Inequality in the Labor Market: Lower Perceived Returns Among - - PDF document

▶

Feb 27, 2024 264 likes •629 views

Inequality in the Labor Market: Lower Perceived Returns Among Marginalized Youths and Girls Manjistha Banerji and Ashwini Deshpande 1 To be presented at XXVIII IUSSP International Population Conference in Session Children, Youth and the Labor

SLIDE 1

Inequality in the Labor Market: Lower Perceived Returns Among Marginalized Youths and Girls

Manjistha Banerji and Ashwini Deshpande1 To be presented at XXVIII IUSSP International Population Conference in Session “Children, Youth and the Labor Market” On Friday 3rd November, 2017

1The authors would like to thank Dr. WilimaWadhwa, Director, ASER Centre for her helpful

comments and suggestions. Usual disclaimers apply.

SLIDE 2

Abstract

In this paper, rather than examining actual wages in the labor market, we look at perceived returns in the labor market among adolescents and their parents. The rationale for carrying out this examination is two folds- previous research has established that lower subjective expectations of labor market returns among parents affects educational investment. Likewise, subjective expectations of children about labor market returns are likely to affect their commitment to their education in terms of effort and time spent. Gender and caste have long been the major axis of discrimination in India. In the labor market, it manifests itself in terms of lower wages for women and persons from marginalized communities. We, therefore, ask if perceived labor market returns among adolescents and their parents vary by caste and

gender. We use a unique dataset on adolescents that has been recently collected (2013) by

ASER Centre, the research and assessment wing of Pratham Education Foundation for our

analysis. Results confirm that girls have lower expected earnings than boys. Caste differences

appear more rigid in Bihar.

SLIDE 3

Section I Introduction Gender and caste have long been the major axis of discrimination in India. In the labor market, it manifests itself in terms of lower wages for women and persons from marginalized castes/tribal communities. In this paper, rather than examining actual wages in the labor market, we examine if perceived labor market returns vary by gender and caste among adolescents and their parents. The rationale for carrying out this examination is two folds- previous research (Maertens, 2010) has established that lower subjective expectations of labor market returns among parents affects educational investment. The same rationale suggests that subjective expectations of children about labor market returns are likely to affect their commitment to education in terms of effort and time spent. In other words, lower subjective expectations among parents and children are likely to translate into lower educational investment and vice-

versa. Furthermore, answers to the question has policy implications. If adolescents of

identical intrinsic ability have different expectations of labor market returns that is based on their gender and caste as do their parents, a vigorous enforcement of civil rights and affirmative action is warranted. On the other hand, if differences in labor market expectations among adolescents are solely due to their intrinsic abilities, then policy recommendation would be to focus on fostering skills through skill development programs (Carneiro, Heckman, & Masterov, 2004). Explorations about expected returns along lines of gender and caste adds another layer to our understanding of systemic biases in the functioning of the labor market and adolescents in India. The paper uses cognition as a measure of an adolescent’s intrinsic

ability. Therein lies the uniqueness of the paper. It brings into the discussion on expected

earnings test scores as a measure of an adolescent’s cognitive ability. It is also unique in that it focuses on adolescents in the age group of 11- 16 years who are likely to join the labor force in few years. Previous discussion of subjective expectations in India did not include any measure to capture cognitive ability and did not focus exclusively on adolescents. To set up the analytical context, Section II surveys the literature on cognitive skills as a predictor of labor market outcomes. It then surveys the state of education keeping in focus the age group under consideration and labor market returns in India. Section III presents the research questions and hypotheses, discusses the data and methods. Results are analyzed in Section IV. Section V concludes.

SLIDE 4

Section II Cognitive skills as a predictor of labor market outcomes Higher cognitive ability is systematically correlated with individual preferences and choices that favour economic success (Burks et al, 2009). Persons with higher test scores have higher earnings (Attanasio and Kaufmann, 2010). Better literacy and numeracy skills increase the likelihood of positive labor market outcomes- wages and employment (Chesters, Ryan and Sinning, 2013)-because basic skills acquired in early childhood and school years, particularly literacy and numeracy, are the necessary foundation for developing higher order skills that contribute to a productive workforce (Treasurer of the Commonwealth of Australia, 2010). Lee and Newhouse (2008) utilize data for up to 315 tested cohorts in 67 countries to establish that both education quality, as measured by performance on international assessments and average educational attainment are important determinants of youth outcomes in labor markets. Higher scores in tests like PISA and TIMSS also correlate with a larger share of youth working in wage and salaried employment, outside the agricultural sector and to some extent in higher status occupation. The effects of test scores for most outcomes continue to hold true even after keeping educational attainment constant

r when the focus is on low and middle income countries; suggesting that the correlations

between test scores and youth employment outcomes are not driven solely by differences in educational attainment, or broad contrasts between the labor markets of high-income and low-income countries. More broadly, cognitive skills of the population are powerfully related to individual earnings, distribution of income, and economic growth (Hanushek and Woessmann, 2008). Further, the earlier in childhood children develop cognitive and social skills, the better are the long-term impacts on their skills and labor market outcomes (Rose, 2005). Data from the National Education Longitudinal Study of 1988 indicates that students who made relatively large test score gains during high school had larger earnings 7 years after high school compared to students whose scores improved little. In other words, test score gains during high school predict subsequent employment status and earnings. There is, however, a gender differential- test score gains increase female earnings by both increasing the likelihood of employment and by increasing earnings once employed. For men, on the other hand, test score gains are not significantly related to employment status or earnings, except for those men who have low initial test scores. It follows from the above discussion then that

SLIDE 5

given the association between cognitive skills and labor market outcomes, adolescents and their parents are likely to factor it in their labor market expectations. Universal enrolment, but “low” quality is an issue The most notable achievement in recent years in the provision of education in India is near universal enrolment as indicated in the ASER surveys over the years (2005- 2016). However, there are numerous quality concerns- poor attendance and high drop out among adolescents, and low learning levels, and that these outcomes vary by background characteristics as gender and caste. In the context of the paper, these deficiencies of the educational system highlight that it is not a system geared to maximizing the educational potentialities of its students, particularly girls and children from marginalized castes. Unlike in developed countries where enrolment is synonymous with attendance, in rural India attendance is far from universal (Deshpande and Banerji, 2017; Bhattacharjea, Banerji and Wadhwa, 2011). Nationally, about 25% children were found to be absent when a team as part of the annual ASER (Annual Status of Education Report) 2016 survey visited the village

school. Data from Wave I of IHDS confirm the findings (Desai et al, 2010). Nationally, 20%
f children were absent for 6+ days in the month preceding the survey, with as many as 26%

children in “less developed villages” reporting an absenteeism in the above reporting period. Second, drop out remains a huge concern, particularly in the adolescent age group and among girls. As per ASER 2016, the percent drop out in the age group 15- 16 years is 13.5% and in the age group 11- 14 years is 3.5%. Discontinuation rates computed for men and women (17+) using IHDS- Wave I suggests that around 50% men dropped out between grades 5 & 10. The corresponding rate for women is 57%. Likewise, discontinuation rates are highest among Adivasis (Scheduled Tribes), followed by Dalits (Scheduled Castes) and lowest among high caste Hindus. There are regional differences- states as Madhya Pradesh, Chhattisgarh and Rajasthan have higher dropouts than other states. Beyond attendance and dropouts, the most important concern that has occupied center stage in the discussion on quality of education in India is poor learning outcomes as highlighted in the annual ASER surveys. As per ASER 2016, only 26% of children in Class V and 43% of children in Class VIII could solve a three- by one digit division (with a remainder) problem. Comparisons over time suggest that learning levels have deteriorated or at best have remained stable. In the above example, percent of Class V children who could solve the division problem was 36.2% in 2010 and it declined to 26.1% in 2014. The corresponding percentages for Class VIII are 68.4% and 43.3% respectively.

SLIDE 6

ASER reports do not give learning levels disaggregated by gender and other background characteristics, but data from India Human Development Survey (2005) which used similar learning tools as ASER surveys suggest that social background characteristics as caste and income levels can act as barriers in educational attainment. About 45% of children belonging to lowest quintile could read a small paragraph, for children from top quintile the corresponding percent is 73%. In terms of social groups, the percentages are as follows- about 45% of Scheduled Castes and Scheduled Tribes as compared to 71% of high caste

Hindus. Just as with ASER surveys, IHDS too finds that children lag in their arithmetic as

compared to reading skills. Differences among SES groups in arithmetic levels parallel differences in reading levels. Gender differences in learning levels are not huge- about 56% boys and 52% girls could read and about 51% boys and 45% girls could subtract. Multivariate analysis, however, confirms gender gap in reading and arithmetic skills among children in ages 8- 11 years (White, Ruther and Kahn, 2015). Household work, particularly care of younger sibling, place girls at a disadvantage than boys in a similar situation with younger sibling. Gender and affluence interact- girls belonging to affluent households have better learning outcomes vis-à-vis girls from poorer households. From the perspective of the analysis presented in this paper, we note that in addition to their cognitive ability, learning levels of children are likely to systematically differ by their gender and background SES. Labor market differentials by caste and gender Caste and gender being the main axis of social differential, it is only expected that participation trends and outcomes in the labor market will vary along these lines as well. Caste Caste differentials in the labor market is visible in the first order itself in types of

employment. Munshi and Rosenzweig (2006) have shown that caste based labor markets,

which persist even today, operate in a manner such that individuals are engaged in the same

ccupations for generations. IHDS I (Desai et al, 2010) documents that the likelihood a

person is in farming or engages in agricultural/ non- agricultural daily labor or business or has salaried employment varies by caste affiliations. Persons from marginalized SC and ST communities are most likely to engage in daily agricultural labor while Muslims are more likely to be in non- agricultural labor as compared to other social groups. Salaried employment is more common among Christians and other religious minorities as well as

SLIDE 7

forward castes vis-à-vis other groups. Among women too, there is a differential- for example, women from SC and ST communities are more likely to be engaged in agricultural labor than women from other backgrounds. IHDS I also indicate differences in wage rate by caste; though here too it varies by the employment sector in which a person is engaged. There is not much variation by social group with respect to agricultural wages, but there are variations in non- agricultural wages by social background- Scheduled Castes and Scheduled Tribes have lower wages than others. It also emerges from IHDS I that returns to education are not visible till secondary education for both men and women. Salary discriminations among persons in the private sector is also apparent- persons from “upper” castes have higher salary than those from marginalized castes and Muslims. Interestingly, salary differentiation by social background is far less in the public sector. Further, there is some evidence of discriminatory hiring practices in the private sector. Thorat and Attewell (2007) document that marginalized groups are discriminated in the private sector at the first stage of the job application process- as compared to an “upper” caste applicant, a Dalit and Muslim have lower chances of being called for an interview. Furthermore, the focus on the mean wage gap hides that there are gaps at various points of the wage distribution (Duraisamy and Duraisamy, 2017). Second, combining social groups into broad categories (for example, combining SCs and STs as one group and comparing it with non- SCs/ STs) also underestimates the wage gap. Analysis focusing exclusively on regular/ salaried workers and using national quinquennial employment/ unemployment surveys (NSS) for the years 1983, 1993- 94, 2004- 05 and 2011- 12 finds that the difference between social groups is smallest among low wage earners (that is, those earning below the mean), but it widens steadily as one moves from the median to the top of the distribution. The gap is highest for Muslims, followed by SCs and STs. Decomposition analysis further reveals that part of the wage gap is indeed due to caste based discrimination. Most importantly, from the point of this paper, discriminations in the labor market is likely to have spill- over effects in terms of educational investment. Investment in education is not a function of average costs and benefits alone; perceived costs and returns to education are important as well (Maertens, 2011; Jensen, 2010; Nguyen, 2008). Based on parental responses to a question on estimated earnings conditional on a child’s background characteristics as gender, caste and perceived ability but unconditional on type of work in three villages in Telangana and south- west Maharashtra, Maertens (2011) constructed a density function of monthly earnings. The results indicate that parents of SC/ ST children

SLIDE 8

consistently underestimate the earnings of their children compared to UC/ OBC parents. Lower estimated earnings in the future may discourage parents from investing in their child’s education. Gender While women in aggregate have lower workforce participation than men, there are two peculiar features of labor force participation of women in India- one: there has been a consistent declining trend in women’s workforce participation since 1983 (Siddiqui, Lahiri- Dutt, Lockie and Pritchard, 2017) and two: it is inversely associated with education, income and caste based affiliations (Das and Desai, 2003; Desai, 2010; Desai et al, 2010). In 1999- 2000, women’s work participation in the age 15- 59 years was around 40% and in 2011- 12, it declined further to 32% (Rawal and Saha, 2015). Second, less educated women are more likely to be in the labor force as compared to women who are highly educated. Likewise, women belonging to less well- off households are more likely to be in the labor force than women from better off households and women from marginalized castes are more likely to be in the work force than women from privileged caste groups or other minority religious

groups. In comparison to women in least developed villages, fewer women from metro cities

are in the labor force. Women’s participation also tends to be underestimated because it is concentrated in the agricultural sector, particularly in livestock care. Women’s participation in wage work is for a fewer days as compared to men and with a significant wage gap. Probit estimates of labor force participation using unit level NSS data confirms that as compared to men with similar background characteristics participation rate of women is significantly lower (Sengupta and Das, 2014). Gender wage discrimination as documented in IHDS- I suggests that it varies by residence (rural or urban) and by type of employment. It is lowest in urban areas and in wage

r salaried employment in the public sector. Urban women in wage or salaried employment

earn only 68 paise per rupee earned by men while rural women earn about 58 paisa less than

men. It is highest for those engaged in daily wage employment- whether agricultural or non-
agricultural. Daily wage for agricultural labor for men is INR 50 and for women is INR 33.

The corresponding daily wage for non- agricultural labor for men is INR 76 and INR 43. NSS data confirms that education does not bridge the gender wage gap. Nominal mean daily wages from NSS 50th (1993- 94), 55th (1999- 2000), 61st (2004- 05) and 66th (2009- 10) rounds of data indicates a gender wage gap at all levels of education (illiterate, literate up to

SLIDE 9

middle, secondary and higher secondary, graduate and above) and in both rural and urban

areas. Further comparisons of relative wage gap across the NSS years indicate that it

increased for all educational groups in rural areas. While it has declined in urban areas over the period 1993- 2010 (with exception of the group secondary and higher secondary), it is not

trivial. In both rural and urban areas, the wage gap is higher among women who have no or

low levels of education and are in the informal sector. In the discussion on labor market differentials by caste in the previous section, we highlighted that employment types among women varies by caste. It, therefore, needs to be asked if the gender wage gap is worse for women from disadvantaged castes and religious

minorities. NSS data suggests that while both Hindu and Muslim women experience a gender

wage gap, it is worse among Muslim women- though the gap has declined over time. STs earned a better wage, while the wage among SC women was only marginally low. Nominal and relative gender wage gap has persisted over time; though calculations of rate of return to education using NSSO data do not indicate similar consistent trends. In the early 1990s, there was no significant difference in the rate of return to education among men and women with primary education. But in 2009- 10, the return for primary education was higher for women than men. In contrast, while return to higher education was higher for women than men in the early 1990s, it declined over the years irrespective of religious/ social group. These gender discriminatory features of the labor market- lower WPR and wage gap- are in turn likely to influence the expectations of adolescent girls and parents of young

daughters. Perceived differences in returns to education between men and women may

motivate parents to differentially invest in the education of their children (Maertens, 2011). Likewise, girls may have less inclination to pursue with their studies if labor market along with social customs (Field and Ambrus, 2008; Foster and Rosenzweig, 2001) are indicative

f lower returns on their education vis-à-vis boys.

Section III Research Questions and Hypothesis The review of the literature suggests that girls and children from marginalized communities are not only disadvantaged in the labor market by their overall lower educational outcomes but market discriminations as well. Thus, the research question posed in the paper is as follows:

SLIDE 10

Is it the case that background characteristics as gender and caste influence an adolescent’s expected labor market returns over and above his or her intrinsic ability as measured by his/ her learning levels? Rational expectations of labor market returns suggest that the same factors that predict actual earnings also predict expectations, conditional on them being in the information set of the individual. Thus, labor market practices systematically discriminating by caste and gender is likely to mean that adolescents from these communities and adolescent girls have lower expected earnings than their peers from privileged communities and adolescent boys respectively. Data and Methods To answer the above research questions, we use a recently collected data (2013- 2015) for a study on access to and quality of middle schooling (henceforth, the Middle School Study) by ASER Centre, the research and assessment wing of Pratham Education Foundation2. The Middle School Study data focused on children in grades 6 to 8 and in the age group 11- 16 years (N= 6,194). The data was collected in three phases- baseline (October 2013- February 2014), midline (July- September 2014) and end line (October 2014- February 2015) and in two districts- Nalanda in Bihar and Satara in Maharashtra (ASER Centre, 2014). The states were selected purposively- while Maharashtra is educationally and economically advanced, Bihar is one of the poorest states in India (Table 1); within the states the choice of districts was based on availability of Pratham Education Foundation’s logistical support to conduct the survey. [Table 1: District profile about here] A two- stage sampling strategy was followed to get a representative sample of 3,336 adolescents in Nalanda and 2,858 adolescents in Satara spread across 60 villages in each of the study districts. The study included both currently enrolled and out-of-school children (drop out and never enrolled). The analytical focus of this paper is on currently enrolled adolescents (N=5,754). The survey collected data on adolescent’s individual and household characteristics, home learning environment and ways in which they have access to information. Most

2Pratham Education Foundation is the largest non- government organization working on education in India.

SLIDE 11

importantly, for the purposes of the paper, sampled adolescents and their parents were asked about their expected earning when they are in the labor market at age 25. The specific question to adolescents was as follows3: “When you are 25 years old, how much you think you will be earning in a month?” Over 80 percent of adolescents in both the districts gave an estimate of expected earnings at age 25. The remaining fifth either responded ‘Don’t Know’ or did not report

anything. Given that not all respondents were able to given an estimate of expected earnings,

the first step in our analysis is whether there are systematic differences by gender and background household characteristics in adolescents reporting or not reporting expected

earnings4. That is,

R1: Is it the case that girls are more likely to not report estimated earnings than boys? Likewise, is it the case that adolescents and parents from marginalized communities are more likely to not report expected earnings as compared to their peers from affluent households? H1: We hypothesize that girls are less likely to report expected earnings than boys as are adolescents from marginalized castes vis-à-vis their peers from general castes. The outcome variable- whether the adolescent knows his or her expected earnings- takes a value ‘1’ if an expected earning is reported and ‘0’ otherwise. Given the binary nature

f outcome variable, we use logistic model to answer the above-mentioned questions (Long,

1997). The structural equation for a binary logit model is: yi = a + β1 * X1i + β2 * X2i + εi where, X1 is a vector of background individual and household characteristics (gender, caste and affluence) and X2 is a vector of all other control variables that possibly determines adolescent’s knowledge of his/her expected labor market returns. This includes the skill level associated with the adolescent’s desired occupation, highest education among adults in the extended family, social networks, access to information and distance from the nearest higher educational facility.

3Parents were asked a corresponding question on what they expect the (sampled) adolescent to earn when (s)he

turns 25 years.

4Questions on social networks and access to information were asked only to adolescents. Responses to these

questions are utilized in the subsequent regression analysis. Hence, parents have been excluded from this analysis.

SLIDE 12

Households are divided into one of the following categories- Scheduled Castes/ Scheduled Tribes, Other Backward Castes and general castes. Historically, scheduled castes/ tribes have been marginalized and general castes are most privileged. In Bihar, there is the additional category of Extremely Backward Castes. This is the most marginalized caste in

Nalanda. The corresponding category in Satara is SC/ ST.

In addition to caste, we control for household affluence in the regression model. This is done because there is considerable overlap between household affluence and caste background- Dalits are more likely to be least well off and vice versa (Desai et al, 2010). Based on ownership of assets households are divided into three categories- most affluent, mid affluent and least affluent (ASER Centre, 2014) 5. It is, however, not clear a priori if affluence is likely to make one aware of labor market returns or if the relationship is the other way around. Surveyed adolescents were asked “If you have an option to become whatever you want to be (and with all resources), what would you like to be?” Following the codes adopted by the Ministry of Labour and Employment (2015), ‘expected occupation’ is recorded as a single digit code ranging from 1 to 9 and an associated skill level ranging from 1 to 46. The higher the level of skill associated with an adolescent’s desired occupation, higher is the likelihood of reporting expected earnings. Parents’ education is expected to positively influence an adolescent’s knowledge of expected earnings. In this paper, given the importance of extended family in the Indian particularly rural context (Shah, 1988), we consider the education level of not only the adolescent’s parents and his or her adult (18+ years) siblings, but also of his or her aunts and uncles; that is, parent's siblings. We consider the level of education among all these different members and take into consideration the highest level of education among them. We hypothesize that higher the level of education in the extended family network, higher is the likelihood that an adolescent has knowledge about expected earnings at age 25. Research (Jha, Rao, and Woolcock, 2007; Szreter and Woolcock, 2004; Krishna, 2002) in developing countries has highlighted the importance of social networks. In broad

5Sampled households were asked whether they possessed a list of 12 consumer durable items. We constructed a

household consumer durable index as a proxy measure of its economic status using the list of 12 consumer durable items. As one might expect the consumer durable index varies from 0 to 12 and household affluence was based on the resulting distribution of consumer durable index. Households were categorised as most affluent if they were in the top 25% of the distribution, mid- affluent if they were in the middle 50% of the distribution and least affluent if they were in the bottom 25% of the distribution.

6 Occupation code 9 is for laborers, for example, and it has the lowest skill level code of 1. Occupation code 2 is

for professionals and it has the highest skill code of 4.

SLIDE 13

terms, knowing a highly placed person, improves one’s informational networks and has potential to improve one’s life chances. In the specific context of the paper, our hypothesis is that access to a wider social network improves the chances of reporting perceived earnings. Social network is captured in the Middle School Study via response to the question: “Do you know anyone who does this work mentioned in [desired occupation]?”. School type (government or private) captures peer network effects i.e. adolescents attending private school are likely to have a wider exposure to occupational avenues and thereby, more likely to report perceived earnings than their peers in government school (Bedi and Garg, 2000). Access to information similarly improves the chances of reporting an income (Maertens, 2011). Access to information is captured at both the individual/ household level by responses to the following survey questions: whether the household has a mobile phone, whether the adolescent reads newspaper, whether the adolescent read book(s) other than textbooks in past one month preceding the survey, whether the adolescent watches news sports and knowledge based television programs and whether the adolescent has ever been

utside the village.

In his book “Broken Ladder”, Krishna (2017) highlights that distance from the nearest urban centre is an important marker of inequality- people living in villages nearest to the urban centre are not only better off, but access to opportunities is also easier for them as compared to people in remote villages. To capture the distance effect, we classify villages into one of the three categories- the nearest degree or higher educational facility is (i) within the village (ii) within 5 km (iii) between 5 km and 10 km (iv) beyond 10 km. We hypothesize that adolescents in remote villages are less likely to report expected earnings than adolescents in villages nearer to educational institutions. Distance to nearest educational facility was not collected in the Middle School Study. We use 2011 census information (Office of the Registrar General, 2011) to bridge this gap. Next, for adolescents who report expected earnings, we examine the following question: R2: Does perceived returns in the labor market vary by caste and gender among adolescents and their parents? Are these systemic biases in perceived labor market returns over and above the adolescent’s intrinsic ability?

SLIDE 14

H2: Based on prevailing labor market differentials, we hypothesize that expected returns would vary by gender and caste. We also expect it to prevail over the adolescent’s intrinsic ability. Since the dependent variable (expected earnings at 25 years) is a continuous variable, we use step-wise Ordinary Least Square model to answer our research question R2. We run separate regressions for each of the two districts considering the background differences in these districts as indicated in Table 1. For each of the two districts, we run two sets of regression- one in which the outcome variable is the adolescent’s expected earnings at 25 years and another one in which the outcome variable is parent’s perception about the earnings for the adolescent at 25 years. We begin with the Restricted Model 2.1, wherein we control for the adolescent’s intrinsic ability as measured by his or her math test scores. The model is expressed as follows: Yi=β0 + β1IAi + β2 X2i + εi (Restricted Model 2.1) where Yi ,the dependent variable, is monthly expected earnings of sampled adolescent (i) and his/her parents (i). It takes values ranging between INR 80 to INR 2,00,000. IAi is adolescent’s cognitive ability as captured by his/ her percentage score in math test. X2i is a vector of control variables which includes expected marriage age7, highest education among adults in the extended family, school type and the category of village the adolescent lives in terms of distance from the nearest higher educational facility. Cognition is about how people perceive, learn, remember, and think about

information. Although other definitions are more encompassing, basic cognitive skills can be

defined as literacy-the ability to read and write in a language- and numeracy- the ability to perform simple mathematical operations (Tazeen, 2008). As a measure of an adolescent’s cognitive ability, we use (his) her math score in learning assessment tests administered to (him) her in the Middle School Study survey. Learning assessments in the survey were carried out in two parts- in the first part, adolescents were administered a basic floor test to assess their foundational reading and arithmetic levels. Adolescents who qualified the floor test were then tested on their higher-grade level (grade 3- grade 7) competencies in four

7The survey asked sampled adolescent “At what age would you like to get married?” The corresponding

question for parents was “At what age will the sampled child get married?”

SLIDE 15

subjects- local vernacular8, math, science and English. While the floor test was a “quick” test administered by the surveyor to the sampled adolescent at the time of the household survey, the latter- a first and only of its kind in India9- was a longer pen and paper test administered to all sampled adolescents who qualified the floor test in the village after the household survey had been completed. The math pen and paper test assessed cognitive levels as recognition, reasoning, analyzing, computing, applying and comparing. Sampled adolescents who did not clear the floor test for math and were not administered the higher order math tests are assumed to score “0” in them. We expect both adolescents and their parents to base their perceived labor market returns on adolescent’s cognitive ability and that there is a positive association between them. Social norms around expected age at marriage is likely to influence perceived

earnings. Lower expected age at marriage10 for girls and associated norms of patrilocal

exogamy (Maertens, 2011) and gendered division of labor within the household is likely to influence choice of desired occupation for girls and thereby, expected earnings. Adolescent girls who expect to be married early know that the burden of household responsibilities would fall on them leaving them with limited time to pursue their own careers. They would be less keen to pursue occupations higher up the scale that are typically heavy in terms of time commitment requirements and often require work outside the usual 9- 5 office hours. A similar rationale is also applicable to parents. Parents who expect to marry their daughters early would be less likely to encourage their daughters to pursue high- end occupations that are associated with higher income. Further, patrilocal exogamy norms would also mean that they will be less willing to invest in their daughter as opposed to their son’s education. Thus, we expect higher the age at marriage, higher would be expected earnings. Likewise, education level in the extended family has a positive relation with expected earnings. Because of the peer effect discussed previously, we expect adolescents in private school report higher perceived earnings than adolescents in government school11. Adolescents in villages without access to educational facilities will have lower expected earnings than adolescents in villages where such facilities are not available. Model 2.1 does not take into consideration adolescent background characteristics as gender, caste, and affluence. It gives an estimate of perceived labor market returns if there

8Hindi in Nalanda, Bihar and Marathi in Satara, Maharashtra 9 The multi- topic and nationally representative, India Human Development Survey (IHDS), too uses learning

tools to assess foundational reading and arithmetic skills. However, unlike the Middle School Study, it does not assess children on higher grade level competencies.

10Both parents and adolescent were asked the expected age at marriage. 11School management type was asked to both parents and adolescents. In 2.28% of cases they do not match.

SLIDE 16

were no market distortions and it was determined solely by an individual’s intrinsic ability, net of other factors. Restricted Model 2.2 considers another extreme situation. It looks at the impact of systemic biases- gender, caste and affluence- captured in the vector X1 on perception of labor market returns among adolescents and parents while controlling for factors included in the vector X2 and excluding a measure of intrinsic ability of adolescents. These biases relate not only to the labor market but also to systematic differences in educational outcomes by these background characteristics as discussed in Section II. Yi=β0 + π1 * X1i + β2 * X2i + εi (Restricted Model 2.2) Model 2.1 and Model 2.2 are for heuristic purposes. In real world both ability and labor market biases are likely to influence perceptions about expected earnings. Thus, the full model is expressed as follows: Yi=β0 + β1IAi + π1 X1i + β2 * X2i + εi (Full Model 2.3) In the full model, we see if the impact of gender, caste and economic class (X1i) on perceived labor market returns exist over and above the adolescent’s intrinsic ability (IAi) while controlling for variables in vector X2i. Our hypothesis is that adolescent girls and Dalits/ Adivasis have lower expected earnings than adolescent boys and privileged OBC and general castes respectively despite having same intrinsic ability. Finally, we have Model 2.4. In this model, we restrict model 2.3 to sample of adolescents and parents whose expected returns are equal to or higher than the median expected return value. We hypothesize that adolescent and parents whose expected earnings at age 25 are at the upper end of the distribution are aiming for occupations or jobs where it is perceived that background characteristics of gender and caste do not make a difference; rather intrinsic ability alone makes a difference. Model 2.4 is expressed as follows: Yi=β0 + β1IAi + π1 X1i + β2 * X2i + εi (Model 2.4 with restricted sample)

SLIDE 17

Section IV Results Differences in reporting of an estimate of expected earnings by adolescents and household characteristics [Table 2 and Table 3 about here] In both the districts, as one might expect, gender makes a difference in terms of whether the adolescent reported expected earnings at age 25. Other factors remaining the same, the log odds of adolescent boys reporting expected earnings is 1.6 times higher than adolescent girls. In Nalanda (but not in Satara), there are significant differences by caste on whether expected earnings are reported. Simple descriptive statistics indicate that as compared to adolescents from EBC (Extremely Backward Caste) background, adolescents from SC/ ST and OBC background are more likely to report expected earnings. In terms of odds ratio, all else remaining the same, the likelihood that a SC/ ST adolescent reports expected earnings is 2.5 times higher than that of an EBC adolescent. Rather curiously, there is not a significant difference between EBC and general caste adolescents in terms of proportions reporting expected earnings. This result holds true in a multivariate framework too. Household affluence appear to make a difference in terms of reporting expected earnings. In Nalanda, in a descriptive framework, more adolescents from the least affluent households reported expected earnings (0.93) as compared to adolescents from mid- affluence level households (0.90) and the most affluent households (0.87). The difference between the least and most affluent households is significant; but the difference between least affluent and mid- affluence level households is not significant. In case of Satara, on the other hand, adolescents from affluent (0.87) and mid- affluence (0.87) level were more likely to report expected earnings than adolescents from least affluent households (0.84) and the difference is statistically significant. However, because there is considerable overlap between caste and household affluence, these differences disappear in a controlled framework in both the districts. Contrary to our hypotheses, reporting a desired occupation associated with a higher skill level, higher level of education in the extended family and studying in a private as

pposed to a government school is not associated with a significant increase in the likelihood
f reporting expected earnings. We hypothesized access to information to improve the

likelihood of reporting expected earnings. The results are, however, not consistent with our

SLIDE 18

hypotheses. In a bivariate framework, reading a newspaper is not likely to improve the

chances of reporting expected earnings in both the study districts. In Nalanda, additionally,

wning a mobile and reading a book does not make a difference in terms of improving

knowledge about labor market returns, but both these variables significant in Satara. Ever been outside the village improves likelihood of reporting expected earnings only in Nalanda and watching knowledge based TV programs is significant in both the districts. In a multivariate framework in Nalanda, none of these variables are significant and in Satara only watching knowledge based TV programs improves likelihood of reporting expected earnings- adolescents who reported watching television are 1.35 times more likely to report expected earnings than those who are not12. Knowing someone from the desired occupation increases the chance of reporting expected earnings only in Satara. Likewise, distance from the nearest educational facility is not a significant variable in Nalanda, but in Satara it makes a significant difference. As compared to villages that have higher educational facilities, staying in villages without such facilities reduces the log odds of reporting expected earnings. In conclusion, results in terms of gender is along hypothesized lines. Differences by caste are not always along expected lines; particularly, general caste adolescents do not appear to be more likely to report labor market returns as compared to adolescents from EBC community in Nalanda. Rather surprisingly, results for control variables are not always consistent with our hypotheses and between districts. Differences in expected to labor market by adolescents and household characteristics [Fig. 1 and Fig. 2 about here] Figure 1 and Figure 2 below gives the kernel density plots of expected earnings of sampled adolescents and their parents. As one might expect, the distribution is positively skewed- bulk of the expected earnings reported by sampled adolescents and their parents are in the lower end of the distribution, but there are a few outliers too. The density plots of adolescents and their parents overlap with each other indicating that adolescents, not yet in the labor market, are neither consistently underestimating or overestimating their earning potential vis-à-vis the (more) realistic expectations of their parents. [Table 4, Table 5, Table 6 and Table 7 about here]

12In separate iteration of the model, we found that adolescents watching television in general which included

films, serials, songs etc. in addition to knowledge based shows are more likely to know their expected income in

Nalanda. Watching television positively impacts adolescent's knowledge about expected labor market earnings

in both the districts.

SLIDE 19

Results of restricted model 2.1 (Table 6) confirm that adolescents scoring higher in math pen and paper test have higher expected earnings. In Nalanda, expected earnings increase by 65 rupees with each percentage point increase in math scores; the corresponding figure for Satara is 258 rupees. Similarly, higher the test score of the adolescent, higher are the parents’ expectations about their adolescent’s earnings (Table 7). Just as with expected earnings as reported by adolescent, math score as a measure of intrinsic ability makes a larger difference for parents in Satara as compared to parents in Nalanda. In Nalanda, each percentage point increase leads to an approximately 25 rupees increase in expected earnings; the corresponding increase in Satara is of about 176 rupees. Moving on, in an uncontrolled framework, there are gender differences in what adolescent expect they would earn monthly when they are of age of 25 in Nalanda and Satara (Table 4). Parents too on an average expect higher monthly earnings for their sons as compared to daughters in both the districts. Kernel density plots also confirm that more girls and their parents are at the lower than the higher end of the distribution. Even after controlling for other individual, household and community level factors but not intrinsic ability (Regression Model 2.2; Table 6), boys expect to earn about 3,000 INR more when they are in labor market at the age of 25 as compared to girls in both the districts. In Nalanda, parents perceive sons to earn more than daughters; however, there is no gender gap in perceived expected earnings among parents in Satara (Table 7). In bivariate analysis (Table 4 and Table 5), there are significant caste differences in adolescent’s and their parent’s expectation about their earnings in Nalanda. Both adolescent and parents belonging to OBCs and General caste expect higher earnings than EBCs. There is no significant difference in earning expectations between adolescents from scheduled caste/tribe and EBC background. In Satara, adolescent and parents from general caste expect higher monthly earnings than scheduled caste/tribes. There is no difference in expected earnings between OBC adolescents and Dalit/ Adivasi adolescents, however, difference in perceived returns exist between parents belonging to these two caste categories. Here too kernel density plots (Figure 3) confirm that in both the districts more adolescents/parents from the SC/ST/ Extremely backward castes are at the lower end of the earnings distribution as compared to general castes. In both the districts, significant differences by household affluence are seen in what adolescents and their parents expect them to earn at the age of 25. In a bivariate framework, adolescents and parents from economically better off households expect a higher labor market earnings as compared to the poorest households. In a controlled framework (Table 6 and Table 7), the difference in perceived earnings between adolescents

SLIDE 20

and parents from middle 50 percent and bottom 25 percent of consumer durable distribution

disappear. However, the difference in expected earnings between adolescents and parents

belonging to the richest (top 25 percent) and the poorest (bottom 25 percent) households

remains. The presence of a variable capturing household affluence in a regression framework

modifies the results related to caste differences in expected earnings. In Satara, caste differences disappear for both adolescents and their parents. In Nalanda, differences in expectations about earnings among adolescents and parents belonging to General category and EBC disappear. However, both adolescents and parents belonging to OBCs expect higher returns than EBCs in controlled framework. Coefficient for expected age at marriage confirms that higher the expected age at marriage, higher are expected earnings for both parents and adolescents. Education level among the extended family makes a difference in expected earnings- higher the education level among adults in the extended family, higher are expected earnings. Each additional year

f education is roughly associated with 500 rupees and 1000 rupees more of expected

earnings in Nalanda and Satara respectively. Studying in a private school does not lead to higher expected earnings except among parents in Nalanda in whose case the difference between perceived earnings among adolescents in private vis-à-vis government school is as much as 5000 rupees. Finally, living in a village without educational facility does not significantly reduce expected earnings in both the districts. Are these systemic biases over and above the adolescent’s intrinsic ability? Results for full model 2.3 (Table 6 and Table 7) indicate that in a situation where cognitive levels are the same between boys and girls, boys perceive significantly higher returns for themselves as compared to girls in both the districts. Parents too expect sons to earn higher than daughters even when boys and girls having similar cognitive levels. This indicates that controlling for adolescent’s performance in pen and paper based test does not bridge the gender gap in expected earnings among adolescent and their parents. Indeed, it is rather surprising that among parents in Nalanda cognitive ability of adolescent as measured by math test scores is not a significant predictor of expected earnings at age 25. In Nalanda, adolescents from OBC category systematically expect significantly higher returns than adolescents from EBC category with similar cognitive ability and so do their parents. On the contrary, the differences in expected earnings between adolescents from OBC category and SC/ST category in Satara become significant only after controlling for adolescent’s math test scores with adolescents from OBC category expecting INR 4687 lower

SLIDE 21

than the adolescent from SC/ ST category. As in model 2.2, parents from different castes in Satara do not perceive different earnings for their children given that they have similar cognitive ability. Marriage age and highest family education among extended family members remains a positive and significant predictor of adolescent’s and parent’s expected earnings in full model 2.3 in both the districts. Adolescents and their parents from richest households (top 25 percent) perceive significantly higher returns than those with similar math test score from poorest households (bottom 25 percent) in both Nalanda and Satara. School type is a significant determinant of adolescent’s expected earnings in Satara and parent’s expected earnings for their children in Nalanda. Is it the case that those adolescents and their parents whose perceived earnings are high base it solely on their (or that of their children’s) intrinsic ability? Do labor market biases by gender and caste disappear in these cases? The final model 2.4 which is restricted to cases equal to or above the median value of expected earnings helps answer this question. In Nalanda, rather surprisingly, in case of both adolescent’s and their parent’s expectations of high labor market returns is not determined by the percentage math score in assessments administered to the children. In this “high expected earnings” scenario, caste differences disappear as do differences by household affluence. But gender continues to be a significant variable. In Satara, both adolescents and their parents associate cognitive ability with labor market returns- a one percentage point increase in math score leads to around INR 184 and INR 106 increase in expected returns for children and parents respectively. Differences in expected returns by gender and household affluence disappear. For parents, caste differences do not determine expected earnings but as in model 2.3 adolescents from OBC category have lower expected earnings than children from SC/ ST category. In terms of other control variables in the regression model, expected marriage age and education among extended family members are significant. Only parents in Nalanda expect higher earnings for adolescents in private schools. Surprisingly, presence of educational facility which has otherwise not been significant, emerges as a significant determinant of expected earnings for this restricted sample of adolescents in Nalanda.

SLIDE 22

Section V Discussion Regression results confirm our hypothesis that gender and caste based distortions in the labor market influence adolescent’s expectations about their earnings when they are old enough to be in the labor force. Adolescents and their parents in both the districts attach a lot

f significance to these systemic biases in perceived potential earnings over and above their

(or that of their children’s) intrinsic ability. Between gender and caste, gender differences appear far more rigid than caste, particularly in Nalanda. Likewise, caste differences appear more rigid in Nalanda than in

Satara. This is perhaps not surprising as development literature indicates that Bihar is among

north Indian states where gender and caste rigidities are stronger (Dreze and Sen, 2002). Caste and gender differences mostly disappear when the sample is restricted to cases where expected earnings are equal to or greater than the median. The occupations associated with these higher earnings are-legislators, senior officials, and managers, clerks, craft and related trade workers in Nalanda and professionals, technicians and associate professionals in Satara in addition to legislators and senior officials. It is not clear that if it is indeed the case that in these occupations background characteristics do not matter and earnings are a sole function

f intrinsic ability. If true, it can potentially provide insights into how the labor markets ought

to function to minimize gender and caste based distortions. Second, the findings indicate the existence of systemic biases in perceived returns in labor market over and above the adolescent’s intrinsic ability among adolescents themselves and their parents. The policy implications that follow is a mix of affirmative actions and policies directed to improve learning outcomes is essential to improve labor market outcomes for young Indians. Relatedly, the findings also indicate that it may not be enough for policies aiming to improve educational outcomes to focus within the education sector alone. Labor market linkages is an important factor to consider when the target age group for increasing learning levels is adolescents. While the results in terms of our independent variables are as expected, other control variables in the models are not along predicted lines and there is often a great deal of inconsistency between the districts. Access to information, for example, did not improve the likelihood of reporting expected earnings. Similarly, studying in private school did not always have the intended peer effect in terms of higher earnings. These findings are contrary to previous research and cannot be explained simply in terms of differences in context

SLIDE 23

between the two districts. Future research ought to provide a more nuanced understanding of how these variables function. This is important because many of these variables are policy amenable (such as informational access) and can play an important role in influencing adolescents’ aspirations and labor market expectations.

SLIDE 24

References Anirudh, K. (2017). The Broken Ladder: The Paradox and the Potential of India's One

Billion. Gurgaon, Haryana, India: Penguin Random House.

ASER Centre. (2016). Annual Status of Education Report (Rural). Delhi: ASER Centre. ASER Centre. (2014). Middle Schools in India: Access and Quality (Baseline Report). Delhi: ASER Centre. Attanasio, O. and K. Kaufmann (2010). Educational Choices and Subjective Expectations of Returns: Evidence on Intra-Household Decisions and Gender Differences. Retrieved from https://www.researchgate.net/publication/228425947_Educational_Choices_and_Subj ective_Expectations_of_Returns_Evidence_on_Intra- Household_Decisions_and_Gender_Differences Bhattacharjea, S., W. Wadhwa and R. Banerji (2011). Inside Primary Schools: A Study of Teaching and Learning in Rural India. Pratham Education Initiative, Mumbai. Bedi, A. and A. Garg (2000). The Effectiveness of Private Versus Public Schools: the Case of

Indonesia. Journal of Development Economics 61, 463-494.

Burks, S., J. Carpenter, L. Goette. and A. Rustichini (2009). Cognitive skills affect economic preferences, strategic behavior, and job attachment. Proceedings of the National Academy of the Sciences of the USA 106 (19): 7745- 50 Chesters, J. and C. Sinning (2013). The returns to literacy skills in Australia. Retreived from https://www.ncver.edu.au/publications/publications/all-publications/the-returns-to- literacy-skills-in-australia Carneiro, P., J. Heckman, and D. Masterov (2004). Labor Market Discrimination and Racial Differences in Premarket Factors. Retrieved from http://www.ifau.se/globalassets/pdf/se/2005/wp05-03.pdf Das, M. B. (2003). Why Are Educated Women Less Likely to be Employed in India? Testing Competing Hypotheses. Washington, DC: World Bank. Desai, S., A. Dubey, B. Joshi, M. Sen, A. Shariff and R. Vanneman (2010). Human Development in India: Challenges for a Society in Transition. New Delhi: Oxford University Press. Deshpande, A. and M. Banerji (2017, April 11). http://www.indiaspend.com. Retrieved from IndiaSpend: http://www.indiaspend.com/cover-story/beyond-universal-enrolment-in- indian-schools-low-attendance-high-dropouts-94333 Dréze, J. and A., Sen (2002). India: Development and Participation. New Delhi: Oxford University Press. Duraisamy, P. and M. Duraisamy (2017, January 28). Social Identity and Wage Discrimination in the Indian Labour Market. Economic and Political Weekly, LII(4), 51- 60. Foster, A. and M. Rosenzweig (1999). Missing Women, the Marriage Market and Economic Growth”. Retrieved from https://sites.hks.harvard.edu/cid/archive/events/cidneudc/papers/sex.pdf Field, E. and A. Ambrus (2008). Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh. Journal of Political Economy, 116(5), 881- 929. Hanushek, E. and L. Woessmann (2008). The Role of Cognitive Skills in Economic

Development. Journal of Economic Literature, 46(3): 607- 668

Jensen, R. (2010). The Perceived Returns to Education and the Demand for Schooling. Quarterly Journal of Economics, 125(2), 515- 48. Lee, J. and D. Newhouse (2008). Cognitive Skills and Youth Labor Market Outcomes. Retrieved from

SLIDE 25

http://siteresources.worldbank.org/EXTNWDR2013/Resources/8258024- 1320950747192/8260293-1320956712276/8261091- 1348683883703/WDR2013_bp_Cognitive_Skills.pdf Long, S. (1997). Regression Models for Categorical and Limited Dependent Variables: Advanced Quantitative Techniques in the Social Sciences. Thousand Oaks, CA: Sage Publications. Maertens, A. (2011). Does Education Pay Off? Subjective Expectations on Education in Rural India. Economic and Political Weekly, XLVI(9), 58- 63. Munshi, K. and M. Rosenzweig (2006). Traditional Institutions Meet the Modern World: Caste, Gender, and Schooling Choice in a Globalizing Economy. American Economic Review, 96(4), 1225- 1252. Nguyen, T. (2008). Information, Role Models and Perceived Returns to Education: Experimen- tal Evidence from Madagascar. MIT. Rawal, V. and P. Saha (2015). Women’s Employment in India: What Do Recent NSS Surveys

f Employment and Unemployment Show? Retrieved from Statistics on Indian

Economy and Society: http://archive. indianstatistics.org/misc/women_work.pdf Rose, H. (2005). Do gains in test scores explain labor market outcomes? Economics of Education Review, 25: 430- 446. Sengupta, A. and P. Das (2014). Gender Wage Discrimination across Social and Religious Groups in India Estimates with Unit Level Data. Economic and Political Weekly, XLIX(21), 71- 76. Siddiqui, Zakaria, K. Lahiri-Dutt, S. Lockie, and B. Pritchard (2017). Reconsidering Women’s Work in Rural India Analysis of NSSO Data, 2004–05 and 2011–12. Economic and Political Weekly, LII(1), 45- 52. Shah, A. (1988). The Family in India: Critical Essays. New Delhi: Orient Longman Limited. Tazeen, F. (2008). Linking Education Policy to Labor Market Outcomes. Washington, DC: The World Bank Thorat, S. and P. Attewell (2007, October 13). The Legacy of Social Exclusion: A Correspondence Study of Job Discrimination in India. Economic and Political Weekly, 4141- 4145. Treasurer of the Commonwealth of Australia (2010). Intergenerational Report. Retrieved from http://archive.treasury.gov.au/igr/igr2010/report/pdf/igr_2010.pdf

SLIDE 26

Table 1: District Profile

Source: * Census 2011 **Niti Aayog

Bihar Nalanda Maharashtra Satara Population (in 000's) 1,03,805 2,873 1,12,373 3,004 SC (percent) 15.9 21.1 11.8 10.8 ST (percent) 1.3 0.1 9.4 1 Sex ratio 916 921 925 986 Female literacy 53.3 54.8 75.5 76.3 Urban population (percent) 8.4 15.9 36.6 19 GSDP (2014-15) at constant prices 189789 NA 947550 NA % Growth of GDSP (at constant prices) over previous year 9.45 NA 5.66 NA Per capita NSDP (2014-15) at constant prices ** 16801 NA 72200 NA

SLIDE 27

Table 2: Percentage of adolescents reporting expected monthly earnings, by district

Nalanda Satara Individual characteristics Gender Female 0.86 0.83 Male 0.91* 0.89* Expected Occupation Skill I & II 0.92 0.80 Skill 3 0.88 0.88 Skill 4 0.89 0.85 Household characteristics Caste Extremely Backward Castes 0.86 Scheduled Castes/ Scheduled Tribes 0.94* 0.85 Other Backward Castes 0.92*** 0.87 General 0.82 0.86 Household affluence Bottom 25 percent 0.93 0.84 Middle 50 percent 0.90 0.87* Top 25 percent 0.87** 0.87* Highest education level in the extended family, including parents', adult siblings and parent's siblings (in years) No education/ illiterate & up to primary education Upper primary education and above 0.89 0.86 School Type Government 0.89 0.85 Private 0.90 0.86 Access to information at individual/ household level Household owns a mobile No mobile in the household 0.93 0.81 Mobile in the household 0.88 0.86* Whether the adolescent reads newspaper or not? Do not read newspapers 0.88 0.85 Read newspapers 0.90 0.87 Whether the adolescent reads book(s) or not? Did not read book in past one month preceeding the survey 0.88 0.84 Read book/s in past one month preceeding the survey 0.90 0.87* Whether the adoleacent have ever been outside the village? Never been outside village 0.85 0.81 Been outside village atleast once 0.89* 0.86 Whether the adolescent watches TV or not? Does not watch TV 0.87 0.83 Watches TV 0.90 0.88* Social Networks Child does not know anyone from desired occupation 0.88 0.84 Child knows someone from desired occupation 0.90 0.89* Village characteristics in terms of education facilities^ Education facility within village 0.93 Education facility within 5 km of village 0.90 0.83* Education facility within 5 km from village 0.88 0.85* Education facility beyond 10 km 0.89 0.86* Observations 1,171 2,217

Note: ^ Education facility includes art, science, engineering, management and medical degree colleges, polytechnic and training institutions and other higher educational institutions. Proportions not reported for cells with no or insufficient observations. Proportions test has been carried out for each variable. The first category is the reference category for every variable. * p<0.01, p<0.05, * p<0.1

SLIDE 28

Table 3: Odds ratio of adolescents who reported expected monthly earnings, by district

Nalanda Satara Individual characteristics Gender (Reference: Female) 1.641 (0.348) 1.659* (0.217) Expected occupation by skill levels (Reference: Skill level 1 and 2) Skill level 3 0.689 (0.313) 1.815 (0.690) Skill level 4 0.881 (0.377) 1.563 (0.576) Household characteristics Caste (Reference: EBC in Nalanda and SC/ST in Satara) Scheduled Castes/ Scheduled Tribes 2.469* (0.768) Other Backward Castes 1.897* (0.462) 1.201 (0.241) General 0.923 (0.244) 1.201 (0.194) Household affluence (Reference: Bottom 25%) Middle 50% 0.894 (0.357) 1.144 (0.197) Top 25% 0.739 (0.306) 1.018 (0.199) Highest education level in the extended family, including parents', adult siblings and parent's siblings (in years) 0.963 (0.0345) 1.025 (0.0309) School Type Private (Reference: Government) 1.054 (0.358) 1.014 (0.143) Access to information at individual/ household level Household has a mobile (Reference: Does not have mobile in the household) 0.720 (0.316) 1.345 (0.408) Reads news paper (Reference: Does not read newspaper) 1.031 (0.215) 1.045 (0.138) Read books in past one month (Reference: Has not read a book in the past one month preceeding the survey) 1.280 (0.258) 1.244 (0.175) Been outside village (Reference: Has never been outside the village) 1.267 (0.355) 1.156 (0.421) Watches television (Reference: Does not watch television) 1.263 (0.254) 1.348 (0.173) Social networks Knows anyone from desired occupation (Reference: Does not know anyone from the desired occupation) 1.249 (0.240) 1.431* (0.183) Village characteristics in terms of education facilities^ (Reference: Education facilities within the village) Education facility within 5 km 0.377* (0.118) Education facility between 5-10 km 0.910 (0.285) 0.427* (0.120) Education facility beyond 10 km 1.033 (0.311) 0.472*** (0.128) Observations 1,171 2,217

Note: ^ Education facility includes art, science, engineering, management and medical degree colleges, polytechnic and training institutions and other higher educational institutions. * p<0.01, p<0.05, * p<0.1

SLIDE 29

Figure 1: Kernel Density plots of adolescents’ expected monthly earnings at the age 25 and parents’ expected monthly earnings for their children at their age of 25

.00001 .00002 .00003 .00004 50000 100000 150000 200000

Adolescents' and Parents' expected monthly earnings at the age 25: Satara

Adolescents' expected monthly earnings: Satara Parent's expected monthly earnings for their child: Satara

Adolescents' and Parents' expected monthly earnings at the age 25: Satara

kernel = epanechnikov, bandwidth = 2.8e+03

Kernel Density Estimate

.00002 .00004 .00006 50000 100000 150000 200000

Adolescents' and Parents' expected earnings at the age 25: Nalanda

Adolescents' expected monthly earnings: Nalanda Parents' expected monthly earnings for their child: Nalanda

Adolescents' and Parents' expected earnings at the age 25: Nalanda

kernel = epanechnikov, bandwidth = 1.4e+03

Kernel Density Estimate

SLIDE 30