An Analysis of Factors Affecting Private Expenditure on Education in India

Abstract

Private expenditure on education is a determining factor for ensuring an individual’s educational progress. Though the government provides a vast infrastructure at minimal cost, the individuals have to incur cost from their pockets for various purposes. In this study, we have analysed the various influences on private expenditure on education based on National Sample Survey 71st round conducted in 2014. We have found that household consumer expenditure, respondent’s age, medium of instruction dummy, private coaching dummy and household computer dummy affect private expenditure on education positively, and household size, rural dummy, female dummy, social group dummies, minority religion dummy and type of school dummy affect private expenditure on education negatively. The important policy implications are the tendency to spend less for the female child needs to be amended and the male and female child needs to be given same preference when it comes to expenditure on education. Family planning should be implemented effectively to keep the household size reasonably small for better educational access of an individual. The weaker social groups such as STs, SCs and OBCs and the minorities should be supported well by scholarships and stipends for furthering their education. The number of government institutions should increase to provide low-cost education to society. English medium schools should be made to offer more seats for the financially weak. Private coaching should be made as redundant as possible by improving teaching in the schools. For having computer in households, the financially weak should get some subsidy or may be community computer centres can cater to their needs at reduced cost or free of cost.

JEL: I25, I22

Keywords

Private expenditure on education national sample survey household expenditure consumer expenditure multiple regression

1. Introduction

The major portion of the Indian education sector is funded by the government. The main bulk of enrolments are in government institutions though private aided and unaided institutions cater to the public in various places.¹ Government institutions make available to the public a vast infrastructure for education at subsidized cost and we can find the expenditure from government budget documents. While the government subsidizes education heavily,² individuals also have to incur costs from their pockets at various stages in the form of course fees, coaching expenses, transport, books and stationery to avail themselves of this government infrastructure. Therefore, the need arises to study the impact of private or household expenditure on education.

Despite a recent surge in India in the number of privately financed educational institutions, the provision of public schooling by the state is still considered indispensable. Regardless of their parental income differences, students should be exposed to uniform curriculum and uniform teaching inputs to be equipped with standard skills and educational ability. Thus, formal schools overseen by the government should be the main source of acquiring education.

While budget documents provide information about the expenditure by the government, there is an increasing role of non-government organisations³ as well as individuals in the education sector. The generation of information about private expenditure on education by individuals through a specialized survey, therefore, has special significance for educational policymakers. The analysis of private expenditure on education can help us gauge the determining factors that go into private educational expenditure and how individuals can avail the best of education by facilitating it from money out of pocket.

The main aim of this study consists of the research question—is private expenditure on education affected by other socio-economic factors and what are those factors affecting the private expenditure on education?

We have formulated the working hypothesis with 13 variables (some of them are dummy variables) and we have hypothesized that these are the variables affecting the private expenditure on education. As we can see, our multiple regression will provide results that will show the validity of our hypothesis.

2. Literature Review

The problem of estimating private expenditure on education has received considerable attention in a few developed countries. Researchers such as Becker (1967, 1981) and McMahon (1984) also examined the impact of household investment decisions on education broadly within the sphere of family investment decisions. McMahon developed a future-oriented family utility function to explain household investment in education in the USA. His investment demand and supply functions included variables on expected non-monetary returns, family dispensable income, tax subsidies, student loans, family size, and order of birth, and the demand function was estimated with the help of academic scores and schooling level of parents. The expertise of children in studies and mothers’ education were found to be significant.

Williams (1983) tried to explain trends in private expenditure on education in Australia with the help of government expenditures, the real price index of the cost of education, real personal disposable income and the demographic term. His paper discusses that government and private expenditure on education cannot be taken into consideration separately because education is supplied by both sectors. Also, he found that the relationship between government and private expenditure on education is usually direct in Australia where the price of private schooling has been lowered substantially by government grants. Annual data have been used to justify the relationship.

For Australia, another significant study comes from Burke (2001) who overviewed expenditure for education by educational institutions in Australia during 1991–2001 using data from the Australian Bureau of Statistics. He found that demographic change has only a minor effect on private expenditure on education. There has been a decline in government expenditure on education and subsequent growth in private expenditure on education for Australia according to Burke. He recommended financial assistance for students based on sectors and tax reduction for educational expenses and donations.

In a paper using the data from the 1988 Family Expenditure Survey in Greece, Kanellopoulos and Psacharopoulos (1997) analysed private expenditure on education in Greece and found such expenditure amounts to 2.1% of total household expenditure. The aggregate private expenditure on education is found to be half of the public expenditure. The private expenditure is dominant for foreign language learning and private crammer schools preparing students for competitive exams to enter higher educational institutes. Private expenditure on education is found to vary substantially according to the location of the household, household’s total consumer expenditure, occupation and educational level of head of the household. The educational infrastructure in Greece is found to be much insufficient and unable to meet the demand for education in the country.

The case of private expenditure on education in Cyprus was examined by Andreou (2012) who analysed factors affecting the level of education using data from family expenditure surveys 1996–1997, 2002–2003 and 2008–2009. It showed the level of private expenditure on education with income across the years. Proportion of household expenditure on education ranged from 60% to 90% at primary and secondary levels. Variations of this proportion over income groups were found almost non-existent. The most profound factors affecting private expenditure on education in Cyprus were income, household’s number of children, region, head’s age and education. Notably, Cyprus has one of the highest percentages of GDP in public expenditure on education within the European Union.

We also have a study for Vietnam by Vu Quang (2012) using the Vietnamese Household Living Standards Survey 2006. Vu Quang investigated factors affecting private expenditure on children’s education. By using a Tobit Model, he found that household income has a significant positive effect on total educational expenditure. Further households where heads have a higher level of education or with professional jobs enhanced the probability of private expenditure. Households with more primary and secondary school level age children spend more, while households with pre-school and college-level age children spend less.

An important work on Egypt was done by data extracted from the 2010–2011 Egyptian household income survey. In this study, Rizk and Afriyie (2014) examined determinants of private expenditure on children’s education by applying the OLS and Generalized method of moment estimation techniques. They found expenditure on children’s education significantly increased with increasing level of household income. Household head’s level of education had increasing positive and significant effect on children’s educational expenditure. They recommended children from financially weak households are given subsidy.

A report released by OECD.org titled Education Indicators in Focus (2012) upheld the question whether increasing private expenditure especially in tertiary education was related to less government spending and less equitable access. The main thing was found that on average across OECD countries private expenditure on education contributes to a large portion but has not increased at the expense of public expenditure. A higher level of private expenditure in tertiary education is not associated with lower chances for backward students in tertiary education.

Private expenditure on education in India was widely overlooked with the argument that government bears the entire expenditure on education until some information was made available to substantiate private expenditure. A survey of the research available on private expenditure (e.g., Kothari, 1966; Panchamukhi, 1965; Shah, 1969) refutes the earlier understanding and establishes that private expenditure on education is sizeable.

The prejudicial view has changed subsequently with the publication of two papers by Tilak (2002) and Kingdon (2005) which have carefully analysed private expenditure on education in India with data from the National Council of Applied Economic Research (NCAER). Tilak (2002) focused on the important share of households in educational expenditure in rural areas, while Kingdon (2005) dealt with the gender bias in the case of private educational expenditure and concluded that much of the discrepancy arises in uneven enrolment for male and female rather than uneven expenditure after being enrolled.

The total costs of education comprising private costs, household costs including opportunity costs were measured by Panchamukhi (1965) and Kothari (1966). According to them, the total cost of education made up about 6% of GNP in 1959–1960. Taking a small sample of students from Baroda in Gujarat, Shah (1969) estimated costs including tuition and without tuition in elementary education incurred by families in the different income groups.

Works on private expenditure on education also emerged from Tilak (1987) through another sample-based survey in Andhra Pradesh that measured private expenditure to be 3.5% of GNP in India in 1979–1980. However, the estimates found from National Accounts Statistics showed private expenditure to be 2% of GNP in 1970–1971 which is considerably lower (see Tilak, 1985). This discrepancy is due to the fact that in the former study observations are based on institutional costs only. The private costs, both maintenance costs and opportunity costs are not taken into account. The later mentioned study estimates a modest household expenditure on education after proper adjustment.

Aggarwal (1998) found that even at lower levels of education households depended on private tuition and incurred extra expenditure thereof to improve the performance of children.

Tilak (2002) contributed a pivotal work which has long-term implications. It was argued in this paper that there is no ‘free education’ in India. Even Scheduled Castes, Scheduled Tribes and other low-income groups incurred considerable private expenditure including in elementary education which is supposed to be provided free by the government. Private expenditure covers the purchase of books, private coaching fees, stationery, uniforms and course fees. Examination fees and other fees are substantial even in government primary and upper primary schools. Children attending government schools incur the minimum cost followed by those who attend government-aided schools and private schools. The determinants of private expenditure are household characteristics, particularly household income, educational level of the head, household size, caste and religion. Generally, gender is believed to be a significant determinant and there is ample evidence in favour of the male child. But it is not found for all age groups. Availability of incentives like mid-day meals, uniforms, textbooks and stationery and the location of the school at a favourable distance are also quite important. Coefficients of elasticity also indicate that government and private expenditure do not substitute but rather complement each other. To mobilize household expenditure on education government needs to increase its own expenditure. Conversely, if government expenditure on education decreases, household expenditure may also decline resulting in severe underinvestment in education.

As far as the responsiveness of households and public bodies to educational need is concerned, Tilak (1991) has furnished important analysis based on time series data of household expenditure in India between 1960–1961 and 1984–1985 from National Accounts Statistics balanced with estimates on public investment in education. Contrary to Schultz (1981), Tilak has found that households are not more responsive to educational needs than public bodies. He found that a small increase in government income level leads to more than a proportional increase in government expenditure on education, while a similar increase in household income leads to less than proportional increase in household expenditure on education.

On the other hand, Shri Prakash and Chowdhury (1994) used a longer Time Series data based on National Accounts Statistics and found higher income elasticity for households (1.03) than for public authorities (0.53) but stated that education was a superior good for both private and public authorities. Apart from this, National Sample Survey Office (NSSO) and NCAER surveys done occasionally come up with important details but do not make possible any systematic comparisons over time.

Important work on the rate of returns to education was done by Kambhampati (2008) who estimated the rate of return to education separately for boys and girls in 33 states and Union Territories. Results clearly indicate that the rate of return on education is highly significant and positively influences the amount of private educational expenditure for both boys and girls. The impact of this variable is much larger at the secondary level and for girls.

Azam and Kingdon (2013) examined the intra-household allocation of private educational expenditure with available data from India Human Development Survey 2005 covering both rural and urban areas. Pro-male gender bias has been found in the primary school age group for several states but the occurrence of gender bias increases with age. It is greater in the middle school age group and still greater in the secondary school age group. The impact of gender bias in favour of males in private educational expenditure is considerably greater in rural than in urban areas. Households send sons to private schools that charge substantial fees and daughters to free of cost government-funded schools.

The lack of literature on private expenditure on education in India is being felt when public budgets are declining and household finances are relied upon with hope. It is argued that households have the ability and willingness to pay for education, not only in higher education but also in elementary education and this should be utilized to depend less on government funds. Public policies are formulated with this vision though counter-arguments are also present there.

Among the literature surveyed by us in the context of India, we have a frequent comparison of public versus private expenditure on education and private expenditure is upheld as a reliable mode of facilitating education, but there has been no exclusive study on the effects of various socio-economic characteristics on the private educational expenditure. It is imperative to address the equity and efficiency issue by understanding the pattern of educational financing. At the same time analysing the effects of regional, demographic and socio-economic household characteristics on private educational expenditure can lead to important policy implications. The contribution of this article is that it has brought up the dependence of private educational expenditure on socio-economically relevant factors. Unless we understand the role of these factors we will not be able to comprehend the full prospect of private expenditure on education.

We have further introduced the medium of instruction, access to private coaching and availability of household computers as dummy independent variables to see their effects on private expenditure on education. English medium schools have gained much popularity in India in recent times and enrolment has shifted to English medium from other medium. Therefore, the English medium dummy is important. Private coaching has come up in India as supplementing formal education and has become part of the educational process. Hence, a private coaching dummy is relevant for the analysis. Household computers have become quite common especially in urban households to aid in children’s education. Hence, we have taken the household computer dummy. These variables were not used in earlier literature.

3. Data and Source

The NSSO since its establishment in 1950 has been carrying out nationwide integrated large-scale sample surveys, employing scientific sampling methods, to generate data and statistical indicators on different socio-economic aspects. The National Sample Surveys are conducted by interviewing sample households selected through a scientific design and cover practically the entire geographical area of Indian Territory. In its 71st round of the survey, conducted during the period 1 January 2014 to 30 June 2014, NSSO carried out a survey on ‘Social Consumption: Education’. The last survey on this subject was conducted during the 64th round of NSS (July 2007–June 2008). The purpose of the 71st round survey was to collect information on the participation of persons aged 5–29 years in the education sector, the usefulness of educational provisions by the government and private sectors and how it affects current attendance in the institutions. Data were also collected on private expenditure on education, educational wastage in the form of dropping out and discontinuity and the reasons. The increasing use of information technology in every sphere of day-to-day life at present seems to be a vital advancement for the country. It is evident that more importance has to be placed on computer literacy as education and computer now complement each other. This survey also captured some information on various facets of ability to operate computer along with possession of computer in the household and access to internet facility.

The schedule of enquiry on Social Consumption: Education (Schedule 25.2) for the 71st round was constructed to gather information on (a) participation of persons aged 5–29 years in the educational sector, (b) private expenditure on education of household members including those who are residents of Students’ hostel at the time of the survey, (c) reasons of dropouts and discontinuity and (d) IT capability of the population of age 14 years and above.

A total of 4,577 villages and 3,720 urban blocks were surveyed as first-stage units (FSUs) in NSS 71st round for the central sample at all-India level. In addition to these, ‘State samples’ were also surveyed by State/UT Governments who participated in this survey. These are solely based on the central samples.

Stratification of the households was done on the basis of having any student (aged 5–29 years) receiving technical/professional or general education. For this particular survey, eight households were selected from each sample village/block. The total number of households, in which Schedule 25.2 was canvassed, was 36,479 and 29,447 in rural and urban India, respectively. The data have been collected from 1 January 2014 to 30 June 2014, a period of six months. Expenditure on education is related to the current academic session of study of a student for the basic course only. If the current academic session spanned over 12 months, then it was restricted to 12 months period.

Education has all along remained an important development priority, necessitating intervention by government for its easy access to different sections of society. Nevertheless, individuals attending educational institutions incur expenditure in the form of payment of course fees (including tuition fee, examination fees, etc.), purchase of books, stationery and uniforms, expenses on conveyance, private coaching and so on. This is referred to as private expenditure on education. In the current survey, information on this had been obtained from all the students.

Status of Private Expenditure on Education

We shall now turn our attention to the private expenditure on education that individuals and households have to bear and we have based our analysis on National Sample Survey 71st round data. Education would be inaccessible without bearing this private expenditure and this expenditure largely varies according to the financial condition of the households. Starting with the average expenditure per student by type of education we find from Table 1 that expenditure in General education is lower than Technical and Vocational streams and Technical/ Professional education demands most of the expenditure. Compared to 2007–2008 (National Sample Survey 64th round), the expenditure is higher in all the streams in 2014 (National Sample Survey 71st round).

Table 1.

Average Private Educational Expenditure (Rs.) Per Student by Type of Education.

Years	Type of Education
Years	General	Technical/Professional (Except Vocational)	Vocational
2014	6,788	61,841	27,676
2007–2008	2,461	32,112	14,881

Source: National Sample Survey 71st Round Report.

We compare the components of expenditure as defined earlier as a percentage to the total expenditure. From Table 2, we get that a major portion of General as well as Professional education is spent on course fee. The course fee for professional education is far greater than general education. A considerable amount is spent on books, stationery, etc., and lesser amount on transport. For general education share of expenditure in private coaching is quite higher in percentage terms than in professional education.

Table 2.

Major Components of Expenditure and Their Share (%) in Total Expenditure.

Component of Expenditure	General Education	Tech./Prof. Education (Including Vocational)
Course Fee	46	73
Books, Stationery, etc.	22	10
Transport	11	6
Private Coaching	15	3
OtherExpenditure	5	8
Total	100	100
Average Expenditure Per course (Rs.)	6,788	57,094

Source: National Sample Survey 71st Round Report.

Region-wise Private Expenditure on Education

We shall also look at the private expenditure region-wise. We have divided all the Indian States into six regions such as Northern (NO), North Eastern (NE), Central (CE), Eastern (EA), Western (WE) and Southern (SO). From the NSS data, it is quite difficult to segregate state-wise expenditure data and the tables will be clumsier also. Therefore, we have divided the Indian Territory into six regions and calculated the data region-wise for simplicity. The following states fall under each category:

Northern (NO): Haryana, Himachal Pradesh, Jammu & Kashmir, Punjab, Uttarakhand, Uttar Pradesh, Chandigarh, Delhi

North Eastern (NE): Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim, Tripura

Central (CE): Chhattisgarh, Madhya Pradesh

Eastern (EA): Bihar, Jharkhand, Odisha, West Bengal, Andaman & Nicobar Islands

Western (WE): Goa, Gujarat, Maharashtra, Rajasthan

Southern (SO): Andhra Pradesh, Karnataka, Kerala, Lakshadweep, Puducherry, Tamil Nadu, Telangana

Table 3 shows the average consumer expenditure for each region and the percentage of total consumer exp spent on education. South has the highest percentage of consumer expenditure spent on education at 11.2% followed by West (8.3%) and North (7.4%). SO and WE regions are the two highest spenders in the percentage of consumer expenditure spent on education. As the occurrence of professional education and private institutions higher in SO and WE the average percentage of expenditure is also higher in these regions. Clearly, professional education is not as widespread as general education due to higher expenses and Southern India leads the race in professional education enrolments.

Table 3.

Region-wise Expenditure on Education.

State Code	Total Yearly Exp. on Education (Mean) (Rs.)	Yearly Consumer Expenditure (Mean) (Rs.)	Percentage of Consumer Exp. Spent on Education (%)
CE	4,612	78,793	5.8
EA	4,958	78,438	6.3
NE	5,595	90,353	6.2
NO	7,761	104,942	7.4
SO	11,951	106,212	11.2
WE	9,753	116, 950	8.3

Source: Authors’ calculation from National Sample Survey 71st round unit-level data.

There are several factors that may affect private expenditure on education. In the 71st round National Sample Survey on education, we have got a list of all those factors and we shall now have an analysis of what really affects private expenditure.

4. Methodology

Model Specification and Estimation

Private expenditure plays a key role in the educational achievement of an individual. Households according to expenditure capacity spend different amounts that take education to various levels. It has been seen that different factors affect the private expenditure capacity of households and National Sample Surveys incorporate a host of indicators for which data are collected.

Private expenditure on education depends on various factors. In the National Sample Survey 71st round survey, we have different indicators based on which we shall run the analysis.

For this analysis, we have done multiple regression with the help of STATA 12 and discussed the regression coefficients for finding out how the dependent variable responds to explanatory variables or the predictors of the regression. In data analysis, we generally use OLS for estimating the unknown parameters in a linear regression model. The aim is to minimize the differences between the collected observations in some arbitrary datasets and the responses predicted by the linear approximation of the data.

Multiple regression generally represents the relationship between one dependent variable and more than one independent variable. Here, a dependent variable is presented as a function of more than one independent variable with matching coefficients and the constant term. Multiple regression derives its name from the fact that it has two or more independent variables. The aim of multiple linear regression is to formulate the linear relationship between the dependent variable and the independent variables. Practically, multiple regression is an expanded form of ordinary least-squares (OLS) regression that has one independent variable. Simple linear regression (OLS) has only one independent variable and one dependent variable whereas multiple regression has more than one independent variable. In simple linear regression, a dependent variable is predicted by one independent variable. In multiple regression, the dependent variable is predicted by two or more independent variables. A multiple regression model is found to be more precise than a simple regression model in predicting the relationships among variables. Multiple regression is utmost helpful in the purpose of planning and prediction.

Here we have further taken a double log model under the framework of multiple regression. We have to take a double log model here because we have to scale down the continuous variables such as total expenditure which is a dependent variable and monthly consumer expenditure, age and household size which are independent variables. All other independent variables are dummies. Therefore, we have log expressions on both sides of the equation. A regression model in which the dependent variable and at least one independent variable are expressed in log form is called a double log linear model. It gives the elasticity of the dependent variable with respect to the independent variable(s). When natural logarithm is applied to an exponential equation, it converts the equation to a linear equation in logarithms.

Here we have taken for regression the following dependent and independent variables:

Dependent variable: log of total yearly educational expenditure

Independent Variables: log of consumer expenditure in a month, log of age, log of household size, rural (dummy), female (dummy), social group (SC, ST and OBC dummy), religion (minority dummy), type of institution (government dummy), medium of instruction (English dummy), whether taking private coaching (dummy), whether the household has computer (dummy)

We have done several regressions taking continuous and dummy explanatory variables in parts. The last one (Regression 6) is the final regression which is of utmost importance.

Regression 1: Only continuous variables

1.
lntot_exp_i ₌ β₁ + β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + u_i

Regression-2: Continuous variables + sector and sex dummies
2.
lntot_exp_i = β₁+ β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + β₅rural_i + β₆female_i + u_i

Regression-3: Continuous variables + sector and sex dummies + social group dummies
3.
lntot_exp_i = β₁+ β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + β₅rural_i + β₆female_i + β₇ST_i + β₈SC_i + β₉OBC_i + u_i

Regression-4: Continuous variables + sector and sex dummies + social group dummies + religion dummy
4.
lntot_exp_i = β₁+ β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + β₅rural_i + β₆female_i + β₇ST_i + β₈SC_i + β₉OBC_i + β₁₀minority_i + u_i

Regression-5: Continuous variables + sector and sex dummies + social group dummies + religion dummy + type of institution dummy
5.
lntot_exp_i = β₁+ β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + β₅rural_i + β₆female_i + β₇ST_i + β₈SC_i + β₉OBC_i + β₁₀minority_i + β₁₁Govt_i + u_i

Regression-6: (Final Regression) All variables
6.
lntot_exp_i = β₁+ β₂lnhh_cons_exp_i + β₃lnage_i + β₄lnhhsize_i + β₅rural_i + β₆female_i + β₇ST_i + β₈SC_i + β₉OBC_i + β₁₀minority_i + β₁₁Govt_i + β₁₂Eng_i + β₁₃PC _i + β₁₄Comp_i + u_i

(where β ₁ _… β ₁₄ are the different regression coefficients and u _i is the disturbance term).

This is the final regression model (6) which is the basis of our results.

The error term in this model is distributed normally.

We have taken the regressions first in parts and then in full because we want to show that with each addition of variables the R-squared (see Table 5) increases gradually. This signifies the improvement in the goodness of fit with the addition of variable specifications.

Factors Affecting Private Expenditure on Education

We shall here discuss in brief the variables in our regression and the reason for choosing those variables for the study.

Here, the dependent variable is ‘lntot_exp’, that is, the log of yearly total private expenditure on education. Total yearly private expenditure on education has been chosen as the Dependent Variable because we want to see how it responds to the changes in various explanatory variables. This will enable us to predict the factors affecting private expenditure on education.

Among the explanatory variables we have taken foremost ‘lnhh_cons_exp’, that is, log of monthly consumer expenditure of the household.⁴ Monthly household consumer expenditure is taken as an explanatory variable since a major portion of it is subjected to educational expenditure. We can evaluate the change in private educational expenditure due to a change in monthly household consumer expenditure and know the proportional expenditure on education. In the literature also, we have found relevant mention of private educational expenditure as a portion of household consumer expenditure (Tilak, 2002). lnhh_cons_exp is expected to have a positive impact on a dependent variable, that is, the sign of the coefficient is hypothesized to be positive (+). This is because with the increase in monthly household consumer expenditure the total expenditure on education is expected to increase.

The next variable we have taken is ‘lnage’, that is, the log of respondent’s age.⁵ A person’s age is taken as an explanatory variable because with the advancement of age expenditure on a person’s education may change. We want to gauge this change through this variable. Here, lnage is expected to have positive effect on the dependent variable, that is, the sign of the coefficient will be positive (+). This is because with the advancement of age a person is assumed to reach higher levels of education and so the total private expenditure on education is expected to increase.

An important variable taken as a determinant is ‘lnhhsize’, that is, log of household size. Household size or the number of persons in a household is taken as an explanatory variable because with the increase of household size the expenditure on an individual’s educational expenditure is due to change as there is an extra burden on the family. Household size has been taken as an explanatory variable by Tilak (2002) also. Here lnhhsize is expected to have negative impact on the dependent variable, that is, the sign of the coefficient is hypothesized to be negative (−). This is because with an increase in household size the educational expenditure of each individual is expected to decrease.

We have taken ‘rural’ sector as a dummy explanatory variable in our regression since according to location or sector the private expenditure on education may vary as there may be differences in financial condition among the urban and rural masses leading to different private educational expenditure. It is assumed, rural = 1 if rural and = 0 if urban. The impact of sector on the dependent variable is expected to be negative, that is the sign of coefficient is hypothesized to be negative (−). This is because for the addition of a person in the rural sector the total expenditure on education of that person is expected to fall compared to a person in urban sector since rural sector has a lower financial capacity for education.

We have also taken a dummy for ‘female’ since according to gender there can be differences in expenditure on a person’s education. Particularly for India, we have found a bias towards male child in private educational expenditure. Dummy for sex has also been used by Tilak (2002). It is assumed that female = 1 if female and = 0 if male. The impact of female dummy on dependent variable is expected to be negative, that is, the sign of coefficient is hypothesized to be negative (−). This is because for the addition of a female the total expenditure in education of that person is expected to be lower than if a male would have been added in her place.

Scheduled Tribe is taken as a dummy explanatory variable as ‘ST’ since this community is one of the weaker sections in the society so we want to see how their participation affects the private expenditure on education. It is assumed that ST = 1 if ST and = 0 if Others. The impact of ST dummy on dependent variable is expected to be negative, that is the sign of coefficient is hypothesized to be negative (−). This means that the addition of an ST person would bring down the total educational expenditure of that person than if a general or upper caste person would have been added.

By the same logic as above Scheduled Castes have been taken as a dummy explanatory variable ‘SC’ to gauge the influence of this weaker section on private educational expenditure. It is assumed that SC = 1 if SC and =0 otherwise. The impact of SC dummy is expected to be negative on dependent variable, that is the sign of coefficient is hypothesized to be negative (−). This means that the addition of an SC person would bring down the total educational expenditure for that person than if a general or upper caste person would have been added.

By the same logic as for ST and SC, ‘OBC’ is also taken as a dummy explanatory variable as another weaker section of the society. We have found social group dummies being used in Tilak (2002). It is assumed that OBC = 1 if OBC and =0 otherwise. The impact of OBC dummy on dependent variable is expected to be negative, that is the sign of coefficient is hypothesized to be negative (−). This indicates that the addition of an OBC person would bring down the educational expenditure for that person than if a general caste person would have been added.

Religious minorities are taken as a dummy explanatory variable ‘minority’ since we want to see the effect of minorities’ participation on private educational expenditure because there is a difference from the majority in financial and social status. Previously religion dummies have been used by Tilak (2002). It is assumed that minority = 1 if minority (other than Hindu) and =0 otherwise (Hindu). The expected impact of minority dummy on the dependent variable is negative, that is the sign of the coefficient is hypothesized to be negative (−). This indicates that the addition of a person from minority community would bring down the educational expenditure than if a majority (Hindu) person would have been added.

Among types of schools, Government Schools are taken as dummy ‘Govt’ because there is much difference in expenditure in government and private and private-aided schools. We want to see the difference for joining a government-run school in educational expenditure. Type of institution dummy has been considered by Tilak (2002). It is assumed that Govt = 1 if Government institution and =0 if otherwise (Private and Private-aided). The expected impact of Govt dummy on the dependent variable is negative, that is the sign of the coefficient is hypothesized to be negative (−). This indicates, with the addition of one person enrolled in a government institution, the educational expenditure will decrease for that person than if a person enrolled in a private institution would have been added.

Among the medium of instructions, English medium is taken as an explanatory dummy variable ‘Eng’ because it is the main medium of standard educational practice and we want to see how it affects the private expenditure on education. It is assumed that Eng = 1 if the medium is English and =0 otherwise. The expected impact of English dummy on the dependent variable is positive, that is the sign of the coefficient is hypothesized to be positive (+). This indicates that, with the addition of one person enrolled in English medium the additional expenditure for that person will increase than if a person enrolled in other medium school would have been added.

Private coaching is a prevalent practice in the Indian educational system. Due to private coaching, extra cost is incurred on an individual’s educational expenditure. Therefore, we have taken access to private coaching as dummy explanatory variable ‘PC.’ It is assumed that PC = 1 if a person avails private coaching and =0 otherwise. The expected impact of PC on the dependent variable is positive, that is the sign of the coefficient is hypothesized to be positive (+). This indicates that, with the addition of one person availing private coaching, the additional expenditure on education for that person will increase than if a person without access to private coaching would have been added.

Having a computer in the household has become quite common after the advent of personal computers. Most of the well-to-do households, particularly in urban areas are having computers for their children to aid in education and it incurs the extra educational expenditure. While in rural areas having a personal computer is rare and students have to let do with community centres or educational institutions. To see how having household computer affects educational expenditure we have taken the availability of household computers as a dummy explanatory variable ‘Comp.’ It is assumed that Comp = 1 if household has computer and =0 otherwise. The expected impact of Comp dummy on the dependent variable is positive, that is the sign of the coefficient is hypothesized to be positive (+). This indicates that, with the addition of one person with household computer, the additional expenditure for that person will increase than if a person without household computer would have been added.
5. Empirical Results

Let us first check the summary descriptive statistics of all the dependent and explanatory variables in Table 4.

Table 4.

Summary Statistics.

Variables	Observations	Mean	Std. Deviation	Min.	Max.
lntot_exp	252,835	8.463422	1.733484	2.302585	14.88469
lnhh_cons_exp	310,724	8.97974	0.616557	4.60517	12.42922
Lnage	307,532	3.093203	0.8623859	0	4.718499
lnhhsize	310,827	1.645501	0.4234064	0	3.465736
Rural (dummy)	310,827	0.5737307	0.4945347	0	1
Female (dummy)	310,827	0.4834007	0.4997252	0	1
ST (dummy)	310,827	0.1357765	0.3425516	0	1
SC (dummy)	310,827	0.1621127	0.3685547	0	1
OBC (dummy)	310,827	0.3941324	0.4886643	0	1
Minority (dummy)	310,827	0.2422216	0.4284284	0	1
Govt (dummy)	310,827	0.4270221	0.4946464	0	1
English (dummy)	310,827	0.3276742	0.4693661	0	1
PC (dummy)	310,827	0.2096986	0.4070942	0	1
Comp (dummy)	310,827	0.2243434	0.4171499	0	1

Source: Calculations by the author.

Let us now present the estimates of the regressions in the form of Table 5.

Table 5.

Regression Estimates. Dependent Variable: lntot_exp

	Regression 1	Regression 2	Regression 3	Regression 4	Regression 5	Regression6 (Final)
const.	−5.45^227**(−86.37)	−3.57^***(−50.27)	−2.56^***(−34.66)	−2.51^***(−34.03)	.63^***(9.05)	1.89^***(27.73)
lnhh_cons_exp	1.73^***(241.83)	1.53^***(191.95)	1.43^***(175.61)	1.43^***(175.42)	1.11^***(144.48)	0.83^***(104.65)
lnage	.17^***(34.10)	.17^***(34.84)	.16^***(34.35)	.16^***(33.86)	.15^***(36.73)	0.13^***(35.03)
lnhhsize	−1.46^***(−127.13)	−1.29^***(−111.29)	−1.24^***(−107.49)	−1.22^***(−105.65)	−1.08^***(−103.94)	−0.78^***(−77.64)
rural (dummy)	–	−.57^***(−68.05)	−.53^***(−64.07)	−.54^***(−65.07)	-.30^***(−41.24)	−0.09^***(−14.22)
Female (dummy)	–	−.03^***(−3.50)	−.03^***(−3.41)	−.03^***(−3.42)	−.03^***(−3.89)	−0.02^***(−3.44)
ST (dummy)	–	–	−.64^***(−45.87)	−.66^***(−46.86)	−.49^***(−37.63)	−0.33^***(−26.91)
SC (dummy)	–	–	−.50^***(−41.33)	−.53^***(−43.23)	−.40^***(−35.33)	−0.25^***(−24.39)
OBC (dummy)	–	–	−.25^***(−27.27)	−.26^***(−28.48)	−.26^***(−32.19)	−0.14^***(−19.36)
Minority (dummy)	–	–	–	−.13^***(−13.97)	−.09^***(−10.10)	−0.11^***(−14.45)
Govt (dummy)	–	–	–	–	−1.25^***(−162.54)	−1.06^***(−139.37)
English (dummy)	–	–	–	–	–	0.71^***(87.77)
PC (dummy)	–	–	–	–	–	0.83^***(123.68)
Comp (dummy)	–	–	–	–	–	0.43^***(47.26)
R-squared	.35	.37	.38	.39	.50	0.58
F	23,298.92	16,862.05	11,712.04	10,534.47	14,207.25	14,558.52
N	250,701	250,701	250,701	250,701	250,701	250,701

Source: Calculated by Author from National Sample Survey 71st round unit-level data.

Notes: ‘t’ values are given in parentheses.

*** Significant at 1% level.

** Significant at 5% level.

* Significant at 10% level.

It should be noted here that from Regression 1 to 6 at every stage the R-squared increases and is the highest for the sixth or final regression. R-squared is a statistical measure indicating how close the data are to the fitted regression line. A low R-squared value indicates that the independent variable is not explaining much in the variation of the dependent variable. The higher the R-squared value the better is the goodness of fit. 100% R-squared indicates perfect fitting with the regression line. Hence, Regression 6 with the highest R-squared among the regressions is the optimal model for us.

Also, in Table 5 F indicates the ratio of the mean regression sum of squares divided by the mean error sum of squares. Its value will range from zero to an arbitrarily large number. The value of Prob(F) is the probability that the null hypothesis for the full model is true (i.e., all of the regression coefficients are zero). In other words, it indicates the probability that all the coefficients in our regression output are actually zero.

Now, our first responsibility is to check whether the model is affected by multicollinearity.

Multicollinearity is the problem associated with high correlations between two or more independent variables. In such a case, one independent variable can be used to predict the other. This creates unnecessary information, biasing the results in a regression model. Examples of correlated independent variables are a person’s height and weight.

Multicollinearity can be measured by the variance inflation factor (VIF), which assesses how much the variance of an estimated regression coefficient increases if the independent variables are correlated. A VIF greater than 5 indicates a high correlation that may not be acceptable. A small VIF value indicates low correlation among variables and under ideal conditions VIF should be less than 5.

From collinearity statistics here, we have found all the variance inflation factors (VIFs) to be less than 5 and mean VIF also less than 5 (Table 6), which indicates the regression model is free from multicollinearity problem.

Table 6.

Variance Inflation Factors.

Variable	VIF	1/VIF
lnhh_cons_exp	1.81	0.553687
lnage	1.03	0.974093
lnhhsize	1.31	0.762419
rural	1.32	0.759899
female	1.00	0.999079
ST	1.34	0.745450
SC	1.52	0.656528
OBC	1.54	0.647362
minority	1.05	0.949638
Govt	1.43	0.699504
English	1.60	0.623860
PC	1.04	0.960729
Comp	1.36	0.736012
Mean VIF	1.34

Source: Calculations by the author.

Further, we have to check the regression model for heteroscedasticity problem.

In statistics, heteroscedasticity arises when the standard deviations of a predicted variable, observed over different values of an independent variable are non-constant. In the case of heteroscedasticity, the representation of the residual errors in a scatter diagram is that they will spread out over time. Heteroscedasticity is the absence of homoscedasticity.

The occurrence of heteroscedasticity is a problematic aspect in regression analysis and the analysis of variance, as it nullifies statistical tests of significance that assume that the modelling errors all have the same variance. Though the ordinary least squares estimator is still unbiased in the occurrence of heteroscedasticity, it is inefficient and generalized least squares should be used in such a case.

Here we have taken a multiple regression model (model 6: final regression) and we have used sampling weights from NSS unit-level data. For such a model, there is no specific econometric test for heteroscedasticity and the standard Breusch-Pagan/Cook-Weisberg test for heteroscedasticity which can be done with the command (hettest) in STATA software is not appropriate here. Therefore, we have used a residual versus fitted plot to identify if there is a heteroscedasticity problem. The residuals of model 6 (final regression) are plotted in Figure 1.

To find out if there is heteroscedasticity present, we have drawn a scatter plot with regression fitted values in the ‘X’ axis and regression residuals in the ‘Y’ axis (Figure 1) and we can see that that the data are densely centred in the scatter plot, which indicates homoscedasticity. Therefore, the model is free from heteroscedasticity.

Figure 1.

Source: Drawn on STATA 12 by the author.

6. The Results of the Regression

We have done the multiple regression (Final Version: Regression 6) and all the explanatory variables have come out significant at 1% level.

Household’s monthly consumer expenditure is positively related to private expenditure on education. This indicates an increase in household’s monthly consumer expenditure leads to an increase in total private expenditure. As household’s monthly consumer expenditure increases a portion of it contributes to private expenditure on education of a member, thus increasing the private expenditure on education. This is in accordance with the findings of Tilak (2002).

Age is positively related to private expenditure on education. With the advancement of individual’s age, the private expenditure on education increases. This is justified as we assume the cost of education increases with advancement of the educational level.

Household size also is negatively related to private expenditure on education. So, with the increasing household size the private expenditure on education of a household member decreases. This signifies the problem of family size control that often disturbs educational pursuits in Indian households. The result is in accordance with Tilak (2002).

Rural sector dummy is negatively related to private expenditure on education. In rural sector, the financial affordability of individuals and households is lower than in the urban sector.

The Female dummy is negatively related to private expenditure on education. For India, we have found a bias in favour of the male child in educational expenditure. Families are more prone to spend on a male child’s education than a female child’s education and female child’s education is often viewed as unnecessary. Our result rightly justifies the previous finding of bias. This result is in accordance with Tilak (2002).

For social groups, all three dummies (SC, ST and OBC) are negatively related to private expenditure on education. The weaker sections of the society represent a vulnerability in educational expenditure that needs to be addressed by the government. The results are similar to those found by Tilak (2002).

In religion we have taken minority dummy and clubbed all the minorities (non-Hindu) there. Minority dummy has a negative coefficient with private expenditure on education. In our case minority dummy represents all minorities and the negative relation with the dependent variable points out the vulnerability of the minority communities in terms of educational expenditure. However, Tilak (2002) has used two dummies for Hindu and Muslim. While the coefficient for Hindu is positive and significant, the coefficient for Muslim is statistically not significant there.

Type of Institution (Government dummy) has negative coefficient with the dependant variable. For government schools, a person incurs less expenditure than private schools and this is an indication of the popularity of government institutions. We have found similar results with the type of school dummy in Tilak (2002).

Medium of Instruction (English dummy) is positively related to the dependent variable. This dummy independent variable introduced by us clearly shows the higher cost of being taught in English medium than other medium.

Private Coaching dummy is positively related to the dependent variable. This dummy independent variable introduced by us shows the increasing burden on household’s educational expenditure due to private coaching.

Household computer dummy has positive coefficient with private expenditure on education. So, the households having personal computers incur more educational costs than those with no computers. This dummy independent variable introduced by us helps us understand the need for community infrastructure of computer and the internet for areas where people’s financial affordability is low.

7. Conclusion and Policy Implications

We find a gender bias when it comes to household’s expenditure on female child’s education compared to male child’s education. The tendency to spend less on the female child needs to be amended and the male and female child needs to be given the same preference when it comes to expenditure on education.

The rural sector households are found to have less financial affordability for education. Therefore, the government needs to offer financial assistance to individuals from the rural sector for pursuing higher education. This assistance can be in the form of scholarships or stipends or fees waiver in educational institutions.

We have also found household size is negatively related to private expenditure on the education of an individual. With increasing household size, the expenditure on an individual’s education suffers. Here comes the effectiveness of family planning and keeping the household size reasonably small for better educational access of an individual.

For SCs, STs and OBCs, we have found negative relation with private expenditure on education. These are the weaker sections of the society and though they also have an inclination towards education they cannot spend as much as the privileged classes. Therefore, they should be supplemented well by scholarships, stipends, etc. for furthering their education.

Minority religion dummy has negative relation with private expenditure on education. Clearly, there is a lack of capability in minorities to spend or education. The minorities should be taken care of on the educational front by the government just as the weaker social groups.

Government institutions have less expenditure (showing negative coefficient) than non-government institutions and therefore are more affordable for learners. More emphasis should be on enrolling in government educational institutions and at the same time the number of government institutions should increase to provide low-cost education to society.

For English Medium, Private Coaching and Household Computer possession the coefficients are positive and significant. Clearly, the availability of these three facilities induces increasing private expenditure on an individual. Here, the disparity between the well-to-do who can spend more in these three areas and the financially weak have to be eradicated. English medium schools should be made to offer more seats to the financially weak, private coaching should be made as redundant as possible by improved teaching in the schools. Having household computer induces higher expenditure on education since it incorporates cost of the computer, internet connectivity and electricity usage. It is seen as aiding formal educational pursuit with an extra cost for the machinery that has high benefits. Of course, it is a matter of affordability for households to keep a personal computer. For having computer in households the financially weak could get some subsidy or may be community computer centres can open to cater to their needs at a reduced cost or free of cost.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Notes

ORCID iDs

Ananda Mukherjee

Sarbajit Sengupta

References

Andreou

S. N.

(2012). Analysis of household expenditure on education in Cyprus. Cyprus Economic Policy Review , 6(2), 17–38.

Azam

, & Kingdon

(2013). Are girls the fairer sex in India? Revisiting intra-household allocation of education expenditure. World Development , 42(C), 143–164.

Becker

G. S.

(1967). Human capital and the personal distribution of income: An analytical approach . Institute of Public Administration.

Becker

G. S.

(1981). Altruism in the family and selfishness in the market place. Economica , 48(189), 1–15.

Burke

(2001). Trends in educational expenditure . In Research to Reality: Putting VET Research To Work. Proceedings of the Australian Vocational Education and Training Research Association (AVETRA) Conference (4th ed, Adelaide, Australia, 28–30 March 2001).

Kambhampati

(2008). Does household expenditure on education in India depend upon the returns to education? ResearchGate.

Kingdon

(2005). The progress of school education in India. Oxford Review of Economic Policy , 23(2),168–195.

Kothari

V. N.

(1966). Factor cost of education in India, Indian Economic Journal , 13(5) (April–June), 631–647.

McMahon

W. W.

(1984). Why families invest in education. In Sudman

& Spaeth

M. A.

(Eds.), The collection and analysis of economic and consumer behaviour data: In memory of Robert Ferber (pp. 75–91). Bureau of Economic and Business Research, University of Illinois at Urbana–Champaign.

10.

Nam

(2007). Sakyoyukbi Kyumoei Sikyeol Choyi Bunseok [Time-series analysis of the scale of private supplementary tutoring]. Kyoyuk Jaejeong Kyongjie Yeonku [Economic study of educational finance], 16(1), 57–79.

11.

NSS 71st round survey report, Key indicators of Social Consumption in India: Education.

12.

Panchamukhi

P. R.

(1965). Educational Capital in India. Indian Economic Journal , 12(3) (January–March), 306–314.

13.

Prakash

, & Chowdhury

(1994). Expenditure on education: Theory, models and growth . National Institute of Educational Planning and Administration.

14.

Psacharopoulos

, & Papakonstantinou

(2005). The real university cost in a ‘free’ higher education country. Economics of Education Review , 24(1), 103–108.

15.

Rizk

, & Afriyie

J. O.

(2014). Determinants of household expenditure on children’s education in Egypt. International Journal of Education Economics and Development , 5(4), 332–360.

16.

Schultz

T. W.

(1981). Investing in people . University of California Press.

17.

Sen

(2002). Introduction. Pratichi education report: A study in West Bengal . Pratichi Research Team.

18.

Shah

K. R.

(1969). Private cost of elementary education. In Pandit

H. N.

(Ed.), Measurement of cost productivity and efficiency of education (pp. 57–74). National Council of Educational Research and Training.

19.

Tilak

J. B. G.

(1985). Analysis of costs of education. Occasional Paper No. 10. National Institute of Educational Planning and Administration.

20.

Tilak

J. B. G.

(1987). Economics of inequality in education . SAGE Publications.

21.

Tilak

J. B. G.

(1991). Family and government investments in education. International Journal of Educational Development , 11(2), 91–106.

22.

Tilak

J. B. G.

(2002). Determinants of household expenditure on education in rural India, Working Paper Series No. 88 , National Council of Applied Economic Research.

23.

Vu Quang

(2012). Determinants of educational expenditure in Vietnam. International Journal of Applied Economics , 9(1), 59–72.

24.

Williams

R. A.

(1983). Interaction between government and private outlays [Discussion Paper No. 79]. Australian National University, Centre for Economic Policy Research.