Parents’ Choice Function for Wards’ School Continuation in Rural India: A Case Study in West Bengal

Abstract

In this article, we present a choice function of a rural household relating to her/his ward’s schooling. It makes an empirical evaluation on the basis of a simple theoretical framework using a primary data set surveyed from two backward districts of West Bengal. It explores the underlying causes of wards’ discontinuation of school by examining the choice function of the parents using ordered probit analysis. The likelihood of dropout is higher at the primary level for low-income households and significantly depends on parents’ attributes, which are mostly endogenous in an educational production function, and other exogenous difficulties in accessing school. It is also triggered by a lack of expectation about the future impact of education on a child’s life.

JEL Classification: C25, C80, D19, I21, I25

Keywords

School Education Dropouts Household’s Choice Ordered Probit Analysis Primary Survey Data

1. INTRODUCTION

Literature suggests that education plays an important role in enhancing capabilities and freedom and paves the way for a quality life (Hoffman, 2006; Jonathan, 2001; Robeyns, 2003; Sen, 1985, 1987, 1992, 1999, 2003). Sen (1992, 1999, 2003) emphasises the importance of capabilities in helping people make normative judgements about equality and well-being; he speculates whether economic wealth and income could be indicators of a country’s quality of life and/or whether human capital arguments could be used for judging the importance of education only by its success to prepare the educated for employment. Despite the theoretical research and well-attempted policy support, there is a significant problem of pupil non-attendance and discontinuation in schools at all levels across the country. The problem is more complex and severe in children from families belonging to the margin. The basic problem is that school attendance by students still continues to be low and inconsistent across the globe, especially in developing countries. The issue of low levels of school attainment in developing countries is usually analysed using the standard approach based on the assumption that parents alone make schooling decisions for their children. This study looks at the issue of school non-attainment, influenced by parent–child differences in preferences over schooling and intra-household dynamics. Thus, it focuses on both the parent and the child, as the interaction of both choice sets plays a central role in schooling decisions.

2. THE PRESENT SCENARIO

School dropout rates in India are extremely high, as over 80 million children do not complete the full cycle of elementary education, while 8 million are out of school over a period of years (The Economic Times, 2014).¹ The dropout rate is skewed towards girls, and the spread of dropouts is far greater in rural areas, when compared with urban pockets.

Based on a nationwide survey, the Annual Status of Education Report or ASER (2011, 2012) reported that around one-quarter of enrolled children in Indian rural schools were absent on any given school day. India ranks 155th among the 207 countries for which relevant data on school attendance have been provided by the United Nations Children’s Fund (UNICEF) (2010). According to UNICEF (2010), 99.5 per cent of young people (out of a total of 125.4 million) between the ages of 15 and 24 are illiterate globally and live in the developing world. Around 51.8 per cent of this illiterate population lives in South Asia, and among this 51.8 per cent, India’s share is around 62 per cent, making India the home of almost one-third (40.4 million, 32.2 per cent) of all the illiterate young people worldwide. A study reveals that of more than 27 million children in India who had joined class 1 in 1993, only 10 million had reached class 10, which is only about 37 per cent of those who had entered the school system; in more than half the states,² only 30 per cent had reached class 10 (Reddy and Sinha, 2010). It is true that with the implementation of The Right of Children to Free and Compulsory Education Act (RTA), 2009, there has been a gradual decline in the annual average dropout rate from 9.1 in 2009–10 to 6.9 in 2010–11 (Daily News and Analysis, 2012). On the other hand, it is also a fact that there were more dropouts in 2010–11 than in 2009–10 in 10 of the 30 states where RTA has been implemented (The Times of India, 2012).

Therefore, there are reasons for concern and a need to investigate the reasons and inner dynamics of trends that are major hindrances to inclusive development. Linkages between poverty and non-participation in schools have been noted by Tilak (1996), Ray (1999), Chaudhri and Wilson (2002), Rustagi et al. (as cited in Chaudhri and Jha, 2011), Chudgar (2010), to mention a few. Studies by Birdsall et al. (2005), Brown and Park (2002), Chugh (2011), Colclough et al. (2000), Dachi and Garrett (2003), Glick et al. (2014), Kotwal and Rani (2007) and Mukudi (2004) showed that the expenditure incurred by parents and households has a positive relationship with dropout and gendered outcomes. Studies also point out to the relationship between child labour, mothers’ literacy and school dropouts (Basu et al., 2010; Bhalotra and Heady, 2003; Francavilla et al., 2012; Kambhampati and Rajan, 2005; Ray, 2000). However, they do not point out the reasons why a child is engaged in income-generating activities and foregoing their future potential of higher earnings and a better life.

In developing countries like India, children typically drop out of school due to income constraints and low expected returns from continuing school. Obviously, this has significant long-term impacts such as low educational attainment and consequently low levels of human capital accumulation, which in turn imply limited future income-earning opportunities. There is also an intergenerational effect: Children born to parents with low levels of education are themselves more likely to end up with low levels of educational attainment. The problem is more complex and severe for people belonging to the margin. The following questions remain: What is the compulsion for non-attendance or what influences students from opting out of the school system irrespective of gender? In the developing world, do people belonging to the margin continue to the next level of education after elementary? If not, what are the binding factors? Do these factors have some pattern in terms of their priority across regions and groups? There is a need to understand the grassroots’ dynamics of these issues to devise possible ‘preventive’ measures.

The purpose of this study is to understand the inherent factors that influence the decision to continue schooling. It attempts to see whether economic factors are more pressing in persuading parents to send their children to school. The views of the actual dropout are taken into consideration for testing the hypotheses proposed. Based on a suitably designed sample survey for two backward regions of West Bengal, the present study tries to address the issues from a socio-economic perspective in order to understand the inner dynamics that might influence attendance decisions.

3. THEORETICAL FRAMEWORK

Based on the model proposed by Drèze and Kingdon (1999), here we develop a simple model of schooling decisions in the cost–benefit framework. Let us assume household income as w_t in the tth period. We further assume that in the same period, household consumption is C_t and expenditure on education is b_t (this may also be termed as the hidden opportunity cost of investing in education). Therefore, the indirect utility function of the household can be written as

W_{t} = C_{t} + I_{t}

where I_t is the amount the household might have invested in some other sector and could expect a gain of I_t_{+ 1} with interest rate r.

That is, I_t_{+ 1} = (1 + r)I_t and if the households invest in education then W_t = C_t + b_t.

If the benefit from education is

B_{t} = B (b_{t}, g_{t} : X_{h}, Z_{k})

where X_h is the vector of household characteristics and Z_k is the vector of school characteristics and g_t is the contribution by the government for continuation of schooling through welfare programmes in the tth period. The perceived benefit from schooling would be gained in the future period and can be shown as

(B_{t ​ + ​ 1}) = B (b_{t} : X_{h}, Z_{k})

We also assume that B(•) is increasing in Z, the component may be thought of as an indicator of ‘school quality’, a function of government intervention.

Let the household’s utility function be

U = f (W, B); U_{W}, U_{B} > 0

Here, it is maintained that the household’s objective function is separable without loss of generality (additively, a separable utility function) with respect to investment in schooling and consumption. Now, if the household chooses to invest in the education of the child, then it will intend to maximise its total expected utility. That is, maximise E(U_t) + E(U_{t +}₁).

Therefore, the household’s total expected utility function will be

E (U ​ | ​_{b_{t} > 0}) = λ^{t} (W_{t} - C_{t} - b_{t} + B_{t}) + λ_{t ​ + ​ 1} (W_{t} - C_{t ​ + ​ 1} + B_{t ​ + ​ 1})

Let $b_{t}^{*} (W_{t}, B_{t}, B_{t + 1} : X_{h}, Z_{k})$ be the substitution of the above maximisation problem. Then V(W_t, B_{t +}₁ : X_h, Z_k) is the maximum value function conditional on the child being enrolled. It is always optimal to spend at least some money on their education (in addition to the fixed cost). The child will get enrolled if and only if

V ​ (W_{t}, B_{t}, B_{t ​ + ​ 1} : X_{h}, Z_{k}) - E ​ (U) > 0.

Further, when a child enrols, the first-order condition for maximisation of equation (6) is

\frac{δ E (U)}{δ b_{t}} = 0 \Rightarrow b_{t}^{*} = {U^{'}}_{w, B} : X_{h}, Z_{k} .

The above simple model leads to several outcomes. They are stated below:

$E (U | _{I_{t}^{}_{> 0}^{}}) = λ^{t} (W_{t} - C_{t} - I_{t}) + λ_{t + 1} (W_{t} - C_{t + 1} + I_{t + 1})$ . The household will invest in education if and only if the total expected return from education is greater than the net expected return from other sources, depending on the household’s characteristics as well as on school quality, $E (U | _{b_{t} > 0}) > E (U | _{I_{t}^{}_{> 0}^{}}) .$

If $\frac{δ V}{δ Z_{k}} > 0$ , it implies that enrolment in school is increasing with respect to school quality. This applies not only to initial enrolment but also for continuation of school participation over time.

If $\frac{δ b^{*}}{δ Z_{k}} = - U_{w, B}^{/ /} : X_{h}, Z_{k},$ it implies that an improvement in school quality raises household expenditure on education if and only if the two are complimentary inputs in the utility function.

Therefore, in general it is plausible to think that a household’s expenditure on education or decision to continue the schooling of a child and school quality are compliments rather than substitutes. Logically, the concerned household belonging to the margin would only invest in school education if the foregone income (investment for continuing secondary schooling) could be regained at a significantly higher amount (or even equal amount) after completion of schooling. If this is not the case, then the present value of costs would be more than the present value of benefits, and one would discontinue education after the completion of elementary school, which is assumed to be free. The higher the discount rate or opportunity cost, the less likely it is that a student will choose to continue schooling. We have modelled a simple inter-temporal choice function of an individual for investment on the education (the decision to continue schooling) of their ward. First-order conditions of maximisation of a constrained inter-temporal utility function of an individual in our theoretical model lead to the prediction that ‘A household will allow its ward to continue schooling if the net expected return from education is positive, where net expected return contains both net monetary return from investing on education over time and current benefit generated from school-quality-related factors. This prediction also pertains to the continuation of school participation at a higher level.’

In order to test the claim, empirical evaluation is must. To examine the determinants of the preference to continue school at the post-elementary level (10–20 years) by testing the relevance of alternative explanations of why children do/do not attend and/or discontinue school in their most formative years, detailed primary data were obtained from two backward districts in the state of West Bengal, Jalpaiguri and Murshidabad, through purposive household sampling.

4. DATA AND FIELD SURVEY

The reasons for selecting these two socio-economically backward districts are twofold. First, the geographical locations of the two districts are fairly sensitive, as one of them (Murshidabad) shares a border with Bangladesh and the other (Jalpaiguri) shares its border with Bhutan. Second, based on the District Information System for Education (DISE, 2009–10) and the Secondary Education Management Information System (SEMIS, 2009–10), it was seen that both Murshidabad and Jalpaiguri show a decline in the gross enrolment ratio (GER). Further, the GER for secondary and higher secondary schools varies within districts. During 2009–10, in Jalpaiguri, the GER for secondary and higher secondary was 63 and 24, respectively, and for Murshidabad it was 60 and 28, respectively. It can be asserted that the GER is inversely proportional to the level of education, which holds for the entire state of West Bengal, as well as for both Jalpaiguri and Murshidabad.

The dropout rates for boys and girls in classes 9–12 for both Jalpaiguri (75.81) and Murshidabad (80.01) are higher than the state average (75.04). Of the 19,206 schools, 6 and 12 per cent have a student classroom ratio (SCR) greater than 60 in Jalpaiguri and Murshidabad, respectively; 4.91 and 5.55 per cent have common toilets in Jalpaiguri and Murshidabad, respectively and approximately 5 per cent of schools have a pupil–teacher ratio (PTR) above 100 in both the districts. For secondary and higher secondary classes, the SCR for Jalpaiguri and Murshidabad ranges between 76 and 101; only 27 per cent of the boys and 25 per cent of the girls pass the class 8 exam with more than 60 per cent marks in Jalpaiguri and this percentage (23 per cent of the boys and 19 per cent of the girls) is lower in Murshidabad. The ‘West Bengal Human Development Report’ (2004) states that in terms of the overall ranking in the human development index, Jalpaiguri ranks 10th and Murshidabad 15th in the list of 17 districts ranked. These low indicators have been given to support our choice of these two districts for the present study.

The study is based on data collected via an appropriately designed questionnaire during 2010–11. As parents are the primary decision-makers in a family, at least on the issue of dropout, we interviewed the fathers of such subjects and where fathers were not available, we interviewed the mothers. In the absence of both, we interviewed the next available head or senior member of the family. The sample of interest was composed of all children who have decided to attend school or not, and whose families are involved in that decision. Under these considerations, the estimate is based on all boys and girls who, at the time of the survey, were in the age between 10 and 20, single, classified as sons or daughters in the household and had discontinued schooling during 2008–10. In this study, the word ‘subject’ indicates the person who has actually discontinued formal schooling during the said period.

In Murshidabad, we interviewed 146 subjects (60 per cent males and 40 per cent females) who had actually dropped out during the period. In Jalpaiguri district, the number was 97 (43 per cent males and 57 per cent females). Table 1 describes the frequency distribution of the sample observations in terms of age, gender, working status of the subjects, fathers’ literacy and mothers’ literacy.

Table 1
Distribution of the Classified Sample as a Percentage of Total Observations

Dropout Category* Never Attended School Dropout at Initial Level Dropout at Middle Level Dropout at Higher Level Total

Frequency 17.28 24.69 37.04 20.99 100.00

Area Cossimbazar 17.28 20.16 16.87 5.76 60.08

Jalpaiguri 0.00 4.53 20.16 15.23 39.92

Gender Female 9.88 14.81 22.22 11.52 58.44

Male 7.41 9.88 14.81 9.47 41.56

Working status of the subjects Idle 4.94 8.64 10.70 7.82 32.10

Working 12.35 16.05 26.34 13.17 67.90

Father’s literacy Literate 7.41 11.52 18.11 11.93 48.97

Not literate 9.88 13.17 18.93 9.05 51.03

Mother’s literacy Literate 3.29 4.94 11.93 8.23 28.40

Not literate 13.99 19.75 25.10 12.76 71.60

	Dropout Category*	Never Attended School	Dropout at Initial Level	Dropout at Middle Level	Dropout at Higher Level	Total
Frequency		17.28	24.69	37.04	20.99	100.00
Area	Cossimbazar	17.28	20.16	16.87	5.76	60.08
	Jalpaiguri	0.00	4.53	20.16	15.23	39.92
Gender	Female	9.88	14.81	22.22	11.52	58.44
	Male	7.41	9.88	14.81	9.47	41.56
Working status of the subjects	Idle	4.94	8.64	10.70	7.82	32.10
	Working	12.35	16.05	26.34	13.17	67.90
Father’s literacy	Literate	7.41	11.52	18.11	11.93	48.97
	Not literate	9.88	13.17	18.93	9.05	51.03
Mother’s literacy	Literate	3.29	4.94	11.93	8.23	28.40
	Not literate	13.99	19.75	25.10	12.76	71.60

Source: Authors’ calculation based on the field survey.

Note: *different responses given by actual dropouts.

5. METHODOLOGY

Based on the theoretical framework and the conceptual background, we first specified some hypotheses for empirical investigation:

Income is a binding factor for the decision to continue schooling;

Additionally, distance to the secondary school (here measured by commutation costs) influences the decision to continue;

Among the parents, the literacy levels of the father or the mother or both act as binding factors for the decision to continue and

The main decision maker’s anticipation of the extent of returns from schooling influences the decision to continue.

Given the data set and the proposed hypotheses, we banked on the ordered probit regression (OPR) model for the econometric analysis. Below, we present a brief account of using the OPR in the present case. Ordered probit technique is a generalisation of the linear regression analysis to cases where the dependent variable is discrete and takes only a finite number of values possessing a natural ordering (Hausman et al., 1991). An ordered probit analysis has been used for the empirical estimation of the underlying reasons for dropping out from the school. This is akin to a generalisation of the linear regression model to cases where the dependent variable is discrete. Underlying the analysis is a ‘virtual’ regression model with an unobserved continuous dependent variable (let us assume as m*), whose conditional mean is a linear function of the observed ‘explanatory’ variables. Although m* is unobserved, y* is observed whose realisations are determined by a select set of independent variables, where m* lies in y* domain. By partitioning the domain into a finite number of distinct levels, m* may be viewed as an indicator function for y* over different levels.

y ​ * = x ​ * β + \in w h e r e, f ​ : m ​ * \to Y ​ *

(1)

This specification is known as ordered probit, a technique used most frequently in cross-sectional studies of dependent variables that take on only a finite number of values possessing a natural ordering.³ Here, the dependent variable is dropouts at different levels of education, as measured by four categories: never attended any school, dropped out at the primary level (class 1–5), dropped out at the secondary level (class 6–8) and dropped out at higher level (class 9–10).

The dependent variable is discrete and naturally ordered as according to the levels of education, quality always moves towards improvement.

\begin{matrix} \begin{matrix} y_{i} = {0 i f n e v e r a t t e n d e d a s c h o o l} \end{matrix} \\ \begin{matrix} = {1 i f d r o p o u t a t p r i m a r y l e v e l} \\ = {2 i f d r o p o u t a t 6 - 8 c l a s s} \\ = {3 i f d r o p o u t a t 9 - 10 c l a s s} \end{matrix} \end{matrix}

(2)

x is the vector of independent variables and β is the vector of regression coefficient which we wish to estimate. There is a disturbance term that follows a standard normal distribution. Like the models for binary data, we are concerned with how changes in the predictors translate into the probability of observing a particular ordinal outcome.

\begin{matrix} P [y_{i} = 0] = P [μ_{1} < y_{i}^{*} \leq μ_{0}], \\ = P [- ​ \infty < y_{i}^{*} \leq μ_{0}] \\ = P [y_{i}^{*} \leq μ_{0}] \end{matrix}

(3)

Substituting from (1)

\begin{matrix} P ​ [y_{i} = 0] = P [x_{i}^{*} b + \in \leq μ_{0}] \\ = P [\in \leq μ_{0} - x_{i}^{*} b], \end{matrix}

(4)

\begin{matrix} \begin{matrix} = f (μ_{0} - x_{i}^{*} b) \\ P ​ [y_{i} = 1] = P [μ_{0} < y_{i}^{*} \leq μ_{1}], \end{matrix} \\ \begin{matrix} = P [μ_{0} - x_{i}^{*} β < \in \leq μ_{1} - x_{i}^{*} b], \\ = f (μ_{1} - x_{i}^{*} b) - f (μ_{0} - x_{i}^{*} b) \end{matrix} \end{matrix}

(5)

Similarly, we get

E ​ (U ​ | ​_{b_{t}^{}_{> 0}^{}}) = λ^{t} (W_{t} - C_{t} - b_{t} - B_{t}) + λ_{t ​ + ​ 1} (W_{t} - C_{t ​ + ​ 1} + B_{t ​ + ​ 1})

(6)

P ​ [y_{t} = 3] = f (μ_{3} - x_{i}^{*} β) - f (μ_{2} - x_{i}^{*} b)

(7)

The estimator which maximises this function will be consistent, asymptotically normal and efficient. It can be shown that this log-likelihood function is globally concave in β, and therefore standard numerical algorithms for optimisation will converge rapidly to the unique maximum.

Here, Table 2 provides the descriptive statistics of the independent variables. To note that the dependent variable is the dropout rate at different level of education measured by four categories: never attended, dropout at the initial level (classes 1–5), dropout at the middle level (classes 6–8) and dropout at the higher level (classes 9–10). The dependent variable is discrete, therefore ordered, as with levels of education quality always moves towards improvement.

Table 2

Descriptive Statistics of the Independent Variables

Sl No.	Variable	Description of the Variable	Mean (Std Dev.)	Min./Max.
1	Sex	Gender of the child (Dummy, male = 1)	0.415 (0.493)	0/1
2	Religion of the subject	Dummy, Hindu = 1	0.473 (0.327)	0/1
3	Working subjects	Dummy, working = 1	0.728 (0.514)	0/1
4	Father’s age	Continuous	62.485 (5.631)	50/73
5	Father’s literacy	Dummy, literate = 1	0.510 (0.500)	0/1
6	Father’s occupation category (according to principal occupation specified by the respondent)	Farming = 1 Service = 2 Small business = 3 Daily wage earner = 4 Other non-labour income = 5	2.304 (0.461)	1/5
7	Mother’s literacy	Dummy, literate = 1	0.716 (0.451)	0/1
8	Working mother (Yes = 1)	Dummy, working = 1		0/1
9	Total income	Category (low income = 1, medium income = 2 and high income = 3)	5,221.811 (2,777.05)	1,200/1,2000
	Reasons Specified as Binding Factors for School Discontinuation by the Subjects
11	No sanitation facility at school	Dummy, binding factor = 1	0.020 (0.142)	0/1
12	Drinking water	Dummy, available = 1	0.065 (0.248)	0/1
13	Exam fear	Dummy, yes = 1	0.551 (0.498)	0/1
14	Torture	Dummy, yes = 1	0.296 (0.457)	0/1
15	Different vernacular	Dummy, yes = 1	0.386 (0.488)	0/1
16	Poverty	Dummy, BPL cardholder = 1	0.843 (0.363)	0/1
17	No predicted return	Dummy, yes = 1	0.662 (0.473)	0/1
18	Neighbourhood effect	Dummy, yes = 1	0.366 (0.482)	0/1
19	Sibling care	Dummy, yes = 1	0.440 (0.497)	0/1
20	Transaction costs	Dummy, high = 1	0.676 (0.468)	0/1
21	Private tuition fees	Dummy, yes = 1	0.806 (0.395)	0/1
22	Aggregate reasons*	Index reflecting all binding factors	6.469 (2.558)	1/11

Source: Authors’ calculations based on the field survey.

Note: *indicates the sum of factors from serial numbers 11 to 21.

6. EMPIRICAL RESULTS

As mentioned, OPR models were estimated for assessing the impact of identified factors for explaining discontinuation from schooling in different grades with a separate set of specifications as shown below in Table 3. It should be noted that, at this point, no variables were removed from the said econometric analysis even if they showed low levels of statistical significance. This is because all chosen variables (selected on the basis of a literature review and prior knowledge in the field) are important in some way or the other, as they are expected to have an impact on the dependent variables, even if small.

Table 3
Determinants of School Attendance—Ordered Probit Results

Dropouts (Dependent Variable)
Coefficient (Std Err)
Coefficient (Std Err)
Coefficient (Std Err)
Coefficient (Std Err)

Specification 1 2 3 4

Sex of the child (boy = 1) 0.282(0.159) 0.082(0.175) 0.369(0.175) 0.023(0.201)

Religion (Hindu = 1) 0.028(0.175) 0.263(0.189) –0.001(0.183) 0.255(0.208)

Father’s age 0.104(0.020) 0.121(0.027) 0.113(0.024) 0.113(0.037)

Father’s literacy (yes = 1) 1.974(0.270) 2.611(0.270) 2.096(0.264) 3.224(0.470)

Father’s occupation category (according to principal occupation specified by the respondent) 0.063(0.063) 0.179(0.085)

Mother’s age –0.102(0.025) –0.102(0.031) –0.068(0.026) –0.074(0.039)

Mother’s literacy (yes = 1) 0.597(0.230) 0.547(0.240) –0.275(0.244) 0.283(0.303)

Working mother (yes = 1) –0.421(0.164) –2.309(0.472)

Total income category (low income = 1, medium income = 2 and high income = 3) –0.577(0.185) 0.103(0.296)

Poor sanitation 2.972(0.452) 3.740(0.743)

No drinking water –0.665(0.405) –1.076(0.614)

Exam fear –1.044(0.243) –0.708(0.325)

Torture by the teacher 0.373(0.221) –2.332–(0.483)

Different vernacular –0.307(0.198) 0.242(0.332)

No predicted return 0.861(0.336) 1.199(0.437)

Neighbourhood effect 0.195(0.236)

High tuition fees –0.216(0.291) 0.622(0.413)

Transaction cost –1.646(0.263) –0.625(0.421)

Aggregate reasons 0.470(0.099) 0.423(0.363)

Poverty (BPL cardholder = 1) –1.036(0.259) –0.613(0.332)

Area (Jalpaiguri = 1) –0.357(0.212) –0.160(0.199)

Dependency ratio (earning member/total members) –3.856(1.548) –14.633***(3.161)

/Cut1 0.518(1.076) 3.317(1.288) 3.436(1.338) –3.003(2.140)

/Cut2 1.700(1.079) 4.889(1.304) 4.727(1.353) –1.186(2.145)

/Cut3 2.875(1.088) 6.179(1.315) 6.097(1.362) 0.624(2.128)

R ² 0.2431 0.3590 0.3031 0.4993

Log likelihood –247.95186 –210.00625 –228.31569 –164.03387

Dropouts (Dependent Variable)	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)
Sex of the child (boy = 1)	0.282(0.159)	0.082(0.175)	0.369(0.175)	0.023(0.201)
Religion (Hindu = 1)	0.028(0.175)	0.263(0.189)	–0.001(0.183)	0.255(0.208)
Father’s age	0.104***(0.020)	0.121***(0.027)	0.113***(0.024)	0.113**(0.037)
Father’s literacy (yes = 1)	1.974***(0.270)	2.611***(0.270)	2.096***(0.264)	3.224***(0.470)
Father’s occupation category (according to principal occupation specified by the respondent)	0.063(0.063)			0.179(0.085)
Mother’s age	–0.102***(0.025)	–0.102***(0.031)	–0.068**(0.026)	–0.074(0.039)
Mother’s literacy (yes = 1)	0.597**(0.230)	0.547(0.240)	–0.275(0.244)	0.283(0.303)
Working mother (yes = 1)	–0.421*(0.164)			–2.309***(0.472)
Total income category (low income = 1, medium income = 2 and high income = 3)	–0.577**(0.185)			0.103(0.296)
Poor sanitation		2.972***(0.452)		3.740***(0.743)
No drinking water		–0.665(0.405)		–1.076(0.614)
Exam fear		–1.044***(0.243)		–0.708(0.325)
Torture by the teacher		0.373(0.221)		–2.332***–(0.483)
Different vernacular		–0.307(0.198)		0.242(0.332)
No predicted return		0.861**(0.336)		1.199**(0.437)
Neighbourhood effect		0.195(0.236)
High tuition fees		–0.216(0.291)		0.622(0.413)
Transaction cost			–1.646***(0.263)	–0.625(0.421)
Aggregate reasons			0.470***(0.099)	0.423(0.363)
Poverty (BPL cardholder = 1)			–1.036***(0.259)	–0.613(0.332)
Area (Jalpaiguri = 1)			–0.357(0.212)	–0.160(0.199)
Dependency ratio (earning member/total members)			–3.856**(1.548)	–14.633***(3.161)
/Cut1	0.518(1.076)	3.317(1.288)	3.436(1.338)	–3.003(2.140)
/Cut2	1.700(1.079)	4.889(1.304)	4.727(1.353)	–1.186(2.145)
/Cut3	2.875(1.088)	6.179(1.315)	6.097(1.362)	0.624(2.128)
R ²	0.2431	0.3590	0.3031	0.4993
Log likelihood	–247.95186	–210.00625	–228.31569	–164.03387

Source: Authors’ calculation based on field survey.

Note: *, ** and *** denotes level of significance.

In assessing the results, we have divided the specifications into three distinct categories:

General information and household characteristics which include sex, religion, father’s age, father’s literacy, mother’s age, mother’s literacy, principal occupational category of the father, working mothers and family income. Based on the literature, school attendance is expected to have a negative relationship with father’s age, mother’s age, being a girl child (the gender factor), being a non-Hindu and varied occupations (multiple sources to generate income). On the other hand, factors such as literacy (for both father and mother), working mother and family income are expected to have a positive relationship with school attendance.

The second specification looks at causes of discontinuation revealed by respondents during the field visit.⁴ The causes include distance of secondary school from home, ‘exam fear’, corporal punishment in school, difference in vernacular between school and home, anticipation of no or low returns from continuing schooling, impact of neighbourhood, unaffordability of private tuition fees and lack of, or unavailability of, safe drinking water. All these factors have a binding impact on the decision to continue schooling, and based on the literature, it can be said that the decision to continue schooling is negatively impacted by each of these factors.

The third specification consists of the variables such as poverty, commutation costs and dependency ratios.⁵ To differentiate between the poverty-ridden families (considered to be a below poverty line (BPL) cardholder), we have treated the positive responses (=1) of the respondents for whom the incidence of poverty was a basic reason for discontinuation. Here too, the incidence of poverty, high commutation costs and high dependency ratios hinder school attendance.

Along with each set of results, for all specifications, three cut-off values, R-squared and likelihood estimates are reported in Table 3 and the other relevant tables. The OPR allows us to separate the effects of various factors that influence the likelihood of occurring from one level to another of the dependent variable. Table 3 discusses the probability of a child discontinuing school. Given the category of the dependent variable and according to the categorisation of the exogenous variables, we obtained four sets of econometric results. However, as OPR does not give a direct relationship between the dependent variables and independent variables, for diagnostic purposes we also estimated the marginal effects after OPR for each specification (Tables 4 and 5).

To estimate the marginal effects, we considered two base values of the dependent variable y_i = {0,1}. Henceforth, we concentrate on Tables 4 and 5 to explain the relationships. Table 2 shows that in all the four specifications, the coefficients of variables like father’s age and father’s literacy show a significant relationship. However, all the coefficients carry a negative sign. Primarily, it is clear that the decision to discontinue school significantly depends on these factors. Now, we also have a negative sign for these coefficients, obtained while estimating the marginal effects.

The presence of the negative sign for the variable ‘father’s age’ is as expected and signifies that the probability of discontinuation of schooling at early grades increases if fatherhood is achieved at an older age. That is, men who become parents at a mature age are more likely to have their child discontinue schooling at the lower grades. This is true for the mother’s age as well. The variable ‘mother’s age’ shows a significant relationship in specifications 1, 2 and 3, with a positive sign and ‘mother’s literacy’ shows a significant relationship with a negative sign for specifications 1 and 2. However, the significance levels along with the value of the coefficients are low.

The OPR exercise yielded some observations, which we briefly discuss. It pointed out the relationship between discontinuation of schooling and father’s literacy. We see that there is a negative correlation between these two variables. The influence of parental education (of both the father and mother) acts as a strong determinant of early school discontinuation. Simply being able to decipher native alphabets and write their names and addresses did not enable them to live life with dignity. That is to say, the literacy missions only equipped them to sign their names in their native languages, and the real benefits from being ‘educated’ are still far from being realised.

Table 4

Marginal Effects after Ordered Probit When Y = Pr (dropouts = never attended)

Dropouts (Dependent Variable)	dy/dx (Std Err)	dy/dx (Std Err)	dy/dx (Std Err)	dy/dx (Std Err)
Specification	1	2	3	4
Sex of the child (boy = 1)	–0.080(0.044)	–0.019(0.040)	–0.096(0.045)	–0.004(0.034)
Religion (Hindu = 1)	–0.008(0.051)	–0.065(0.050)	0.000(0.049)	–0.046(0.041)
Father’s age	–0.030***(0.006)	–0.028***(0.007)	–0.031***(0.007)	–0.019**(0.007)
Father’s literacy (yes = 1)	–0.523***(0.066)	–0.576***(0.061)	–0.521***(0.064)	–0.590***(0.093)
Father’s occupation category (according to principal occupation specified by the respondent)	–0.018(0.018)			–0.030(0.015)
Mother’s age	0.029***(0.008)	0.024**(0.008)	0.018**(0.007)	0.013(0.007)
Mother’s literacy (yes = 1)	–0.163**(0.060)	–0.121*(0.051)	0.076(0.070)	–0.046(0.048)
Working mother (yes = 1)	0.122*(0.049)			0.444***(0.109)
Total income category (low income = 1, medium income = 2 and high income = 3)	0.167**(0.055)			#
Bad sanitation		–0.269***(0.044)		0.291(0.217)
No drinking water		0.197(0.143)		0.139(0.077)
Exam fear		0.282***(0.078)		0.585***(0.129)
Torture by the teacher		–0.081(0.048)		–0.041(0.058)
Different vernacular		0.071(0.046)		0.089(0.044)
No predicted return		–0.264(0.122)		–0.327(0.155)
Neighbourhood effect		–0.045(0.054)
High tuition fees		0.047(0.058)		0.114(0.085)
Transaction cost			0.468***(0.078)	–0.136(0.113)
Aggregate reasons			–0.126***(0.029)	0.027(0.034)
Poverty (BPL cardholder = 1)			0.227***(0.050)	–0.018(0.050)
Area (Jalpaiguri = 1)			0.099(0.061)	2.479***(0.672)
Dependency ratio (earning member/total member)			1.037*(0.427)	–0.068(0.055)
Y	0.211	0.151	0.187	0.095

Source: Authors’ calculation based on the field survey.

Notes: (1) # here, we have controlled the variable ‘income’ to get the effect of ‘dependency ratio’.

(2) *, ** and *** denotes level of significance.

(3) Y implies the dependent variable.

(4) Pr = probability

Table 5

Marginal Effects after Ordered Probit when Y = Pr (dropped out at primary)

Dropouts (Dependent Variable)	dy/dx (Std Err)	dy/dx (Std Err)	dy/dx (Std Err)	dy/dx (Std Err)
Specification	1	2	3	4
Sex of the child (boy = 1)	–0.025 (0.018)	–0.009 (0.021)	–0.040 (0.024)	–0.004 (0.037)
Religion (Hindu = 1)	–0.002 (0.014)	–0.023 (0.016)	0.000 (0.018)	–0.040 (0.030)
Father’s age	–0.009 (0.004)	–0.013 (0.006)	–0.011* (0.005)	–0.021 (0.009)
Father’s literacy (yes = 1)	–0.127** (0.042)	–0.182** (0.055)	–0.154** (0.045)	–0.270*** (0.065)
Father’s occupation category (according to principal occupation specified by the respondent)	–0.005 (0.006)			–0.032 (0.017)
Mother’s age	0.008 (0.004)	0.011 (0.006)	0.007 (0.004)	0.013 (0.008)
Mother’s literacy (yes = 1)	–0.060 (0.032)	–0.072 (0.042)	0.024 (0.020)	–0.054 (0.063)
Working mother (yes = 1)	0.033 (0.017)			0.241*** (0.068)
Total income category (low income = 1, medium income = 2 and high income = 3)	–0.025 (0.018)			–0.629*** (0.050)
Bad sanitation		–0.540*** (0.053)		–0.019 (0.128)
No drinking water		–0.007 (0.058)		0.092** (0.037)
Exam fear		0.037 (0.041)		–0.005 (0.074)
Torture by the teacher		–0.051 (0.035)		–0.043 (0.058)
Different vernacular		0.035 (0.027)		0.136 (0.088)
No predicted return		0.030 (0.064)		0.028 (0.099)
Neighbourhood effect		–0.022 (0.029)
High tuition fees		0.030 (0.050)		0.098 (0.059)
Transaction cost			0.060 (0.043)	–0.055 (0.030)
Aggregate reasons			–0.046* (0.018)	0.029 (0.037)
Poverty (BPL cardholder = 1)			0.163** (0.057)	–0.019 (0.054)
Area (Jalpaiguri = 1)			0.030 (0.019)	2.651** (0.920)
Dependency ratio (earning member/total member)			0.381 (0.202)	–0.082 (0.079)
Y	0.436	0.554	0.469	0.599

Source: Authors’ calculation based on the field survey.

Notes: (1)*, ** and *** denotes level of significance.

(2) Y implies dependent variable.

(3) Pr = probability

Again, we observe that the numeric value of the coefficient shows that among all the significant explanatory variables, ‘father’s age’ scored the highest. There are studies (Al Samarai and Peasgood, 1998; Chugh, 2004; Mahmud, 2003) that show how the presence of an older father and many siblings influences (negatively by providing minimum resources to the child who according to the father is incapable of fetching a return from his investment in education) the decision to continue schooling of the wards. This essentially confirms our view that the ‘decision to continue schooling’ is extensively controlled by the ‘father’ in general—for both male and female school-goers. The father, as the principal bread earner of a family, dominates the major decisions in our society—patriarchy still remains a strong feature of the Indian society. In addition, this result confirms the view that a comparatively older father (with many siblings) is unlikely to continue schooling of his children as by then, his income is stretched among many members.

Another important constraint that hinders schooling of a child is household or family income of the child’s family. In our sample, we have categorised income into three distinct percentile values (low income = 1, medium income = 2 and high income = 3). The result shows a positive and significant relationship in the marginal effects estimated by the OPR. This indicates that if income rises from a low level to medium to high level for the responding families, the probability of discontinuation at early grades decreases accordingly.

In our sample, for children with working mothers, the probability of discontinuation rises only at higher grades. That is, for children whose mothers also earn and add to the pool of family income, parents prefer to continue schooling as far as it is possible for them to do so. This may be related to two issues: the rise in family income which supports the cost of schooling and what has been described as the rise in ‘bargaining power’ of such women in the household.

Estimation of the marginal effects after the OPR reveals a significant relation with bad sanitation. Here, the variable ‘bad sanitation’ (which has a negative sign) means dry toilets/latrines or in some cases, it may be an open-air grassland or bushes in the school compound or at a distance away. The variable ‘exam fear’ has a positive sign—as the children move to higher grades, many develop a fear of two specific subjects, English and mathematics (as was revealed in the field discussion notes), which eventually results in their retention in the same class for 2–3 years. Eventually, the parents withdraw their wards and engage them either at home or in some income-earning activity. This study shows that these fears originate from various factors, such as, the curriculum followed, mode of teaching, the atmosphere in which the child lives and the inner ability of the child to accept challenges and overcome them. These can gradually diminish a student’s confidence in their ability to understand a subject, which may spill over to other subjects as well, and ultimately end with the decision to discontinue.

The unaffordability of commutation costs reveals a significant relationship, with a positive sign. This indicates that as the influence of this factor rises, it becomes impossible for the parents to continue sending their child to school. It should be mentioned that commutation costs, which refer to the costs associated with travelling from home to school and back, increase as the child goes to higher grades. For example, as the child goes to the higher grades, her/his friends or peer groups also expand, and the need for money for extra spending also goes up. Further, with the problems of a different vernacular and low-quality teaching, they have to fall back on private tutors for support. Though the utility of a single teacher teaching all subjects may not be beyond debate, for families with low incomes, they get the chance to send their children to a private tutor in the hope that they will at least be able to cope with the curriculum.

7. CONCLUSION

In this article, we present a choice function of a rural household relating to their ward’s schooling or discontinuation of schooling at different levels, using an ordered probit analysis. This has been used for the empirical inference of the probable reasons for school dropout in different classes, taking into account their parents’ preference, which has been claimed by theoretical analysis. As the OPM considers the number of dropouts at different levels of schooling at a single point in time, it automatically captures the continuation of schooling at different levels and analyses the inter-temporal choice function of an individual as a parent.

We have considered income, dependency ratio, father’s occupation, mother’s occupation, transaction costs (total monetary cost to access school) and net returns from education as a dummy to capture the net monetary benefit from the ward’s regular and meaningful participation in schooling. In order to investigate how investment in education is intrinsically linked to the expected net benefit from participation in formal schooling from the household’s perspective and how the valuation of school-related factors affect access to schools in target locations, we have tested the significance level of each control variable on the probabilistic outcome of the categorical dependent variable.

According to the results, the unaffordability of commutation costs has a significant relationship with the decision to discontinue, with a positive sign. This reflects the fact that as the influence of the factor rises, the parents’ ability to continue schooling decreases further. It should be mentioned that although commutation costs here refer to the cost associated with travelling from home to school and back, in reality the costs associated increase as the child goes up the academic ladder. Two dummy variables on school-related factors, namely, bad sanitation facilities and the lack of drinking water, became significant determinants of the discontinuation of schooling. Therefore, it can be claimed that our proposition perceived from theoretical formulation is satisfactorily validated by the empirical estimation.

As observed, the reasons for discontinuation are varied across income classes and are complex in nature. The decision to discontinue schooling of the ward significantly depends on the father’s literacy, father’s age, nature of occupation of the head of the household, dependency ratio and average family income. Moreover, the mother’s participation in the decision-making process for continuation of schooling for a child largely depends on her education, participation in the workforce as well as her financial contribution towards the family. For the lower-income group, poverty and unaffordability of commutation costs are the main constraints of access to education. Apart from these basic causes, some external factors, such as distance to the school, exam fear and low-quality teaching, trigger the decision to discontinue, which primarily comes from poverty and unaffordability. The findings reveal that both family- and school-related factors were responsible and closely linked with each other. It was also seen that dropout is an outcome of not merely financial constraints but also the lack of information about future benefits from education and need for education in life. Identifying these binding factors and characteristics, however, is only a first step. This is crucial for understanding how decisions are made, the push and pull factors, how parents and schools approach child discontinuation in school and whether the interaction between these can be enhanced to pull children back into schools.

Footnotes

Acknowledgements

The authors are grateful to Saumen Chattopadhyay (ZHCES/JNU) for his useful comments and insights. User’s disclaimer applies.

Notes

References

Aitchison

, & Silvey

(1957). The generalization of probit analysis to the case of multiple responses’. Biometrika, 44, 131–40.

Al Samarai

, & Peasgood

(1998). Education attainments and household characteristics in Tanzania. Economics of Education Review, 17(4), 395–417.

ASER (2011). Annual Status of Education Report 2010. Mumbai: Pratham Resource Centre.

ASER (2012). Annual Status of Education Report 2011. Mumbai: Pratham Resource Centre.

Ashford

J.R.

(1959). An approach to the analysis of data for semi-quantal responses in biological essay. Biometrics, 15, 573–581

Basu

, Sanghamitra

, & Bhakra

(2010). Child labor and household wealth: Theory and empirical evidence of an inverted-U. Journal of Development Economics, 91(1), 8–14.

Bhalotra

, & Heady

(2003). Child farm labour: The wealth paradox. World Bank Economic Review, 17(2), 197–227.

Birdsall

, Levine

, & Ibrahim

(2005). Towards universal primary education: Investments, incentives, and institutions. European Journal of Education, 40(3), 337–49.

Brown

, & Park

(2002). Education and poverty in rural China. Economics of Education Review, 21(6), 523–41.

10.

Chaudhri

D.P.

, & Wilson

(2002). Nutritional poverty, school education and supply of child labour. In Ramachandran

& Massum

(Eds), Coming to grips with rural child work (pp. 113–42). Delhi: Institute of Human Development and United Nations World Food Programme.

11.

Chaudhri

D.P.

, & Jha

(2011). Child poverty and compulsory elementary education in India: Policy insights from household data analysis. Retrieved 20 April 2014, from https://crawford.anu.edu.au/acde/asarc/pdf/papers/2011/WP2011_04.pdf

12.

Chudgar

(2010). Female headship and schooling outcomes in rural India. World Development, 39(4), 550–60.

13.

Chugh

(2004). Why children drop out: Case study of a metropolitan city. New Delhi: Bookwell.

14.

Chugh

(2011). Dropout in secondary education: A study of children living in the slums of Delhi. Occasional Paper No. 37. NUEPA, New Delhi.

15.

Colclough

, Rose

, & Tembon

(2000). Gender inequalities in primary schooling: The roles of poverty and adverse cultural practice. International Journal of Educational Development, 20(1), 5–27.

16.

Dachi

H.A.

, & Garrett

R.M.

(2003). Child labour and its impact on children’s access to and participation in primary education: A case study from Tanzania. London: Department of International Development (DFID).

17.

Daily News and Analysis. (2012, April 1). India [online]. Retrived 30 November 2012, from http://www.dnaindia.com/india/report_rte-report-carddropout-rate-in-schools-falls_1669959

18.

DISE (2009–10). District Information System for Education, National University of Educational Planning and Administration (NUEPA). New Delhi and Secondary Education Management Information System (SEMIS: 2009–10).

19.

Drèze

, & Kingdon

G.G.

(1999). School participation in rural India. The Development Economics Discussion Paper Series. The Suntory and Toyota International Centres for Economics and Related Disciplines, London School of Economics.

20.

Francavilla

, Giannelli

G.C.

& Grilli

(2012). Mothers’ employment and their children’s school: A joint multilevel analysis for India. World Development, 2013, 41(C), 183–95.

21.

Glick

P.J.

, Sahn

D.E.

, & Walker

T.F.

(2014). Household shocks and education investment in Madagascar. IZA Discussion Paper No. 8731. December, Bonn.

22.

Gurland

, Lee

, & Dahm

P.A.

(1960). Polychotomous quantal response in biological assay. Biometrics, 22, 382–98.

23.

Hausman

A.J.

, Lo

, & Mackinlay

A.C.

(1991). An ordered Probit analysis of transaction stock prices. Draft Working Paper. Rodney L. White Center for Financial Research, University of Pennsylvania.

24.

Hoffman

A.M.

(2006). The capability approach and educational policies and strategies: Effective life skills education for sustainable development. Paris: Agence Française Du Développement (AFD).

25.

Jonathan

(2001, Summer). Higher education transformation and the public good. Kagisano Higher Education Discussion Series, 1(1), 28–63.

26.

Kambhampati

, & Rajan

(2005). Economic growth: A panacea for child labor. World Development, 34(3), 426–45.

27.

Kotwal

, & Rani

(2007). Causes of school dropouts among rural girls in Kathua district. Journal of Human Ecology, 22(1), 57–59.

28.

Maddala

G.S.

(1983). Limited dependent and qualitative variables in Econometrics. Cambridge, UK: Cambridge University Press.

29.

Mahmud

(2003). Female secondary stipend project in Bangladesh: A critical assessment, BIDS-Dhaka. Retrieved 30 June 2008, from http://portal.unesco.org/education/es/file_download.php/6c1807a68c58613a407957a6adbd8cc7Female+secondary+school+stipend+programme+in+Bangladesh.doc

30.

McCullagh

(1980). Regression models for ordinal data. Journal of the Royal Statistical Society, Series B, 42, 109–42.

31.

Mukudi

(2004). Effects of user-fee policy on attendance rates among Kenyan elementary school children. International Review of Education, 50(5–6), 447–61.

32.

Ray

(1999). Household size and child welfare in India. Working Paper. School of Economics, University of Tasmania, Australia.

33.

Ray

(2000). Child labor, child schooling and their interactions with adult labor: Empirical evidence for Peru and Pakistan. The World Bank Economic Review, 14(2), 347–67.

34.

Reddy

, & Sinha

(2010). School dropouts or pushouts? Overcoming barriers for the right to education: Create pathways to access. Research Monograph No. 40. NUEPA, New Delhi.

35.

Robeyns

(2003). Sen’s capability approach and gender inequality: Selecting relevant capabilities. Feminist Economics, 9(2–3), July–November, 61–92 (special issue on Gender and the Work of Amartya Sen, edited by B. Agarwal, J. Humphries & I. Robeyns).

36.

Rustagi

, & Mitra

R.G.

(2008). South Asia regional child poverty network meeting. Sponsored by UNICEF, New Delhi.

37.

Sen

(1985). Commodities and capabilities. Amsterdam: Oxford University Press.

38.

Sen

(1987). On ethics and economics. Oxford: Blackwell.

39.

Sen

(1992). Inequality re-examined. New York and Oxford: Clarendon.

40.

Sen

(1999). Development as freedom. New Delhi: Oxford University Press.

41.

Sen

(2003). The importance of basic education. Speech to the Commonwealth Education Conference, Edinburgh, as quoted in The Guardian, Tuesday, 28 October 2003. Retrieved 30 June 2008, from http://www.cis.ksu.edu/~ab/miscellany/basiced.html

42.

The Economic Times (2014, May 11) School dropout scenario in India ‘extremely high’: UNICEF. Retrieved 15 May 2015, from http://articles.economictimes.indiatimes.com/2014-03-11/news/48118325_1_dropout-unicef-full-cycle

43.

The Times of India (2012, April 1). Retrieved 30 November 2012, from http://articles.timesofindia.indiatimes.com/2012-04-01/india/31269828_1_rte-provisions-dropout-rate-teacher-student-ratio

44.

Tilak

J.B.G.

(1996). How free is ‘free’ primary education In India? Economic and Political Weekly, 31(6), 355–66.

45.

UNICEF (2010). The state of the world’s children. UNICEF.