Who Gains More from Education? A Comparative Analysis of Business, Farm and Wage Workers in India

Abstract

The economics literature on returns to education has focused largely on wage workers, thereby ignoring a sizable section of the workforce which is self-employed. This paper presents the estimates of private returns to education for business, farm and wage workers in India using a nationally representative household survey. The paper addresses the sample-selectivity issue arising due to endogenous sector allocation in the earnings equation using the multinomial-selection approach. Our results show that the average rate of return to education is higher for wage workers followed by business and farm workers. Focusing only on wage workers would provide an overestimate of returns by 30 per cent for business workers and by 40–50 per cent for farm workers. Further, the profile of returns across the education ladder varies perceptibly for the three type of workers with higher education being more rewarding for wage workers.

Acknowledgements

We are grateful to two anonymous referees of this journal for their valuable comments. We also thank the IHDS Team at NCAER, New Delhi. Institutional support provided by the Indian Institute of Management Udaipur, Udaipur and Indian Institute of Technology Delhi, New Delhi is gratefully acknowledged. The data used in this paper is publicly available at the ICPSR website (ICPSR 36151; http://www.icpsr.umich.edu/icpsrweb/DSDR/studies/36151), and the code is available on request. The usual disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. In developing economies, the better jobs are usually in wage employment and not in self-employment. Within wage employment, regular jobs are considered better than casual ones (Fields, 2011). In India, 53.5 per cent of the workforce was self-employed in 2009–2010 and the remaining, 46.5 per cent, was in wage employment (Chowdhury, 2011). Of the latter, only 8.5 per cent were regular wage earners and 38 per cent were casual workers.

2. The informal sector comprised the workers with poor working conditions and lack of social security and included both wage workers and self-employed entrepreneurs.

3. The changes in India’s occupational profile resemble the typical structural change witnessed by emerging economies. With increases in per capita income, the share of wage and salaried workers in the workforce also goes up (see also, Gindling & Newhouse, 2014).

4. The IIA assumes that the odds for any pair of outcomes are determined independently of the other available outcomes.

5. The IHDS is one of the first nationally representative surveys in India to gather data on income of individuals.

6. Trimming distributions at the extremes is common in the development literature. For instance, to avoid outliers the income distribution is truncated at the top (0.5 percentile) for the purpose of Inequality-adjusted HDI calculations (UNDP, 2010, p. 232).

7. If an individual works in more than one employment category, total hours is the sum of hours spent in all employment categories.

8. This assumes that an individual works eight hours a day at least for 25 days in a year.

9. This is based on the assumption that an individual starts schooling at the age of five (Duraisamy, 2002).

10. States are administrative units in India, and the country is divided into 29 states and seven Union Territories.

11. The difference in the average earnings of business and wage workers is marginal for males.

12. A nationally representative survey of Indian farmers indicated volatility of farm yield as a major source of discontent with farming as a profession (Agarwal and Agrawal, 2017).

13. The need for selectivity-corrected estimates arises from the selection built in the problem and we get evidence indicating the selection.

14. For female business workers, the OLS estimate is smaller than the corresponding selectivity-corrected.

15. The coefficient in Table 2, multiplied by 100, is the percentage increase in earnings for an additional year of schooling.

16. Occupational segregation by gender, that is, ‘males being disproportionately attracted to manual occupations where schooling is poorly valued and females being disproportionately attracted to professional ones where it has a high return’ significantly explains the gender differential in returns (Dougherty, 2003, p. 21).

17. Results with all control variables (selectivity-corrected) are reported in Appendix Table A2.

18. Since the dependent variable is in the logarithmic form, the marginal effect of a dummy variable is (e^c°^efficient − 1).

19. Among females, farm workers have higher returns than wage workers for the first two educational levels.

20. The marginal returns for different levels can be obtained by comparing the coefficients of the adjacent dummy variables: $r_k=({\beta }_k-{\beta }_{k-1})/\mathrm{\Delta }{n}_k$ where, β_k is the coefficient of kth education level, β_k-1 is the coefficient of the education level preceding k, and ∆n_k is the difference in years of schooling between kth and (k-1)th schooling levels. Following the literature, we have used ∆n_k = 3 for Primary, Middle, and Graduation and 2 for Secondary and Higher Secondary levels. Since primary school children do not forego earnings during their entire study-period, it is not advisable to assign them six (or five, depending on country) years of forgone earnings (Psacharopoulos, 1994). Therefore, ∆n_k is taken to be three years for computing marginal returns for the Primary.

21. As Lazear (2005) puts it, ‘entrepreneurs differ from specialists in that entrepreneurs have a comparative disadvantage in a single skill but have more balanced talents that span a number of different skills’ (p. 650).

22. Van der Sluis et al. (2008) find that non-whites tend to earn lower in the United States.

23. We are grateful to anonymous referees of the journal for this suggestion.

24. Of the 40 coefficients on education variables (five educational categories for eight specifications in Table 3, namely, male and female specifications separately for business, farm, wage, and all workers), 35 changed by less than 0.009 and the maximum change for the remaining five was 0.029.

25. Of the 40 coefficients on education variables, 15 changed by less than 0.009 and the maximum change was less than 0.05.

26. Education could be endogenous if it is correlated with some unobserved characteristics of individuals, such as, ability, influence earnings (Card, 1999). In such cases, the OLS estimator may overestimate the returns because of positive correlation between (i) education and ability, and (ii) earnings and ability (Wooldridge, 2002). The studies typically resort to instrument variables (IV) to address this issue. The IV estimator needs to satisfy certain crucial assumptions and may not necessarily yield an asymptotically unbiased estimate of the average return (Card, 1999). We are, however, constrained by data availability in the use of IV and recognise this limitation.

27. The average rate of return is much higher, 8.22 per cent for the world (in the year 2005) and 10.58 per cent in South East Asia (Caselli, Ponticelli, & Rossi, 2017) though different methods can yield different estimates of returns (Barouni & Broecke, 2014).