 12 and 11.6 years of age, respectively and approximately 49% are female in the short-term sample whereas on average 45% were female in the long-term sample. A majority of the sample identified as belonging to Other Backward Classes (OBCs), making up 32.6% of the short-term sample and approximately 42% of the long-term sample. In the short-term impact sample, only about 21.2% aspired to study to complete an undergraduate degree, whereas this figure was nearly 41% in the long-term impact sample. The scores on the socio-emotional skills varied, but adolescents in the short-term sample scored very highly on resilience (nearly 33 on average of a maximum possible of 36). In the long-term impact sample, however, the average resilience score was 16.25. Similarly, in the long-term sample, the scores on perceived self-efficacy were much higher on average (31), compared to the short-term sample where average scores only 22.14. The balance test for randomization (Table 3) for the long-term project data suggests that there are two dimensions along which there are statistically significant differences: gender and religious identity. Thus, it is plausible that more boys were sampled in the treatment group relative to the control group, which could threaten the overall validity of the identification strategy. As a result, we report results using an interaction for gender, to better understand heterogenous treatment effects, and how much these diverge from the average treatment effects. To control for any other household or individual-level factors, we include these variables as additional controls in the estimation framework as outlined below.

Table 2.

Short-term impact data summary statistics (baseline only)

	(1)	(2)	(3)	(4)
Variables	Mean	SD	Min	Max
Age (years)	12.01	0.951	9	15
Proportion female	0.492	0.500	0	1
Proportion belonging to Other Backward Class (OBC)	0.280	0.449	0	1
Proportion Scheduled Caste (SC)	0.248	0.432	0	1
Proportion Scheduled Tribe (ST)	0.0818	0.274	0	1
Gender attitudes score	17.37	4.443	7	28
Resilience score	33.04	4.906	12	36
Perceived self-efficacy score	19.46	5.893	0	31
Proportion attending school at least five times a week	0.889	0.314	0	1
Grade currently studying in	7.060	0.968	5	9
Proportion aspiring to study an undergraduate degree	0.107	0.309	0	1
Proportion aware of Right to Education (RTE) Act	0.390	0.488	0	1
Observations	1898

Table 3.

Long-term impact data summary statistics (difference between control and treatment groups at baseline)

	Intervention/Control
	Control	Intervention	Total	Test
Age (years)	11.711 (1.615)	11.637 (1.919)	11.676 (1.767)	0.120
Gender attitudes score	16.452 (4.477)	16.469 (4.349)	16.460 (4.416)	0.885
Resilience score	16.245 (2.918)	16.255 (3.119)	16.250 (3.016)	0.900
Perceived self-efficacy score	31.027 (5.316)	31.115 (5.663)	31.069 (5.485)	0.551
Current grade	6.847 (7.722)	6.381 (7.313)	6.624 (7.532)	0.021
Gender
Male	0.527	0.580	0.553	<0.001
Female	0.473	0.420	0.447
Caste
Scheduled Caste	0.340	0.363	0.351	0.129
Scheduled Tribe	0.101	0.100	0.101
Other Backward Class	0.438	0.408	0.424
Upper Caste	0.121	0.129	0.125
Religion
Hindu	0.939	0.914	0.927	<0.001
Muslim	0.045	0.053	0.049
Sikh	0.002	0.001	0.001
Christian	0.000	0.002	0.001
Buddhist	0.014	0.031	0.022
Aspire to study undergrad	0.414	0.407	0.411	0.582
Regularity in attending school
Did not attend	0.006	0.007	0.006	0.274
1 – 2 days in a week	0.002	0.004	0.003
3 – 4 days in a week	0.020	0.026	0.023
5 or more days in a week	0.972	0.963	0.968
N	2,910	2,672	5,582

For continuous variables, p-values reported from main effects test from a linear regression. For factor variables, p-values reported from Pearson's chi-squared test

To form the panel, we first generated unique child IDs and a variable ($Post$) that took a value of 1 for all measures at the midline and zero at the baseline. These can be interpreted as the short-term effects of the program. The final dataset had data on all variables between baseline and midline for 1,802 children across five projects. We adopt a panel multiple linear regression variable framework for the short-term studies, where the following reduced form equation was estimated:

$${Y*}_{\mathit{ihvt}}=a+yPos{t}_{\mathit{ihvt}}+{\beta }_1\mathit{F}\mathit{e}\mathit{m}\mathit{a}\mathit{l}{\mathit{e}}_{\mathbf{i}\mathbf{h}\mathbf{v}\mathbf{t}}+{\mathit{\beta }}_{\mathbf{2}}\mathit{S}\mathit{c}\mathit{a}\mathit{l}{\mathit{e}}_{\mathbf{i}\mathbf{h}\mathbf{v}\mathbf{t}}\mathbf{+}{\mathit{\beta }}_{\mathbf{3}}{\mathit{X}}_{\mathbf{i}\mathbf{h}\mathbf{v}}\mathbf{+}{\mathbf{\in }}_{\mathbf{i}\mathbf{h}\mathbf{v}\mathbf{t}}$$ 1

$$\mathit{S}\mathit{c}\mathit{a}\mathit{l}{\mathit{e}}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}=\mathit{\omega }+\mathit{\kappa }\mathit{P}\mathit{o}\mathit{s}{\mathit{t}}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}+{\mathit{\delta }}_1\mathit{F}\mathit{e}\mathit{m}\mathit{a}\mathit{l}{\mathit{e}}_{\mathit{i}\mathit{h}\mathit{v}}+{\mathit{\delta }}_2{\mathit{X}}_{\mathit{i}\mathit{h}\mathit{v}}+{\mathit{\eta }}_{\mathit{i}\mathit{h}\mathit{v}}$$ 1.1

Where Y^* is the underlying outcome variable (regularity of attendance, aspirations, RTE awareness, and liked attending school). The dummy variable regularity of attending school uses a logistic regression, whereas ordered logistic regressions are used for the aspirations variable that is an ordered variable for the i^th child residing in the h^th household in the v^th study at time t; Post is the variable that indicates whether the data is at the baseline or midline and is the primary measure of program effects; Female is a dummy variable that takes a value of 1 if the child is female and zero otherwise. Scale is a vector of continuous variables that consist of total scores on the perceived self-efficacy, resilience, and gender attitudes scales.6X is a vector of individual, household, and location-level characteristics such as age, sex, class of education, religion, caste of child, and $\in$ is the error term.

In all estimations, we report odds ratios for ease of interpretation. Similar to the baseline, we also report data on changes in the behavioural parameters (self-efficacy, resilience, and gender attitudes)7 as a result of participation in the Life Skills program. To examine sub-group effects, we interacted the program dummy variable with age groups (Less than or equal to 11 years, 12–13 years, and 14 and above) and gender. These results are reported separately.

We round off this analysis by running estimations similar to Eqs. 1 and 1.1 on panel data that had baseline and endline measures, where children were assigned randomly to treatment and control groups for the long-term program and tracked over two time periods. The equations estimated using difference-in-differences (DiD) methods are as below:

$${{\mathit{Y}}^{\mathbf{\ast }}}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}=\alpha +\gamma Pos{t}_{\mathit{ihvt}}+{\beta }_1{\mathit{Treat}}_{\mathit{ihv}}+{\beta }_{2}Trea{t}_{\mathit{ihv}}\times Pos{t}_{\mathit{ihvt}}+{\beta }_3\mathit{S}\mathit{c}\mathit{a}\mathit{l}{\mathit{e}}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}+{\beta }_4{\mathit{X}}_{\mathit{i}\mathit{h}\mathit{v}}+{\mathbf{\in }}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}$$ 2

$$\mathit{S}\mathit{c}\mathit{a}\mathit{l}{\mathit{e}}_{\mathit{i}\mathit{h}\mathit{v}\mathit{t}}=\mathit{\omega }+\kappa Pos{t}_{ihvt}+{\delta }_1{\mathit{Treat}}_{\mathit{ihv}}+{\delta }_{2}Pos{t}_{\mathit{ihvt}}\times Trea{t}_{\mathit{ihv}}+{\delta }_3{\mathit{X}}_{\mathit{i}\mathit{h}\mathit{v}}+{\mathit{\eta }}_{\mathit{i}\mathit{h}\mathit{v}}$$ 2.1

Where, ${\mathit{Treat}}_{\mathit{ihvs}}$ takes a value of 1 if the child was in the treatment group and zero otherwise, and Post _ihvst takes a value of 1 if the data was recorded at endline, and 0 if it was at the baseline. The remainder of the covariates are the same as those described following 1 and 1.1, except that we also include data on religion. These are estimated using panel OLS techniques with standard errors clustered at the household level. Equation (2) estimates the impact of the treatment on outcomes operating over and above the impact changes in socioemotional scores rather than the total effect of the treatment.

Additional outcome variables of interest in attending school and participation in class (“Do you like to go to school?”; “Do you participate in class?”) are also included in a logistic regression framework. These are converted from categorical variables (that take values of always, sometimes, or never) into binary variables (takes a value of 1 if response is sometimes or always, and zero otherwise). The regularity of attending school and liked attending8 school variables were recoded to binary variables (see Table 1).

Results and discussion

Baseline results

We first present results from the baseline only (without the variable Post) to draw associations between socio-emotional skills, school-related outcomes, and individual and household characteristics. The results can be found in the appendix, Tables A.2 to A.4. We find no statistically significant associations between scores on socio-emotional skills and gender (Table A.2), with the notable exception of current grade positively being associated with higher scores on the resilience and egalitarian gender attitudes scale, particularly for girls having a better problem-solving ability. Table A.3 presents the results of the ordered logistic regression of child characteristics, behavioural parameters, and other socio-demographics on regularity in attending school. At the baseline, girls have lower odds of attending school regularly, with stepwise models (except the last) indicating a nearly 75% reduction in regular attendance at the baseline for girls in the sample. Second, being in a higher grade (class) is associated with having higher odds of regularly attending school. In the full model (column 4), a change in one year of schooling (e.g., going from 5th class to 6th class) translates into a 26% increase in the odds of regularly attending school. The results on caste grouping show significant disadvantage for SCs (log-odds are 40% lower), and STs (log-odds are lower by 66%). When we disaggregate the estimates by gender (results available on request), we find that compared to girls from the SC community, boys from the same community have greater odds of liking to attend school. Other notable intersections with caste identity suggest that girls belonging to the OBC community report significantly lower scores on the egalitarian gender attitudes scale compared to OBC boys. Finally, Table A.4 shows that having a higher score on resilience was associated with a 32% reduction in the odds of regularly attending school at the baseline. Taken together, these suggest that (a) socio-emotional skills are indeed closely linked with other school-related outcomes; and (b) these associations are likely heterogeneous along the lines of age and gender of the adolescents.

Short-term program effects

Table 4 contains the results of the consolidated baseline and midline data from the above-mentioned studies on school-related outcomes (columns 1–3) from an OLS estimation of Eq. 1. Since program uptake may not have been randomly assigned, we cannot directly infer the program effects (life skills intervention) on the intended outcome variables from the coefficients, but instead infer statistical associations. Additional results from sub-group analyses are reported alongside the main model (columns 4–6).

Table 4.

Short-term program effects panel regression on school outcomes

VARIABLES	(1)	(2)	(3)	(4)	(5)	(6)
	Aspiration to study	Attend school regularly	Awareness of RTE	Aspiration to study till graduation	Attend school regularly	Awareness of RTE
Program Effect	2.120***	0.930	21.88***	3.313***	0.857	26.23***
	(0.159)	(0.109)	(2.327)	(0.740)	(0.206)	(5.595)
Age (in years)	0.992	0.921	1.242***
	(0.0365)	(0.0530)	(0.0606)
Age group 2 (12-13yrs)				1.189	0.940	1.370***
				(0.220)	(0.159)	(0.155)
Age group 3 (>13yrs)				0.892	1.032	2.105***
				(0.332)	(0.351)	(0.440)
Program X Age group 2 (12-13yrs)				0.873	0.849	0.909
				(0.195)	(0.209)	(0.206)
Program X Age group 3 (>13yrs)				1.368	0.643	0.684
				(0.563)	(0.270)	(0.248)
Female	0.933	0.900	1.112	1.076	0.716**	1.164
	(0.0673)	(0.0985)	(0.0985)	(0.180)	(0.113)	(0.118)
Program X Female				0.972	1.548**	0.835
				(0.190)	(0.333)	(0.166)
Current Grade	0.973	1.082	0.971	0.962	1.080	0.988
	(0.0357)	(0.0634)	(0.0476)	(0.0475)	(0.0631)	(0.0479)
Gender attitudes score	1.038***	1.045***	1.054***	1.039***	1.044***	1.055***
	(0.00674)	(0.0113)	(0.00972)	(0.00882)	(0.0113)	(0.00976)
Perceived self-efficacy score	0.993	1.078***	1.020**	0.994	1.078***	1.020**
	(0.00668)	(0.0112)	(0.00829)	(0.00946)	(0.0113)	(0.00829)
Resilience score	1.050***	0.956**	0.996	1.083***	0.957**	0.996
	(0.0139)	(0.0207)	(0.0174)	(0.0219)	(0.0208)	(0.0174)
Other Backward Caste	1.636***	0.856	1.476***	1.661***	0.852	1.496***
	(0.140)	(0.117)	(0.158)	(0.192)	(0.117)	(0.161)
Scheduled Caste	1.329***	0.716**	1.278**	1.567***	0.713**	1.280**
	(0.118)	(0.0983)	(0.143)	(0.191)	(0.0983)	(0.143)
Scheduled Tribe	1.060	0.984	0.920	1.314	0.998	0.915
	(0.150)	(0.225)	(0.154)	(0.263)	(0.229)	(0.153)
Constant		1,203***	0.00535***	0.0113***	521.5***	0.0531***
		(1,774)	(0.00496)	(0.00977)	(709.8)	(0.0418)
Mean at baseline (proportion)				0.106	0.889	0.389
Observations	4,196	4,196	4,152	4,186	4,196	4,152

Column 1 is an ordered logit regression, column 4 is a logistic regression on a binary dependent variable (whether the child responded that they wish to study till graduation). Odds ratios and standard errors of odds ratios reported in parentheses. All estimates include study fixed effects. *** p<0.01, ** p<0.05, * p<0.1

First, we find a robust statistically significant and positive association between the program and increased aspiration to study (column 1), as well as a three-fold associated increase in the odds of studying till graduation after the intervention. There are very large and statistically significant impacts on awareness of the RTE program as well that accrue after the program. Relative to the youngest age group (less than or equal to 11 years), we do not find that this impact varies by age group. However, the impact varies by gender: a girl that participated in the life skills intervention has a 55% increase in the odds of regularly attending school and this finding is in line with that of Edmonds et al., [7] where, after two years of the life-skills intervention ‘Girls’ Education Programme (GEP)’, a 30% reduction in dropout among adolescent girls was documented. This is particularly encouraging since at the baseline, being a girl child was associated with lower odds of regularly attending school.9

Having more egalitarian gender attitudes (potentially on account of the life skills program) is associated with an increase in the odds of aspiring to study further (3.9%), regularly attending school (4.4%), and awareness of the RTE (5.5%). Similarly, a higher self-efficacy score was linked to a 7.8% increase in the odds of regularly attending school, and 2% increase in the odds of being aware of the RTE policy. These findings are corroborated by Roy et al., [20] where they establish a strong relationship between self-efficacy and the number of years adolescent girls desire to study. Having a higher resilience score increased the odds of aspiring to study further by 5%, but was associated with a reduction in the odds of regularly attending school by 4.3%.

Table 5 reports the key results of program impacts on socio-emotional skills in the short term as per Eq. 1.1. These are measured by the scale variables on egalitarian gender attitudes, perceived self-efficacy and resilience. In general, we find a positive and statistically significant impact of the program on improved and egalitarian gender attitudes (by 0.6 points approximately) as well as perceived self-efficacy (by approximately 4 points). The strongest increase as a result of the program comes in the oldest age groups for egalitarian gender attitudes improvements; whereas it appears to originate from improvements in perceived self-efficacy among the youngest age group who participated in the program. This implied that older children (14 years or above) had an increase in their scores of gender attitudes (by more than 1 unit), and 0.36 unit increase in the resilience score. To disentangle these effects further, we report the difference in the coefficients to derive the “pure” programme effect within this age group. The strongest effect remains for improvements in gender attitudes score for the older age group, and a smaller reduction of the perceived self-efficacy score for those in the oldest age group. This suggests that changes in gender attitudes can be attributed to the programme, it is more likely that reduced self-efficacy is not purely on account of developmental changes. Girl children in particular develop more egalitarian gender attitudes after having gone through the programme, and this is associated with a 1.5 unit increase on average in their gender attitudes score. This finding is in line with the findings made by Edmonds et al., [7] from the GEP that treatment girls display more positive gender norms.

Table 5.

Program effects on socio-emotional skills and sub-group analyses

	(1)	(2)	(3)	(1)	(2)	(3)
VARIABLES	Gender attitudes score	Resilience score	Perceived self-efficacy score	Gender attitudes score	Resilience score	Perceived self-efficacy score
Program Effect (γ)	0.571***	0.0124	3.928***	−0.497	0.133	4.447***
	(0.162)	(0.0814)	(0.168)	(0.347)	(0.175)	(0.359)
Age (in years)	0.106	0.129***	0.0350
	(0.0906)	(0.0416)	(0.0853)
Age group 2 (12-13yrs) (δ1)				0.142	0.377***	0.558*
				(0.233)	(0.137)	(0.299)
Age (>13yrs) (δ2)				−0.456	0.234	1.000*
				(0.420)	(0.253)	(0.555)
Program X Age (12-13yrs) – (β1)				0.274	−0.0898	−0.722*
				(0.370)	(0.181)	(0.379)
Program X Age (>13yrs) - (β2)				1.478**	0.128	−1.588**
				(0.581)	(0.303)	(0.633)
Female	2.547***	−0.0427	−0.172	1.686***	0.0456	−0.238
	(0.166)	(0.0783)	(0.160)	(0.210)	(0.120)	(0.271)
Program X Female				1.501***	−0.147	0.134
				(0.312)	(0.160)	(0.329)
Current Grade	−0.128	−0.0748	−0.124	−0.118	−0.0728	−0.128
	(0.0893)	(0.0465)	(0.0860)	(0.0885)	(0.0463)	(0.0853)
Other Backward Caste	0.285	−0.000930	−0.626***	0.239	0.000825	−0.606***
	(0.206)	(0.0986)	(0.196)	(0.206)	(0.0982)	(0.196)
Other Caste	−2.457***	0.213	−5.011**	−2.652***	0.211	−4.901**
	(0.832)	(0.791)	(2.229)	(0.847)	(0.783)	(2.201)
Scheduled Caste	−0.507**	0.0249	−0.395*	−0.523**	0.0195	−0.394*
	(0.217)	(0.0976)	(0.210)	(0.217)	(0.0976)	(0.210)
Scheduled Tribe	−0.260	−0.132	−0.941***	−0.243	−0.145	−0.955***
	(0.330)	(0.179)	(0.365)	(0.329)	(0.179)	(0.365)
Constant	18.66***	32.48***	24.45***	20.20***	33.76***	24.59***
	(1.219)	(0.563)	(1.123)	(0.791)	(0.366)	(0.711)
Program Effect $\left({\beta }_1,-,{\delta }_1\right)$				0.132	−0.467	−1.280**
Program Effect $\left({\beta }_2,-,{\delta }_2\right)$				1.934**	−0.105	−2.588**
Mean at baseline	17.37	33.04	19.45	17.37	33.04	19.45
Observations	4,208	4,196	4,208	4,208	4,196	4,208

Coefficients from panel OLS estimates reported for short-term impacts. Standard errors of coefficients reported in parentheses. All estimates include study fixed effects. *** p<0.01, ** p<0.05, * p<0.1

Long-term program effects

We now present findings from the long-term program impacts estimated as per Eqs. 2 and 2.1. There is some difference in the outcome variable, since data on active class participation and RTE were not available. We thus restrict our focus to the aspiration to study further and regularity of attending school as outcome variables as well as changes in socio-emotional skills as measured by the three behavioural parameters (egalitarian gender attitudes, perceived self-efficacy, and resilience).

All results can be found in Table 6. We find a strong and positive statistically significant impact of the program at the endline relative to those who did not attend the program on all life skills parameters (Fig. 1). The largest increase comes for the perceived self-efficacy measure, with an average increase of 2.5 points. The resilience score increased by 1.1 points on average, and the gender attitudes score improved by 0.45 points. In terms of school-related outcomes, we also see a large and statistically significant increase in the odds of regularly attending school by 66.5% points, and an increase of 27.1% points in the odds of aspiring to study further (Fig. 2). Although the results are not driven by females for aspirations, the long-term program effect is particularly strong for girls who attended the program at the endline: they more than double their odds of regularly attending school. Similar to findings from other studies, there was a strong positive association between having more egalitarian gender attitudes and likelihood of attending school regularly (87.4% point increase in odds) and aspiring to study further (15% point increase in odds). Similarly, an increase in resilience scores was associated with a nearly 50% increase in the odds of regularly attending school on average, but a sharp reduction in the aspiration to study further. An increase in the perceived self-efficacy score was also associated with a 47% point increase in the odds of wanting to study further, but not significantly with regularly attending school. These findings are in line with the United Kingdom’s Social and Emotional Aspects of Learning (SEAL) program, which found that attendance increased for both, the staff and students [25].

Table 6.

Long-term program effects on school outcomes and socio-emotional skills

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Gender attitudes score	Perceived self-efficacy score	Resilience score	Regularly attend school	Aspiration to study to graduation
Post	−0.710***	0.810***	17.58***	0.254***	2.924***
	(0.138)	(0.162)	(0.0965)	(0.0862)	(0.403)
Treated	0.0533	0.0336	0.0288	0.932	0.833***
	(0.117)	(0.147)	(0.0801)	(0.184)	(0.0555)
Post x Treated	0.450**	2.496***	1.089***	1.665**	1.271**
	(0.185)	(0.212)	(0.120)	(0.430)	(0.129)
Perceived self-efficacy				0.966	1.467***
				(0.0529)	(0.0295)
Resilience				1.496***	0.731***
				(0.228)	(0.0452)
Gender attitudes				1.874***	1.150***
				(0.101)	(0.0211)
Post x Female				0.676	1.089
				(0.201)	(0.118)
Treated x Female				0.571*	1.192*
				(0.185)	(0.119)
Post x Treated x Female				2.320**	0.952
				(0.960)	(0.141)
Observations	10,412	10,412	10,413	10,232	10,412

Columns 1–3 report normalized scores on gender attitudes, resilience, and perceived self-efficacy and are estimated using panel OLS, whereas 4 and 5 report ordered variables on regularity of attending school and aspiration to study further and are estimated using ordered logistic regression techniques, reporting ordered risk ratios. All estimations included additional controls on age of the child, squared age, gender of the child, the class in which they are currently studying, religion, and caste grouping. All regressions use time and state fixed effects. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Fig. 1.

Coefficient plot for program impacts on socio-emotional skills

Note: Plot depicts point estimates of coefficients and 95% confidence intervals from separate panel ordinary least squares regressions of the program and timing of survey on socio-emotional skills separately for boys and girls. All regression specifications contain additional covariates on age (in years), squared age, caste grouping, religion, current grade. State (sub-national) fixed effects included in all estimates. Heterogenous-robust standard errors estimated

Fig. 2.

Coefficient plot for program impacts on educational outcomes

Note: Plot depicts point estimates of coefficients and 95% confidence intervals from separate panel ordinary least squares regressions of the program and timing of survey on school-related outcomes separately for boys and girls. All regression specifications contain additional covariates on age (in years), squared age, caste grouping, religion, current grade. Estimations also contain additional controls of the socio-emotional skills – perceived self-efficacy, resilience, and gender attitudes. State (sub-national) fixed effects included in all estimates. Heterogenous-robust standard errors estimated

Conclusions

This research presents substantial evidence that the Magic Bus India Foundation’s program on developing life skills among adolescents has important effects for children’s school-related outcomes as well as in building important socio-emotional skills. At the baseline itself, we found that socio-emotional skills (especially resilience) are important for enabling regular school attendance. If a child (especially a female child) scored higher on resilience, then she was 25.8% more likely to be regularly attending school. Thus, targeting the development and improvement of such socio-emotional skills is an important aspect of the program. Similarly, we found that having good problem-solving skills improves the aspirations to study further as well as class participation. Since there is not much India-specific research on the association between these variables and school-related outcomes, this indicates that such evidence may be very relevant in terms of overall education policy as well.

We presented findings from short-term and long-term analysis of the program across studies. In the short-term (i.e., between baseline and midline, typically 18 months duration), we find robust effects of participating in the intervention on increased aspiration to study further and awareness of educational policies such as the Right to Education Act. Although these program effects do not appear to vary by age, we find that girl children that participate in the program increase their regularity of attending school. As mentioned previously, the development of socio-emotional skills is critical to school-related outcomes, and this finding is borne out in the analyses. Having more egalitarian gender attitudes and improved perceived self-efficacy results in more regular school attendance, an aspiration to study further, and greater awareness of educational policy in India. Notably, a higher resilience score is associated with a reduced regularity of attendance at school, highlighting that there are pathways through resilience that we do not yet fully understand.

The program also has significant effects on socio-emotional skills; participation was associated with increased egalitarian gender attitudes and perceived self-efficacy. In examining these effects by different groups, we found that improvements in gender attitudes and resilience are particularly strong and positive for older children in the same cohort (> 11 years of age). The largest increase in egalitarian gender attitudes comes for female children who participated in the program, resulting in an 8.2% increase in their scores. These effects are further ameliorated in the longer-term, with endline results suggesting that the program boosted egalitarian gender attitudes and self-efficacy substantially. With a larger sample (when endline data are available), this finding can be made more robust to inform future interventions. However, it is important to note that we cannot generalize the impact of SEL training on school-related outcomes from this study, since it does not usually nationally representative data or consider the same sample for long and short-term results.

Lastly, the findings offer some important implications for India’s education system with the implementation of the new National Education Policy (NEP) by the Government of India in 2020. The NEP aims to provide a more holistic approach to education in India, with a focus on improving the development of life skills among adolescents. Our findings suggest that developing SEL has significant links with other school-related outcomes that are broadly of policy interest (e.g., attendance, especially among girls). India can also do well to adapt SEL training material from other countries such as Colombia [26]. Honduras, Argentina, Chile, and Mexico, who have constituted training programs for teachers to develop awareness around SEL [27]. In a socio-culturally diverse country such as India, adapting such material to local contexts may be critical to cementing the links between SEL and school-related outcomes.

Authors' contributions

AT contributed to study design and conceptualization, formal analysis, report writing, editing, and funding acquisition. HK contributed to study design and conceptualization, report editing and funding acquisition. AB contributed to report writing and formal analysis.

Funding

This study was funded by a research grant from the Magic Bus India Foundation (MB#22).

Data availability

The data that support the findings of this study were made available by the Magic Bus India Foundation, Mumbai. Restrictions apply to the availability of these data, which were used under agreement for this study.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the guidelines prescribed by the Indian Council for Medical Research for Biomedical and Health Research, and the Belmont Report. This study received ethics approval from the Monk Prayogshala Institutional Review Board with reference #045 − 020 in July 2020. All participants provided consent to participate to the Magic Bus India Foundation. Informed written consent to participate was obtained from the parents or legal guardians for all participants under the age of 18 and all minors provided oral assent to participate to the Magic Bus India Foundation. No identifying information (names, addresses, phone numbers) was received in the data that was shared by Magic Bus India Foundation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.