Abstract
Using data from the Indian Human Development Survey, we examine evidence of caste and religion-based discrimination in the Indian private and public sector. Both Dalits and Adivasis show significant results of discrimination in the private sector, and benefit disproportionately from working in the public sector. This is strong evidence that at least some of the affirmative action policies in the public sector are proving effective. The policy implications are relevant: should similar affirmative action policies be implemented in the private sector? Further, this research suggests a path for further research to understand why protected castes do not benefit from affirmative action programs to the same extent as Dalits and Adivasis.
Keywords
Introduction
Caste and religion-based discrimination continues to be a problem in India. Despite legislative efforts, discrimination in education, employment, and politics persists as a human-rights concern. There are important efficiency considerations as well. Researchers have documented positive effects of diversity in the workplace while other scholars have argued that discrimination has resulted in an inefficient allocation of labor, which would have negative effects on economic growth as a whole (Milliken and Martins 1996; Ely and Thomas 2001). The Indian government has used legislation to increase the representation of scheduled castes (SCs), scheduled tribes (STs) and other backward castes (OBCs) in the public sector. Scheduled castes (SCs), or Dalits, and scheduled tribes (STs) are two groups of historically disadvantaged people recognized in the Indian Constitution. Other backward castes (OBCs, are other disadvantaged groups that receive favorable treatment from the Indian government. The public sector is subject to a reservations policy that sets aside a percentage of government positions for members of underserved groups. The private sector in India has no such requirement (Deshpande 2006). As a result, the representation of SCs, STs, and OBCs at private companies continues to lag (Census of India 2011).
Previous studies have shown that caste and religion-based wage discrimination exists, but there is no definitive answer on whether the degree of discrimination differs between the public and the private sector (Thorat et al. 2005). The policy implications of this work are important. If caste and religion-based wage discrimination is worse in the private than in the public sector, this may force the government to consider affirmative-action legislation for the private sector. It also begs the question as to why India's private sector has failed to take advantage of the desegregation benefits that U.S. firms have already realized.
We intend to examine whether caste and religion-based discrimination is worse in the private than in the public sector. We hypothesize that the caste-based affirmative-action programs in the public sector and the lack of a well-structured program in the private sector, will lead to a larger caste wage differential in the private than the public sector. We use data from the Indian Human Development Survey (IHDS) conducted in 2005. Using cross-sectional data, we conduct a multivariate regression with fixed effects to control for variation in the type of occupation.
Literature Review
India has a history of discriminations towards several disadvantages castes in addition to certain minority religions. To increase the job and education opportunities for groups that face serious discrimination, the government promotes an affirmative action program called the reservation system that is targeted to benefit SC/ST/OBCs (Deshpande 2006). The reservation program is mandatory for the public sector and stipulates that the employment quota for bottom castes should be equal to their share of the total population. The reservation program is not compulsory for the private sector, but private firms may use it for different reasons, for example when faced with political pressure. Furthermore, the government continues to consider expanding this policy to the private sector (Prakash 2009).
The literature on India's caste system—and particularly caste-based inequality/discrimination—is extensive, but a comprehensive assessment of how caste-based discrimination may affect the public versus private wage differential remains largely unexplored. Hnatkovska, Lahiri, and Paul explore the gap that exists between SC/ST/OBCs and other castes in terms of educational attainment, labor/industry mobility, and wages (2010). They find that the discrimination gap has begun to converge, particularly since the 1980s. They argue that the major structural changes that India has undergone— particularly in the last three decades with numerous reforms and innovative policies— have coincided with a systematic removal of caste-based barriers. This, in turn, has lead to more inclusive socioeconomic mobility in the country. Although a trend toward decreasing caste-based inequalities may be underway, it would be helpful to further explore the extent to which inequalities and discrimination still exist and in what ways they continue to impact the lives of the SC/ST/OBCs. Even if the trend is significant, what is the magnitude of remaining gaps and how long will it take for the gaps to close at the current pace?
We now turn to the literature that specifically evaluates the effects of caste-based inequalities on employment opportunities for the disadvantaged castes. Borooah (2005) analyzed inequalities that affect the poorest population in India. By using a log-linear model to differentiate between in average income by caste, a multinomial probability model to estimate the probability of a household being in a particular income percentile, as well as a decomposition of income inequality and poverty across castes, Borooah estimated that at least one-third of the difference between Hindu caste households and SC/ST households in average income could be due to the discrimination faced by members of SC/ST groups.
Takahiro also took on the task of quantifying caste-based discrimination in the labor market (2007). By distinguishing between wage discrimination (the differences in wages within a specific occupation) and job discrimination (unequal access to specific jobs/industries) he sought to estimate the wage differential between SC/ST/OBCs and other non-backwards castes. Takahiro (2007) found that there is indeed evidence that barriers to employment are a greater problem for SC/ST/OBC individuals. However, he did not find significant evidence of wage discrimination. This led him to the conclusion that while wage discrimination—which has been illegal in India for decades—may not be a major problem, caste-based job discrimination is a persistent barrier that continues to oppress members of SC/ST/OBCs. This could be why we see members of SC/ST/OBCs underrepresented in certain higher-level positions. This does not manifest itself in other studies that solely examine wage discrimination because the occupation or type of occupation is usually controlled for.
Finally, we find that far fewer papers exist in the literature that specifically explore the wage gap between SC/ST/OBCs and other castes within the context of comparing private and public sectors. An exception is a study by Madheswaran and Attewell (2007) using data from India's National Sample Survey. They analyze caste-based income and employment disparities particularly among highly educated employees. Using several decomposition methods they find that discrimination in the public sector has decreased over the last few decades, but that it has remained stable (and high) in the private sector. By holding constant specific household and individual characteristics and estimating the SC/ST/OBC effect on earnings as well as the non-scheduled caste effects on earnings, they find that non-scheduled castes are overpaid while SC/ST/OBCs are underpaid. They also estimate that approximately 15% of the earnings differential between non-scheduled castes and SC/ST/OBCs is due to discrimination. Consistent with Borooah's findings, Madheswaran and Attewell conclude that job discrimination continues to plague the labor market, particularly the private sector.
Our work attempts to expand on Madheswaran and Attewell's findings, which are somewhat limited in the data they use. Their data set, the National Sample Survey, only includes urban data. The IHDS data is mostly rural, so this alone will be a significant expansion on their findings. Further, they were only able to divide castes into two groups: SC/ST and non- SC/ST. Naturally, there are limitations to those findings which we hope to develop through our variable which includes information on caste as well as religion. The model we use is also a multivariate model as opposed to the logit model used by Madheswaran and Attewell.
Data and Methodology
We use data from the India Human Development Survey (IHDS), a nationally representative, multi-topic survey of 41,554 households in 1503 villages and 971 urban neighborhoods across India. The data we use in this paper is from the first round of interviews, which were completed in 2004–2005. A second round of interviews was completed in 2011–2012, but the data is not yet publicly available. The data set contains detailed economic and employment information in addition to carefully collected variables on social groups.
India has a number of different castes and religions, and the IHDS database captures them in one variable formed by 8 categories: Brahmin, High Caste, OBC (Other Backward Classes), Dalit (Scheduled Castes), Adivasi (Scheduled Tribes), Muslim, Sikh-Jain, and Christian. A review of occupations reveals that there is considerable overlap between the most common occupations in the public and the private sector. This suggests that an analysis that controls for occupation would have a large enough sample size and would be able to reflect the differences between the public and private sector wages.
Below is a tabulation of the distribution of individuals that work in the government. SGRY is the Universal Rural Employment Guarantee, a rural employment program that guarantees 100 days of work to any eligible adult, and Food for Work is another government program for low-income workers that gives food vouches in exchange for work. Because these are welfare programs it is difficult to compare them to private sector employment. For the purposes of our analysis we will exclude these two categories and focus exclusively on individuals employeed in government positions outside of SGRY and Food for Work. Approximately 12% of respondents work in regular government positions, compared with 86% of respondents who do not have any kind of government work (Table 1).
Tabulation of employment in government
Next, we examine some summary mean-wage statistics for the key variables in our study, divided by public vs. private sector. As a whole, income is significantly higher in the public sector, which fits well with the hypothesis we are investigating. This also supports the work done by Glinskaya and Lokshin, who found that the difference of wages between workers in the public sector was about 2.1 times that in the organized private sector, and about 3.8 times that in the private-informal sector (2005).
Not surprisingly, income is higher among men than women, while the other categories show the expected outcomes as well; The higher castes show higher mean incomes, OBCs and Dalits show lower mean incomes. Adivasi and Muslims both have mean incomes similar to OBCs and Dalits, while Christians have a mean income closer to the high castes. The Sikh/Jain population has a mean income that is greater even than the average for the Brahmin caste, part of which may be explained by their relatively small sample size, but this is still a result worth noting (Table 2 and Fig. 1).
Mean wage comparison between public and private sector
Mean wage by group, public vs private
Summary statistics show that mean age is higher in the public than in the private sector. This could be explained by a tendency to look for a government job with more security later on in life or the amount of education required for many public sector jobs. The gender breakdown shows that the proportion of men vs. women is higher in the public sector than the private sector. This may be because many women in India are self-employed, which would be counted under the private sector. Respondents from the public sector worked more hours each year, which may be connected to the greater security and consistency in a public sector job. They also completed almost six more years of education, on average, than respondents from the private sector. In most countries employees in the civil service are, on average, better educated than employees in the private sector so this result is not surprising.
Table 3 shows the total number of observations for each group, divided by public/private. Sikh/Jain and Christian are the two groups with the lowest number of observations. All other groups have more than 1000 observations in the data set. Worth noting, the majority of observations within each group do work in the public sector.
Public vs. private, examined by caste
In summary, the regression model looks to explain the annual wage based on urban dummy to differentiate between areas with fundamentally different economic systems, years of education, a female dummy, and age. Age-squared is also included to model wage as having decreasing returns to age. Additionally, the regression includes dummies for individual castes and religions, and a sector dummy if the participant is working in the private sector. We are especially interested in seeing the significance of interactions terms between caste/religion and the public sector. This can indicate whether certain castes/religions benefit, and by how much, from working in the public as opposed to the private sector. In the specific case of this model, the dependent variable will be the natural logarithm of annual wage.
Wages among occupations differ widely in India. For example, the annual-average wage of an engineer ($114,669) is almost six times the annual average wage of a barber ($20,756). For this reason, it is important to analyze wage discrimination while controlling for occupation. For the model, we divide occupations into quintiles based on average earnings. Originally, we had used a fixed-effects regression that controlled for each individual occupation, but we were concerned that outliers in individual occupations were skewing our results and that occupations were occasionally inconsistent across the public and private sector. The bottom quintile includes the occupations that fall into the bottom fifth of professions in terms of average wage (see Appendix). We then create dummy variables for each quintile to effectively use a fixed-effects regression that will control for the anticipated earnings differences among occupations. Our regression equation is included below. The base for the model would be a male individual from Brahmin caste, employee in the private sector and living in the rural area.
Results and Discussion
We use variations on the normal multivariate regressions in all of our models. Three of the most relevant regressions we examined are included in the table below. Regression I does not include interaction terms between caste/religion and public sector or any of the occupation controls. The purpose of this regression is to see the overall effect of caste/religion on earnings. Differences between this regression and regressions that control for occupation may give some indication of the extent to which wage differentials are based on job discrimination as opposed to wage discrimination. Regression II includes interaction terms between caste/religion and public sector but none of the occupation wage controls. Regression III includes all of the previous variables as well as the different controls for the level (wage-quintile) of occupation (Table 4).
What is the effect of caste and public sector work on annual wage (measured as log wage)?
p < 0.05
p < 0.01
In the three regression models, the measure and direction of the coefficients are close. For the castes/religions the omitted base is Brahmin, so the castes and religions should be treated as a comparison to Brahmin. The relationships among the basic caste/religion dummies are consistent with our hypothesis. High Caste is negative, OBC is very negative, as is Dalit and Adivasi, and all relationships are significant at the 1% level. Muslim is slightly negative, but this relationship is no longer significant once the level of occupation is controlled for. Sikh/Jain and Christian are both positive.
The urban dummy shows a strongly positive and significant relationship in all models, as expected. The female dummy is strongly negative and significant, which likely reflects the prevalent job and wage discrimination against women in India. As expected, the variable years of education is very positive and significant. Age is positive, and age-squared is negative, which supports the expected relationship of decreasing wage returns to age. The return on expected wages to another year of education would be close to 5% for the model that excludes the occupation variables, and by 3% when occupation is controlled.
Our most important findings are the positive and significant coefficients on the interaction terms between public sector, and Dalit and Adivasi. This implies that being a member of the Dalit or Adivasi group yields higher returns in the public than in the private sector. This is the strongest evidence we have that affirmative action programs are effectively making the public sector a better place to work than the private sector for certain protected groups.
Table 5 shows a comparison between public and private sector when Brahmin caste is used as a base. When using the model that includes interaction terms but none of the occupation wage controls (Regression II), the Brahmin and Dalit wages have a difference of 24% in the private sector and a slight difference of 7.9% in the private sector. When controlling for occupation (Regression III), the Brahmin and Dalit wage differences are not as large. The private sector wages for Dalits are 8.1% below the Brahmin wages, and their public sector wages are only 3.8% lower.
Public vs. private sector wages - regression III coefficients and Brahmin as base
Wages for the Adivasi, or Scheduled Tribes, are around half of Brahmin wages in the private sector (both with and without occupation controls), but in the public sector this difference is quite small. For OBCs, wages in the private sector are below Brahmin's wages by 36.4% in Regression II and 22.2% in Regression III., Wages are around one third of those percentages for the public sector in both regressions. When compared with Brahmins in the private sector, Christians' wages are higher when the model controls for occupations; but the difference is almost zero for the public sector in both models.
The difference within each caste/religion between public and private sector is shown in Table 6, for the regression II (the model that includes interactions terms) and regression III (interaction terms and control for occupation). While in regression II Dalits' wages are 109% higher in the public vs the private sector, when controlling for occupation (regression III) this difference decreases to 77%. As expected, this result is similar for all groups; the difference is smaller when the model controls for occupations.
Differences among public vs. private sector wages
Conclusion
Our results have important policy implications. The interaction terms between public sector, and Dalit and Adivasi groups were both positive and statistically significant. This suggests that the Dalits and Adivasis benefit more from working in the public sector than do other groups and is strong evidence that at least some of the affirmative action policies in the public sector are proving effective. Further, this implies that wage discrimination in the private sector is still worse than in the public sector.
A number of questions still need to be answered. First, why are other protected castes not benefitting from public sector reservation programs to the same extent as the Dalit and Adivasi groups? There is a growing literature on understanding how different sub-groups have been affected by reservation policies; a closer examination of which protected groups most need and most benefit from affirmative-action policies would be beneficial to inform public policy. Second, the fact that certain groups reap higher rewards from the public sector than the private sector implies that the private sector may be inefficiently recruiting labor. As referenced in the introduction, there are documented benefits to a diverse labor force. Further, labor force discrimination may be resulting in an inefficient allocation of labor, which could have a detrimental effect on economic growth.
The Indian government should consider whether mandatory affirmative-actions programs should exist in the private sector. The economic benefits of ending caste/religion-based discrimination, along with the human rights and constitutional justification provide a compelling argument for reservation-type policies in the private sector.
Footnotes
Appendix
Definition of occupation quintiles
| Quintiles | Occupations |
|---|---|
| Q1 Low average wages | Ag labour, Barbers, Construction, Cooks/waiters, Fishermen, Food, Forestry, House keepers, Labour nec, Launderers, Maids, Other farm, Plantation lab, Potters, Stone cutters, Textile, Tobacco |
| Q2 Low-medium average wages | Artists, Carpenters, Hotel/restaurant, Loaders, Other farmers, Painters, Paper, Production nec, Professional nec, Rubber/plastic, Sales nec, Salesshop, Service nec, Shoe makers, Sweepers, Tailors, Wood/paper |
| Q3 Medium average wages | Assemblers, Boilermen, Chemical, Cinema op, Drivers, Farm manager, Jewellery, Machine tool op, Mail distributors, Metal workers, Miners, Money lenders, Performers, Plumbers/welders, Printing, Shopkeepers, Tanners |
| Q4 Medium-high average wages | Book-keepers, Clerical nec, Computing op, Elected officials, Electrical, FIRE sales, Journalists, Lawyers, Manuf agents, Nursing, Police, Teachers, Technical sales, Telephone op, Transp conductors, Typists, Village officials |
| Q5 High average wages | Accountants, Air/ship officers, Clerical supe, Eng tech, Engineers, Govt officials, Managerial nec, Mgr finance, Mgr manf, Mgr service, Mgr transp/commun, Mgr Whsl/retail, Other scientific, Physical sci tech, Physical scientists, Physicians, Social scientists, Transp/commun supe |
| No wages registered | Cultivators, Economists, Hunters, Life science tech, Life scientists, Statisticians |