Political Reservations and Rural Public Good Provision in India
Nandini Krishnany
Boston University
November 2007
Abstract
This paper asks whether political representation for historically disadvantaged groups
can serve as an e¤ective means to redress historical inequities. The Indian Constitution mandates reservations for two such groups, Scheduled Castes (SC) and Scheduled
Tribes (ST), in federal and state legislatures. While this policy has been successful in
ensuring the political presence of minority legislators, evidence on their performance in
serving the interests of their electorate is limited to the analysis of aggregate expenditure patterns. This paper is the …rst to establish a causal link between the reservation
of seats in state legislatures and rural public good provision, and to explore its implications for the inter- and within-district pattern of provision. The sample is drawn from 9
Indian states, and uses village census data aggregated to 64 districts and 610 electoral
constituencies. The empirical strategy exploits features of the process of reservation to
identify an exogenous source of variation that a¤ects reservation discontinuously. With
a few exceptions, I …nd no evidence that ST legislators perform any di¤erently than legislators elected from unreserved constituencies. However, SC legislators perform better,
providing greater access to educational facilities, in particular, primary schools, within
their constituencies. In addition, these schools are located so as to favor of their own
community. At the same time, I do …nd evidence that other minority groups bene…t from SC reservation as well. Overall, these …ndings do not support the prediction
that political reservation adversely a¤ects the responsiveness to minority interests or
politician quality, at least insofar as these a¤ect the provision of public goods.
I am indebted to Dilip Mookherjee for his constant support and invaluable guidance, and Kevin Lang
and Robert Margo for their comments and insights. I gratefully acknowledge the Registrar-General of India
and the Election Commision of India for allowing the use of their data. I also thank Nethra Palaniswamy,
Rashmi Barua, Neha Rati Kumar, Chun-Yu Ho, Vivek Mani and seminar participants at Boston University
for suggestions and encouragement. Any errors are solely mine.
y
Ph.D. Candidate, Department of Economics, Boston University, Boston MA 02215; [email protected]
1
1
Introduction
One of the main objectives of the Constitution of India at the time of Independence was to
reduce economic and social disparities through a program of a¢ rmative action in favor of
historically disadvantaged groups (Beteille, 1992). An important component of this policy
consists of mandatory political reservations in federal and state legislatures for two groups,
the Scheduled Castes (SC) and the Scheduled Tribes (ST)1 . The term Scheduled Castes is
intended to encompass groups isolated and disadvantaged by their low status in the caste
hierarchy (Galanter, 1984). The Scheduled Tribe category includes groups distinguished
by their geographic isolation from the rest of the population, as well as their linguistic
and cultural distinctiveness. Scheduled Castes and Tribes make up about 16% and 8% of
India’s population respectively (Census of India, 2001). While this reservation policy has
been successful in ensuring the political presence of minority legislators, evidence on their
performance in serving the interests of their electorate is limited to the analysis of the e¤ect
of reservations in state legislatures on state government expenditure patterns (Pande, 2003).
This paper uses a panel dataset to examine the e¤ect of this guaranteed political representation in state legislative assemblies in India on the provision of rural public goods.
First, I ask whether reservations for Scheduled Castes and Scheduled Tribes lead to di¤erences in the quantity and pattern of public good provision at the level of the district, the
most important administrative unit below the state. If ‘reserved’legislators do provide different kinds of or quantities of public goods in their own districts, these goods may bene…t
constituents other than the ones these legislators represent. Therefore, I examine whether
these public goods are located within their own electoral constituencies, and in favor of
members of the group that bene…ts from reservation. Finally, by examining the e¤ect of the
policy at both the administrative as well as the electoral unit, this paper aims to shed light
on the ability of minority legislators to in‡uence the inter- and intra-district allocation of
resources, in terms of the provision of public goods.
My empirical strategy exploits institutional features of the process of reservation of
electoral constituencies to identify the e¤ects of mandated representation on public good
outcomes. The number of constituencies deemed reserved for a group within a district is
a discontinuous and non-linear function of past census population shares of the Scheduled
Castes and Scheduled Tribes. Within a district, the constituencies deemed reserved are
those where, as far as feasible, the group’s past census population share in the constituency
is the highest amongst constituencies in the district2 . Controlling for current population
1
Articles 330 and 332 of the Constitution of India guarantee political representation in federal and state
legislatures respectively.
2
Seats for the STs "are to be reserved in the constituencies in which the percentage of their popu-
2
shares, this generates a potentially exogenous, non-linear source of variation at the level of
the electoral constituency that can be used to estimate the e¤ects of political reservation.
This paper contributes to the literature on the economic impact of political reservations
in various ways. First, it goes beyond current empirical work on the e¤ect of political
reservations in state legislatures in India by linking the reservation policy to the provision
of public goods as opposed to identifying its e¤ect on aggregate expenditure patterns. Given
the often tenuous link between expenditures and actual public good provision in developing
country contexts, this approach provides a more meaningful measure of the impact of the
reservation policy. As shown in Pande (2003), state level reservations lead to increases in
transfers to groups that bene…t from reservation. However, if reservation adversely a¤ects
politician quality, or if minority legislators lack political clout, these expenditures may not
translate into public good outcomes. Alternatively, the e¤ectiveness of reservation as a
means to redress economic inequality may be captured better by measures other than state
level expenditures. Even if minority legislators are successful in directing funds to their own
districts, they may be unable to in‡uence the allocation of these funds across constituencies
within the district or across di¤erent groups of constituents. This paper …lls this gap in
the literature by linking the state level reservation policy to the provision of public goods
across and within districts. In addition, the use of village census information enables me
to identify the e¤ect of reservation on the allocation of public goods across constituencies
and in favor of certain groups or communities. Finally, this paper provides insights into
the e¤ect of reservation on the pattern of allocation of resources at two important units of
analysis, one administrative and the other electoral.
In theory, mandatory representation for disadvantaged groups should lead to a shift in
the distribution of public goods if the identity of the decision maker a¤ects the distribution
of public goods and if the policy maker favors members of her own group. For the latter to be
true, the preferences of di¤erent groups over public goods must di¤er, and members of these
groups must be underrepresented in the absence of reservation (Du‡o, 2005). Historically,
both Scheduled Castes and Scheduled Tribes have been characterized by lower literacy rates
and higher poverty rates relative to the rest of the population and continue to lag behind
in terms of economic and social indicators of welfare. There is also empirical evidence that
in the 1970s, these groups were poorly served in terms of access to public goods (Banerjee
and Somanathan, 2007). Moreover, both groups are under-represented in the absence of
reservation (Galanter, 1984; Mandelsohn and Vicziany, 1998). For instance, Scheduled
lation to the total population is the largest". The "constituencies for SCs are to be distributed in different parts of the State and seats are to be reserved for SCs in those constituencies where the percentage of their population to the total population is comparatively large". (http://www.delimitationindia.com/Constitution/ConstProv332.pdf)
3
Caste and Scheduled Tribe representation in the upper houses of parliament, where there is
no reservation of seats, is disproportionately low. In addition, candidates from these groups
are rarely …elded in unreserved constituencies.
In India, the reservation policy does not involve separate electorates for disadvantaged
groups. The candidates who stand for election from ‘reserved’constituencies must belong to
the relevant group, but the entire electorate votes to decide the winning candidate. While
there is little doubt reservations have been crucial in ensuring political representation for
hitherto underrepresented groups (Galanter, 1984), there is much debate on its e¤ectiveness
as a means to in‡uence policy and reduce economic disparities. Elected members of state
legislative assemblies or MLAs (Members of Legislative Assembly) can in‡uence rural public
good provision through their role in the legislative assembly, as members of government,
heads of government departments, nominated members of local government bodies and
as representatives of their constituency (Singh, 1997). In light of the recent trends of
increasing political participation and mobilization among the lower castes in India (Yadav,
2002; Hasan, 2002), and the accompanying demands for more widespread reservations in
employment, it is important to determine the economic e¤ects of political representation.
These features of the Indian experiment with guaranteed political representation generate implications for the e¤ect of reservations on the provision of public goods. In particular,
to the extent that minority groups have di¤erent preferences over public goods relative to the
rest of the population and have historically su¤ered from poor provision of goods, one may
expect that areas represented by minority MLAs get a di¤erent bundle of goods. However,
the ability of Scheduled Caste and Scheduled Tribe MLAs to respond to the preferences
of their own community is mitigated by the incentives created by a policy of reservation
that does not involve separate electorates. For instance, Scheduled Castes are a demographic minority in the constituencies that are reserved for them, and therefore, Scheduled
Caste MLAs may have to cater to the preferences of non-Scheduled Caste groups as well.
Another factor in‡uencing the ability of minority legislators to respond to group-speci…c
preferences and relative deprivation is the political mobilization and clout of these groups.
Greater political clout for the group as a whole implies legislators may be more e¤ective
in parliament in acting en bloc to implement policy, and in securing resources and implementing programs within their own constituency. Thus, the estimated e¤ect of reservation
on the actual provision of public goods re‡ects di¤erences in group preferences and in the
existing pattern of access to public goods, the political clout of minority legislators and the
incentives generated by the design of political reservations in India.
My …rst set of results estimates the e¤ect of political reservation on the inter-district
provision of public goods, speci…cally the proportion of villages in the district which have
4
a particular public good. My results indicate that the reservation of an additional constituency for Scheduled Castes in a district increases the probability that a village in the
district has an educational facility. In particular, districts represented by a greater number
of reserved Scheduled Caste MLAs have a higher proportion of villages with access to a
primary school. In contrast, districts with a higher number of seats reserved for Scheduled
Tribes have lower measures of access to public health facilities such as primary health care
centres and maternal and child health care centres. However, some of the results on the
e¤ects of Scheduled Tribe reservation are not robust to alternative speci…cations.
To see if these public goods are being located in constituencies where the reserved seat
was placed, I examine the e¤ect of reservation on the pattern of public good provision
across assembly constituencies within a district. The reservation of a constituency for a
Scheduled Caste candidate also increases the probability that a village in the constituency
has an educational facility, as measured by the presence of a school and by the presence
of a primary school. The results for Scheduled Tribe reservation at the constituency level
show no discernible di¤erences in the pattern or quantity of public good provision relative
to unreserved constituencies.
There are at least two explanations for these …ndings. The …rst is that members of
Scheduled Tribes have similar preferences over public goods as the rest of the population.
Given the distinct di¤erences between Scheduled Tribes and the rest of the population,
this seems unlikely. The second explanation has to do with the relative deprivation of the
two groups with respect to di¤erent public goods and di¤erences in the degree of political
mobilization between the two groups. If Scheduled Castes have been historically underprovided in terms of educational facilities, Scheduled Caste MLAs may be responding to
this need by providing primary schools. In addition, the Scheduled Castes have grown
increasingly successful in the political arena in recent decades. As a result, Scheduled Caste
MLAs may be much more e¤ective as legislators and as representatives of their community.
The second set of results relates to the e¤ect of reservations on the inter-district and
inter-constituency allocation of public goods across di¤erent groups of constituents. For
instance, Scheduled Caste legislators could be locating primary schools in villages with
predominantly non-Scheduled Caste populations. My …ndings suggest that this is not the
case. In districts with greater representation for Scheduled Castes, Scheduled Caste MLAs
locate schools so as to favor their own community, while bene…ting other groups as well.
Scheduled Tribe reservation may also lead to the location of schools in favor of Scheduled
Tribes within the district, although this e¤ect is imprecisely estimated.
The results on the allocation of public goods within constituencies show a somewhat
similar pattern. It is interesting that the reservation of constituencies for Scheduled Castes
5
results in greater per capita access for Scheduled Tribes to educational facilities, including
primary schools. Thus, relative to an unreserved constituency, a Scheduled Tribe person is
more likely to live in a village with a school if he or she lives in a constituency reserved for
Scheduled Castes. My point estimates also consistently indicate a larger e¤ect on Scheduled
Caste access than on access for the population as a whole. However, the estimates on
Scheduled Caste access are imprecisely estimated at the constituency level, and fall short
of statistical signi…cance at conventional levels.
This paper contributes to the literature on the e¤ect of political reservations in India on
policy outcomes and the distribution of public goods. Evidence is currently limited to the
analysis of aggregate expenditure patters, or to the distribution of public goods provided by
local governments often constrained by the limited devolution of funds, functions and functionaries. Chattopadhyay and Du‡o (2004) evaluate the e¤ect of reservations for minorities
in local government, and …nd that while reservations do not a¤ect the types of public goods
provided, they do lead to greater public investment in Scheduled Caste hamlets. Besley et
al (2004) …nd that reservation of the village council head position improved within-village
targeting for low spillover goods in favor of the minority group. Pande (2003) examines
the e¤ect of reservations on state government expenditures, and …nds that an increase in
state-level reservations for Scheduled Castes and Scheduled Tribes leads to an increase in
targeted transfers (job quotas and Scheduled Tribe welfare spending respectively). In this
paper, I provide evidence that state level reservations for Scheduled Castes do alter the
pattern of public goods provided across districts and constituencies. Overall, my …ndings
do not suggest that political reservations adversely a¤ect the interests of minority groups.
2
Political Reservations in India
Political reservations are an important component of a policy of compensatory discrimination that as been followed in India since Independence. The Indian Constitution guarantees
seats in national and state legislatures for Scheduled Castes and Tribes (Articles 330 and
332). Despite being time-bound, these provisions have been extended …ve times, and are
currently in place until 20103 . Prior to elections, speci…ed jurisdictions are deemed reserved
for Scheduled Castes and Scheduled Tribes. The extent of political reservation in state
legislatures re‡ects each group’s share in the state population. States are divided into districts, which are the most important lower administrative unit, and each district is typically
comprised of a number of electoral units called assembly constituencies (Figures 1 and 2).
3
Article 334, Constitution of India
6
Each assembly constituency elects one member to the state legislature, according to the
‘…rst past the post’ rule. In assembly constituencies reserved for Scheduled Castes, only
candidates belonging to this group can contest elections, while the entire electorate votes
to choose the winning candidate.
The demarcation of constituency boundaries and the designation of constituencies as
reserved falls under the purview of a quasi-constitutional body, the Delimitation Commission. Constituency boundaries can be redrawn and constituency status changed only when
new census population estimates become available. The aim of this exercise is to ensure
as far as possible that each constituency is equal in terms of population. While the census
exercise is undertaken every ten years, the delimitation exercise has been carried out only
four times since Independence–1952, 1963, 1973, and 2002. Between 1976, when the third
delimitation exercise was implemented, and 2001, there has been no change in constituency
boundaries or reservation status. Thus, for the time period considered in this paper, the
decade 1991-2001, constituency boundaries and reservation status were determined based
on the 1971 Census of India.
In selecting constituencies for reservation for Scheduled Tribes, the only criteria is the
concentration of the Scheduled Tribe population. Since Scheduled Tribes also tend to be
geographically concentrated, a large proportion of the Scheduled Tribe population falls
in constituencies reserved for Scheduled Tribe, while a large part of the electorate in a
constituency reserved for Scheduled Tribes consists of Scheduled Tribes (Galanter, 1984).
This has implications for the ability of an Scheduled Tribe MLA to re‡ect the preferences
of his or her own community in public good provision. To the extent that Scheduled Tribes
have di¤erent preferences from the rest of the population, the design of reservation combined
with their geographic concentration implies that it may be relatively easier for Scheduled
Tribe legislators to cater to the preferences of their own community.
In the case of seats reserved for Scheduled Castes, the Delimitation Commission uses two
criteria. Constituencies in which seats are ‘reserved’for these groups must be distributed
in di¤erent parts of the state, and located, as far as practicable, in those areas where the
proportion of their population to the total is comparatively large. Unlike the Scheduled
Tribes, Scheduled Castes are much more geographically dispersed, and therefore, in the
constituencies reserved for them, Scheduled Castes comprise a relatively smaller proportion
of the electorate. This implies that Scheduled Caste MLAs may have to cater more to
the demands of the majority non-Scheduled Caste members of their electorate. To the
extent that they do so, the pattern of public good provision in Scheduled Caste reserved
constituencies may be less likely to mirror the preferences of the Scheduled Caste community.
7
3
Empirical Strategy
The main di¢ culty in estimating the causal e¤ect of reservation is in separating the e¤ect
of systematic di¤erences between areas with reservation and areas without reservation from
the e¤ect of reservation itself. In order to estimate the e¤ect of reservation for Scheduled
Castes and Tribes on the amount and pattern of public goods provided, while simultaneously
controlling for any smooth e¤ects of current demographic shares, one could estimate the
following speci…cation:
ydst =
0
+
+
SC
1 (#ACd )
5 Xdst
+
+
s State
+
ST
2 (#ACd )
t T ime
+
3 f (SCsharedst )
+
4 g(ST sharedst )
+ "dst
where ydst is a measure of public good provision in district d in state s at time t, #ACdi
(i = SC; ST ) is the number of constituencies deemed reserved for group i within a district
d, f (:) and g(:) are polynomials, SCsharedst and ST sharedst are current group population
shares, Xdst represents a vector of district controls, and State and T ime represent state
and time dummies respectively.
OLS estimates of
1
and
2
may be biased since the number of constituencies deemed
reserved for each group within a district is not chosen at random. The number of constituencies reserved for Scheduled Castes and Scheduled Tribes within a district is determined by
each group’s 1971 census population share in the district and the number of electoral constituencies assigned to the district. Thus, 1971 census population estimates are correlated
with both the number of constituencies deemed reserved within a district as well as the
current group population shares.
The main concern in identifying the impact of reservations is that the e¤ects of lagged
population measures that are omitted may bias the estimates of the e¤ect of reservation.
This is because districts with a larger number of constituencies reserved for a group are
more likely to have had relatively higher concentrations of that group in the past. Thus,
districts with a larger number of constituencies reserved for Scheduled Tribes may be more
geographically isolated than other districts, given the characteristics of Scheduled Tribes.
Moreover, past demographic shares of these groups may be correlated with historical patterns of public good provision. Based on their work on the location of public goods across
rural India, Banerjee and Somanathan (2007) suggest that "group identities are strongly
correlated with access to public goods" and that Scheduled Castes and Scheduled Tribes
had poorer access to almost all goods in the 1970s. If patterns of provision have been historically biased against areas with higher populations of Scheduled Castes and Tribes, then
OLS estimates of the e¤ect of reservation will be biased downward. On the other hand, since
8
government programs for poverty alleviation and rural infrastructure creation are based on
measures of deprivation, districts with higher Scheduled Caste or Tribe populations may
be more likely to have bene…tted from such programs. This may bias OLS estimates of
reservation upward.
One possible strategy to identify the e¤ect of reservation may be to include district
dummies or use …rst-di¤erences to eliminate the e¤ect of unobserved and persistent district
e¤ects. However, the district level reservation variable is time-invariant in the period considered in this paper, the decade 1991-2001, which implies that any of the former approaches
will not allow me to separately estimate the e¤ect of reservation.
However, these concerns could be addressed by directly controlling for the variable that
causes reservation, the 1971 census population shares of each of these groups in the district.
This would yield the ideal speci…cation:
ydst =
0
+
+
SC
1 (#ACd )
5 Xdst
+
+
ST
2 (#ACd )
6 h(SCshareds71 )
+
+
3 f (SCsharedst )
7 l(ST shareds71 )
+
+
4 g(ST sharedst )
s State
+
t T ime
+ "dst
However, these census population shares, SCshareds71 and ST shareds71 are only available for districts as de…ned by their boundaries in 1971. About half the districts in my
sample have been reorganized since that time. While it is possible to calculate the 1971
population shares of the two groups according to the current boundaries of districts, and
I do so, this process is not error-proof. The inclusion of these potentially mismeasured
population shares in the above speci…cation in place of the actual population estimates
may lead to inconsistent estimates of the e¤ect of reservation. Moreover, my estimated
population shares and current population shares are very highly correlated. As a result,
this speci…cation yields very imprecise estimates of the e¤ects of reservation. Allowing the
1971 estimated population shares to proxy for current population e¤ects does not solve the
problem because the reservation variable is correlated with the variables that are measured
with error. Therefore, I instrument for reservation status, while allowing current population
shares to proxy for the e¤ects of 1971 population shares. If the instrument is valid, this
strategy will yield consistent estimates of the e¤ect of reservation.
In order to construct my instruments for reservation status, I exploit certain features of
the process of reservation. These features induce a discontinuity in the relationship between
1971 group population shares and the district level reservation variable at unit intervals (0.5,
1.5, 2.5 and so on). The number of reserved constituencies within a district are assigned as
the following discontinuous function of group census population shares and the number of
9
Assembly Constituencies in the district.
#ACdi = I[
i P opulationd
T otal P opulationd
#ACd ]
where I[:] is the nearest integer function, #ACd denotes the number of Assembly Constituencies in district d, #ACdi denotes the number of Assembly Constituencies reserved for
group i in district d, and i = SC; ST .
Consider two hypothetical districts each allocated 5 Assembly Constituencies, but with
di¤erent Scheduled Caste population shares in 1971. By applying the above rule, I would
predict that D1 will have a single assembly constituency deemed reserved for Scheduled
Castes while D2 will have none.
District Scheduled Caste Number of AC
Population Share in the district
D1
0.11
5
D2
0.09
5
Number of AC reserved
for SC in the district
1
0
Had district boundaries remained unchanged between 1971 and 2001, using this rule and
1971 census population information, it would be possible to perfectly predict the number
of Assembly Constituencies reserved in each district. As mentioned above, about half
of the districts in my sample have been reorganized, almost entirely along pre-existing
administrative boundaries (tehsil or block), but not along Assembly Constituency lines.
Thus, while the reorganization of districts has been exogenous to Assembly Constituency
boundaries and reservation status, I still need to predict the 1971 populations of districts
in accordance with their 2001 boundaries. This procedure is explained in detail in the
appendix.
I estimate the equation of interest using two-stage least squares regressions that exploit
a ‘fuzzy’ regression discontinuity design. I instrument for the number of constituencies
reserved for each group within a district using the instrument generated by applying the
reservation rule to the predicted 1971 group population shares. While past population
shares may have persistent and smooth e¤ects on public good provision, it is unlikely that
1971 census population shares discontinuously a¤ect the provision of public goods except
through the reservation variable. By using the instruments generated by the reservation
rule, I identify the e¤ect of reservation solely through the discontinuous e¤ect of 1971 group
population shares on the probability of reservation. Tables 1 and 2 report the number of
constituencies deemed reserved within a district in my sample for Scheduled Castes and
Scheduled Tribes respectively, as well as the predicted number of reserved constituencies.
To sum up, I examine the causal e¤ect of reservation on district level outcomes by
10
estimating equations of the following form by 2SLS:
ydst =
0
+
+
SC
1 (#ACd )
5 Xdst
+
+
s State
ST
2 (#ACd )
+
t T ime
+
3 f (SCsharedst )
+
4 g(ST sharedst )
+
4 g(ST sharedst )
+ "dst
where the "…rst-stage" relationship of interest is:
#ACdi =
0
+
+
di
1 (#AC d )
5 Xdst
+
+
s State
dj
2 (#AC d )
+
t T ime
+
3 f (SCsharedst )
+
dst
i
d d are the predicted number of constituencies reserved for group i (i; j =
and #AC
SC; ST ; i 6= j) in district d, the instruments generated by the reservation rule.
This empirical strategy is valid if current population shares are su¢ ciently good proxies
for 1971 population shares, and if there are no unobserved and unexpected changes in
district population shares over time that are correlated with both reservation and public
good outcomes. One may be concerned for instance, that the omitted 1971 population
shares a¤ect outcomes in some way that is not controlled for by current population shares,
which may be wrongly attributed to the e¤ect of reservation. Given the high correlation
between current population shares, and my estimated 1971 population shares, this is not
likely to lead to a large bias. Moreover, this bias, if any, is downward.
Consider two districts with similar current population shares, one of which is predicted to
have one assembly constituency reserved, and the other to have no assembly constituencies
reserved. This will be the case of my estimated 1971 population shares for the …rst district
are higher than those for the other. If the actual 1971 population shares are unusually
large given current population shares, then the omitted e¤ect of this additional population
could be wrongly attributed to the e¤ect of reservation. In the Indian context, areas with
unusually high past group population shares are also likely to be those with poorer provision
of public goods, and more likely to have a greater number of reserved seats. The negative
e¤ect of unusually higher past population shares on public good outcomes would then be
wrongly attributed to the e¤ect of reservation. However, this would only bias my estimates
of the e¤ect of reservation downward. To check that my empirical strategy is valid, I
also report estimates using alternative speci…cations and instruments as robustness checks.
These indicate that the bias, if any, is likely to be very small.
My empirical strategy to estimate the e¤ect of reservation on constituencies within a
11
district is similar. I estimate constituency-level equations of the following form:
yACdst =
0
+
+
SC
1 (RAC )
5 XACdst
+
+
ST
2 (RAC )
d District
+
+
3 f (SCshareACdst )
s State
+
t T ime
+
+
4 g(ST shareACdst )
ACdst
i
where yACdst is a measure of public good provision in constituency AC, RAC
is a dummy
variable that takes the value 1 if the constituency is reserved for group i (i = SC; ST ),
XACdst is a vector of constituency level controls, District, State and T ime represent a full
set of district, state and time dummies respectively.
Identi…cation of the reservation e¤ect at the level of the electoral constituency poses
very similar challenges as the district level analysis. Controlling for the variable that causes
reservation would yield the ideal speci…cation. In the period considered in this paper,
reservation of a constituency for Scheduled Tribes depends on the 1971 census population
share of Scheduled Tribes in the Assembly Constituency and on the number of Scheduled
Tribe reservations allocated to the district. In the case of Scheduled Caste reservations, an
additional standard is also applied–the geographical dispersal of reservations. Consider a
hypothetical district with 6 Assembly Constituencies, ranked in terms of the share of Scheduled Tribes in the Assembly Constituency population. Then, if two constituencies within
the district are to be reserved for Scheduled Tribes, the two with the highest concentrations
of Scheduled Tribes in 1971 will be reserved. The following table illustrates this example.
Assembly Constituency Scheduled Tribe share
in AC Population
AC1
0.4
AC2
0.5
AC3
0.35
AC4
0.23
AC5
0.45
AC6
0.28
Rank
3
1
4
6
2
5
Reservation
Status
G
ST
G
G
ST
G
To identify the e¤ect of reservation, my empirical strategy isolates the discontinuous
e¤ect of these 1971 population shares on the reservation variable, while simultaneously controlling for the smooth e¤ects of current population shares. To my knowledge, census population shares for 1971 are not publicly available at the level of the assembly constituency.
Therefore, I calculate 1971 group population shares in each constituency. This procedure
is explained in detail in the appendix. As before these cannot be directly included in the
regressions due to concerns about measurement error in these variables, and the resulting
inconsistency in the estimates of reservation. To address this issue, I allow current population shares to proxy for the e¤ects of 1971 shares, and instrument for reservation status at
12
the level of the assembly constituency. Using my predicted 1971 AC population and group
shares, a constituency is predicted as reserved for group i if
RankACwithin district [P redicted share of group i in AC population]
N umber of AC reserved f or group i in the district
Thus, the reservation variable is a discontinuous and non-linear function of past census
population shares. Identi…cation stems from the assumption that 1971 census population
shares discontinuously a¤ect public good provision only through the reservation variable.
Tables 3 and 4 report the reservation status of constituencies and their predicted reservation
status.
Thus, my equation of interest at the constituency level is estimated by two-stage least
squares:
yACdst =
0
+
+
SC
1 (RAC )
5 XACdst
+
+
ST
2 (RAC )
d District
+
+
3 f (SCshareACdst )
s State
+
t T ime
+
+
4 g(ST shareACdst )
ACdst
where the "…rst-stage" relationship is:
i
RAC
=
0
+
+
^i
1 (RAC )
5 XACdst
+
+
^j
2 (RAC )
d District
+
+
3 f (SCshareACdst )
s State
+
t T ime
+
+
4 g(ST shareACdst )
ACdst
^ i is the instrument generated by the reservation rule, i; j = SC; ST ; i 6= j:
where R
AC
In order to check for the validity of this speci…cation, I also report estimates from other
speci…cations and instruments as robustness checks.
4
Data
The dataset for this analysis has been compiled from various sources. Districts in my
sample are from 9 Indian states– Andhra Pradesh, Kerala, Karnataka, Madhya Pradesh,
Maharashtra, Orissa, Rajasthan, Tamil Nadu, and Uttar Pradesh. Within these states,
districts corresponding to villages in the ARIS- REDS (Additional Rural Income and Rural
Economic Development Surveys) household panel were chosen, to facilitate an extension
of this work to analyze the e¤ect of reservations on rural household welfare. The ARISREDS data were designed to constitute a nationally representative rural sample of Indian
households. While the ARIS-REDS covers all of India’s major states, my sample is restricted
to those for which I was able to aggregate village census information at both the district
13
and the constituency level. The …nal sample consists of 64 districts, and 610 Assembly
Constituencies. Data on district and constituency level controls, as well as information on
public good outcomes is from a panel of rural villages (1991 and 2001) from the decennial
Indian censuses, treated as a pooled cross-section.
Dependent variables on the availability of various public goods are aggregated to district and constituency level using village-level information. The aggregation of village-level
information to the level of the electoral constituency uses GIS information that maps census villages to their Assembly Constituencies4 . The availability of this information was the
main constraint in limiting the analysis to 9 states. Information about constituency boundaries and reservation status of each constituency is from various orders of the Delimitation
Commission.
I examine the e¤ect of political reservations on a range of public good outcomes, including education facilities, public health facilities, roads, power supply, and drinking water
facilities. Reserved constituencies do not show a consistently di¤erent pattern of provision
as far as roads, power supply and drinking water facilities are concerned. For expositional
ease, I report one set of regression results for these variables at each level of aggregation.
The …rst set of variables measure the amount and pattern of rural public goods. The
dependent variables considered are the proportion of villages within the district or constituency that have the relevant public good. The second set of variables measure the
location of public goods and whether they favor certain groups. These per capita access
variables measure the proportion of people who live in a village with the public good relative
to the rural population of the district or constituency. Per capita group access variables
measure the proportion of people belonging to group i who live in a village with the public
good relative to the population of group i in the district or the constituency.
Tables 5–8 report descriptive statistics for education and public health facilities as well
as controls at the district level for each year, by the number of constituencies reserved
within the district. In both periods in my sample, as the number of constituencies in a
district reserved for Scheduled Castes increases from 0 through 3, current population shares
of Scheduled Castes increase from 12 percent to 26 percent. The variation in population
shares is larger for Scheduled Tribes. In districts where no constituencies are reserved for
Scheduled Tribes, their population share is only 5 percent. However, in districts with more
than 4 constituencies reserved for this group, their population proportion is more than 50
percent.
Tables 9–10 report descriptive statistics at the level of the assembly constituency for each
4
This data has been obtained from Geographic Enterprises, LLC, based on GIS information
(http://www.geo-e.com/products/india-gis-data/)
14
year, by the reservation status of the constituency. In constituencies that are not reserved
for Scheduled Castes, the latter make up around 17 percent of the population on average,
compared to constituencies reserved for Scheduled Castes, where they constitute about 24
percent of the population. Similar to the pattern in the district data, while Scheduled
Tribes make up on 6 percent of the population in constituencies that are not reserved for
Scheduled Tribes, they constitute a majority in constituencies reserved for them. Figures 3
and 4 show the distribution of Scheduled Castes and Tribes across di¤erent states in India
according to the 2001 Census of India.
5
Results
5.1
District Level Outcomes
In this section, I present results for the e¤ects of reservation on district level measures
of the provision and pattern of public goods. Table 11 reports the …rst stage regressions
for each of the endogenous district level reservation variables. These show that there is a
very strong correlation between the instruments predicted by the rule and the number of
constituencies reserved for each group. The …rst stage coe¢ cient on the instrument for the
number of constituencies reserved for Scheduled Castes in a district is 0.85, while that on
the instrument for the number of constituencies reserved for Scheduled Tribes in a district
is 0.75. These coe¢ cients are highly statistically signi…cant, with …rst-stage F-statistics of
98.01 and 32.94 respectively.
In table 12, I report the estimated e¤ect of reservation on the percentage of villages
in the district that have access to an educational facility. A village is reported to have an
educational facility if it has a school (primary, middle, secondary or senior secondary school),
a college, an adult literacy centre, or a vocational or industrial training school. Column
1 reports OLS estimates and includes controls for linear, squared and cubic terms in the
population shares of the two groups, as well as the ratio of males to females in the district,
and the district population density. I also control for state and year e¤ects. Standard
errors are corrected for heteroskedasticity and clustered at the district level. Column 2
shows IV estimates with the same set of controls. These show that the reservation of
an additional constituency for Scheduled Castes in a district leads to an increase in the
proportion of villages with an educational facility by 7 percent. The results on the e¤ect
of current Scheduled Caste population shares are also interesting. In both the OLS and
IV speci…cation, district Scheduled Caste population shares are negatively correlated with
the provision of educational facilities, suggesting that Scheduled Castes may be relatively
deprived with respect to these public goods.
15
Table 13 shows OLS and IV estimates of the e¤ect of the district level reservation
variables on three education-related outcomes. For expositional ease, I only report the
coe¢ cients on the reservation variables and on district Scheduled Caste and Scheduled
Tribe population shares. Each district level regression includes the full set of controls
mentioned above. In columns 1 and 2 of Table 13 , the dependent variable is the percentage
of villages with a school. Columns 3 and 4 show OLS and IV estimates when the outcome
is the percentage of villages with a primary school. IV estimates con…rm what I …nd in
Table 12, and show that in districts with an additional constituency reserved for Scheduled
Castes, the percentage of villages with a primary school increases by 7.1 percent. The
results for educational facilities, schools and primary schools are very similar in magnitude,
which indicates that Scheduled Caste reservation in a district results in greater provision
of primary schools rather than in other types of educational facilities. In columns 5 and
6 of table 13, the dependent variable is a measure of the average distance to the nearest
educational facility. This shows that additional Scheduled Caste reservation in a district
lowers the average distance to the nearest educational facility by 1.73 kilometers. In contrast
to these results, district level Scheduled Tribe reservation seems to have no signi…cant impact
in terms of the provision of educational facilities.
My data set also includes measures of access to public health facilities, roads, power
supply, and drinking water. Table 14 reports the results of regressions for measures of
access to public health facilities. Dependent variables include the proportion of villages
with access to a maternal and child health care facility, to some primary health care facility
(primary health centres, primary health sub-centres and health centres), to a health center
primary, and to a health sub-centre. For each of these measures, IV estimates show that a
unit increase in the number of constituencies reserved for Scheduled Tribes within a district
decreases the proportion of villages with access to the public health facility.
Reservation for Scheduled Castes and Scheduled Tribes have no consistent or robust
e¤ects on access to roads, power supply or drinking water. A representative set of results
is reported in Table 15.
Table 16 presents estimates of the e¤ects of reservation in terms of the allocation of
educational facilities in favor of di¤erent groups, as well as the implied distance to the
nearest educational facility. These include measures of per capita access–the proportion of
the district rural population that lives in a village with the public good, and measures of
per capita group access– the proportion of the district rural Scheduled Caste, Scheduled
Tribe and non-Scheduled Caste/Scheduled Tribe population that lives in a village with the
public good. These measures capture the probability that a given person with a certain
group identity lives in a village with the public good.
16
For each of the dependent variables considered in table 16, the results show a similar
pattern of location of public goods. Columns 1 and 2 in panels A, B and C present OLS and
IV estimates of the e¤ect of reservation at the district level on the probability that a given
person in the district lives in a village with an educational facility, a school and a primary
school respectively. These outcome variables provide a per capita measure of the provision
of public goods. A unit increase in the number of constituencies within the district reserved
for Scheduled Castes results in a 18 percent increase in the proportion of district population
living in a village with an educational facility, and a 19 percent increase in the proportion
living in a village with a school or with a primary school. In columns 1 and 2 of panel D,
the dependent variable is the average distance to the nearest educational facility for a given
person in the district. IV estimates show that a unit increase in the district level reservation
for Scheduled Castes lowers the distance to the nearest educational facility for an average
person in the district by 5.3 kilometers.
Higher district level reservation for Scheduled Castes also bene…ts their own group–it
increases the proportion of Scheduled Caste people in the district living in a village with
an educational facility by 17 percent and with a school or primary school by 18 percent.
This is shown in column 4 of panels A, B and C. Column 4 of panel D shows that a unit
increase in reservations for Scheduled Castes in a district reduces the average distance to
the nearest educational facility by 4.9 kilometers for a Scheduled Caste person.
The results also suggest that Scheduled Tribes may bene…t from reservation for Scheduled Tribes at the district level. While some of the IV estimates are imprecisely estimated,
they are larger in magnitude than the signi…cant OLS estimates. These estimates are presented in columns 5 and 6 of table 16. Finally, columns 7 and 8 show that in terms of
all these measures, district level reservation for Scheduled Castes bene…ts non-Scheduled
Caste/Scheduled Tribe persons as well.
5.2
Constituency Level Results
This section reports my results for the e¤ects of reservation on constituency level measures of
the provision and pattern of public goods. In table 17, I report the …rst stage regressions for
each of the endogenous reservation variables. The …rst stage coe¢ cient on the instruments
generated by the reservation rule, the predicted reservation dummies, are 0.21 for Scheduled
Caste reservation, and 0.24 for Scheduled Tribe reservation. Both coe¢ cients are highly
statistically signi…cant, with …rst-stage F-statistics of 38.01 and 85.19 respectively.
Table 18 presents OLS and IV estimates for the regression on the percentage of villages
in the Assembly Constituency with an educational facility. Controls include current group
population shares, the ratio of males to females in the constituency, population density of the
17
constituency, and the number of Assembly Constituencies in the district. My speci…cation
also includes a full set of district, state and time dummies. Standard errors are corrected
for heteroskedasticity and clustered at the level of the assembly constituency. IV estimates
show that the reservation of a constituency for Scheduled Castes increases the percentage
of villages within the constituency with an educational facility by 13.8 percent.
Table 19 reports estimates for the other three measures of the provision of education
facilities. Columns 2, 4 and 6 show the IV estimates for each of these measures respectively. The reservation of a constituency for Scheduled Castes increases the proportion of
villages with a school by 13.7 percent, the proportion of villages with a primary school by
14.3 percent, and reduces the average distance to the nearest educational facility by 4.3
kilometers. As in the case of the district, the main e¤ect of Scheduled Caste reservation on
educational facilities seems to be working through the provision of primary schools. While
the above results con…rm the …ndings of the district regressions for the e¤ect of Scheduled
Caste reservations, the results for the e¤ects of Scheduled Tribe reservation on public health
are not con…rmed in terms of the within-district provision of goods.
In table 20, I report results for a few measures of availability of drinking water, pucca/metalled
approach road, and power supply at the constituency level. Reservation for either group
does not signi…cantly alter provision of these goods.
Table 21 report OLS and IV estimates for the e¤ect of constituency reservation on the
allocation of public education facilities across people with di¤erent group identities, and the
implied access to di¤erent groups. The measures considered are the same as in the district
regressions above, but aggregated to the assembly constituency level. Columns 5 and 6
show that for all four measures, the reservation of an assembly constituency for Scheduled
Castes bene…ts Scheduled Tribes. It increases the proportion of Scheduled Tribes who live
in a village with a primary school by 13 percent, and reduces the distance to the nearest
educational facility for a Scheduled Tribe person by 3.73 kilometers. There is also some
evidence that the reservation of constituencies for Scheduled Castes bene…ts the latter as
well. Comparing columns 2 and 4, I …nd that IV point estimates are consistently higher
for Scheduled Caste access than for per capita access measures. However, the former are
imprecisely estimated and fall short of statistical signi…cance.
5.3
Robustness Checks
This section reports a series of robustness checks for both the district and the constituency
level analyses. First, I report results from restricting my sample to points around the
discontinuity generated by the reservation rule. Table 22 reports IV estimates for the
district level education outcomes from a
15= + 15 discontinuity sample for Scheduled
18
Caste reservation. Since the discontinuity in the district reservation variable occurs at
unit intervals of 0:5, 1:5 and 2:5, districts in this sample are restricted to those for which
the predicted number of Assembly Constituencies reserved for Scheduled Castes (before
applying the integer rule) fall in the intervals [0:35; 0:65], [1:35; 1:65] and [2:35; 2:65] : While
these estimates are more imprecisely estimated than in the full sample, the point estimates
of the e¤ects of the number of Assembly Constituencies reserved of Scheduled Castes have
the same sign and a larger magnitude than in the full sample. Moreover, these e¤ects remain
signi…cant. This con…rms that the district level results on the e¤ects of Scheduled Caste
reservation are not driven by districts that are far from the discontinuity.
Table 23 reports IV estimates for public health measures at the district level from a
20= + 20 discontinuity sample for Scheduled Tribe reservation. Districts in this sample are
restricted to those for which the predicted number of Assembly Constituencies reserved for
Scheduled Tribes (before applying the integer rule) fall in the intervals [0:3; 0:7], [1:3; 1:7],
[2:3; 2:7] and so on. For two of the four measures, the coe¢ cient on the Scheduled Tribe
reservation variable changes sign from negative in the full sample to positive in the discontinuity sample. This suggests that my results on the e¤ect of an additional assembly
constituency being reserved for Scheduled Tribes is not robust to changes in speci…cation.
The main speci…cation presented in this paper uses instruments constructed from predicted 1971 population shares, while controlling for the smooth e¤ects of current population
shares. This speci…cation is valid if the included measures of current population shares suf…ciently proxy for the e¤ects of 1971 census population shares. I check for the robustness
of my estimates to changes in speci…cation in two ways.
First, I present estimates using instruments constructed by following the reservation
rule, using 1991 census population shares instead of estimated 1971 population shares.
In this speci…cation, shown in table 24, I also allow for any smooth e¤ects of current
population shares. The estimates of the e¤ect of reservation using this speci…cation are
very similar in magnitude and signi…cance to those obtained from the main speci…cation.
This suggests that measurement error in the construction of predicted population estimates
is unlikely to be driving my results. It also indicates that the included measures of current
population shares are good proxies for any e¤ects of 1971 population shares. In addition, I
also present estimates using instruments constructed from predicted 1971 population shares,
while controlling for estimated 1971 population shares. The results from this regression are
shown in table 25. In this speci…cation, none of the 1971 population terms are signi…cant.
Since the latter variable is measured with error, and is correlated with the reservation
variable, standard errors in this speci…cation are higher. However, the point estimates on
reservation remain similar in sign and magnitude.
19
Tables 26 and 27 present additional robustness checks for the district education results.
Since my main speci…cation involves a relatively small number of clusters, and relatively
few observations within each cluster, Table 26 presents IV estimates that allow for random
district e¤ects. Again, the e¤ects for additional Scheduled Caste reservation in a district
are similar in magnitude for all four measures, and in signi…cance for all but the distance
measure. Table 27 shows IV estimates when the speci…cation includes quartic population
share terms. The e¤ects of reservation for Scheduled Castes at the district level is robust
to this speci…cation as well.
The next set of tables report robustness and speci…cation checks for the assembly constituency results. Table 28 shows IV estimates for a discontinuity sample for assembly constituency reservation for Scheduled Castes. The sample is restricted to those constituencies
for which the rank of the Assembly Constituency in terms of its predicted Scheduled Caste
population share is equal to the number of Assembly Constituencies reserved for Scheduled
Castes in the district, and to those for which the rank is one higher than the number of
Assembly Constituencies reserved for Scheduled Castes in the district. While the estimates
are imprecisely estimated, the point estimates of the e¤ect of Scheduled Caste reservation
are higher in magnitude and have the same sign as in the full sample.
The next two tables report speci…cation checks. Table 29 shows estimates of the e¤ect
of the reservation of constituencies using instruments constructed from 1991 population
estimates. These estimates have are similar in magnitude and signi…cance, except for the
distance measure, which loses signi…cance. This suggests that measurement error in calculating 1971 census shares for constituencies is not likely to be driving my results. In table 30,
I present estimates using instruments constructed from 1971 predicted population shares,
while simultaneously controlling for any smooth e¤ects of estimated 1971 population shares.
The latter terms are not signi…cant in each of the regressions. However, the estimates are
imprecisely estimated, although they are similar in sign and magnitude to those presented
in the main speci…cation.
Table 31 presents IV estimates that allow for random assembly constituency e¤ects.
My results are robust to this speci…cation as well. In fact, allowing for random e¤ects
yields higher estimates for the e¤ects of constituency level reservation for Scheduled Castes.
Finally, table 32 includes squared and cubic population share terms in the speci…cation.
While it is hard to distinguish between the e¤ect of reservation and of a cubic term in
population shares, my point estimates retain a similar magnitude.
20
6
Discussion
This paper asks whether political representation for historically disadvantaged groups can
serve as an e¤ective means to redress historical inequities. While the presence of Scheduled
Castes and Scheduled Tribes in legislative bodies is largely explained by mandated political reservations, evidence on the performance of these legislators in serving the interests
of their electorate is limited. This paper provides evidence that Scheduled Caste MLAs
provide greater access to educational facilities, in particular, primary schools, within their
constituencies, and in districts that have a larger number of Scheduled Caste legislators
elected on reserved seats. With a few exceptions, I …nd no evidence that Scheduled Tribe
MLAs perform any di¤erently than legislators elected from unreserved constituencies. These
…ndings do not support the theoretical prediction that political reservation adversely a¤ects
politician quality, at least in terms of the e¤ect of these factors on the provision of public
goods.
In theory, political reservations for hitherto under-represented communities may also
lead to di¤erences in the pattern of public good provision, by re‡ecting the relative preferences of di¤erent groups over these goods. While I cannot directly infer preferences of
di¤erent groups over public goods, Scheduled Caste population shares consistently show a
negative correlation with the presence of educational facilities. This suggests that Scheduled
Castes may be relatively deprived in terms of these goods, and that Scheduled Caste MLAs
may be responsive to their needs.
In the Indian context, di¤erences in the characteristics of the two groups, combined with
the design of reservation, imply that the e¤ect of reservations may be di¤erent for the two
groups. MLAs elected from constituencies reserved for Scheduled Tribes face a majority
Scheduled Tribe electorate. In contrast, Scheduled Caste MLAs must satisfy the needs of
the majority non-Scheduled Caste electorate in their constituency, as well as of their own
community. This paper provides evidence that it is the Scheduled Caste legislator, who is
far more dependent upon a constituency made up of members not from his own group, who
provides a greater proportion of villages in his or her constituency with primary schools.
This very dependence on a non-Scheduled Caste electorate explains my results on the e¤ects
of reservation on the location of educational facilities. A randomly selected person or a
randomly selected Scheduled Tribe person in a constituency reserved for Scheduled Castes
is more likely to belong to a village with a primary school, relative to one in an unreserved
constituency. In addition, my point estimates consistently indicate higher Scheduled Caste
access to educational facilities compared to per capita access measures, but fall short of
statistical signi…cance. Overall, these results do not show strong evidence of any locational
e¤ects–Scheduled Caste MLAs locate public goods in villages so as to cater to a broad set
21
of voters.
This paper also provides indirect evidence of the di¤erences in the degree of political
mobilization of the two groups. Unless I assume that members of the Scheduled Tribes have
the same preferences over public goods as the rest of the population, the performance (or
lack thereof) of Scheduled Tribe MLAs may be representative of their lack of political clout
or of relatively similar patterns of public good access compared to the rest of the population.
Given the cultural distinctiveness and relative isolation of these tribes, it is very likely that
their preferences over public goods are di¤erent from those of the rest of the population.
Moreover, Scheduled Tribe population shares do not show a strong negative correlation with
the provision of many public goods. This suggests that the absence of a political identity for
Scheduled Tribes independent of the major political parties, at least in the states considered
in my sample, may explain why there is no distinguishable di¤erence between the public
goods provided by a Scheduled Tribe MLA and one elected from an unreserved constituency.
In contrast, the 1990s has seen an increase in the political participation and political voice of
the lower castes, and this may be re‡ected in the performance of Scheduled Caste legislators.
22
References
[1] Banerjee, Abhijit & Somanathan, Rohini, 2007. "The political economy of public goods:
Some evidence from India," Journal of Development Economics, Elsevier, vol. 82(2),
pages 287-314, March.
[2] Besley, Timothy, Rohini Pande, Lupin Rahman and Vijayendra Rao, 2004. "The Politics of Public Good Provision: Evidence from Indian Local Governments", Journal of
European Economic Association Papers and Proceedings, Vol. 2, 2/3, pp. 416-426.
[3] Beteille, Andre, 1992. The Backward Classes in Contemporary India, Oxford University
Press.
[4] Chattopadhyay, Raghabendra and Du‡o, Esther, 2004. "The Impact of Reservation in
the Panchayati Raj: Evidence from a Nationwide Randomized Experiment", Economic
and Political Weekly, pp.979-986, February.
[5] Du‡o, Esther, 2005. “Why Political Reservations?”Journal of the European Economic
Association, Vol. 3, No. 2-3, Pages 668-678, April/May.
[6] Galanter, Marc, 1984. Competing Equalities, University of California Press.
[7] Hasan, Zoya, 2002. "Representation and Redistribution: the new lower caste politics of
North India" in Transforming India: Social and Political Dynamics of Democracy, Eds.
Francine Frankel, Zoya Hasan, Rajeev Bhargava and Balveer Arora, Oxford University
Press.
[8] Mandelsohn, Oliver & Vicziany, Marika, 1998. The Untouchables: Subordination,
Poverty and the State in Modern India, Cambridge University Press.
[9] Pande, Rohini, 2003. "Can Mandated Political Representation Provide Disadvantaged
Minorities Policy In‡uence? Theory and Evidence from India", American Economic
Review, Vol. 93 (4), pp. 1132-1151.
[10] Singh, Nirvikar, 1997. “On Federalism and Reform in India” Seminar, No. 459, pp.5462, October.
[11] Yadav, Yogendra, 2002. "Understanding the Second Democratic Upsurge: trends of
bahujan participation in electoral politics of the 1990s" in Transforming India: Social
and Political Dynamics of Democracy, Eds. Francine Frankel, Zoya Hasan, Rajeev
Bhargava and Balveer Arora, Oxford University Press.
23
7
Appendix
STATES
State Legislative Assembly
DISTRICTS
Administrative unit
Reservation variable:
Number of Assembly
Constituencies reserved
for SC and ST
ASSEMBLY
CONSTITUENCIES
Electoral unit
Reservation variable:
Reservation status of
Assembly Constituency
Figure 1: Administrative and Electoral units
Figure 2: Administrative and Electoral Divisions–An Example
24
Figure 3: Distribution of Scheduled Castes in India, 2001
25
Figure 4: Distribution of Scheduled Tribes in India, 2001
7.1
Predicting 1971 District Population Estimates For 2001 District Boundaries
The change in district boundaries between 1971 and 2001 implies that I need to calculate
the 1971 populations of districts in my sample in accordance with their 2001 boundaries.
About half of the districts in my sample have been reorganized since 1971, usually along
pre-existing lower administrative boundaries. These smaller administrative units are known
as tehsils, taluks or blocks. Figure 5 shows the taluks of Bidar district as it is de…ned today.
26
Figure 5: Administrative Divisions
Taluks of Bidar District
In order to calculate the 1971 populations of districts as de…ned in 2001, I sum up
the 1971 populations of those tehsils and blocks that lay within the district in 2001. For
instance, Aurangabad district in the state of Maharashtra was comprised of twelve di¤erent
tehsils in 1971. In 1981, Aurangabad was bifurcated and four tehsils were transferred to
form a new district, Jalna. To estimate the 1971 census population of Aurangabad district as
it existed in 2001, I aggregate the 1971 tehsil populations of the eight tehsils that remained
in Aurangabad.
7.2
Predicting 1971 Population Estimates for Assembly Constituencies
In order to predict the 1971 census population of an assembly constituency, I use the fact
that while assembly constituencies lie within a district, they generally straddle more than
one tehsil within the district. The tehsil is the lowest level of disaggregation for which 1971
census …gures were available to me. Tehsil boundaries are drawn by District and State administrations, whereas Assembly Constituency boundaries are decided by the Delimitation
Commission and have been …xed since 1971 and throughout my period of analysis. The
procedure for calculating the 1971 population shares for each assembly constituency is as
follows. First, since I have census population estimates for each tehsil in 1971 as well as
current census estimates, I calculate the population growth rates for each tehsil. Next, I
use information on the current distribution of the assembly constituency population across
tehsils, and attribute the tehsil growth rate to that part of the population of the assembly
constituency that currently lives in the tehsil. This gives me an estimate of 1971 population
and group shares for each assembly constituency.
27
Number of AC
reserved for SC
0
1
2
3
Total
0
16
6
0
0
22
Predicted number of AC reserved for SC
1
2
3
8
0
0
48
6
0
8
22
0
0
6
8
64
34
8
Total
24
60
30
14
128
Table 1: Number of Assembly Constituencies reserved for Scheduled Castes in a district-Actual and Predicted
Number of AC
reserved for ST
0
1
2
3
4
7
Total
0
92
2
0
0
0
0
94
Predicted number of AC reserved for ST
1
2
3
10
0
0
10
2
0
4
2
0
2
0
0
0
0
2
0
0
0
26
4
2
4
0
0
0
0
0
2
2
Total
102
14
6
2
2
2
128
Table 2: Number of Assembly Constituencies reserved for Scheduled Tribes in a district-Actual and Predicted
Reservation Status of Assembly
Constituencies
Not reserved for SC
Reserved for SC
Total
Predicted Reservation Status of Assembly Constituencies
Not reserved for SC
Reserved for SC
Total
482
47
529
47
34
81
529
81
610
Table 3: Reservation and Predicted Reservation of Assembly Constituencies for Scheduled Castes
Reservation Status of Assembly
Constituencies
Not reserved for ST
Reserved for ST
Total
Predicted Reservation Status of Assembly Constituencies
Not reserved for ST
Reserved for ST
Total
574
8
583
9
19
27
583
27
610
Table 4: Reservation and Predicted Reservation of Assembly Constituencies for Scheduled Tribes
28
Number of AC reserved for SC
0
Mean
1
Std. Dev.
Mean
2
Std. Dev.
Mean
3
Std. Dev.
Outcomes (1991)
% of villages with an educational facility
0.82
0.11
0.81
0.15
0.85
0.14
% of villages with a school
0.81
0.14
0.79
0.17
0.84
0.14
Average distance to nearest educational facility
0.61
0.38
0.72
0.89
0.47
0.51
% of villages with a primary school
0.81
0.14
0.79
0.17
0.83
0.15
% of villages with a MCH facility
0.07
0.19
0.11
0.13
0.08
0.12
% of villages with a PHC, HC or PHS
0.09
0.08
0.18
0.21
0.14
0.17
% of villages with a HC
0.02
0.05
0.02
0.03
0.01
0.01
% of villages with a PHS
0.04
0.05
0.10
0.14
0.09
0.12
District Controls (1991)
Current share of SC in district population
0.12
0.05
0.17
0.05
0.19
0.05
Current share of ST in district population
0.16
0.17
0.08
0.14
0.06
0.08
Ratio of males to females in the district
1.04
0.04
1.04
0.07
1.07
0.07
Rural population density of the district
3.09
2.78
3.69
2.82
3.01
1.85
Observations
12
30
16
MCH-Maternal and child health care; PHC-Primary health centre;HC-health centre;PHS-primary health sub-centre
Mean
Std. Dev.
0.81
0.81
0.61
0.80
0.05
0.10
0.01
0.06
0.09
0.09
0.36
0.09
0.05
0.07
0.01
0.08
0.26
0.06
1.06
2.67
0.08
0.06
0.08
0.79
7
Table 5: Descriptive Statistics: District Outcomes and Controls, 1991, by SC reservation
Number of AC reserved for SC
0
Mean
1
Std. Dev.
Mean
2
Std. Dev.
Mean
3
Std. Dev.
Outcomes (2001)
% of villages with an educational facility
0.86
0.10
0.84
0.15
0.90
0.11
% of villages with a school
0.86
0.11
0.83
0.15
0.90
0.11
Average distance to nearest educational facility
0.44
0.28
0.43
0.38
0.30
0.35
% of villages with a primary school
0.86
0.11
0.83
0.15
0.90
0.11
% of villages with a MCH facility
0.08
0.12
0.16
0.17
0.09
0.15
% of villages with a PHC, HC or PHS
0.21
0.18
0.26
0.27
0.26
0.21
% of villages with a HC
0.02
0.02
0.05
0.07
0.03
0.05
% of villages with a PHS
0.19
0.19
0.20
0.19
0.23
0.17
District Controls (2001)
Current share of SC in district population
0.12
0.05
0.17
0.05
0.19
0.06
Current share of ST in district population
0.16
0.17
0.09
0.15
0.07
0.09
Ratio of males to females in the district
1.04
0.04
1.03
0.07
1.06
0.07
Rural population density of the district
3.48
3.32
4.37
3.12
3.79
2.40
Observations
12
30
16
MCH-Maternal and child health care; PHC-Primary health centre;HC-health centre;PHS-primary health sub-centre
Table 6: Descriptive Statistics: District Outcomes and Controls, 2001, by SC reservation
29
Mean
Std. Dev.
0.87
0.87
0.36
0.86
0.12
0.20
0.02
0.18
0.07
0.07
0.16
0.07
0.17
0.11
0.01
0.10
0.26
0.06
1.05
3.19
0.08
0.06
0.07
1.08
7
Number of AC reserved for ST
0
Mean
1
Std. Dev.
Mean
Std. Dev.
Mean
2 or 3
Std. Dev.
Outcomes (1991)
% of villages with an educational facility
0.82
0.15
0.87
0.07
0.83
% of villages with a school
0.81
0.16
0.87
0.06
0.82
Average distance to nearest educational facility
0.66
0.75
0.38
0.22
0.56
% of villages with a primary school
0.80
0.16
0.86
0.07
0.81
% of villages with a MCH facility
0.11
0.14
0.01
0.01
0.02
% of villages with a PHC, HC or PHS
0.17
0.19
0.06
0.02
0.06
% of villages with a HC
0.02
0.03
0.01
0.01
0.01
% of villages with a PHS
0.10
0.13
0.02
0.02
0.03
District Controls (1991)
Current share of SC in district population
0.19
0.06
0.14
0.05
0.15
Current share of ST in district population
0.05
0.06
0.15
0.09
0.27
Ratio of males to females in the district
1.05
0.07
1.03
0.05
1.05
Rural population density of the district
3.64
2.60
2.11
0.54
1.87
Observations
52
7
4
MCH-Maternal and child health care; PHC-Primary health centre;HC-health centre;PHS-primary health sub-centre
Mean
>=4
Std. Dev.
0.07
0.07
0.19
0.08
0.01
0.01
0.01
0.02
0.73
0.72
0.85
0.70
0.02
0.08
0.01
0.04
0.04
0.05
0.17
0.08
0.02
0.04
0.02
0.04
0.05
0.14
0.05
0.94
0.06
0.58
1.03
1.47
0.01
0.03
0.01
0.02
2
Table 7: Descriptive Statistics: District Outcomes and Controls, 1991, by ST reservation
Number of AC reserved for ST
0
Mean
1
Std. Dev.
Mean
Std. Dev.
Mean
2 or 3
Std. Dev.
Outcomes (2001)
% of villages with an educational facility
0.85
0.13
0.94
0.03
0.86
% of villages with a school
0.85
0.14
0.94
0.03
0.86
Average distance to nearest educational facility
0.41
0.36
0.19
0.09
0.46
% of villages with a primary school
0.84
0.14
0.94
0.03
0.86
% of villages with a MCH facility
0.14
0.17
0.05
0.05
0.04
% of villages with a PHC, HC or PHS
0.26
0.25
0.21
0.08
0.19
% of villages with a HC
0.04
0.06
0.01
0.01
0.01
% of villages with a PHS
0.21
0.20
0.20
0.09
0.17
District Controls (2001)
Current share of SC in district population
0.19
0.07
0.13
0.05
0.15
Current share of ST in district population
0.05
0.07
0.15
0.08
0.30
Ratio of males to females in the district
1.04
0.07
1.03
0.05
1.04
Rural population density of the district
4.33
3.01
2.58
0.94
2.21
Observations
52
7
4
MCH-Maternal and child health care; PHC-Primary health centre;HC-health centre;PHS-primary health sub-centre
Table 8: Descriptive Statistics: District Outcomes and Controls, 2001, by ST reservation
30
Mean
>=4
Std. Dev.
0.13
0.13
0.41
0.13
0.02
0.07
0.01
0.08
0.89
0.88
0.35
0.88
0.08
0.23
0.03
0.21
0.01
0.01
0.05
0.01
0.07
0.03
0.01
0.05
0.05
0.15
0.06
0.82
0.06
0.60
1.02
1.82
0.01
0.03
0.01
0.05
2
AC reservation status
Outcomes (1991)
% of villages with an educational facility
% of villages with a school
Average distance to nearest educational facility
% of villages with a primary school
AC Controls (1991)
Current share of SC in AC population
Current share of ST in AC population
Ratio of males to females in the AC
Rural population density of the AC
Observations
1991
2001
Not reserved for SC Reserved for SC Not reserved for SC Reserved for SC
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
0.86
0.85
0.48
0.84
0.15
0.16
0.67
0.16
0.83
0.82
0.61
0.81
0.18
0.19
1.04
0.19
0.90
0.89
0.28
0.89
0.13
0.13
0.34
0.13
0.88
0.88
0.33
0.88
0.14
0.14
0.38
0.14
0.17
0.08
1.04
0.43
0.09
0.15
0.07
0.30
0.24
0.05
1.07
0.79
0.09
0.09
0.08
3.13
0.16
0.09
1.04
0.37
0.08
0.15
0.07
0.25
0.24
0.05
1.05
0.96
0.09
0.09
0.07
5.32
529
81
529
81
Table 9: Descriptive Statistics, Constituency Outcomes and Controls, by SC reservation status
AC reservation status
Outcomes (1991)
% of villages with an educational facility
% of villages with a school
Average distance to nearest educational facility
% of villages with a primary school
District Controls (1991)
Current share of SC in AC population
Current share of ST in AC population
Ratio of males to females in the AC
Rural population density of the AC
Observations
1991
2001
Not reserved for ST Reserved for ST Not reserved for ST Reserved for ST
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
0.86
0.85
0.50
0.84
0.15
0.16
0.74
0.16
0.81
0.81
0.60
0.79
0.12
0.13
0.45
0.13
0.89
0.89
0.29
0.89
0.13
0.13
0.35
0.13
0.90
0.89
0.35
0.89
0.08
0.08
0.27
0.08
0.18
0.06
1.05
0.47
0.09
0.09
0.08
1.20
0.09
0.55
1.03
0.64
0.07
0.20
0.05
0.21
0.18
0.06
1.04
0.44
0.09
0.09
0.07
1.99
0.09
0.56
1.02
0.52
0.07
0.19
0.05
0.17
582
28
582
Table 10: Descriptive Statistics, Constituency Outcomes and Controls, by ST reservation status
31
28
Number of AC reserved for SC Number of AC reserved for ST
First Stage Regressions
Coefficient
8.69
2.92
-36.90
-5.06
59.00
2.11
1.39
-0.02
0.85
0.02
Current SC share in district population
Current ST share in district population
Square of current SC share in district population
Square of current ST share in district population
Cube of current SC share in district population
Cube of current ST share in district population
Ratio of males to females in the district
District population density
Predicted number of AC reserved for SC
Predicted number of AC reserved for ST
R-squared
Adjusted R-squared
Observations
Std. Error
10.09
2.70
53.52
12.32
82.73
14.87
1.15
0.03
0.09
0.14
Coefficient
-11.98
-3.47
54.31
30.83
-76.28
-29.60
0.14
0.00
0.01
0.75
0.76
0.72
Std. Error
9.23
2.46
48.94
11.27
75.65
13.59
1.05
0.03
0.08
0.13
0.87
0.85
128
Table 11: First-stage regressions--Number of reserved constituencies in the district
Percentage of villages with an educational facility
(1) OLS
0.041***
(0.012)
0.026
(0.021)
-3.731*
(2.162)
-0.834
(0.513)
20.298*
(11.805)
2.184
(2.552)
-36.623*
(18.493)
-2.367
(3.106)
0.401
(0.298)
-0.007
(0.005)
(2) IV
0.070***
(0.025)
-0.032
(0.047)
-4.740*
(2.559)
-0.801
(0.558)
22.892*
(13.238)
3.310
(2.936)
-39.345*
(20.396)
-3.057
(3.285)
0.329
(0.316)
-0.006
(0.005)
State dummies
Yes
Yes
Year dummy
Yes
Yes
Number of AC reserved for SC
Number of AC reserved for ST
Current SC share in district population
Current ST share in district population
Square of current SC share in district
population
Square of current ST share in district
population
Cube of current SC share in district
population
Cube of current ST share in district
population
Ratio of males to females in the
district
District population density
0.705**
0.854**
(0.346)
(0.404)
Observations
128
128
R-squared
0.62
0.57
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
Constant
Table 12: District regression--Percentage of villages with an educational facility
32
Dependent Variable
Number of AC
reserved for SC
Number of AC
reserved for ST
Current SC share in
district population
Current ST share in
district population
Percentage of villages with a
school
(1) OLS
0.046***
(0.013)
0.026
(0.022)
-3.523
( 2.193)
-0.854
(0.525)
(2) IV
0.071***
(0.026)
-0.032
(0.049)
-4.524*
(2.595)
-0.819
(0.573)
Percentage of villages with a
primary school
(3) OLS
0.046***
(0.013)
0.026
(0.022)
-3.513
(2.183)
-0.853
(0.511)
Observations
128
128
128
R-squared
0.64
0.59
0.65
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
(4) IV
0.071***
(0.026)
-0.030
(0.048)
-4.476*
( 2.563)
-0.819
(0..555)
128
0.61
Average distance from a
village to the nearest
educational facility
(5) OLS
(6) IV
-0.121**
-0.173**
(0.048)
(0.081)
-0.068
-0.010
(0.066)
(0.170)
12.471
13.478
(9.827)
( 9.512)
2.076
2.064
(1.689)
(1.693)
128
0.51
128
0.50
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female
ratio, District population density, State dummies, Year dummies and a constant term
Table 13: District regressions--Percentage of villages with a school, primary school and the Average
distance from a village to the nearest educational facility
Dependent Variable
% of villages with maternal
and child health care facility
% of villages with a primary
health centre/sub-centre or
health centre
% of villages with a health
centre
% of villages with a primary
health sub-centre
(1) OLS
(2) IV
(3) OLS
(4) IV
(5) OLS
(6) IV
(7) OLS
(8) IV
Number of AC
0.004
0.021
0.007
0.023
0.001
-0.000
0.002
0.010
reserved for SC
(0.008)
(0.020)
(0.013)
(0.030)
(0.002)
(0.004)
(0.013)
(0.029)
Number of AC
-0.016
-0.054**
-0.031*
-0.105**
-0.009**
-0.020**
-0.027*
-0.092**
reserved for ST
(0.011)
(0.026)
(0.017)
(0.047)
(0.004)
(0.009)
(0.016)
(0.044)
Current SC share in
-3.856
-4.519*
-4.585*
-5.845**
-1.806**
-1.997***
-4.291*
-5.383**
district population
(2.635)
( 2.511)
( 2.528)
( 2.301)
(0.743)
(0.713)
( 2.396)
(2.296)
Current ST share in
-0.155
-0.132
-0.837**
-0.778
-0.099
-0.086
-0.505
-0.448
district population
( .225)
(0..315)
(0.385)
(0.506)
(0. .073)
(0.094)
( 0.357)
(0 .448)
Observations
128
128
128
128
128
128
128
128
R-squared
0.80
0.78
0.81
0.79
0.70
0.69
0.62
0.59
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population
density, State dummies, Year dummies and a constant term
Table 14: District regressions--Public health facilities
33
Dependent Variable
% of villages with a drinking
water source
% of villages with a
% of villages with some power
pucca/metalled approach road
supply
(1) OLS
(2) IV
(3) OLS
(4) IV
(5) OLS
(6) IV
Number of AC
0.00
0.00
-0.00
0.00
-0.00
0.01
reserved for SC
(0.00)
(0.00)
(0.01)
(0.01)
(0.02)
(0.02)
Number of AC
0.00
-0.00
-0.02
-0.03
-0.02
-0.05
reserved for ST
(0.00)
(0.01)
(0.01)
(0.02)
(0.02)
(0.03)
Current SC share in
-0.36
-0.44
-1.75
-1.93
0.44
0.02
district population
(0.52)
(0.45)
(1.09)
(1.23)
(2.52)
(2.64)
Current ST share in
-0.06
-0.06
-0.52*
-0.52*
-0.01
0.00
district population
(0.07)
(0.07)
(0.31)
(0.30)
(0.56)
(0.55)
Observations
128
128
128
128
128
128
R-squared
0.41
0.41
0.56
0.56
0.63
0.62
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female
ratio, District population density, State dummies, Year dummies and a constant term
Table 15: District regressions--Other public goods
34
A: Educational
Facility
Number of AC
reserved for SC
Number of AC
reserved for ST
Observations
R-squared
B: Schools
Number of AC
reserved for SC
Number of AC
reserved for ST
Observations
R-squared
C: Primary schools
Number of AC
reserved for SC
Number of AC
reserved for ST
Observations
R-squared
D: Distance to the
nearest educational
facility
Number of AC
reserved for SC
Number of AC
reserved for ST
Observations
R-squared
% of district population living
in a village with an educational
facility
(1) OLS
0.013**
(0.006)
0.012
(0.011)
128
0.53
(2) IV
0.018**
(0.009)
-0.007
(0.020)
128
0.51
% of district population living
in a village with a school
(1) OLS
0.017***
(0.006)
0.013
(0.011)
128
0.56
(2) IV
0.019**
(0.009)
-0.005
(0.022)
128
0.55
% of district population living
in a village with a primary
school
(1) OLS
0.018***
(0.006)
0.012
(0.011)
128
0.57
(2) IV
0.019**
(0.009)
-0.005
(0.021)
128
0.56
Average distance to the nearest
educational facility for an
average person in the district
(1) OLS
-0.037**
(0.017)
-0.034
(0.029)
128
0.54
(2) IV
-0.052**
(0.025)
0.006
(0.058)
128
0.53
% of district SC population
living in a village with an
educational facility
(3) OLS
0.011*
(0.006)
0.012
(0.011)
128
0.52
(4) IV
0.017*
(0.009)
-0.011
(0.021)
128
0.48
% of district SC population
living in a village with a
school
(3) OLS
0.014**
(0.007)
0.013
(0.011)
128
0.54
(4) IV
0.018*
(0.010)
-0.011
(0.022)
128
0.52
% of district SC population
living in a village with a
primary school
(3) OLS
0.015**
(0.006)
0.011
(0.011)
128
0.55
(4) IV
0.018*
(0.010)
-0.011
(0.022)
128
0.52
% of district ST population
living in a village with an
educational facility
(5) OLS
0.025*
(0.015)
0.048**
(0.021)
128
0.34
(6) IV
0.028
(0.026)
0.095
(0.063)
128
0.32
% of district ST population
living in a village with a
school
(5) OLS
0.027*
(0.015)
0.049**
(0.022)
128
0.35
(6) IV
0.028
(0.026)
0.094
(0.064)
128
0.33
% of district ST population
living in a village with a
primary school
(5) OLS
0.025*
(0.014)
0.050**
(0.021)
128
0.37
(6) IV
0.025
(0.026)
0.100
(0.063)
128
0.34
% of district non-SC/ST
population living in a village
with an educational facility
(7) OLS
0.013**
(0.006)
0.009
(0.010)
128
0.56
(8) IV
0.018**
(0.009)
-0.011
(0.019)
128
0.54
% of district non-SC/ST
population living in a village
with a school
(7) OLS
0.016***
(0.006)
0.010
(0.010)
128
0.59
(8) IV
0.019**
(0.009)
-0.009
(0.021)
128
0.57
% of district non-SC/ST
population living in a village
with a primary school
(7) OLS
0.017***
(0.006)
0.009
(0.010)
128
0.595
(8) IV
0.019**
(0.009)
-0.010
(0.020)
128
0.58
Average distance to the nearest Average distance to the nearest Average distance to the nearest
educational facility for an
educational facility for an
educational facility for an
average SC person in the
average ST person in the
average non-SC/ST person in
district
district
the district
(3) OLS
-0.029*
(0.017)
-0.038
(0.029)
128
0.53
(4) IV
-0.049*
(0.025)
0.016
(0.060)
128
0.51
(5) OLS
-0.071
(0.066)
-0.243**
(0.114)
128
0.45
(6) IV
-0.105
(0.138)
-0.519*
(0.291)
128
0.41
(7) OLS
-0.034*
(0.017)
-0.030
(0.027)
128
0.56
(8) IV
-0.053**
(0.026)
0.022
(0.055)
128
0.54
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population
density, State dummies, Year dummies and a constant term
Table 16: District Regressions--Allocation of educational facilities across groups
35
AC reserved for SC
First Stage Regressions
Current SC share in district population
Current ST share in district population
Ratio of males to females in the district
District population density
Number of AC in the district
AC predicted to be reserved for SC
AC predicted to be reserved for ST
R-squared
Adjusted R-squared
Observations
AC reserved for ST
Coefficient
Std. Error
Coefficient
Std. Error
1.05
0.19
-0.03
0.08
-0.08
0.14
1.11
0.06
0.52
0.33
-0.32
0.13
0.03
0.01
0.00
0.00
0.01
0.01
0.00
0.00
0.21
0.03
0.02
0.01
0.24
0.03
-0.04
0.06
0.18
0.64
0.13
0.62
1219
Table 17: First stage regressions--Reservation status of Assembly Constituencies
Percentage of villages with an educational facility
(1) OLS
-0.004
(0.011)
0.027
(0.024)
-0.183
(0.160)
-0.208***
(0.056)
0.062
(0.192)
-0.007***
(0.002)
0.019***
(0.002)
(2) IV
0.138*
(0.074)
-0.048
(0.113)
-0.413*
(0.211)
-0.091
(0.163)
-0.072
(0.181)
-0.011***
(0.003)
0.018***
(0.002)
District dummies
Yes
Yes
State dummies
Yes
Yes
Year dummies
Yes
Yes
AC reserved for Scheduled Castes
AC reserved for Scheduled Tribes
Current SC share in AC population
Current ST share in AC population
Ratio of males to females in the AC
AC population density
Number of AC in the district
0.676***
(0.183)
Observations
1219
R-squared
0.64
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
Constant
0.837***
(0.178)
1219
0.53
Table 18: Assembly Constituency regression--% of villages with an educational facility
36
Dependent Variable
% of villages with a school
% of villages with a primary
school
Average distance to the nearest
educational facility
(1) OLS
(2) IV
(3) OLS
(4) IV
(5) OLS
(6) IV
AC reserved for Scheduled
-0.004
0.137*
-0.005
0.143*
0.018
-0.432*
Castes
(0.011)
(0.074)
(0.011)
(0.075)
(0.040)
(0.242)
AC reserved for Scheduled
0.029
-0.044
0.025
-0.054
-0.089
0.463
Tribes
(0.025)
(0.114)
(0.025)
(0.114)
(0.086)
(0.529)
Current SC share in AC
-0.178
-0.407*
-0.177
-0.415**
0.251
0.975
population
(0.161)
(0.212)
(0.160)
(0 .211)
(0.469)
(0.628)
Current ST share in AC
-0.210***
-0.097
-0.201***
-0.077
0.795***
-0.008
population
(0 .058)
(0.164)
(0 .058)
(0.164)
(0.209)
(0.727)
Observations
1219
1219
1219
1219
1219
1219
R-squared
0.65
0.56
0.65
0.55
0.65
0.47
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number
of AC in the district, State dummies, District dummies, Year dummies and a constant term
Table 19: Assembly Constituency regressions--Education measures
% of villages with a drinking
% of villages with a
% of villages with some power
water source
pucca/metalled approach road
supply
(1) OLS
(2) IV
(3) OLS
(4) IV
(5) OLS
(6) IV
AC reserved for Scheduled
0.00
0.01
-0.01***
-0.03
-0.02
0.08
Castes
(0.00)
(0.01)
(0.00)
(0.03)
(0.01)
(0.06)
AC reserved for Scheduled
-0.00
-0.01
-0.03
-0.01
-0.06*
-0.35
Tribes
(0.00)
(0.01)
(0.02)
(0.08)
(0.03)
(0.23)
Current SC share in AC
0.02**
0.02
0.00
0.03
0.16**
0.01
population
(0.01)
(0.02)
(0.04)
(0.06)
(0.07)
(0.11)
Current ST share in AC
0.00
0.02
-0.00
-0.03
-0.13**
0.27
population
(0.01)
(0.02)
(0.05)
(0.11)
(0.07)
(0.31)
Observations
1219
1219
1211
1211
1219
1219
R-squared
0.46
0.45
0.57
0.57
0.64
0.54
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number
of AC in the district, State dummies, District dummies, Year dummies and a constant term
Dependent Variable
Table 20: Assembly Constituency regressions--Other public goods
37
A: Educational Facility
AC reserved for
Scheduled Castes
AC reserved for
Scheduled Tribes
Observations
R-squared
B: Schools
AC reserved for
Scheduled Castes
AC reserved for
Scheduled Tribes
Observations
R-squared
% of AC rural population living
in a village with an educational
facility
(1) OLS
(2) IV
-0.006
0.065*
(0.006)
(0.039)
-0.005
-0.06
(0.008)
(0.063)
1219
1219
0.43
0.33
% of AC rural SC population
living in a village with an
educational facility
(3) OLS
(4) IV
-0.005
0.087
(0.008)
(0.064)
-0.004
-0.068
(0.009)
(0.071)
1217
1217
0.42
0.25
% of AC rural ST population
living in a village with an
educational facility
(5) OLS
(6) IV
0.018**
0.123**
(0.009)
(0.051)
-0.002
-0.069
(0.013)
(0.100)
1112
1112
0.4
0.31
% of AC rural population living % of AC rural SC population
% of AC rural ST population
in a village with a school
living in a village with a school living in a village with a school
(1) OLS
-0.005
(0.006)
-0.005
(0.008)
1219
0.47
(2) IV
0.062
(0.039)
-0.057
(0.062)
1219
0.39
(3) OLS
-0.005
(0.008)
-0.005
(0.009)
1217
0.46
(4) IV
0.086
(0.064)
-0.069
(0.071)
1217
0.32
(6) IV
0.128**
(0.053)
-0.060
(0.100)
1112
0.31
AC reserved for
Scheduled Castes
AC reserved for
Scheduled Tribes
Observations
R-squared
% of AC rural population living
in a village with a primary
school
(1) OLS
(2) IV
-0.006
0.066*
(0.007)
(0.039)
-0.008
-0.052
(0.010)
(0.061)
1219
1219
0.47
0.39
D: Average distance to
the nearest educational
facility
Average distance to the nearest Average distance to the nearest Average distance to the nearest
educational facility for an
educational facility for an
educational facility for an
average person in the AC
average SC person in the AC
average ST person in the AC
C: Primary schools
AC reserved for
Scheduled Castes
AC reserved for
Scheduled Tribes
Observations
R-squared
(1) OLS
0.022
(0.025)
0.013
(0.027)
1219
0.36
(2) IV
-0.189*
(0.111)
0.302
(0.269)
1219
0.29
% of AC rural SC population
living in a village with a
primary school
(3) OLS
(4) IV
-0.005
0.092
(0.008)
(0.065)
-0.009
-0.058
(0.011)
(0.070)
1217
1217
0.46
0.32
(5) OLS
0.019**
(0.009)
0.001
(0.013)
1112
0.4
(3) OLS
0.025
(0.030)
0.012
(0.027)
1217
0.35
(4) IV
-0.223
(0.173)
0.219
(0.218)
1217
0.27
% of AC rural ST population
living in a village with a
primary school
(5) OLS
(6) IV
0.018**
0.130**
(0.009)
(0.053)
-0.001
-0.061
(0.014)
(0.099)
1112
1112
0.4
0.31
(5) OLS
-0.067***
(0.024)
0.037
(0.086)
1112
0.47
(6) IV
-0.373**
(0.169)
0.437
(0.516)
1112
0.42
% of AC rural non-SC/ST
population living in a village
with an educational facility
(7) OLS
(8) IV
-0.010*
0.006
(0.006)
(0.023)
-0.005
-0.033
(0.007)
(0.036)
1218
1218
0.47
0.46
% of AC rural non-SC/ST
population living in a village
with a school
(7) OLS
(8) IV
-0.010*
0.001
(0.006)
(0.023)
-0.005
-0.027
(0.007)
(0.035)
1218
1218
0.52
0.51
% of AC rural non-SC/ST
population living in a village
with a primary school
(7) OLS
(8) IV
-0.010*
0.003
(0.006)
(0.023)
-0.008
-0.017
(0.009)
(0.038)
1218
1218
0.52
0.51
Average distance to the nearest
educational facility for an
average non-SC/ST person in
the AC
(7) OLS
(8) IV
0.034
-0.024
(0.025)
(0.075)
0.012
0.127
(0.021)
(0.124)
1218
1218
0.38
0.37
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number of AC in the district, State dummies, District
dummies, Year dummies and a constant term
Table 21: Assembly Constituency regressions--Allocation of educational facilities across groups
38
Dependent Variable
Number of AC reserved for SC
Number of AC reserved for ST
Observations
R-squared
% of villages with an
educational facility
% of villages with a
school
% of villages with a
primary school
Average distance to the
nearest educational facility
(1)
(2)
(3)
(4)
0.102**
0.104**
0.114**
-0.304*
(0.045)
(0.046)
(0.052)
(0.163)
-0.265*
-0.273*
-0.300*
0.469
(0.144)
(0.144)
(0.166)
(0.752)
36
36
36
36
0.93
0.93
0.92
0.92
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population density, State
dummies, Year dummies and a constant term
Table 22: 15% Discontinuity Sample for District level SC reservation: IV regressions
Dependent Variable
Number of AC
reserved for SC
Number of AC
reserved for ST
Observations
R-squared
Percentage of villages with
Percentage of villages with a
maternal and child health care primary health centre/sub-centre or
facility
health centre
Percentage of villages with a
health centre
Percentage of villages with a
primary health sub-centre
(1)
(2)
(3)
(4)
-0.173
0.495
-0.009
0.594
(0.263)
(1.132)
(0.048)
(1.306)
-0.667
1.579
0.053
1.895
(0.676)
(3.550)
(0.113)
(4.058)
26
26
26
26
.
0.05
0.79
.
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population density,
State dummies, Year dummies and a constant term
Table 23: 20% Discontinuity Sample for District level ST reservation: IV regressions
39
Dependent Variable
Percentage of villages with
an educational facility
Percentage of villages
with a school
(1)
0.069**
Number of AC reserved for SC
(0.028)
-0.035
Number of AC reserved for ST
(0.034)
Observations
128
R-squared
0.57
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
Percentage of villages with Average distance to the
a primary school
nearest educational facility
(2)
0.070**
(0.028)
-0.032
(0.035)
128
0.59
(3)
0.069**
(0.028)
-0.031
(0.035)
128
0.61
(4)
-0.187**
(0.076)
0.096
(0.100)
128
0.48
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population density, State
dummies, Year dummies and a constant term
Table 24: District regressions--1991 IV specification
Dependent Variable
Percentage of villages with an
Percentage of villages with a Average distance to the nearest
Percentage of villages with a school
educational facility
primary school
educational facility
(1)
0.063*
(0.036)
Number of AC
-0.002
reserved for ST
(0.035)
Observations
128
R-squared
0.48
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
Number of AC
reserved for SC
(2)
0.058
(0.038)
0.008
(0.036)
128
0.50
(3)
0.057
(0.038)
0.004
(0.035)
128
0.52
(4)
-0.150
(0.101)
0.004
(0.110)
128
0.45
All regressions include linear and squared 1971 population share terms for SC and ST, District male to female ratio, District population density, State
dummies, Year dummies and a constant term
Table 25: District regressions--1971 IV specification with controls for 1971 estimated population shares
40
Dependent Variable
Number of AC reserved for SC
Number of AC reserved for ST
Percentage of villages with
an educational facility
Percentage of villages
with a school
(1)
0.062**
(0.031)
-0.004
(0.054)
128
64
(2)
0.063**
(0.032)
-0.005
(0.056)
128
64
Observations
Number of Districts
Standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Percentage of villages with Average distance to the
a primary school
nearest educational facility
(3)
0.065**
(0.031)
-0.008
(0.055)
128
64
(4)
-0.165
(0.105)
-0.024
(0.197)
128
64
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population density, State
dummies, Year dummies and a constant term
Table 26: District regressions--Random effects specification
Dependent Variable
Percentage of villages with an
Percentage of villages with a Average distance to the nearest
Percentage of villages with a school
educational facility
primary school
educational facility
(1) IV
0.060**
(0.024)
Number of AC
-0.031
reserved for ST
(0.047)
Observations
128
R-squared
0.59
Robust standard errors in parentheses, clustered at the district level
* significant at 10%; ** significant at 5%; *** significant at 1%
Number of AC
reserved for SC
(2) IV
0.062**
(0.025)
-0.029
(0.048)
128
0.61
(3) IV
0.063**
(0.025)
-0.027
(0.047)
128
0.62
(4) IV
-0.151**
(0.073)
-0.018
(0.171)
128
0.51
All regressions include linear, square and cubic current population share terms for SC and ST, District male to female ratio, District population density,
State dummies, Year dummies and a constant term
Table 27: District regressions--Specification includes quartic terms in population shares
41
Dependent Variable
Percentage of villages
with an educational
facility
Percentage of villages
with a school
Percentage of villages
with a primary school
Average distance to the
nearest educational
facility
(1)
(2)
(3)
(4)
AC reserved for
0.319
0.282
0.309
-1.276
Scheduled Castes
(1.015)
(0.882)
(0.964)
(4.537)
AC reserved for
1.357
1.157
1.264
-6.214
Scheduled Tribes
(3.713)
(3.331)
(3.609)
(15.584)
Observations
207
207
207
207
R-squared
.
.
.
.
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number of
AC in the district, State dummies, District dummies, Year dummies and a constant term
Table 28: Assembly Constituency Discontinuity Sample--IV regressions
Dependent Variable
Percentage of villages
with an educational
facility
Percentage of villages
with a school
Percentage of villages
with a primary school
Average distance to the
nearest educational facility
(1)
(2)
(3)
(4)
AC reserved for Scheduled
0.119*
0.119*
0.124*
-0.329
Castes
(0.070)
(0.070)
(0.071)
(0.216)
AC reserved for Scheduled
0.076*
0.079*
0.079*
-0.257*
Tribes
(0.041)
(0.042)
(0.041)
(0.142)
Observations
1219
1219
1219
1219
R-squared
0.56
0.58
0.58
0.51
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number of AC in
the district, State dummies, District dummies, Year dummies and a constant term
Table 29: Assembly Constituency--1991 IV specification
Dependent Variable
Percentage of villages with
an educational facility
Percentage of villages
with a school
Percentage of villages
with a primary school
Average distance to the
nearest educational facility
(1) IV
(2) IV
(3) IV
(4) IV
AC reserved for
0.120
0.115
0.119
-0.358
Scheduled Castes
(0.083)
(0.084)
(0.086)
(0.291)
AC reserved for
-0.078
-0.076
-0.079
0.387
Scheduled Tribes
(0.060)
(0.061)
(0.060)
(0.236)
Observations
1219
1219
1219
1219
R-squared
0.54
0.57
0.57
0.49
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include 1971 population share terms for SC and ST, AC male to female ratio, AC population density, Number of AC in
the district, State dummies, District dummies, Year dummies and a constant term
Table 30: Assembly Constituency--1971 IV with controls for estimated 1971 population shares
42
Dependent Variable
Percentage of villages with Percentage of villages
an educational facility
with a school
Percentage of villages
Average distance to the
with a primary school nearest educational facility
(1) IV
(2) IV
(3) IV
(4) IV
AC reserved for
0.185***
0.184***
0.187***
-0.437**
Scheduled Castes
(0.056)
(0.057)
(0.058)
(0.207)
AC reserved for
0.033
0.036
0.021
0.456
Scheduled Tribes
(0.102)
(0.103)
(0.104)
(0.366)
Observations
1219
1219
1219
1219
Number of AC
610
610
610
610
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current population share terms for SC and ST, AC male to female ratio, AC population density, Number
of AC in the district, State dummies, District dummies, Year dummies and a constant term
Table 31: Assembly Constituency--Random effects specfication
Dependent Variable
Percentage of villages with an Percentage of villages with Percentage of villages with
educational facility
a school
a primary school
Average distance to the
nearest educational facility
(1) IV
(2) IV
(3) IV
(4) IV
AC reserved for
0.088
0.086
0.095
-0.340
Scheduled Castes
(0.061)
(0.060)
(0.063)
(0.249)
AC reserved for
-0.114
-0.110
-0.14
0.899
Scheduled Tribes
(0.203)
(0.202)
(0.217)
(1.139)
Observations
1219
1219
1219
1219
R-squared
0.6
0.62
0.61
0.49
Robust standard errors in parentheses, clustered at AC level
* significant at 10%; ** significant at 5%; *** significant at 1%
All regressions include current linear, squared and cubic population share terms for SC and ST, AC male to female ratio, AC population
density, Number of AC in the district, State dummies, District dummies, Year dummies and a constant term
Table 32: Assembly Constituency--Specification includes squared and cubic terms
43