Caste, Affirmative Action and Discrimination in India*

Ira N. Gang
Department of Economics, Rutgers University
gang@economics.rutgers.edu

Kunal Sen
IDPM, University of Manchester
kunal.sen@manchester.ac.uk

Myeong-Su Yun
Department of Economics, Tulane University
msyun@tulane.edu

ABSTRACT

Affirmative action has been at the heart of public policies towards the socially disadvantaged in
India. Compensatory discrimination policies which have been adopted for the Scheduled Castes (SC)
and Scheduled Tribes (ST) since independence are now available to Other Backward Classes (OBC).
We examine why OBC have lower living standards, as measured by per capita household
consumption expenditures, relative to the mainstream population, and whether these reasons are
similar to those observed for SC and ST. We find that while the causes of the living standard gap for
the OBC are broadly similar to those for the SC and ST, the role of educational attainment in
explaining the gap is higher in importance for the OBC.

Keywords: living standards, caste, reservation policy, decomposition

JEL Classifications: I32, O12, J15

* We thank Richard Palmer-Jones for helping us with the data.

 1

1. Introduction

Affirmative action policies to increase access to education and employment have been at the core

of public policies towards historically disadvantaged or non-dominant groups in both developed and

developing countries (Weisskopf 2004, Mcharg and Nicolson 2006, Yuill 2006). Among developing

countries, India has had perhaps the longest history of affirmative action to counter caste and ethnic

discrimination (Revankar 1971). Article 46 of the 1950 Constitution pronounces “The State shall

promote with special care the educational and economic interests of the weaker sections of the

people, and, in particular, of the Scheduled Castes and the Scheduled Tribes, and shall protect them

from social injustice and all forms of exploitation.”

Articles 341 and 342 of the Constitution included a list of castes and tribes entitled to such

provision, and the castes and the tribes included in these two lists were known as Scheduled Castes

(SC) and Scheduled Tribes (ST) respectively. As a consequence of this provision, a policy of

compensatory discrimination via reserved positions was implemented both at the national and

subnational levels in the allocation of university places and public service appointments (Galanter

1984).1 While there were references in the Constitution to an undefined wider category of

‘depressed’ or ‘socially and educationally backward classes of citizens’, the identity of these groups

– the Other Backward Classes (OBC) – was left unclear (Bayly 1999), and no special measures such

as job reservations or quotas in educational institutions were implemented for these groups.

Recently several papers have examined the causes of economic disparities among social

groups in India. These studies attempt to answer why living standards among SC and ST are much

1
SC and ST were also favored by the Indian government in land redistribution policies, loan
allocations, and a large number of other official development programs. Quotas were also introduced
for SC and ST in state legislatures and local governments.
 2

lower than the rest of the population (Bhaumik and Chakrabarty 2006, Kijima 2006, Borooah 2005,

and Gang, Sen and Yun 2008). Gang, Sen and Yun (2008) show that along with differences in

educational attainment between SC and Other households, the occupational characteristics of SC

households places them at a disadvantage compared to Other households, while for ST households,

locational factors rather than occupational characteristics are more important in explaining their

higher poverty status. Borooah, Dubey and Iyer (2007) explicitly address the effectiveness of job

reservations on the economic opportunities of persons belonging to the SC and ST, finding that the

boost provided by job reservations in raising the proportion of SC and ST individuals in regular

salaried employment is 5 percent.

The issue of whether the compensatory discrimination that have been provided by the Indian

state to SC and ST should also be provided to OBC has remained at the heart of the affirmative

action debate in India, with recent proposals to extend the reservation policies for OBC to elite

institutions of higher education and to the private sector (Thorat 2004, Thorat, Aryamma and Negi

2005). So far, there have been no studies that have examined the economic status of the OBC, and

compared the determinants of the economic conditions of the OBC to the those of SC and ST social

groups and the remainder of the population.

This paper examines the determinants of living standards for OBC and other social groups

in India. Living standards in SC, ST and OBC households are much lower than the mainstream

population, comprising the Hindu ‘forward castes’ and other religions, including those belonging

to the Christian, Muslim and Sikh religious faiths. We call these households ‘Advantaged Classes
 3

and Castes’ (ACC) households.2

We use real average monthly per capita expenditure (PCE) as the measure of living

standards3. In urban areas. overall, real average monthly PCE is 978.568 rupees. This varies widely

by group, with PCEs at 686.897, 772.917, 849.743, and 1143.760, respectively for SC, ST, OBC

and ACC households. In rural areas, monthly PCE is 521.563. By group we see PCEs of 449.924,

412.533, 511.520, and 624.774, respectively for SC, ST, OBC and ACC households.

We study differences in living standards among social groups in India using Oaxaca

decomposition analysis (Oaxaca 1973). Our approach allows us to develop a better grasp of the

sources of the differences in living standards among different social groups in India, in particular

whether differences in the amounts of schooling, occupational choice and demographic

characteristics hold the key to understanding the living standard gap, and whether the income

generating strength of household or individual characteristics (e.g., education and occupation) are

different for each group.

We investigate whether the causes of lower living standards among OBC households are

similar to SC and ST households. If a similar set of causes explain the lower living standards of

OBC as well as SC/ST households than ACC households, it provides a stronger justification for the

Mandal Commission’s finding that OBC shares the same economic status in Indian society as the

2
To make our social groups comparable, we do not confine our analysis of the ACC households
to the ‘forward caste’ Hindu population, as several SC and OBC households were also classified as
belonging to religions other than Hinduism. The individuals in these households may have been
originally Hindu, but have converted to a different religion.
3
We deflate nominal per capita expenditures by state and sector specific poverty lines obtained from
Indian Planning Commission to take into account the variations in costs of living across Indian states
and between the rural and urban sector.
 4

SC and ST. Based on our findings on the causes of lower living standards - and in particular, the

roles of education and occupation- among OBC households in comparison to SC and ST households,

we evaluate the specific reservation policies.

In the next section, we summarize the historical background of the debate in India on the

extension of compensating discrimination to the OBCs. Section 3 discuss the mean characteristics

of the social groups in our analysis - Other Backward Classes, Scheduled Castes, Scheduled Tribes

and the ACC household group. Section 4 investigates the determinants of living standards for these

social groups, examining the relative influence of various socio-economic variables on monthly per

capita expenditures. Section 5 employs decomposition analysis using regression coefficients to

examine and explain living standard gaps between Other Backward Classes, Scheduled Castes,

Scheduled Tribes on one hand and the mainstream population on the other. Finally, Section 6

provides a summary of our study and its main conclusions.

2. Extending Affirmative Action to Other Backward Classes

In 1979, a commission under the chairmanship of B.P.Mandal – popularly known as the Mandal

Commission – was established by the ruling Janata Party under the Prime Ministership of Morarji

Desai with the objective of identifying the Other Backward Classes (OBC). In 1980, it published its

findings, placing a total of 3428 ‘communities’ in the OBC category, comprising 54.4 percent of the

country’s population (Bayly 1999). The Mandal Commission recommended that there should be

employment quotas in public sector organizations (including nationalized banks and private sector

undertakings which received financial assistance from the government in one form or the other) and

reserved places in higher educational institutions of 27 percent for OBC in addition to the 22.5
 5

percent job quotas and seats in higher educational institutions that were already in place for SC and

ST (Ramaiah 1992). The figure of 27 percent was arrived at as the Supreme Court limited total

reservations to under 50 percent.

Due to a change in the government in 1979, the report from the Mandal Commission was

shelved. In 1989, a successor party to the Janata Party – the Janata Dal – achieved power as the

leading element of a national coalition government under the Prime Ministership of V.P. Singh and

announced plans to implement the Mandal Commission recommendations, significantly increasing

quotas in public sector employment and in university admissions for the communities which had

been classified as OBC by the Mandal Commission.4 The announcement led to significant violent

resistance in many parts of India, including a series of widely publicized self-immolations by high-

caste students (Bayly 1999). In recent years, with the coming to power of the Congress-led

Government of Manmohan Singh in 2004, there have also been proposals to extend the job quotas

to private sector jobs and to certain privileged institutions of higher education. In April 2008, the

Supreme Court upheld the Mandal Commission recommendation that 27 per cent of seats in

Government funded institutions (including the prestigious Indian Institutes of Technology and of

Management) be reserved for OBCs.5 However, the Supreme Court did not agree to extend the

reservations for OBCs to educational institutions in the private sector.

4
It should be noted that several state governments in India (such as the governments of the Southern
states of Karnataka and Tamil Nadu) had already extended reservation policies to OBCs for jobs in
state governments
5
The judgement, known as the Ashoka Kumar Thakur versus the Union of India case, excluded OBC
children from households above a certain income level, or where one of the parents was employed
in certain professions (such as doctors, engineers and bureaucrats) from the ambit of the reservation
policy (this was known as the ‘creamy layer’ exclusion).
 6

Unlike the case of compensatory discrimination policies for SC and ST implemented in the

immediate post-independence period which were widely regarded as justified and did not cause

much controversy, there has been significant criticism of the Mandal Commission findings and

recommendations. Firstly, the methods and criteria adopted by the Mandal Commission to define

a ‘backward class’ were widely regarded as flawed raising skepticism whether the communities

determined to be OBC by the Commission were truly socially disadvantaged or deserving of the

massive welfare programs subsumed under reservation policy (Beteille 1992, Radhakrishnan 1996).

Secondly, several observers felt that the reasons successive governments tried to implement the

Mandal Commission recommendation had more to do with political factors than economic and social

justification, as several of the communities included in the Mandal Commission’s list of OBC

formed important ‘vote banks’ for political parties both in power and in opposition (Sivaramayya

1996, Bayly 1999).

3. Data and Descriptive Statistics

For our analysis we use the 55th round of India’s National Sample Survey (NSS) on consumer

expenditure in rural and urban areas collected in 25 states and 7 Union Territories. The survey period

extended from July 1999 to June 2000. The 55th round was the first time the NSS demarcated OBC

from other non-scheduled caste Hindus in the expenditure survey. In previous rounds, expenditure

data on OBC were combined with other non-scheduled caste Hindus, making earlier examination

of differences in living standards between OBC and ACC households impossible.6

6
It is possible that the discrimination against OBC had already been reduced when the 1999-2000
survey was conducted as compared to the time of the Mandal Commission study, because several
Indian states may have implemented the Mandal Commission recommendations before the survey.
 7

The NSS data are a cross-section of a geographically distributed random sample of

households. Besides information on household consumer expenditure and demographic behavior,

the NSS contains detailed questions on other household characteristics such as the educational level

and occupation of the head of the household. Since the NSS provides expenditure data by household,

our estimates of monthly per capita expenditures are at the level of the household, not at the level

of the individual.7 We restrict our sample to households where the age of the head of the household

is between 20 and 70 years.

Tables 1 and 2 show the mean characteristics of the sample rural and urban households,

respectively. We first describe the mean characteristics of rural households (Table 1). Considering

the demographic characteristics of the four groups of households, we find that OBC, SC and ST

households have a lower mean age for the head of the household compared to ACC households.

OBC, SC and ST households are also smaller than ACC households – the mean household size for

OBC, SC and ST households are 5.10, 4.88 and 4.96 respectively, compared with a mean household

size of 5.17 for ACC households. ACC households own the most land (0.97 hectares), followed by

ST households (0.82) and OBC households (0.71). The average land holding of SC households at

0.31 hectares is substantially less than the average land holdings of the other three social groups.

A much higher proportion of SC and ST households are not literate (61.7 percent and 65.3

percent, respectively), compared with OBC households (50.5 percent) and ACC households (35.9

percent). With respect to occupation, 15.5 percent of OBC households are self-employed in non-

agriculture, 31.3 percent as agricultural laborers, 7.9 percent as non-agricultural laborers, 35.2

7
This distinction becomes important when there are significant differences in the intra-household
consumption of food and other necessities across the SC, ST, OBC and ACC households.
 8

percent are self-employed in agriculture while 10.2 percent are classified in a residual category which

we term ‘miscellaneous’. In the case of SC households, 12.1 percent of SC households are self-

employed in non-agriculture, 54.0 percent as agricultural laborers, 9.6 percent as non-agricultural

laborers, 16.4 percent are self-employed in agriculture while 7.8 percent are classified as

‘miscellaneous’. For ST households, 5.4 percent are self-employed in non-agriculture, 44.1 percent

are agricultural laborers, 8.1 percent are non-agricultural laborers, 36.7 percent are self-employed

in agriculture while 5.7 percent are in ‘miscellaneous’ occupations. Finally, for ACC households,

15.0 percent are self-employed in non-agriculture, 20.9 percent are agricultural laborers, 6.4 percent

are non-agricultural laborers, 41.4 percent are self-employed in agriculture while 16.3 percent are

in ‘miscellaneous’ occupations. Thus, a greater proportion of OBC households are self-employed

in agriculture as compared to SC households. A lower proportion of OBC households are agricultural

laborers as compared to SC and ST households.

We next describe the mean characteristics of urban households (Table 2). Similar to rural

households, we find that OBC, SC and ST households have a lower mean age for the head of the

household compared to ACC households. Similarly, OBC, SC and ST households are also smaller

than ACC households – the mean household size for OBC, SC and ST households are 4.60, 4.81 and

4.56 respectively, compared with a mean household size of 4.50 for ACC households.

The proportion of ACC households who are illiterate at 12.3 percent is less than half of the

next literate group – OBC households - where the illiteracy rate is 25.0 percent. The illiteracy rate

for SC and ST households are 37.1 percent and 34.2 percent respectively. With respect to

occupation, 37.9 percent of OBC households are self-employed, 36.6 percent are salaried workers,

18.6 percent are casual laborers and 6.9 percent are classified in a residual category termed
 9

‘miscellaneous’. In the case of SC households, 26.8 percent are self-employed, 39.4 percent are

salaried workers, 28.9 percent are casual laborers and 4.8 percent are in ‘miscellaneous’ occupations.

In the case of ST households, 23.8 percent are self-employed, 40.5 percent are salaried workers, 25.4

percent are casual laborers and 10.3 percent are in ‘miscellaneous’ occupations. Finally, for ACC

households, 36.4 percent are self-employed, 47.6 percent are salaried workers, 7.7 percent are casual

laborers, and 8.3 percent are in ‘miscellaneous’ occupations. The proportion of SC and ST

households who are casual laborers are higher than for OBC and ACC households. The proportion

of OBC households where the head of the household is a salaried worker is not very different than

SC and ST households.

4. Determinants of Living Standards

To study determinants of living standards, we regress log real average monthly per capita

expenditure on various household characteristics in addition to state/region fixed effects. The

regression equation is , where y, , and are, respectively, an N×1 vector of log real

average monthly per capita expenditure, an N×K matrix of independent variables, and a K×1 vector

of coefficients. We now discuss the specification of our regression equation, which we estimate

using ordinary least squares for households in the ACC group, other backward classes (OBC),

scheduled castes (SC) and scheduled tribes (ST), separately. We also discuss the implications of the

estimated coefficients on the determinants of living standards. Though the reported coefficients

for each of the independent variables are broadly similar across all four social groups,

likelihood ratio tests (not reported) show that the coefficients for each group are significantly
 10

different from the other groups. 8

Our focus is on education and occupation. To capture the effect of education on the

household’s living standards, we use dummy variables corresponding to the highest educational

level completed by the head of the household. We include dummy variables corresponding to

‘literate, below primary level’, ‘literate, below secondary level’, ‘literate up to secondary level’, and

‘literate, higher secondary and above’ (the reference group is households where the head of the

household is not literate). With respect to occupation, we include dummy variables corresponding

to four occupational groups for rural households – self-employed in non-agriculture, self-employed

in agriculture, agricultural labor and non-agricultural labor – and three occupational categories for

urban households – self-employed, waged/salaried workers, and casual labor (with the reference

group for both rural and households being the occupational category we termed the ‘miscellaneous’

category).9

Besides the explanatory variables capturing occupation and educational levels, we include

in our analysis a number of background and demographic variables. We include the generational

impact reflected by the age of the person. We use two variables: age (number of years), and age-

squared (number of years of age-squared divided by 100), to reflect the non-linear effects of age on

8
This likelihood ratio test supports our approach of studying the OBC, SC and ST separately.
9
The NSS classifies rural and urban households in occupational categories according to the main
source of income reported for each surveyed household. This is called the “principal occupation
code” of the household. The principal occupation is defined to be that which contributes at least 50%
of household income. The category we term ‘miscellaneous’ includes those where no one income
source exceeds 50% or more of total income. Thus, the households in this category have very
diversified income sources or more than one earning member.
 11

living standards. We incorporate the effect of household size on living standards, as previous studies

have noted a negative relationship between per capita expenditures and the size of the household

(Krishnaji 1984). Given the possible presence of economies of scale in household consumption, we

include household size squared as an additional control variable. We also include total cultivated

land owned by the household as a measure of the household’s wealth status for rural households.

We include controls for the location of the household. There are significant differences in

rural living standards across Indian states, with states in North-Western India (Haryana, Punjab)

along with the state of Kerala having higher living standards than the national average (Datt and

Ravallion 1998). In contrast, the living standards in Assam, Bihar and Orissa are much lower than

the national average.10 Furthermore, there is non-negligible variation in rural living standards within

Indian states across NSS regions and these variations are crudely associated with differences in agro-

ecological conditions which may be vastly different within a state, parts of which may be more

similar to those prevailing in geographically contiguous states (Palmer-Jones and Sen 2003).11

Similar differences in living standards across Indian states are also evident for urban

households. More industrialized states such as Tamil Nadu and West Bengal have higher urban

living standards than less industrialized states such as Orissa and Uttar Pradesh. Dubey,

Gangopadhyay and Wadhwa (2001) find that per capita expenditures of urban households are

increasing in the size of the city in which the households are located, which could be attributed to

the presence of stronger agglomeration economies in the larger cities. This suggests that there may

10
There are 32 states and union territories in the 55th round of the NSS consumer expenditure
survey.
11
NSS regions are groupings of contiguous districts within states. There are 82 regions in the 55th
round of the NSS consumer expenditure survey.
 12

also be significant living standards differences across NSS regions within Indian states, with regions

that contain larger cities having higher living standards on average than regions with smaller cities.

The omission of dummy variables to capture the location of the household may bias the results if the

OBC households are mostly residing in Indian states and NSS regions where living standards are

lower, and if these lower living standards are due to state-level and sub-state NSS region-level

factors exogenous to the household such as the nature of state-level public policies toward poorer

households, agro-climactic factors or the presence of agglomeration economies. We present our

results with the inclusion of NSS region fixed effects.

The OLS estimates of the regression equation for the rural sample are reported in Table 3, and

for the urban sample in Table 4. We first describe our results for the rural sample. The estimated

coefficients show that greater educational attainment is associated with a statistically significant

increase in per capita expenditure, with everything else held constant. This is true for all four

household groups. However, higher educational attainment from the secondary level up seems to lead

to a greater increase in per capita expenditures for ACC households when compared with OBC, SC

and ST households.

We now turn our attention to occupation and its impact on per capita expenditure. Compared

with the occupational category ‘miscellaneous’, all other occupational categories have lower per

capita expenditure, i.e., those households are more likely to have lower living standards for all four

social groups. Agricultural laborer household are more likely to be poorer among all occupational

groups, controlling for other determinants. Interestingly, OBC households who are self-employed in

agriculture are relatively better off compared to those in SC, ST and ACC households, as evident in

a smaller magnitude in the coefficients on self-employed in agriculture for OBC households as

 13

compared to SC, ST and ACC households. With respect to demographic factors, higher aged heads

of households are associated with a higher per capita expenditure. However, this relationship is non-

linear, with further increases in age leading to less than proportionate increases in per capita

expenditures. A non-linear relationship is also found between living standards and household size;

per capita expenditure first decreases then increases. The possession of cultivable land seems to have

a positive effect of similar magnitude on per capita expenditure for OBC, SC and ST households,

though not for ACC households.

The results imply that rural households that are larger, where the head of the household is not

literate, is an agricultural laborer, is younger in age, and possess a smaller amount of land have lower

living standards. We also find that the effects of explanatory variables on per capita expenditure vary

over social groups.

Turning now to the results for urban households in Table 4, we find that as in the case of rural

households, higher educational achievements have a monotonically increasing effect on per capita

expenditures for all social groups. However, the return to education in terms of increasing living

standards is most evident among ST and ACC households as compared to OBC and SC households.

With respect to occupation, casual laborers are most likely to have lower living standards, and this

is observed for all social groups. Salaried workers are likely to have higher living standards than other

occupations in the case of OBC and SC households, though this is not the case with ST and ACC

social groups, where the coefficients on the dummy variable for workers who are salaried for these

two groups are not statistically significant. As in the case of rural households, age and household size

have similar non-linear effects on per capita expenditures. While per capita expenditures increase

and then decrease with the age of the head of the household, per capita expenditures decrease then
 14

increase with the size of the household.

5. Accounting for Differences in Living Standards

In this section, we seek to explain why living standards are lower among OBC households as

compared to ACC houeholds, and see whether the determinants in the living standards gap between

OBC and ACC households is different than those for SC versus ACC households, and ST versus

ACC households.

Sources of differences in living standards can be found from differences in household

characteristics across groups (characteristics effect) and from differences in returns to household

characteristics across groups (coefficients effect). The characteristics effect relies on the possibility

that the characteristics or attributes of households contributing to living standards may differ among

groups. For example, one group may have less education than another group, or be in“bad” jobs. The

characteristics effect reflects how differences in the attributes of households among groups affect

differences in living standards.

The coefficients effect relies on the possibility that the effectiveness of household

characteristics, reflected in regression estimates, may vary across groups. For example, education may

be less effective in raising living standards in scheduled and OBC households compared with ACC

households. The coefficients effect reflects how differences in the regression coefficients across

groups affect living standards.

As argued by Gang, Sen and Yun (2008), interpreting these two effects is always difficult and

controversial as shown in studies decomposing wage differentials. The popular interpretation is that

the characteristics effect is not due to discrimination while the coefficients effect may be related to
 15

an outcome of unequal treatment by society (discrimination). Though differences in characteristics

are supposed to reflect differences in income generating qualifications and credentials possessed by

various groups, it is possible that the disparity in attributes might result from widespread

discrimination against the scheduled and OBC groups in terms of educational opportunity and

occupational choice. On the other hand, it is not clear that discrimination is the only source for the

existence of the coefficients effect. For example, educational quality may differ between scheduled

and non-scheduled households for reasons not due to discrimination. Hence, the differences in the

coefficients on education may also capture differences in education quality between the scheduled and

non-scheduled in addition to capturing discrimination. Therefore, our interpretation is that the

coefficients effect captures the amount of the living standard gap caused by the differences in the

effectiveness of characteristics in enhancing living standards between the comparison groups. These

caveats should be kept in mind in interpreting decomposition results.

Using regression estimates of living standards, we can decompose the average differences in

living standards measured in terms of per capita monthly expenditure between group A and B as

follows:

where the first and the second components represent the characteristics effect and the coefficients

effect, respectively, and “over bar” represents the value of the sample average. Since OLS is used,

the residuals effect ( ) disappears.12

12
A decomposition equation with a different parameterization, that is,
possible; our results with it are not substantially different from those presented here and are available
from the authors upon request. Another issue when interpreting the decomposition results is that the
 16

We now discuss our empirical findings from the decomposition analysis. We focus on the

percentage share that tells us what percentage of the total living standard gap is accounted for by that

particular element or group of elements. We discuss the overall effects first, and then break down the

overall effects into smaller subgroups. We discuss the living standard gap of OBC relative to ACC

households in Table 5, for SC relative to ACC households in Table 6, and that of ST compared with

the ACC households in Table 7. In Tables 5, 6 and 7 we provide the results of the aggregate

breakdown, and of key groups of variables, for both rural and urban samples for the each of the paired

comparisons.

We proceed by first discussing the aggregate effects and sub-aggregate effects for rural OBC

households (Table 5, first half of the table). The Aggregate Effects row shows the overall effects of

characteristics versus coefficients in explaining differences in living standards. The top panel shows

that 54.7 percent of the difference in living standards between the OBC and ACC households is

explained by the differences in the levels of characteristics possessed by the two groups, while 45.3

percent by the differences in the regression coefficients. Both aggregate characteristics and

coefficients effects are significant at the 1 percent level of significance. If in both groups the various

variables influencing living standards had the same strength (their coefficients had been equal), then

coefficients effect in the detailed decomposition is not invariant to the choice of omitted groups
when dummy variables are used (see Oaxaca and Ransom 1999, for details of this issue). We follow
a solution suggested by Yun (2005) that, if alternative reference groups yield different estimates of
the coefficients effects for each individual variable, it is natural to obtain estimates of the
coefficients effects for every possible specification of the reference groups and take the average of
the estimates of the coefficients effects with various reference groups as the “true” contributions of
individual variables to differentials. While appearing cumbersome, this can be accomplished with
a single estimation. We can transform our regression estimates into a normalized equation and use
the normalized equation for our decomposition (see Yun 2005 and Gang, Sen and Yun 2008, pages
68-69).
 17

45.3 percent of the lower living standards in OBC households compared to ACC households would

disappear. On the other hand, if both groups had the same characteristics, 54.7 percent of the living

standard gap would disappear.

In the first half of Table 5, we also see the breakdown of characteristics and coefficients

effects into important variable groupings for the rural sample. We see the importance of the education

effect for occupation in determining the living standard gap, contributing 22.5 percent. The

coefficients effect of education is a negative 3.8 percent. Thus, it is the characteristics effect of

education rather than its coefficients effect, which explains why OBC rural households have a much

lower level of living standards than ACC households. Occupational structure is also important, but

not as much as education. The characteristics and coefficients effects of education contributes 12.4

and -2.4 percent respectively to the living standard gap.

Among the control variables, land owned, household size, and age do not contribute

significantly to the living standard gap. The characteristics and coefficients effects of land owned

contribute 3.8 and - 4.8 percent respectively to the living standard gap. The coefficients effect of age

structure (age and age-squared taken together) is not significant while the characteristics effect in

positive and significant, though small. For household size we find the characteristics effect is

negative, and the coefficients effect is positive and large. Household size differences reduce the living

standard gap, but differences in coefficients increase the living standard gap.13 Finally, locational

factors are important in explaining the large gap in living standards between rural OBC and ACC

households - the characteristics and coefficients effects of the NSS region dummy variables contribute

13
As seen in Table 1, OBC households are smaller in size than ACC households, and our analysis
suggests that the likelihood of a lower living standards is positively related to household size.
 18

24.3 percent to the living standard gap.

Moving now to urban households (second half of Table 5), we find that 72.2 percent of the

difference in living standards between the OBC and ACC households is explained by the differences

in the levels of characteristics possessed by the two groups, while 27.8 percent by the differences in

the regression coefficients. Both aggregate characteristics and coefficients effects are significant at

the 1 percent level of significance. The characteristics effect of education along with the coefficients

effect of age dominate other effects in explaining the living standard gap between OBC and ACC

households. The characteristics effect of education explains 46.2 of the living standard gap while the

coefficients effect of age explains 82.7 percent. The coefficients effect of education explains only a

negative 1.0 percent of the living standard gap. Interestingly, occupational structure has a very limited

role to play in explaining the urban living standard gap for OBC households – the characteristics and

coefficients effects of education are 9.1 and -2.5 percent, respectively. Locational factors (as captured

by the NSS regional dummies) are less important in explaining the living standard gap for urban

households compared to rural households.

We now examine the determinants of differences in living standard between SC and ACC

households in Table 6. For the rural sample, the aggregate characteristics and coefficients effects for

SC households are very similar to OBC households (first half of Table 6). The aggregate

characteristics effect contributes 56.2 percent of the living standard gap. The coefficients effect

contributes 43.8 percent of the living standard gap. Similar to OBC households, a large proportion

of the difference in living standards can be explained by the characteristics effects of education and

occupation - these contribute 21.8 percent each of the living standard gap. However, the contribution

of the coefficients effects for these two variables is small, -8.1 percent for education and -5.4 percent
 19

for occupation. The characteristics and coefficients effects of other variables are small, except the

coefficients effect of age at 11.0 percent, and the combined characteristics and coefficients effects of

the NSS Region dummy variables at 29.0 percent.

As in the case with rural households, the aggregate characteristics and coefficients effects for

urban SC households is very similar to urban OBC households - the characteristics and coefficients

effects contribute 70.6 and 29.4 percent of the living standard gap respectively (second half of Table

6). The characteristics effect of education explains much of the living standard gap, contributing 42.4

percent to the latter. The coefficients effect of age also contributes 39.3 percent. The characteristics

effect of occupation contribute 11.5 percent of the urban living standard gap for SC households. Other

variables have little role to play in explaining the living standard gap. Locational factors have a minor

role ro play in explaining the living standard gap.

Finally, examining the determinants of the living standard gap for ST vs ACC households,

the aggregate characteristics and coefficients effects for rural ST households is 53.5 and 46.5 percent

respectively (first half of Table 7). The characteristics effects of education and occupation contribute

18.4 percent and 12.1 percent respectively of the rural living standard gap. The contribution of the

coefficients effects for these two variables is small, –5.2 percent for education and 0.1 percent for

occupation. Locational factors explain much more of the ST rural living standard gap than they do

of the OBC and SC rural living standard gap, with the combined characteristics and coefficients

effects of NSS Region dummy variables explaining 38.4 percent of the rural living standard gap for

ST households. Among other variables, the coefficients effect of household size at 13.0 percent is

large.

In the case of urban households, the aggregate characteristics and coefficients effects of ST
 20

households contribute 66.8 and 33.2 percent of the living standard gap (second half of Table 7). As

in the case of SC households, the characteristics effect of education and the coefficients effect of age

dominate all other effects, contributing 36.3 percent and 32.2 percent of the urban living standard gap

respectively. The coefficients effect of education explains only 0.7 percent of the living standard gap.

The characteristics and coefficients effects of occupation contributes 11.2 percent and -5.4 percent

of to the living standard gap of urban ST households.14

The overall findings suggest a rather murky picture of social justice in India. In urban areas,

the decomposition analysis suggests that the characteristics effects dominates the coefficients effect,

explaining around 65-70 percent of the difference in living standards between SC, ST and OBC

households on one hand and ACC households on the other, while in rural areas, both characteristics

and coefficients effects have roughly equal impacts in explaining the difference in living standard gap.

This may be due a lower level of discrimination against ST, SC and OBC in urban labor markets than

rural labor markets. Differences in educational attainment are the most important source of the

characteristics effect in urban areas, explaining between 36 to 46 percent of the differences in living

standards for these social groups. Differences in occupational characteristics are relatively less

14
One may wonder what happens if occupation variables are omitted, as occupations are often
considered endogenous. Once occupation variables are excluded from the specification of the
regression and the computation of the decomposition equation, then, roughly speaking, the
characteristics and coefficients effects previously attributed to occupations are shifted to differences
in intercepts while the two effects of the other variables are not changed substantially. As the
constant term is included in the coefficients effect, for decomposing differences in poverty incidence
between the OBC, SC or ST and ACC, this increases the size of the aggregate coefficients effect and
decreases the size of the aggregate characteristics effect. Note that the characteristics and coefficients
effects attributed to occupations are positive when occupations are included, so that excluding them
increases the differences in the intercepts and increasing the size of the coefficients effect. This may
be a natural consequence of not controlling for occupations in the regression. The results of the
regression and the decomposition without occupation variables are available from authors upon
request.
 21

important in determining the urban living standard gap, explaining around 9-11 percent of the

difference in living standards. Differences in educational attainment are more important for OBC

(46.2%) than for SC (42.4%) and ST (36.3%).15 Locational factors, characteristics and coefficients

effects combined, play a small role in explaining the urban living standard gap for OBC, SC and ST

households, explaining around 11-12 percent of the gap. Interestingly, differences in returns to

education do not explain the lower living standards of OBC, SC and ST households as compared to

ACC households.

With respect to the rural living standard gap, the characteristics effects of both education and

occupation matter in explaining the former for all three social groups. The characteristics effect of

education matters slightly more for OBC (22.5%) and SC households (21.8%) as compared to ST

households (18.4%). The characteristics effect of occupation matter much more for SC households

(21.8%) as compared to OBC (12.4%) and ST households (12.1%). Surprisingly, the coefficients

effect of education is negative for all three social groups, suggesting that the returns to education lead

to higher living standards for OBC, SC and ST households as compared to ACC households. The

coefficients effect of occupation is relatively unimportant in explaining the rural standard gap.

Differences in locational distribution are more important in explaining the living standard gap for

ST households as compared to SC and OBC households.

15
The detailed results (available on request) on the characteristics effect for educational suggest that
for the urban sample, a lower level of higher education attainment (higher secondary and above) in
the three disadvantaged social groups is the main reason why the characteristics effect of education
explains a large proportion of the living standard gap for OBCs, SCs and STs - the characteristics
effect related to higher secondary education and above explain 28.9, 23.9 and 18.8 per cent of the
gap for OBC, SC and ST households relative to ACC households respectively. In contrast, for the
rural sample, lower level of attainment in the below primary level explains17.8, 16.7 and 14.8 per
cent of the living standard gap between OBC, SC and ST households relative to ACC households
respectively.
 22

Our aggregate results indicate that the causes of the living standard gap between the OBC and

ACC s are broadly similar to those for the SC and ST – the characteristics effects explain much of

the urban living standard gap, while the coefficients and characteristics effects have approximately

equal roles in explaining the rural living standard gap. The disaggregated results indicate that the

causes of lower living standards for OBCs in urban areas are broadly similar to the other two social

groups and can be attributed mainly to lower levels of educational attainment among all three social

groups as compared to the rest of the population. However, in rural areas, the causes of the lower

living standards of the OBC as compared to the ACC s are not identical to those for SC and ST, with

locational and occupational characteristics playing a less important role as compared to differences

in educational characteristics.

6. Summary and Conclusions

Recently, compensating discrimination has been extended in India to the OBC for admission into

higher education institutions. This follows, partly, the recommendations of the Mandal Commission,

set up in 1979 by the Janata Party government under Prime Minister Morarji Desai with a mandate

to identify and suggest policy towards the socially or educationally backward. We use the 55th round

of India’s National Sample Survey (NSS) on consumer expenditure which for the first time

distinguishes Other Backward Classes from other castes and tribes to examine whether the Other

Backward Classes deserves recognition as a distinct social group requiring special social welfare

programs.

By employing regression estimates of per capita expenditure and an Oaxaca decomposition

analysis, we study how these differences in living standards arise. We undertake the decomposition
 23

analysis separately for rural and urban households, as the underlying causes for the differences in

living standards may be different rural and urban areas. Using a decomposition equation we can

explain differences in living standards in terms of differences in characteristics (characteristics effect)

and differences in the coefficients (coefficients effect).

We find that the causes of the living standard gap between the Other Backward Classes and

ACCs are broadly similar to those for the Scheduled Castes and Scheduled Tribes – the

characteristics effects explain much of the urban living standard gap, while the coefficients and

characteristics effects have approximately equal roles in explaining the rural living standard gap. In

particular, for Other Backward Classes households, lower levels of educational attainment as

compared to ACC households are crucial in explaining the living standard gap for Other Backward

Classes households, whether in rural or urban areas. In the case of Scheduled Castes and Scheduled

Tribes households, along with lower levels of educational attainment, occupational structure (in the

case of Scheduled Castes households) and locational characteristics (in the Scheduled Tribes

households) also seem to matter in explaining living standard gap, especially in rural areas. Our

findings suggest that an important role for policy in redressing the gap in living standard between

OBCs and ACCs is to increase the educational attainment of OBC households, both in rural and urban

areas. In this respect, the recent judgement of the Indian Supreme Court to extend compensatory

discrimination to OBCs for entry to higher education may help. There is also a role for the

government to expand the provision of schooling to rural areas, specifically targeted to the socially

disadvantaged groups in India.

 24

REFERENCES

Bayly, S. (1999), Caste, Society and Politics in India, Cambridge: Cambridge University Press

Beteille, A.(1992), The Backward Classes in Contemporary India. Delhi: Oxford University Press.

Bhaumik, S.K. and M. Chakrabarty (2006), Earnings Inequality in India: Has the Rise of Caste and

Religion based Politics in India had an impact?, IZA Discussion Paper 2008.

Borooah, V.K. (2005), Caste, Inequality and Poverty in India, Review of Development Economics

9(3), 399-414.

Booroah, V., A. Dubey and S. Iyer (2007), The Effectiveness of Job Reservations: Caste, Religion

and Economic Status in India, Development and Change 38(3), 423-445.

Datt, G. and M. Ravallion (1998), Why Have Some Indian States Done Better Than Others at

Reducing Rural Poverty?, Economica 65(257), 17-38.

Dubey, A., S. Gangopadhyay and W. Wadhwa (2001), Occupational Structure and Incidence of

Poverty in Indian Towns of Different Sizes, Review of Development Economics 5(1), 49-59.

Galanter, M. (1984), Competing Equalities: Law and the Backward Classes in India, Berkeley:

University of California Press.

Gang, I., K. Sen and M.-S.Yun (2008), Poverty in Rural India: Caste and Tribe, Review of Income and

Wealth 54(1), 50-70.

Kijima, Y. (2006), Caste and Tribe Inequality: Evidence from India, 1983-1999, Economic

Development and Cultural Change 54(2), 369-404.

Krishnaji, N. (1984), Family-size, Levels of Living and Differential Mortality in Rural India - Some

Paradoxes, Economic and Political Weekly 19(6), 248-258.

Mcharg, A and D. Nicolson (2006), Debating Affirmative Action, London: Blackwells.

 25

Oaxaca, R. (1973), Male-Female Wage Differentials in Urban Labor Markets, International Economic

Review 14 (3), 693-709.

Oaxaca, R.and M. R. Ransom (1999), Identification in Detailed Wage Decompositions, Review of

Economics and Statistics 81(1), 154-157.

Palmer-Jones, R. and K. Sen, (2003), What has luck got to do with it? A regional analysis of poverty

and agricultural growth in rural India, Journal of Development Studies 40(1), 1-33.

Radhakrishnan, P. (1996), Mandal Commission Report: A Sociological Critique, in M.N. Srinivas

(ed.), Caste: Its Twentieth Century Avatar, Delhi: Penguin India.

Ramaiah, A. (1992), Identifying Other Backward Classes in India, Economic and Political Weekly,

6th June: 1203-1207.

Revankar, R.G.(1971), The Indian Constitution - A Case Study of Backward Classes. Rutherford:

Fairleigh Dickinson University Press.

Sivaramayya, B. (1996), The Mandal Judgement: A Brief Description and Critique, in M.N. Srinivas

(ed.), Caste: Its Twentieth Century Avatar, Delhi: Penguin India.

Thorat, S.K. (2004), Caste System in India: Social and Economic Exclusion and Poverty, Indian

Institute of Dalit Studies, New Delhi.

Thorat, S.K., Aryamma, and P. Negi (2005), Reservations and the Private Sector: Quest for Equal

Opportunity and Growth, Delhi: Rawat Publications.

Weisskopf, T. (2004), Affirmative Action in the United States and India: A Comparative Perspective,

New York: Routledge.

Yuill, K. (2006), Richard Nixon and the Rise of Affirmative Action: The Pursuit of Racial Equality

in an Era of Limits, New York: Rowman and Littlefield.

 26

Yun, M.-S. (2005), A Simple Solution to the Identification Problem in Detailed Wage

Decomposition, Economic Inquiry 43(3), 766-772.

 27

Table 1: Sample Means, Rural Households

Variables Other Backward Scheduled Castes Scheduled Tribes ACCs
Classes
Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Monthly Per Capita Expenditure 511.520 342.603 449.924 346.036 412.533 221.478 624.774 403.991
Log of Monthly Per Capita Expenditure 6.123 0.451 6.007 0.424 5.919 0.438 6.306 0.485
Demographic and Other Control Variables
Age 43.958 12.423 42.654 12.215 41.996 12.786 44.699 12.454
Household Size 5.103 2.557 4.884 2.271 4.964 2.280 5.174 2.625
Land Owned (hectares) 0.710 1.502 0.313 0.806 0.817 1.296 0.971 2.172
Education Variables
Not Literate 0.505 0.500 0.617 0.486 0.653 0.476 0.359 0.480
Literate, below primary 0.155 0.361 0.144 0.351 0.137 0.344 0.154 0.361
Literate, below secondary 0.231 0.421 0.171 0.376 0.152 0.359 0.277 0.448
Literate, secondary 0.061 0.239 0.040 0.197 0.028 0.164 0.113 0.316
Literate, higher secondary and above 0.049 0.216 0.028 0.165 0.029 0.169 0.098 0.297
Occupation Variables
Self-employed in non-agriculture 0.155 0.362 0.121 0.327 0.054 0.225 0.150 0.357
Self-employed in agriculture 0.352 0.478 0.164 0.370 0.367 0.482 0.414 0.493
Agricultural labor 0.313 0.464 0.540 0.498 0.441 0.497 0.209 0.406
Non-agricultural labor 0.079 0.269 0.096 0.295 0.081 0.272 0.064 0.244
Miscellaneous 0.102 0.303 0.078 0.269 0.057 0.232 0.163 0.370
Number of Observations 23630 12316 9706 22171
Notes: Sample means are calculated using the individual household multiplier.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS.
 28

Table 2: Sample Means, Urban Households

Variables Other Backward Scheduled Castes Scheduled Tribes ACCs
Classes
Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Monthly Per Capita Expenditure 849.743 2339.620 686.897 447.908 772.917 635.202 1143.760 1054.780
Log of Monthly Per Capita Expenditure 6.546 0.551 6.388 0.514 6.452 0.599 6.841 0.607
Demographic and Other Control Variables
Age 42.375 12.066 41.137 11.613 40.866 11.235 43.366 12.015
Household Size 4.598 2.342 4.807 2.325 4.560 2.202 4.499 2.347
Education Variables
Not Literate 0.250 0.433 0.371 0.483 0.342 0.475 0.123 0.329
Literate, below primary 0.133 0.339 0.143 0.350 0.120 0.325 0.086 0.280
Literate, below secondary 0.301 0.459 0.263 0.440 0.221 0.415 0.229 0.420
Literate, secondary 0.153 0.360 0.107 0.310 0.113 0.317 0.188 0.391
Literate, higher secondary and above 0.163 0.370 0.115 0.319 0.203 0.402 0.374 0.484
Occupation Variables
Self-employed 0.379 0.485 0.268 0.443 0.238 0.426 0.364 0.481
Regular wage/salary earning 0.366 0.482 0.394 0.489 0.405 0.491 0.476 0.499
Casual labor 0.186 0.389 0.289 0.453 0.254 0.435 0.077 0.267
Miscellaneous 0.069 0.253 0.048 0.215 0.103 0.304 0.083 0.276
Number of Observations 12626 5670 2927 23671
Notes: Sample means are calculated using the individual household multiplier.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS.
 29

Table 3: Determinants of (log) Monthly Per Capita Expenditure for Rural Households
Other Scheduled Scheduled ACCs
Backward Castes Tribes
Classes
Intercept 6.208*** 6.231*** 6.092*** 6.350***
(0.048) (0.062) (0.094) (0.063)
Demographic Control Variables
Age 0.015*** 0.014*** 0.012*** 0.013***
(0.002) (0.002) (0.003) (0.002)
*** *** ***
Age Square -0.012 -0.012 -0.009 -0.009***
(0.002) (0.003) (0.003) (0.003)
*** *** ***
Household size -0.128 -0.147 -0.138 -0.126***
(0.005) (0.006) (0.009) (0.007)
*** *** ***
Household size squared 0.005 0.006 0.005 0.004***
(0.0003) (0.001) (0.001) (0.0005)
Land Owned (hectares) 0.039*** 0.036*** 0.034*** 0.027***
(0.004) (0.006) (0.005) (0.003)
Education Variables – Reference Group: ‘Not Literate’
Literate, below primary 0.067*** 0.069*** 0.090*** 0.072***
(0.009) (0.011) (0.014) (0.010)
Literate, below secondary 0.153*** 0.124*** 0.146*** 0.149***
(0.008) (0.011) (0.015) (0.009)
*** *** ***
Literate, secondary 0.260 0.207 0.239 0.274***
(0.014) (0.021) (0.032) (0.014)
Literate, higher secondary and above 0.382*** 0.340*** 0.324*** 0.414***
(0.019) (0.032) (0.035) (0.018)
Occupation Variables - Reference Group: ‘Miscellaneous’
Self-employed in non-agriculture -0.023 -0.080*** -0.112*** -0.063***
(0.014) (0.022) (0.035) (0.015)
** ***
Self-employed in agriculture -0.020 -0.048 -0.130 -0.050***
(0.014) (0.021) (0.031) (0.014)
*** *** ***
Agricultural labor -0.176 -0.192 -0.228 -0.230***
(0.014) (0.020) (0.031) (0.016)
*** *** ***
Non-agricultural labor -0.066 -0.120 -0.146 -0.103***
(0.017) (0.025) (0.037) (0.018)
R-squared 0.401 0.384 0.485 0.458
Notes: a) Observations are weighted by the individual household multiplier. b) Dependent variable is the
natural logarithm of monthly per capita expenditure. c) Standard errors in parentheses are robust to
heteroskedasticity and clustered residuals within villages. d) *** and ** denote significance at the 1 and
5 percent respectively. e) Though estimates are not reported, regional dummy variables are included in the
regression estimation.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS; our calculations.
 30

Table 4: The Determinants of (log) Monthly Per Capita Expenditure for Urban Households
Other Scheduled Scheduled ACCs
Backward Castes Tribes
Classes
Intercept 6.613*** 6.359*** 6.292*** 6.443***
(0.071) (0.098) (0.191) (0.071)
Demographic Control Variables
Age 0.011*** 0.016*** 0.017** 0.020***
(0.003) (0.004) (0.008) (0.003)
Age Square -0.008** -0.013*** -0.014 -0.013***
(0.003) (0.004) (0.009) (0.003)
Household size -0.204*** -0.176*** -0.212*** -0.188***
(0.008) (0.013) (0.019) (0.007)
Household size squared 0.009*** 0.007*** 0.011*** 0.008***
(0.001) (0.001) (0.002) (0.001)
Education Variables – Reference Group: ‘Not Literate’
Literate, below primary 0.110*** 0.092*** 0.124** 0.100***
(0.017) (0.019) (0.051) (0.018)
Literate, below secondary 0.197*** 0.166*** 0.240*** 0.191***
(0.014) (0.018) (0.039) (0.017)
Literate, secondary 0.370*** 0.249*** 0.394*** 0.374***
(0.018) (0.024) (0.049) (0.015)
Literate, higher secondary and above 0.596*** 0.541*** 0.704*** 0.672***
(0.019) (0.029) (0.050) (0.023)
Occupation Variables - Reference Group: ‘Miscellaneous’
Self-employed 0.034 0.041 0.038 -0.002
(0.027) (0.041) (0.088) (0.023)
Waged/salaried workers 0.082*** 0.190*** 0.133 -0.003
(0.028) (0.038) (0.085) (0.024)
Casual labor -0.135*** -0.105*** -0.142* -0.248***
(0.030) (0.038) (0.086) (0.036)
R-squared 0.480 0.507 0.615 0.503
Notes: a) Observations are weighted by the individual household multiplier. b) Dependent variable is the
natural logarithm of monthly per capita expenditure. c) Standard errors in parentheses are robust to
heteroskedasticity and clustered residuals within block. d) *** and ** denote significance at the 1 and 5
percent respectively. e) Though estimates are not reported, regional dummy variables are included in the
regression estimation.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS; our calculations.
 31

Table 5: Decomposition of the Gap in Per Capita Expenditures Between Other Backward
Classes vs. ACCs: Aggregate and Sub-Aggregate Effects
Characteristics Effect Coefficients Effect
Estimate Share (%) Estimate Share (%)
Rural Households
Aggregate Effects 0.100*** 54.7 0.083*** 45.3
(0.008) (0.010)

Intercept - - 0.100 54.6

(0.074)
Land Owned 0.007*** 3.8 -0.009*** -4.8
(hectares) (0.001) (0.003)
Age 0.004*** 2.0 -0.016 -9.0
(0.0002) (0.070)
Household Size -0.004*** -2.2 0.005 2.7
(0.0001) (0.024)
Education 0.041*** 22.5 -0.007 -3.8
(0.001) (0.005)
Occupation 0.023*** 12.4 -0.004 -2.4
(0.001) (0.004)
NSS Region 0.030*** 16.2 0.015 8.1
Dummy Variables (0.007) (0.010)
Urban Households
Aggregate Effects 0.213*** 72.2 0.082*** 27.8
(0.009) (0.012)
Intercept - - -0.190** -64.3
(0.094)
Age 0.009*** 3.0 0.244*** 82.7
(0.0005) (0.094)
Household Size 0.012*** 4.0 0.031 10.5
(0.0003) (0.029)
Education 0.136*** 46.2 -0.003 -1.0
(0.004) (0.002)

Occupation 0.027*** 9.1 -0.007 -2.4

(0.003) (0.006)
NSS Region 0.029*** 9.9 0.007 2.3
Dummy Variables (0.008) (0.012)
Note: a) Standard errors in parentheses. b) ***, **, and * denote significance at the 1, 5 and 10 percent
respectively. c) Share is the percentage of 0.183 log-points in rural, and of 0.295 log-points in urban.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS.
 32

Table 6: Decomposition of the Gap in Per Capita Expenditures Between Scheduled

Castes vs. ACCs: Aggregate and Sub-Aggregate Effects
Characteristics Effect Coefficients Effect
Estimate Share(%) Estimate Share(%)
Rural Households
Aggregate Effects 0.168*** 56.2 0.131*** 43.8
(0.008) (0.011)
Intercept - - 0.030 10.2
(0.079)
Land Owned (hectares) 0.018*** 5.9 -0.003 -0.9
(0.002) (0.002)
Age 0.010*** 3.4 0.033 11.0
(0.001) (0.075)
Household Size -0.016*** -5.2 0.050* 16.6
(0.001) (0.026)
Education 0.065*** 21.8 -0.024*** -8.1
(0.002) (0.008)
Occupation 0.065*** 21.8 -0.016*** -5.4
(0.004) (0.006)
NSS Region Dummy Variables 0.026*** 8.6 0.061*** 20.4
(0.006) (0.010)

Urban Households
Aggregate Effects 0.320*** 70.6 0.133*** 29.4
(0.006) (0.012)
Intercept - - 0.016 3.5
(0.111)
Age 0.019*** 4.1 0.178* 39.3
(0.001) (0.105)
Household Size 0.037*** 8.1 -0.037 -8.1
(0.001) (0.040)
Education 0.192*** 42.4 -0.022*** -4.8
(0.006) (0.005)
Occupation 0.052*** 11.5 -0.034*** -7.5
(0.005) (0.009)
NSS Region Dummy Variables 0.020*** 4.5 0.031*** 6.9
(0.003) (0.011)
Note: a) Standard errors in parentheses. b) ***, **, and * denote significance at the 1, 5 and 10 percent
respectively. c) Share is the percentage of 0.299 log-points in rural, and of 0.453 log-points in urban.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS.
 33

Table 7: Decomposition of the Gap in Per Capita Expenditures Between Scheduled

Tribes vs. ACCs Aggregate and Sub-Aggregate Effects
Characteristics Effect Coefficients Effect
Estimate Share(%) Estimate Share(%)
Rural Households
Aggregate Effects 0.207*** 53.5 0.180*** 46.5
(0.012) (0.015)
Intercept - - 0.043 11.1
(0.085)
Land Owned (hectares) 0.004*** 1.1 -0.006 -1.5
(0.0004) (0.005)
Age 0.013*** 3.3 0.050 13.0
(0.001) (0.082)
Household Size -0.009*** -2.4 0.045 11.6
(0.001) (0.032)
Education 0.071*** 18.4 -0.020** -5.2
(0.003) (0.010)
Occupation 0.047*** 12.1 0.0003 0.1
(0.003) (0.008)
NSS Region Dummy Variables 0.081*** 21.0 0.067*** 17.4
(0.011) (0.015)

Urban Households
Aggregate Effects 0.260*** 66.8 0.129*** 33.2
(0.010) (0.018)
Intercept - - -0.014 -3.7
(0.185)
Age 0.020*** 5.1 0.125 32.2
(0.001) (0.178)
Household Size 0.012*** 3.2 0.032 8.2
(0.001) (0.056)
Education 0.141*** 36.3 0.003 0.7
(0.004) (0.007)
Occupation 0.043*** 11.2 -0.021 -5.4
(0.004) (0.013)
NSS Region Dummy Variables 0.043*** 11.1 0.005 1.3
(0.009) (0.013)
Note: a) Standard errors in parentheses. b) ***, **, and * denote significance at the 1, 5 and 10 percent
respectively. c) Share is the percentage of 0.387 log-points in rural, and of 0.389 log-points in urban.
Source: 55th round (1999/2000) of the consumer expenditure survey of the NSS.