World Development 111 (2018) 113–129

Contents lists available at ScienceDirect

World Development
journal homepage: www.elsevier.com/locate/worlddev

Historical legacies of colonial indirect rule: Princely states and Maoist

insurgency in central India q
Shivaji Mukherjee
University of Toronto, Canada

a r t i c l e i n f o a b s t r a c t

Article history: What are the long term effects of colonial institutions on insurgency? The literature on civil wars has not
Accepted 13 June 2018 explored the historical legacies of colonial institutions for insurgency. I address this gap in the literature,
Available online 17 July 2018 by exploiting sub-national variation in the most important internal security threat in the world’s largest
democracy—the Maoist insurgency in India. Within India, I focus on the crucial case of the Maoist rebels
Keywords: in the tribal state of Chhattisgarh in central India which epitomizes the causal mechanism of indirect rule
Maoist insurgency through native princely states creating enclaves of weak state capacity, low development and tribal grie-
Indirect rule
vances due to natural resource exploitation. I test my theory on a new dataset at the sub district level
Colonial legacies
India
within Chhattisgarh, and use instrumental variable regression to address endogeneity due to selection
Bastar bias, combined with historical analysis and interview data to demonstrate path dependence. This study
Princely state demonstrates historical origins of weak state capacity and ethnic grievances due to natural resource
Chhattisgarh exploitation, which are important explanations for civil war onset. It also sets the agenda for further
research on other cases where colonial indirect rule creates conditions for insurgency, like the Taliban
in FATA in Pakistan, the ethnic insurgencies in Burma’s peripheries, and leftist insurgencies in Nepal,
Peru and Colombia.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction
What is the long term effect of colonial institutions in creating
historical legacies that influence spatial variation in rebel control in
insurgency? Why is it that the Taliban in Pakistan was more suc-
cessful in areas of former British indirect rule in the FATA area of
NWFP, than in the British direct rule districts (Naseemullah,
2014)? Some of the secessionist insurgencies in India’s North East
like Mizoram, Nagaland, and Manipur can be traced back to discon-
tent and identity formation emerging from indirect rule and
chieftaincy system set up by the British (Baruah, 2005). The leftist
FARC insurgency in Colombia occurs in areas of historically low
q
I would like to thank the following for their comments and advise on this
project—Noel Anderson, Catherine Boone, Melani Cammett, Mario Chacon, Kanchan
Chandra, Thad Dunning, Aditya Dasgupta, Roberto Foa, Kishore Gawande, Guy
Grossman, Ronald Herring, Devesh Kapur, Anirudh Krishna, Atul Kohli, Matthew
Lange, Alex Lee, Jason Lyall, James Mahoney, Jason Miklian, Anoop Sarbahi, Shankar
Satyanath, Paul Staniland, Kenneth Scheve, Jake Shapiro, Jan Pierskalla, James
Robinson, Manny Teitelbaum, Tariq Thachil, Ashutosh Varshney, Steven Wilkinson,
Joseph Wong, Elisabeth Wood, and two anonymous reviewers. I would also like to
thank participants in the Comparative Politics Workshop at Yale university, 2013,
the Workshop on Maoist Insurgency at Princeton, February 2014, and Comparative
Colonialisms Workshop at OSU, April 2016. All errors remain my own.
E-mail address: shivaji.mukherjee@utoronto.ca
state penetration (Robinson, 2013). The type of land tenure system
created is partly the result of colonial institutions of direct/indirect
rule, and could explain the level of sons of the soil land related con-
flict in Africa (Boone, 2017). This article sets a new research agenda
by proposing that historical institutions are omitted factors that
need to be taken seriously by the civil war literature to explain
the persistence and recurrence of conflict.
There are various reasons for taking historical institutions
seriously for the study of civil war and insurgency. First, the civil
war literature does not explain very well why sometimes there is
persistence and recurrence of conflict from colonial and pre-
colonial times in certain regions of countries. Dell (2010) finds that
districts in Peru which were part of the exploitative mita labor
system under the Spanish tend to have lower levels of public goods
today, and Guardado (2016) shows that these districts with mita
system also had more anti colonial rebellions and also more
support for the Sendero Luminoso insurgency. Besley and Reynal-
Querol (2014) find persistence of conflict from pre-colonial times
in certain zones in Africa. Focusing on the role of historical institu-
tions and also past conflict allows us to analyze why certain con-
flicts recur in different time periods.
Also, there is potential omitted variable bias in different
quantitative models of civil war onset using cross national
datasets, since some omitted factors like institutional quality

https://doi.org/10.1016/j.worlddev.2018.06.013
0305-750X/Ó 2018 Elsevier Ltd. All rights reserved.
114 S. Mukherjee / World Development 111 (2018) 113–129
may be driving both economic/ethnic outcomes, as well as conflict
(Miguel, Satyanath and Sergenti (2004: 726). Colonial institutions
are one such potential omitted variable that influence both the
sub-national variation in proximate levels of state capacity and
ethnic exclusion, and also the chances of conflict (Acemoglu,
Johnson, & Robinson, 2001), so including them in models of civil
war will partly reduce such omitted variable bias.
Another analytical advantage of bringing historical institutions
back into the study of insurgencies is that the cross-national liter-
ature on civil war does not address well the issue of endogeneity of
socio-economic factors to the process of conflict (Hegre and
Sambanis (2006: 513-514). By analyzing the effects of historical
institutions which are causally and temporally prior to current
conflict dynamics, and using an instrument for such historical
institutions to address possible selection bias, my study addresses
such endogeneity issues.1
Given the dearth of research on this less understood link
between colonial institutions and their long term effects on insur-
gency, it would be profitable to focus on one type of colonial insti-
tution and carefully study its effect on some important cases of
insurgency using fine-grained sub-national data, before extending
the analysis to a cross-national level.2 A particularly fruitful direc-
tion of research would be to analyze the colonial institution of indi-
rect rule, which is an important colonial institution with long term
effects. Studies like Lange (2009); Mahoney (2010); Kohli (2004);
Mamdani (1996) analyze the effects of colonial indirect rule on devel-
opment, state capacity, and ethnic genocide, but not on insurgency.
I fill this gap by exploring the effects of colonial indirect rule on
the most important internal security threat in the world’s largest
democracy—the Maoist insurgency in India.3 I focus on the crucial
case (Gerring, 2007) of the Maoists in the tribal state of Chhattisgarh
in central India where formal indirect rule through princely states
created weak state capacity and natural resource extraction and con-
flict (Sundar, 2007). While indirect rule through princely states was
abolished following Indian independence in 1947, the effects of
these institutions in the forms of weak state capacity, extraction of
natural resources, and rebellious tribes excluded from the process
of modern nation-state formation, persisted through path dependent
processes into the 1980s in these areas in central India. These histor-
ically created state weakness and tribal grievances were exploited by
the People’s War Group (PWG) Maoists, paving the way for very high
levels of Maoist rebel control in these areas by late 1990s.
I combine quantitative analysis of an original dataset at the sub
district level within Chhattisgarh, with qualitative analysis show-
ing path dependence, to demonstrate how historical institutions
create current levels of state capacity, and ethnic inequalities in
central India. While my larger project presents historical evidence
for process tracing and path dependence of mechanisms, this arti-
cle focuses on addressing selection bias by using an instrument for
the British choice of indirect rule through princely state. This also
addresses the issue of endogeneity of more proximate factors like
state capacity, tribal grievances, to ongoing conflict which affects
other econometric studies of Maoist insurgency, and moves
beyond correlations to search for long term causal effects of histor-
ical institutions on insurgency.4
1
Wucherpfennig et al. (2015) similarly address the criticism that ethnic exclusion
is affected by reverse causality from conflict, by developing an instrument based on
the different types of relations with indigenous elites used by the British and French
colonial rule.
2
A few studies like Lange & Dawson (2009) have developed broad measures of
colonial institutions to show its effect on rebellion using cross-national analysis.
3
The ex-Indian Prime Minister, Manhoman Singh, called it India’s ‘‘number one
internal security threat” in 2006.
4 5
The other study on Maoists addressing endogeneity is Gawande et al. (2017)
which uses rainfall variation as an instrument for changes in forest cover and
therefore employment in these forested Maoist affected areas of India.
In the rest of the paper, I first engage with complementary and
alternate explanations for Maoist insurgency in India. I then out-
line theory and mechanisms through which colonial indirect rule
sets up the structural conditions for Maoist insurgency. I then dis-
cuss the history of indirect rule and its effects on the Maoist move-
ment in Chhattisgarh. Finally, I present IV-2SLS analysis to show
that princely state has a positive correlation with Maoist control,
even after addressing selection effects, and conclude by discussing
generalizability of the theory.
2. Complementary and rival explanations for Maoist insurgency
2.1. Complementary explanations for Maoist insurgency
There are several excellent emerging studies of India’s Maoist
insurgency using district level econometric analyses. Gawande,
Kapur and Satyanath (2017) and Vanden Eynde (2017) explain
variation in patterns of Maoist and state violence by focusing on
how rainfall shocks change forest cover and have an adverse effect
on opportunity costs to rebel recruitment. Dasgupta, Gawande and
Kapur (2017) analyze the effects of government employment
programs like the National Rural Employment Guarantee
Scheme (NREGS) on Maoist recruitment, and find that it reduces
violence in strong states like Andhra Pradesh and Chhattisgarh
by reducing opportunity costs and grievances of unemployed
youth. Gomes (2015) finds that districts with landlord tenure also
tend to have more Maoist conflict, building on Banerjee and Iyer
(2005)’s theory, and also that districts with more land inequality,
forest cover, and scheduled castes (SCs) and scheduled tribes
(STs) create conditions for Maoist insurgency. Echoing these find-
ings, Hoelscher, Miklian, and Vadlamannati (2012) find that min-
ing, scheduled castes/tribes, and NREGS all have an effect on
Maoist violence.
My article makes several contributions which make it distinct
from the micro-econometrics literature on India’s Maoist insur-
gency (Gawande et al., 2017; Dasgupta et al., 2017; Vanden
Eynde, 2017; Gomes, 2015; Hoelscher et al., 2012; Chandra &
Garcia-Ponce, 2014). First, while these studies use Maoist violence
as their dependent variable, my paper uses a different dependent
variable of Maoist control measured using novel data from State
Election Commission.5 Second, my theory focuses on how colonial
indirect rule creates structural constraints for rebel leaders before
the insurgency suddenly expands following unification of the PWG
and MCC factions in 2004, and is distinct from these studies which
explain violence in the latter phase of insurgency (2005–2012).
Third, most of these recent studies of Maoist insurgency use district
level datasets. In contrast, I develop a sub-district Assembly Con-
stituency (AC) level dataset for the state of Chhattisgarh, which is
the first such micro level dataset.
My paper is also complementary to these other studies which
analyze the effects of proximate opportunity factors like forest
cover, lower caste and tribal grievances, ethnic parties, land
inequalities and NREGS on levels and patterns of violence. In con-
trast, my study analyzes the omitted variable of colonial indirect
rule which created structural conditions that enabled Maoists to
mobilize more successfully in areas of former indirect rule. Once
the British chose colonial institutions of indirect rule for particular
areas, these institutions had an independent effect on probability
of leftist insurgency, beyond that created by proximate opportu-
nity structures like forest cover and hilly terrain. Colonial indirect
rule created tribal grievance due to natural resource extraction,
This data is not publicly available and was collected through field-trips to the
State Election Commission, and also by filing official requests through the Right to
Information Act (RTIs), and represents new data.
 S. Mukherjee / World Development 111 (2018) 113–129
lower levels of state capacity and public good distribution, which
then became the intermediating mechanisms that are exploited
later by Maoist rebels. So colonial indirect rule in conjunction with
these more proximate factors explains spatial variation in Maoist
control.
2.2. Addressing rival explanations based on colonial institutions
A working paper by Teitelbaum and Verghese (2016) suggests
that districts under former colonial direct rule have more Maoist
insurgency in India. Verghese (2016a,b) analyzes the Bastar prin-
cely state which he interprets as British direct rule and has Maoist
insurgency. These papers suffer from several empirical and theo-
retical problems.
First, Teitelbaum and Verghese (2016) cannot explain why the
People’s War Group (PWG) Maoists were successful in creating
some of the strongest core areas of guerilla zones in the southern
epicenter of the insurgency in former princely state (indirect rule)
areas of Hyderabad (Telangana) in Andhra Pradesh, and Bastar/
Kanker states in Chhattisgarh and former feudatory states in
Orissa.6 Verghese’s (2016a,b) suggestion that Bastar princely state
was direct rule because the British intervened often is not satisfac-
tory, since that was the case with most princely states where the Bri-
tish used the Residency system to control and intervene (Fisher,
1991). This does not imply that these are direct rule like in Madras
and Bombay presidency where the British used their own bureau-
cracy, police, and judiciary, and allowed provincial elections since
1920s (Tudor, 2013).
The second empirical problem with the Teitelbaum and
Verghese (2016) paper is that there is relatively higher develop-
ment and no Maoist insurgency in the western and southern states
of Maharashtra, Tamil Nadu which were formerly colonial direct
rule with ryotwari land tenure systems in Bombay and Madras
presidency respectively, which contradicts the paper’s prediction.
A third conceptual problem is that Teitelbaum and Verghese
(2016) and Verghese (2016a,b) interpret the zamindari landlord
tenure in the northern epicenter of the Maoist insurgency in Ben-
gal/Bihar/Jharkhand zone as British direct rule.7 This interpretation
is incorrect because it was the British indirect revenue collection sys-
tems through landlords or zamindars used in Bengal/Bihar/Jharkhand
that created land and caste inequalities and led to Maoist insurgency.
Kohli (2004: 225–26), Dirks (2001), Lange (2009) and Mahoney (2010)
acknowledge the zamindari tenure system in Bengal/Bihar as another
shade of indirect rule, in which the overburdened colonial state
depended on native intermediaries to collect land revenues on their
behalf, instead of collecting it directly as in the ryotwari (non landlord)
tenure system in Bombay and Madras. In contrast, areas in Bombay or
Madras provinces in the south-west of India, where the British used
direct and equitable ryotwari land tenure system which were truly
direct rule and direct revenue collection, did not see tribal/peasant
rebellions in colonial or post-colonial times.
Summarizing, there are two problems with these studies. The
first is that they treat indirect rule vs. direct rule as a dichotomous
category, when in reality, as Lange (2009), Mahoney (2010),
Gerring, Daniel, Van Gorp, and Julian (2011) and Wilkinson
(2017) have suggested it is a more continuous category and there
are different types of indirect rule. Second, they assume all princely
states as uniformly having benevolent rulers, when in reality there
6
Similarly, Verghese (2016, p. 1623, footnote 9) mentions ‘‘Bastar is not the only
former princely state that experiences tribal conflict – the Naxalites are also active in
Orissa and Telangana”, but does not have an explanation for why these areas of
former indirect rule through princely states, had and continue to have tribal rebellion.
7 9
Verghese (2016: 1621) notes that these ‘‘tribal conflicts continued to occur . . .
especially in former areas of direct British rule (emphasis mine) like Bengal, Bihar, and
Jharkhand”.
115
is variation between different types of princely states, as recent
studies by Foa (2016) and Lee (2017) propose.8 In this paper, move
in the direction of disaggregating the concept of indirect rule by first
pointing out two different types of indirect rule broadly within India,
and secondly by analyzing the feudatory princely states in Chhattis-
garh which allows us a glimpse into this diversity of princely state
types.
3. Variation in colonial indirect rule in India and historical
legacies for Maoist insurgency
In his classic study of the effects of colonialism on development,
Mahoney (2010) suggests that were 4 types of British colonial
rule – (1) settlement, (2) indirect, (3) direct, and (4) hybrid. While
most African countries like Nigeria, Uganda, Sierra Leone were low
population density and almost fully indirect rule, the largest British
colony of India had higher population density with complex pre-
colonial institutions and was in the hybrid category, since it had
a mix of direct and indirect rule within its territory. Lange (2009)
notes that while the British colonizers chose direct rule in certain
areas of India where they set up their own institutions of revenue
collection, administration, and laws, in other areas they chose to
rule indirectly through pre-existing native rulers called princes,
nizams, or rajas, and this resulted in official (de jure) colonial
indirect rule through princely or native states.
I build on the suggestion of Mahoney (2010), Gerring et al.
(2011: 378), and Wilkinson (2017) to think of direct vs. indirect
rule as a non-dichotomous category. To explain both the northern
and southern epicenters of the Maoist insurgency which studies
like Verghese (2016a,b), Teitelbaum and Verghese (2016) fail to
do, in a different paper I propose that there were two different
forms of indirect rule within India, which coincided with the two
epicenters of Maoist insurgency in India, and created separate cau-
sal pathways leading to insurgency.9
3.1. Type 1: Informal (de facto) indirect rule/indirect revenue collection
through zamindari land tenure system
One causal pathway coincides with the northern districts of
Bihar/Bengal/Jharkhand, where the British used an informal or de
facto form of indirect rule or revenue collection (Kohli, 2004;
Lange, 2009; Mahoney, 2010), by depending on upper caste land-
lords, called zamindars or malguzars, to collect land revenue on
their behalf. While officially part of direct rule, zamindari and other
such landlord or intermediary based land revenue systems were de
facto indirect rule or revenue collection, since revenue was not col-
lected directly through British officials, but indirectly through
landlords (Lange, 2009; Dirks, 2001; Kohli, 2004: 225–26). This
resulted in land/caste inequality which persisted and created
structural conditions for Maoist insurgency in the northern
Bihar-Jharkhand epicenter (Gomes, 2015; Banerjee & Iyer, 2005).
3.2. Type 2: Formal (de jure) indirect rule through princely states
The second causal pathway to Maoist insurgency in India, which
is the focus of this paper, coincides with the southern epicenter
near Chhattisgarh-Andhra Pradesh-Orissa, where there was formal
or de jure colonial indirect rule through various princely states,
which were allowed substantial autonomy in the sphere of admin-
8
Foa (2016) develops a typology of indirect rule and pre-colonial states in South
Asia. Lee (2017) also makes this point of outsider rulers not being benevolent towards
subjects.
Mukherjee (2018). ‘‘Colonial Origins of Maoist Insurgency in India: Historical
Institutions and Civil War”, Journal of Conflict Resolution, published online September
2017 at http://journals.sagepub.com/doi/full/10.1177/0022002717727818.
116 S. Mukherjee / World Development 111 (2018) 113–129
istration, taxation, land revenue and law, while keeping foreign
policy under the direct control of the colonial administration
(Fisher, 1991; Lee-Warner, 1910).
There are several mechanisms created by colonial indirect rule which
could shape structural conditions for rebellion
First, colonial indirect rule through princely states tended to pro-
duce lower levels of development, weaker bureaucratic and police
capacity (Acemoglu et al., 2001; Kohli, 2004; Lange, 2009).10 Sec-
ond, as Mahoney (2010, p. 240) notes, ‘‘In the indirectly ruled areas,
the British actively protected oppressive local elites, who otherwise
likely would have faced major rebellion from the rural producers
they exploited. . ..” and this allowed despotic extraction of natural
resources, revenue, and labor from peasants and tribals (Boone,
1994; Boone, 2013; Mamdani, 1996), which would not have been
possible in areas of British direct rule.
Further variation within formal indirect rule
It is difficult to generalize across such a large variation in size
and type of princely states, and there is further heterogeneity
between princely states. Building on Ramusack (2004), Foa
(2016) suggests that the tributary/feudatory states in Orissa, and
Chhattisgarh areas often had outsider Rajput princes who were
protected by the British and tended to be extractive. This was in
contrast to warrior states like Travancore and Mysore which were
ruled by kings of same ethnicity/region who challenged the British
and Dutch, built armies and bureaucracies, provided more public
goods to their subjects. Also, there were the successor states of
Hyderabad and Bengal which had Muslim rulers who were for-
merly vassals of the Mughal emperor, but later became suzereign
rulers, and were also despotic.
In this paper, I focus on the feudatory princely states of Bastar,
Kanker, in Chhattisgarh, which exemplify the mechanisms of lower
state capacity and development (Lange, 2009; Kohli, 2004;
Mahoney, 2010), as well as land and natural resource exploitation
from their ancient tribal lands (Boone, 1994, 2017; Mamdani,
1996; Banerjee & Iyer, 2005).
4. Colonial indirect rule and Maoist insurgency in Chhattisgarh
4.1. Case selection – the crucial pathway case of Chhattisgarh for
testing this theory
I test the larger theory of different types of indirect rule using an
all-India district dataset based analysis in a different paper
(Mukherjee, 2018). Nested within this broader all India district
level analysis (Lieberman, 2005), in this paper I focus on the prin-
cely states in Chhattisgarh, as a crucial pathway case (Gerring, 2007)
to trace the less studied causal pathway of how the more formal
type of indirect rule through princely states can create weak state
capacity (Fearon & Laitin, 2003), and land and natural resource
extraction (Boone, 1994).11
4.2. Legacies of colonial indirect rule in Chhattisgarh
Formal indirect rule through the feudatory princely states in
Chhattisgarh created weak state capacity, and exploitation of land
10
In contrast, Iyer (2010) suggests indirect rule produces better developmental
outcomes in India, though her IV2SLS analysis using an all India dataset has flaws,
which I explain below.
11
The indirect revenue collection system through zamindars in the northern
epicenter in Bihar/ Bengal, have been more studied (Bhatia, 2005, Banerjee & Iyer,
2005).
and natural resources from Gond tribes, which created grievances
that were exploited by Maoist rebels to create zones of rebel control
in the heart of India (Sundar, 2007). In Chhattisgarh, the insurgency
occurred in the districts which were part of the princely states of
Kanker and Bastar in the south and Surguja, Koriya, Changbhakar,
Udaipur, and Jashpur in the north. The central districts of Raipur,
Bilaspur, were annexed in 1854 into British direct ruled Central
Provinces, and do not have insurgency (See Fig. 1B below).
I briefly outline the two mechanisms:
(1) Indirect rule created low state capacity and development
Using indirect rule allowed the British to avoid governing these
areas which had different Gond tribal groups in the Dandakaranya
forests in south Chhattisgarh, who preferred to maintain their own
indigenous customs and institutions, and had historically been
resistant to centralized state-formation, similar to Scott’s (2009)
zomia lands in South East Asia.12 The renowned human rights acti-
vist, K. Balagopal, mentioned this decision by the British to avoid
using direct rule in this area: ‘‘ . . . Abujhmarh in Chhattisgarh area
. . . was experimentally allowed to be ungoverned, and unsurveyed
also. . . . And the neighboring areas of today’s Kanker, Bastar and
Dantewada are so backward that administering was possible by
the Maoists . . . ”13
This effect of British indirect rule can be seen from the fact that
at the time of independence of India from British rule in 1947, the
southern region of Bastar and the northern region of Surguja prin-
cely states in Chhattisgarh areas were really low state penetration,
and low levels of governance and public goods, while the central
regions of Bilaspur and Raipur districts which had been part of
direct colonial rule as Central provinces was much higher state
penetration. Fig. 1A from the Census of India (1951) shows that
the spread of railway lines and electricity stations soon after India’s
independence in Madhya Pradesh state which was earlier Central
Provinces. It can be compared with the historical map of colonial
districts and princely states of Central Provinces in Fig. 1B to show
that railway and electricity stations were lowest in the highly
backward areas of Bastar and Surguja princely state areas in the
south and north respectively of the eastern part of Madhya Pra-
desh, which later became Chhattisgarh state.
Further evidence for the long term difference in effects of direct
vs. indirect rule comes in Tables 1A & 1B below, which shows that
the constituencies that were part of direct rule tend to have more
roads and bus facilities, and more schools, water tanks and tube-
wells, and more access to electricity, which are all crucial public
goods (2001 Census data).
(2) Indirect rule caused exploitation of forest and natural resources
from tribes
One of the main interests of the British colonial state was the
rich natural resources in the Bastar region, and indirect rule
through the prince and local landlords would allow the colonial
state to get these revenues on the cheap. The British introduced
the Forest Act, 1878, which gave the state the right to reserved for-
ests and prevented the original sons of the soil tribes living in these
areas the right to access to minor forest products which they had
historically used for their livelihood.14 Sundar (2007: 114) men-
tions that while the official reason for reserving almost two-thirds
of the total forest areas of the state was to ‘‘conserve commercially
12
There were tribal rebellions against the British in earlier periods, for example the
Bhumkal rebellion of 1910 under the leadership of Gunda Dhar (Sundar, 2007).
13
Interview with K. Balagopal, Hyderabad, February 8, 2008.
14
Ramachandra Guha (1983). ‘‘Forestry in British and Post-British India: A
Historical Analysis.” Economic and Political Weekly 18: 45/46.
 S. Mukherjee / World Development 111 (2018) 113–129 117

Fig. 1A. Map showing road, railway, and electricity generating stations in Madhya Pradesh (earlier Central Provinces) in 1951. Source: Census of India, 1951, Volume VII,
Madhya Pradesh, Part 1-A Report. Nagpur, Government Printing, 1953.

Fig. 1B. Map of Princely states and British direct rule in Central Provinces: princely states are colored yellow, while British direct rule districts are colored pink. Source: The
Imperial Gazetteer of India: Volume 26—Atlas, Plate 39. 1909. Available at Digital South Asia Library, U Chicago, http://dsal.uchicago.edu/reference/gaz_atlas_1909/fullscreen.
html?object=45 . Also at http://enacademic.com/pictures/enwiki/73/IGI1908CPandBerar2.jpg. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)

valuable forests before they were destroyed”, the real purpose Besides timber, there was also attempts by various companies
behind forest reservation was revenue generation, and exporting to prospect for and find iron ore and other mineral resources in
timber for meeting needs of adjoining states and British districts. the Bastar area. One of the major companies which started looking
118 S. Mukherjee / World Development 111 (2018) 113–129

Table 1A
Summary statistics and difference in means by princely state.

Variable No. of constituencies Mean Difference in Means (s.e)

Total British Empire princely States Total British Empire Princely States (British Empire)-(princely State)
Maoist Rebel Control
% High Sensitive 90 52 38 10.66 2.17 22.26
20.09*** (3.30)
Polling Stations
Geography
Forest Cover % 90 52 38 30.83 19.18 45.68
25.70*** (5.52)
Elevation (meters) 90 52 38 389.52 318.49 486.71
168.22*** (24.83)
Demography
Scheduled Caste% 90 52 38 11.88 15.51 6.93 8.58*** (1.34)
Scheduled Tribe% 90 52 38 33.24 20.03 51.31
31.28*** (4.03)
Population Density 90 52 38 293.93 366.47 194.69 171.78*** (50.73)
Rural % 90 52 38 84.70 81.48 89.12
7.64 (4.83)
Socio-Economic
Total Literacy 90 52 38 63.622 67.33 58.55 8.77*** (2.39)
Male Literacy 90 52 38 76.65 81.21 70.41 10.79*** (2.19)
Female Literacy 90 52 38 50.61 53.46 46.69 6.77** (2.65)
Male Unemployment 90 52 38 56.88 59.44 53.38 6.07*** (0.79)
Per Capita Income 90 52 38 603.29 747.55 405.89 341.66 (493.32)
State Capacity
Post Office 90 52 38 0.18 0.19 0.15 0.04** (0.02)
Dist from Town (km) 90 52 38 31.36 22.58 43.39
20.81*** (4.69)
Paved Road 90 52 38 0.36 0.41 0.30 0.11*** (0.04)
Railway Facility 90 52 38 0.007 0.008 0.005 0.003 (0.002)
Bus Facility 90 52 38 0.17 0.19 0.15 0.04** (0.02)
Education
Middle School 90 52 38 0.29 0.31 0.26 0.05** (0.02)
Secondary School 90 52 38 0.09 0.11 0.07 0.04*** (0.01)
Water
Tanks 90 52 38 0.384 0.518 0.202 0.316*** (0.094)
Tubewells 90 52 38 0.306 0.465 0.087 0.377*** (0.102)
Power
Electricity 90 52 38 0.344 0.408 0.256 0.152*** (0.043)

Source: All public goods and demography data from Census of India 2001.
*
p < 0.10, **p < 0.05, ***p < 0.01. Standard Errors in parentheses.

Table 1B
OLS estimates of impact of colonial institutions on development and state capacity.
Dependent Variable: Various measures of development and state capacity.

Development and public good measures State capacity measures

Model 1A – Model 1B – Model 1C – Model 1D – Model 2A – Model 2B – Model 2C – Dist. Model 2D –
Male Literacy Electricity Tubewells Middle Schools Paved Road Post Office from Town Railway Station
Princely State
10.80**
0.152**
0.377*
0.0511**
0.105**
0.0432* 20.81***
0.00308
(0.029) (0.013) (0.081) (0.049) (0.046) (0.093) (0.006) (0.202)
Constant 81.21*** 0.408*** 0.465** 0.310*** 0.408*** 0.198*** 22.58*** 0.00853***
(0.000) (0.000) (0.031) (0.000) (0.000) (0.000) (0.000) (0.000)
Observations 90 90 90 90 90 90 90 90
R2 0.216 0.122 0.135 0.048 0.088 0.060 0.182 0.022
Adjusted R2 0.207 0.112 0.125 0.037 0.078 0.049 0.173 0.011

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
Robust Std. Errors adjusted for 17 clusters by district.

and mining for iron ore deposits was Tata Sons who got the first
mining lease to prospect for minerals and iron ore in the area
around 1916.15 There was no railway tracks linking the Bastar
region, and the iron ore deposits were inaccessible, but since it ‘‘pro-
vides some revenue for the Bastar state in license fees”, the Depart-
ment of Industry and Labor continued to renew the prospecting lease
for the Tatas in 1920–30s.16 In the 1960s, there were large deposits
15
‘‘Grant of prospecting license to the Tata Iron and Steel Company Limited, to
prospect for iron ore in the Bastar State.” NAI, Foreign and Political Department,
Branch Internal-B, Progs. January 1917, Nos. 126.
16
‘‘Renewal of the prospecting license for iron ore in the Bastar State to Messrs. Tata
Sons, Limited, Bombay for a period of one year” NAI, Foreign and Political Department,
Reforms Branch, File No. 281-R/1932.
of bauxite and iron ore found in this area, and so the foundation for
its extraction were already laid in the colonial period.
4.3. Persistence of the effects of these institutions into post-colonial
times
Following India’s independence in 1947, the princely states and
zamindars were officially abolished by the 1950s. However, due to
path dependence of the effects of these institutions, low state pene-
tration and extraction of rich iron ore resources in these areas con-
tinued. As a consequence, the northern and southern parts of
Chhattisgarh, which were former princely states, remained areas of
low levels of development, with continuing exploitation of natural
resources.
 S. Mukherjee / World Development 111 (2018) 113–129
Path dependent mechanism 1: areas treated as administrative
backwaters
One mechanism through which the low levels of colonial bureau-
cratic and police capacity continued into post-colonial times, was
because these areas are still considered ‘punishment postings’ to
which poor quality officers, or those being punished for their recal-
citrant political positions were sent by politicians.17 According to a
police officer who works in Maoist affected south Chhattisgarh:
‘‘. . .in a batch of ASIs, those who are the toppers in the batch or best
officers, . . . are not sent to the Naxal areas, the others who are not
the best officers are sent.”18 Such interviews suggested that the colo-
nial policy of treating these areas as administrative backwaters was
continued by the Indian bureaucracy and police.19
Path dependent mechanism 2: continuation of resource exploitation
creating grievances
Following independence, the princes and zamindars were abol-
ished, and there was a Congress Party government in the state of
Madhya Pradesh for most of the 1950–80s period. However, much
like the colonial state, there has been a continuing interest of the
post-colonial Indian state in these peripheral areas to exploit the
rich deposits of bauxite and iron ore in southern Chhattisgarh, thus
showing another continuity in state policy (Kennedy &
Purushotham, 2012: 840). This extraction occurred through the
National Mineral Development Corporation since the 1960s, and
more recently through MNCs like Essar in the Kirandul/Bacheli
area in Dantewada district, without building adequate roads and
providing sufficient development.20 These iron ore plants near
Kirandul release a lot of pollution and negatively affect health out-
comes of neighboring villages.21 Human rights activists note that
the controversial Salwa Judum vigilante movement was used to
make land easily available for mining projects by private companies
who were interested in the rich iron ore resources in the Dantewada
area.22
Another continuing form of exploitation is in minor forest prod-
ucts like tendu leaves which is used to make local cigarettes or
bidis, and which has a very strong lobby of contractors and traders
with a lot of influence on the state politicians. The tribals were paid
really low wages for picking tendu leaves and delivering them to
the non tribal contractors (PUCL, p. 13).23
4.4. Maoist rebel agents exploit these structural conditions of weak
state capacity and inequality
I briefly mention the history of the Maoists in southern
Chhattisgarh here.24 The People’s War Group (PWG) Maoists
emerged under the leadership of K. Sitaramaiah in Telangana region
of Andhra Pradesh in 1980. In early 1980s, some PWG squads from
Andhra Pradesh crossed the state border into Dandakaranya forest
zone in south Chhattisgarh area.25 They found areas of deep forests
and Gond tribes with grievances due to exploitation of their
17
Interview with Superintendent of Police, (district not revealed), Chhattisgarh,
2009.
18
Interview with Deputy Superintendent of Police, (district not revealed), Chhat-
tisgarh, 2009.
19
Miklian (2009: 443-44) mentions that Chhattisgarh police ‘‘lags far behind other
Indian states in both quantitative and qualitative terms”, and most policemen view
Chhattisgarh positions as bad for their career.
20
Interview with human rights activist Dr. Binayak Sen, Raipur, Chhattisgarh, 2008.
21
Focus group discussion with villagers, Village X (pseudonym), near Kirandul
town, Dantewada district, 2009.
22
Interview with PUCL activist Y (pseudonym), Bhilai, Chhattisgarh, 2009. Also see
PUCL, Where the State Makes War on its Own people, p. 7–8.
23
(PUCL, p. 13).
24
For a longer description, see Online Appendix H.
25
Dandakaranya Special Zonal Committee, CPI-Maoist. 2005. Jan Jagran Nahi, Jan
Daman Abhiyan, p. 6.
119
Fig. 2. Annual fatalities from Maoist insurgency in Chhattisgarh: 1990–2008.
Source: Data from 1990 to 99 is from Madhya Pradesh police records and from 2000
to 08 is from Chhattisgarh police records.
forest and natural resources. These Maoist rebels started addressing
these grievances of these tribals, and started recruiting and
mobilizing them to form rebel groups.26
Once the PWG settled into the area of Bastar/Kanker in South
Chhattisgarh in 1980–90s, they started the Dandakaranya Adivasi
Kisan Majdoor Sangh (DAKMS) or the Tribal peasant and workers
association, and the Krantikari Adivasi Mahila Sangh (KAMS) or
the Revolutionary tribal women’s association, through which they
recruited the adivasi tribals.27 An important contribution of the
Maoists was to force the government and contractors to increase
the price that the tribals received for picking tendu leaves. This was
an important form of livelihood, and the PWG Maoists pressurized
the traders and the government to increase the wages, else threat-
ened labor strikes. While for over two decades, ‘‘the wages remained
at 3 paise to 8 paise per gaddi (a bundle of 50 leaves)” . . . from about
‘‘1982 it has risen dramatically to 30 paise in 1988, while the rest of
the state remained at 9 paise.”28
By 1990s, the PWG succeeded in developing rebel control deep
inside the Abujhmarh forest areas of former Bastar princely state
in the south of Chhattisgarh, by exploiting these historically cre-
ated structural conditions of weak state capacity and ethnic grie-
vances. Except for a few short escalations in the 1990s, violence
remained low in this area which had become a Maoist stronghold.
It was only when there was sudden escalation of violence in 2005
due to the emergence of an anti-Maoist vigilante movement called
the Salwa Judum, led by the Congress leader Mahendra Karma, that
the central government took notice of the Maoists in this corner of
India. The center launched a counter insurgency operation called
Green Hunt in 2009 with the goal of eliminating Maoist power in
their core areas of Maoist control, but this was not fully successful
(see Fig. 2 for violence patterns).
4.5. Scope conditions: timing of onset of insurgency, and theory not
applicable to post 2004 period
Historical institutions like colonial indirect rule do not explain
the timing of onset of insurgency, but only explain the spatial vari-
ation in rebel control after insurgency emerges. The timing of emer-
gence of the Maoists in Chhattisgarh was related to the ideological
and strategic re-thinking within the broader radical leftist move-
ment in India in 1977–80 (Sinha, 1989), which led to emergence
26
P. Shankar (2006), p. 222–226; PUCL, Bastar: Development and Democracy, July
1989, p. 13.
27
P. Shankar (2006), p. 222–226; PUCL, Bastar: Development and Democracy, July
1989, p. 13.
28
PUCL, Bastar: Development and Democracy, 1989, p.10.
120 S. Mukherjee / World Development 111 (2018) 113–129
of the PWG in Telangana region of former princely state of Hyder-
abad in the neighboring state of Andhra Pradesh in 1980, and a
strategic decision by the PWG leadership to expand to neighboring
Bastar region of Chhattisgarh in 1980s.29
Also, colonial institutional legacies may not be useful to explain
rapid expansion of insurgency once a rebel organization develops
sufficient strength to be able to expand even to areas of higher
state capacity. So this theory can only explain spatial variation in
rebel control before the unification of the two most powerful Mao-
ist groups, the MCC and PWG, to form CPI-Maoist in 2004, which
suddenly increased the organizational capacity and resources of
the Maoists and resulted in a phase of expansion all over India.
5. Empirical analysis of spatial variation in Maoist control in
Chhattisgarh
5.1. Addressing the selection issue – using an instrument for colonial
indirect rule
In her anthropological study of Bastar princely state, Sundar
(2007), p. 103, writes: ‘‘Indirect rule had several advantages for
the British. First, it was useful for areas which were not immedi-
ately open to exploitation, like Bastar. In 1884, the Chief Commis-
sioner explicitly concluded that Bastar was too poor to justify
direct management by the state. Secondly, it was cheaper. . . . ”
The British possibly selected districts for indirect rule based on
whether they were more difficult to govern due to hilly terrain,
high forest cover, intrinsic tribal rebelliousness, low revenues,
and other unobserved qualities.
If this is true, then the pre-colonial qualities of these districts
made them more prone to tribal rebellion in the future, and so
OLS estimates would be potentially biased upward. However,
selection was not universal and historical contingencies sometimes
prevented the British from being selective, and they chose districts
like Khairagarh, Kawardha, Nandgaon in the western border of
Chhattisgarh for indirect rule, even though these were more agri-
culturally productive, and had no tribal rebellion, which goes
against the expected direction of bias.30 In the OLS analysis below,
I control for some observable determinants of indirect rule choice,
like forest cover and elevation. To better address the issue of unob-
servable qualities of districts causing selection effects, I then use an
instrument for the choice of indirect rule through princely state,
based on the random death of the ruler of Nagpur, which resulted
in the central districts of Chhattisgarh switching to British direct
rule.
5.1.1. Instrument for indirect rule through princely states
(instru_NagpurLapse)
Following the defeat of Appa Sahib of the Marathas in 1817 to
the British, the Nagpur State was managed by the British on behalf
of the minor Bhonsla Raja Raghuji III, who ‘‘succeeded to the
estates in 1830, but died without heirs in 1853, and the province
lapsed to the British Government.” (Baden Powell, 1892, Vol. 1,
p. 47) 31 The reason for lapse of Nagpur State, ruled by the Bhonsla
clan of the Marathas, to the British government due to death of the
ruler was because of the existence of the policy of Doctrine of Lapse
implemented by the governor general Lord Dalhousie (1847–56)
under which any Indian ruler who died without a male heir would
have to forfeit his kingdom to the British, and adoption of heirs
would not be recognized. The switch to British direct rule was due
to the random circumstance of the death of the ruler of Nagpur state
29
See Section 2.1 above and Appendix H for details of history of the movement in
Chhattisgarh.
30
E. A de Brett, Chhattisgarh Feudatory States, p. 3–4.
31
Baden Powell, Land Tenure Systems of British India, Volume 1, p. 47.
without a male heir, and was for reasons exogenous to the quality of
the districts within Nagpur state. The central districts of current day
Chhattisgarh—Raipur, Bilaspur, Durg, Dhamtari and Mahasamund—
were part of Nagpur state in 1854, when they formed only two large
districts of Raipur and Bilaspur. Since the reason for the switch of
these districts to colonial direct rule were due to random circum-
stances of the natural death of the ruler of Nagpur state, and were
exogenous to the quality of these districts which could make them
more prone to rebellion then or in the future, therefore it is used
as an instrument for the British choice of direct/indirect rule in the
Chhattisgarh area.
Exclusion restriction: The only way by which the death of the
Nagpur ruler is related to the outbreak of Maoist insurgency in
the future is through the conversion of these districts to direct rule.
This satisfies the requirement of the instrument being exogenous
to the outcome being studied, and fulfills the exclusion restriction
condition of a good instrument. While it could be argued that the
British wanted Nagpur state for its resources and there was selec-
tion involved, the British administrators could not have predicted
the death of Raghuji III in 1854 beforehand. So the actual timing
of the switch to direct rule was randomly caused by this death,
due to the prevailing official policy of Doctrine of Lapse of the gov-
ernor general Lord Dalhousie, and thus satisfies the exclusion
restriction condition. The only reason this was not exogenous
would be if the British planned the death of the ruler, but there
is no evidence of this. Also, the British chose not to annex Nagpur
state in 1817 after defeating the Marathas in battle due to geo
political circumstances, and this too goes against the argument
that they selected it for direct rule for its resources suddenly in
1854, something they could have easily done in 1817 itself, but
did not.
Coding of the instrument: See Appendix E (online appendix) for
details of coding of instrument. The instrument is coded as those
constituencies in current day Chhattisgarh state which were part
of the colonial era districts of Bilaspur and Raipur, since these
two districts in the colonial era were part of erstwhile Nagpur state
and today are part of current state of Chhattisgarh. In 2001 census,
the erstwhile Raipur and Bilaspur districts have been divided into
the following districts of current day Chhattisgarh—Raipur, Bilas-
pur, Durg, Dhamtari and Mahasamund. These districts are thus
coded as the instrument since they were part of Nagpur state in
1854, when they formed only two large districts of Raipur and
Bilaspur, and switched to British direct rule and became part of
Central provinces.
However, the districts of Rajnandgaon, Kawardha etc. which
were part of the colonial era princely state of Nandgaon in the
west, did not switch to British direct rule, and neither did the prin-
cely states of Surguja, Koriya, Jashpur in the north, and the princely
states of Bastar, Kanker in the south. Also, there were some dis-
tricts like Korba (3 constituencies) and Janjgir-Champa (6 con-
stituencies) which were British direct rule and not part of
Nagpur princely state and did not become direct rule because of
this random switch due to death of Nagpur ruler, and so are not
coded as part of the instrument.
Comparing with Iyer (2010)’s IV2SLS: This instrument is similar
to the instrument in Iyer (2010)’s analysis of the effects of colonial
British rule on development outcomes, but the IV2SLS analysis
below contrasts with the Iyer (2010) IV2SLS results which suggest
that on average districts with indirect rule in India have more
development. This is because Iyer (2010) uses a truncated sample
for her IV2SLS analysis for districts annexed only in the 1847–56
period of Lord Dalhousie’s reign, thus not being able to address
selection in the northern epicenter of insurgency in Bihar/Bengal
which had already been annexed in 1765 in the Battle of Buxar.
Also there are some coding errors of princely state in the Iyer
(2010) dataset, i.e. Bastar and Rajnandgaon districts of 1991
 S. Mukherjee / World Development 111 (2018) 113–129
Census are coded as britdum = 1, when actually they are prin-
cely states and britdum = 0 (Imperial Gazetteer of India, v. 26,
Political Divisions map, p. 20). This is explained in greater detail
in Online Appendix K.
5.2. OLS and IV-2SLS regression analysis
The following hypothesis is proposed based on the theory out-
lined above.
Hypothesis: Areas which had indirect colonial rule through princely
states tend to have higher probability of Maoist insurgency in the
future, even after controlling for various proximate measures of rebel
opportunity and ethnic grievances.
5.2.1. Data and variables
Mapping Census data into Assembly Constituencies:. Unlike most of
the econometric analysis of Maoist insurgency in India which use
district level datasets, this paper uses the Vidhan Sabha (State
Assembly) constituency as the unit, to achieve sufficient disaggre-
gation in measurement of rebel control. Since the constituency is
the basic political unit for which MLAs are elected and are account-
able to the public, it makes sense to use this as the unit, instead of
the district, which is too aggregate and masks sub-district variation
in Maoist control.32 This is the first study on the Maoist insurgency in
India to use the sub-district Assembly Constituency as the unit of
analysis.
The state of Chhattisgarh has 17 districts subdivided into 97
tehsils, and 90 State assembly constituencies (ACs).33 Each district
has one or more ACs. I develop a dataset for these 90 ACs which is
very detailed and the first constituency level dataset for the state
of Chhattisgarh. The dependent variable of Maoist control as well
as the independent variable of princely state are measured at the
AC level directly. Unfortunately, the Census 2001 socio economic,
public good and demographic data is in administrative units like dis-
trict/tehsils, and there is not a perfect match geographically of tehsil/
district boundaries with Assembly constituency (AC) electoral
boundaries.34 The geographical matching of constituencies with
administrative boundaries is a problem for several well-known stud-
ies of Indian politics.35 To match the 90 constituencies to the 97 teh-
sils, a map of district/tehsil boundaries of Chhattisgarh in 2001 was
compared to constituency boundaries from an electoral map from
2003 State Assembly Elections.36 The boundaries of tehsils and
assembly constituency boundaries was matched to the extent possi-
ble, following the procedures of Banerjee and Somanathan (2007)
and Thachil (2011). The coding procedure and data sources are
described in Online Appendix B.
Measuring Maoist control as the dependent variable. In this paper, I
explain spatial variation in Maoist rebel control, instead of killings
32
For example, the Indian state considers a district like Kanker in Chhattisgarh as
under high Maoist influence, but it is the western areas of Kanker which are under
higher Maoist control, while the eastern areas are not.
33
There are three levels of elections within Indian democracy. The first is the Lok
Sabha or national legislative assembly elections. The second is state legislative
assembly or Vidhan Sabha elections for individual provinces or states. The third is
elections to panchayats of village level bodies.
34
The administrative unit comprises the state followed by the district, and then the
mandal or taluk or tehsil (different states have different names for this sub district
level unit), followed by block, town or village. The electoral unit within a state is the
assembly constituency (AC) followed by polling stations.
35
Chandra (2004, p. 159–160) notes ‘‘. . . government data, including data from the
census, are reported by administrative subdivisions (district, tehsil, and block), which
do not coincide with constituencies.”
36
Map 7: ‘Chhattisgarh Administrative Divisions 20010 in Census of India 2001
Chhattisgarh Administrative Atlas, p. 15, with Table 7: ‘List of Assembly Constituencies
District-wise’, pages 15–17, and ‘Map of Assembly Segments’, p. 18 of Chhattisgarh
State Assembly Election – 2003.
121
or fatalities. It is important to distinguish between these concepts,
since in those areas where the state is completely absent and the
rebels have full control and set up alternate institutions of gover-
nance, there is possibly less violence, and yet insurgency is most
successful (Kalyvas, 2006). In the case of Chhattisgarh, this is even
more the case because before the start of the Salwa Judum move-
ment in 2005 which led to escalation of violence, the levels of rebel
control were quite high in the southern districts like Narayanpur
and Dantewada, and yet levels of violence were relatively low, as
shown in Fig. 2. So using insurgency violence will not correctly mea-
sure the areas of rebel control in early 2000s before violence esca-
lated in Chhattisgarh in 2005. A second reason is that media
based events data may not record violent events deep inside rebel
territories due to lack of access to information from such zones of
high rebel control (Zukerman-Daly, 2012). For both these reasons,
using total violence could mis-measure rebel control and lead to
conceptual stretching (Sartori, 1970).
While Kalyvas (2006) had used rebel control as an independent
variable to explain selective violence, he does not have a good expla-
nation for why rebel control emerges in some places and not others,
and it is important to explain spatial variation in rebel control as the
dependent variable. Following the precedence of Mitchell (1968),
Kalyvas and Kocher (2009), I measure Maoist control based on gov-
ernment perception of rebel influence.37 Electoral data gives the %
of polling stations within each Assembly constituency that were
Highly Sensitive, Sensitive, or Normal for the 2003 State Assembly
elections, and was collected during field work.38 I use the % of Highly
Sensitive polling stations within each constituency as a proxy of
Maoist control in that constituency. This is a measure of how polit-
ically sensitive a particular constituency is perceived to be, based on
reports that the individual district administrations send to the Chief
Election Commissioner of the state prior to elections. This is a con-
tinuous measure and provides more information than a categorical
measure as used by Kalyvas (2006) or Mitchell (1968).39 It is a valu-
able empirical contribution, given the challenges of collecting good
data in an active conflict zone with the Indian state often reluctant
to share data, and field work being risky.40 See Online Appendix C
for details.
Possible bias in measure of rebel control does not bias regression
coefficients: There are various possible biases in this measure of
Maoist control and these are discussed in detail in Online Appendix
C, and I briefly mention these here.
One possible bias occurs in a well-known previous study using
measures of rebel control based on government perception, i.e.
Kalyvas and Kocher (2009: 342) use the Hamlet Evaluation System
(HES) data created by the US military during the Vietnam war to
measure rebel control, but note that there is ‘‘optimistic bias” in
the HES data in favor of showing more government control.41 It
is possible that in Chhattisgarh, the district police chief intentionally
37
Mitchell (1968) tries to predict spatial variation in government control in the
Vietnam war using data on 26 South Vietnamese provinces measured as ‘‘the
percentage of hamlets in a province classified as government controlled according to
a detailed map . . . which was derived from US government sources.” Kalyvas and
Kocher (2009) and measure rebel control in the Vietnam insurgency based on the
Hamlet Evaluation System (HES) data created by the US government through regular
surveys of US and Vietnamese officials.
38
Table 22: ‘Assembly Segment Wise Highly Sensitive, Sensitive & Normal Polling
Stations’, Chhattisgarh State Assembly Election – 2003, p. 52.
39
Analysis is also carried out by converting the % of Highly Sensitive polling stations
into a categorical measure of Maoist control, on a scale of 0 to 4, like Kalyvas’ (2006)
categorization of rebel control. The results are similar.
40
Various human rights activists and scholars have been harassed during field work
in Maoist areas, and the author faced suspicion from the police.
41
Kalyvas and Kocher (2009: 342) find that ‘‘Bole & Kobata (1975) find evidence
that the HES was biased in the direction of government control (the so-called
‘optimistic bias’), perhaps in response to bureaucratic incentives to show ‘progress’ in
the pacification effort.”
122 S. Mukherjee / World Development 111 (2018) 113–129
underestimates some areas under Maoist control to show progress to
superior officers, thus creating this ‘optimistic bias’.42 Bias could go
in the opposite direction, if the police chief intentionally overesti-
mates the number of sensitive polling stations in their district to
ensure that the Election Commission provides extra police and
paramilitary troops to that district, to make it easier for them to
organize peaceful elections.43 However, the error in the measure of
political sensitivity is not correlated with the independent variable
of colonial indirect rule, since the police have number of Maoist
squads, and maybe level of violence in mind when perceiving the
level of Maoist control, but are not thinking of historical institu-
tions.44 So, the OLS estimators are unbiased and consistent, though
there will be larger variance in the error terms.45
Independent variables and control variables.
Independent variable: The main independent variable is colonial
indirect rule through princely states, in the area which is today part
of Chhattisgarh state in central India.
Princely state: Princely state dummy was created by matching
the 2003 Chhattisgarh Assembly Constituency Election Map46 with
the map of direct and indirect rule in Chhattisgarh from The
Imperial Gazetteer of India: Volume 26—Atlas, Plate 39 (see Fig. 1B
above). This is coded as 1 if the constituency was earlier part of
any of the princely states within Chhattisgarh area, and 0 if it was
part of the British governed areas of Raipur, Bilaspur and Durg,
which is possible since there are no constituencies today which were
partly within British India and partly within princely states.
Wilkinson (2017) criticizes studies like Verghese (2016a,b),
Banerjee and Iyer (2005), and Iyer (2010) for clubbing different
types of colonial land tenure and indirect rule which are really
quite dissimilar into similar categories.47 While in future research
it would also be useful to have a continuous measure of princely
state, for this paper I consider that using the same binary measure
of princely state as in Iyer (2010) has the following advantages. First,
it allows my study to be comparable to Iyer (2010) by using similar
binary measure of princely state and still showing that the conclusion
is the opposite, i.e. indirect rule leads to lower development and
insurgency. Second, unlike in an all India analysis where there would
be a lot of variation between types of princely states, within the state
of Chhattisgarh, there is not much variation in types of princely
states between Kanker, Bastar, Surguja, Koriya, Nandgaon which
are all Chhattisgarh feudatory states with Rajput rulers. Third, I
cluster using colonial era princely state boundaries in Table 4C, and
that takes care of heterogeneity between princely states in
Chhattisgarh.
Control variables: Electoral data at the constituency level were
directly obtained from the Chhattisgarh State Election Commis-
sion. The socio-economic and demographic data like literacy,
unemployment, percent of scheduled castes and tribes, provision
of public goods etc. were converted from the Census of India
(2001) data, using the procedure described above. These variables
and their sources are described in Online Appendix D.
42
Interview with Superintendent of Police, Chhattisgarh (district not revealed),
2009.
43
Interview with Chief Election Commissioner of Chhattisgarh, Raipur, November
2008.
44
The Ministry of Home Affairs, India, use various criteria like number of
underground Maoist squads, number of overground front organizations of Maoists
etc. to determine which districts are under Maoist control at the national level for the
Security Related Expenditures (SRE) scheme, and similar criteria are probably being
used here.
45
See Woolridge, Introductory Econometrics, p. 302–303.
46
‘Map of Assembly Segments’, p. 18 of Chhattisgarh State Assembly Election – 2003.
47
See Steven Wilkinson, ‘‘Looking Back at the Colonial Origins of Communal and
Caste Conflict in India,” February 21, 2017, The Wire, at https://thewire.in/
110535/communal-violence-caste-colonialism/.
5.2.2. OLS estimates of the impact of indirect colonial rule on Maoist
insurgency
(a) Difference in means shows effects of colonial institutions on
intermediary mechanisms of state capacity, development
Table 1A gives the difference in means between direct rule and
indirect rule constituencies of each variable for all constituencies
in Chhattisgarh, and shows significantly higher mean values of
Maoist control for the constituencies that were under indirect rule
through princely states. It shows that constituencies that were for-
merly princely states had more forest cover, and higher terrain,
which indicates possible selection.
However, Table 1A also show that princely state constituencies
have lower access to many public goods like schools, water tanks
and tubewells, electricity, while British direct rule areas have
higher levels of access to paved roads, buses, and less average dis-
tance of villages from main town, i.e. higher state penetration in
these rural areas (Census 2001 data). Direct rule constituencies
also seem to have higher literacy and population density (levels
of urbanization). Table 1B presents OLS regressions that show sim-
ilar results that princely state is a statistically significant predictor
of all these various contemporary measures of public goods, devel-
opment and state capacity. This provides evidence for the long
term negative effects of the Chhattisgarh princely states, and pos-
itive effects of colonial direct rule on development, state capacity,
and ethnic inequalities, which acted as intermediating mecha-
nisms that were exploited later by Maoist rebels.
(b) Model specification: The OLS regressions are of the form
Y i ¼ a þ b princelystatei þ dX i þ ei ;
where Yi is the dependent variable of Maoist control for con-
stituency i, princelystatei is the dummy for whether the constituency
was earlier part of a princely state or British India, and Xi stands for
other constituency characteristics controlled for, including forest
cover, elevation, literacy, rurality, % of tribes and lower castes,
and other variables usually included in civil war onset models. ei
represents the error term.
The OLS model results are presented in Tables 2A, and 2B. In all
models, standard errors are clustered at the district level to
account for unobserved district level characteristics, because the
district is the administrative unit within a province, and electoral
constituencies in the same district are similar in terms of adminis-
tration, counter insurgency, and development policy, and other
unobserved qualities.48 In Table 2A, the baseline specification is
Model 1A, which has only the main independent variable, princelys-
tate, which is statistically significant at the 1% error level in predict-
ing Maoist control. In Models 1B to 1D, princely state remains
significant even after controlling for forest cover and elevation, which
are two measures of terrain that could provide opportunity for rebel-
lion, and could also be considered pre-colonial qualities of these
areas affecting selection by British. Combined with the results in
Table 1B which shows that princely states has lower levels of devel-
opment and state capacity, the results in Models 1A to 1D in
Table 2A suggests that princely state has an effect on Maoist control
through these causal mechanisms.
48
It is not possible to do district fixed effects to control for any possible district level
unobservables, because there is little within-district variation in the princely state
variable, preventing Fixed Effects estimation for princely state due to almost no within
group variation (See Gujarati, Basic Econometrics, p. 640–51). This is because post-
colonial district boundaries continued to follow the colonial era district boundaries
without much change, and there are no current districts in Chhattisgarh which were
part of both British direct rule and princely states. However, Random Effects
estimation shows princely state dummy to be still statistically significant. Results
available from author.
 S. Mukherjee / World Development 111 (2018) 113–129 123

Table 2A
OLS estimates of impact of colonial institutions on Maoist rebellion.
Dependent Variable: Maoist Control (% Highly Sensitive Polling Stations in each Constituency).

Pre colonial controls Post colonial controls

Model 1A Model 1B Model 1C Model 1D Model 2A Model 2B
Princely State 20.09*** (0.001) 13.42*** (0.004) 17.83** (0.044) 14.67** (0.042) 15.55*** (0.005) 11.93*** (0.002)
Forest Cover 0.260** (0.015) 0.277** (0.039) 0.281** (0.014) 0.118 (0.220)
Elevation 0.0134 (0.569)
0.0101 (0.724)
0.0304 (0.173)
Popln Density 0.000250 (0.960) 0.00238 (0.588)
Rural % 0.0285 (0.636) 0.0688 (0.178)
Per Capita Income
0.000320 (0.476) 0.000172 (0.483)
Male Unemployment 0.490 (0.260) 0.852* (0.064)
Scheduled Caste%
0.505** (0.038)
Scheduled Tribe% 0.450*** (0.009)
Electricity 26.93** (0.040)
ST% * Electricity
1.029*** (0.001)
Male Literacy
0.481* (0.090)
Constant 2.173* (0.061)
3.013 (0.136)
2.110 (0.774)
0.151 (0.984)
34.74 (0.203)
13.86 (0.749)
Observations 90 90 90 90 90 90
R2 0.296 0.429 0.304 0.433 0.439 0.651

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
Robust Std. Errors adjusted for 17 clusters by district.

Table 2B
OLS estimates of impact of colonial institutions on maoist rebellion.
Dependent Variable: Maoist Control (% Highly Sensitive Polling Stations in each Constituency).

Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Princely State 12.10*** 11.34*** 11.46*** 11.62*** 12.14*** 12.87***
(0.001) (0.002) (0.003) (0.002) (0.002) (0.001)
Forest Cover 0.129 0.0916 0.105 0.116 0.115 0.109
(0.194) (0.384) (0.226) (0.227) (0.230) (0.241)
Elevation
0.0310
0.0289
0.0287
0.0287
0.0314
0.0265
(0.167) (0.164) (0.190) (0.180) (0.157) (0.226)
Popln Density
0.000349 0.00784 0.00217 0.00781 0.00266 0.00454
(0.952) (0.325) (0.624) (0.114) (0.560) (0.259)
Rural % 0.0672 0.0811 0.0562 0.0799 0.0683 0.0853
(0.181) (0.181) (0.312) (0.130) (0.203) (0.136)
Per Capita Income 0.000177 0.000418
(0.484) (0.137)
Male Unemployment 0.839* 0.833* 0.763* 0.819* 0.869* 0.565
(0.061) (0.052) (0.063) (0.061) (0.068) (0.111)
Scheduled Caste%
0.431*
0.454**
0.449*
0.556**
0.492*
0.741**
(0.059) (0.040) (0.065) (0.037) (0.057) (0.022)
Scheduled Tribe% 0.462*** 0.493*** 0.393** 0.408** 0.461** 0.313**
(0.008) (0.006) (0.031) (0.017) (0.011) (0.046)
Electricity 26.34** 29.53** 24.15** 25.19** 26.04** 25.03**
(0.033) (0.023) (0.032) (0.047) (0.026) (0.024)
ST% * Electricity
1.032***
1.047***
0.960***
0.968***
1.017***
1.001***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.000)
Male Literacy
0.479*
0.498*
0.491*
0.526*
0.502
0.476*
(0.086) (0.078) (0.091) (0.075) (0.126) (0.096)
Paved Road 7.410
(0.270)
Post Office
24.27
(0.334)
Distance from Town 0.0691
(0.515)
Railway Station
170.3
(0.110)
VoterTurnout 1985 0.0568
(0.760)
ENPV 1998 6.374**
(0.030)
Constant
16.25
12.00
7.445
8.116
15.70
11.82
(0.708) (0.775) (0.859) (0.846) (0.720) (0.770)
Observations 90 90 90 90 90 90
R2 0.653 0.657 0.655 0.655 0.652 0.687

p-values in parentheses.
Robust Std. Errors adjusted for 17 clusters by district.
*
p < 0.10, **p < 0.05, ***p < 0.01.

In Table 2A, Model 2A adds proximate post-colonial controls electricity, and the interaction term of scheduled tribes with access
like population density, rural percent, per capita income, and male to electricity. Table 2B has Models 3 to 6 replace per capita income
unemployment, and Model 2B includes proximate factors like male which Fearon and Laitin (2003) use as their proxy for state capac-
literacy, percent of scheduled caste and scheduled tribes, access to ity, with other proxies for state capacity like access to paved road,
124 S. Mukherjee / World Development 111 (2018) 113–129

Table 3A
First stage of IV-2SLS estimates of impact of colonial institutions on Maoist rebellion.
Dependent Variable: Princely State.

Pre colonial controls Post colonial controls

Model 1A Princely Model 1B Princely Model 1C Princely Model 1D Princely Model 2A Princely Model 2B Princely
State State State State State State
Nagpur Lapse (1854)
0.776*** (0.000)
0.719*** (0.001)
0.645*** (0.002)
0.641*** (0.004)
0.576*** (0.008)
0.604*** (0.004)
Forest Cover 0.00229 (0.454) 0.000280 (0.902)
0.00110 (0.621)
0.00250 (0.262)
Elevation 0.000995 (0.114) 0.000970* (0.089) 0.000694 (0.134) 0.000446 (0.214)
Popln Density
0.000124 (0.430) 0.00000304 (0.981)
Rural %
0.000635 (0.778)
0.000100 (0.962)
PerCapita Income 0.00000732 (0.148) 0.00000500 (0.517)
Male Unemployment
0.0249 (0.103)
0.0224 (0.285)
Scheduled Caste%
0.00880 (0.463)
Scheduled Tribe%
0.00389 (0.562)
Electricity
0.382 (0.138)
ST% * Electricity 0.00612 (0.340)
Male Literacy
0.00658* (0.099)
Constant 0.776*** (0.000) 0.679*** (0.006) 0.329 (0.381) 0.328 (0.386) 1.950** (0.029) 2.696* (0.060)
Observations 90 90 90 90 90 90
R2 0.611 0.626 0.676 0.676 0.707 0.728
Adjusted R2 0.607 0.617 0.668 0.665 0.682 0.685
Weak Instrument Diagnosis:
Partial R2
Angrist Pischke F statistic 37.27 17.44 12.95 11.36 9.08 11.63
Angrist Pischke p value (0.000) (0.000) (0.0024) (0.0039) (0.0082) (0.0036)
Kleibergen-Paap Wald 37.27 17.44 12.95 11.36 9.08 11.63
statistic
Cragg-Donald Wald F 138.49 99.56 89.38 82.18 61.25 60.97
statistic

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
Robust Std. Errors adjusted for 17 clusters by district.

access to post office, railway station, average distance of villages from
nearest town in a constituency.49
These OLS models 2A, 2B, and 3 to 6 show that princely state
remains a statistically significant predictor of Maoist control even
after controlling for more proximate measures of ethnic grievances,
poverty and literacy, state capacity, which are themselves the
result of colonial rule institutions. This suggests that the colonial
indirect rule institutions have its own separate effect, over and
above the effects of contemporary factors like state capacity,
ethnicity, development on rebellion. However, the OLS results
may possibly be biased due to selection, so the next section pre-
sents the IV-2SLS analysis.
5.2.3. 2SLS instrumental variable regression model estimation:
In this section, I replicate the Models from the OLS regressions
in Tables 2A and 2B, using IV-2SLS. I instrument for the potentially
endogenous independent variable, princelystate, using instru_Nag-
pur Lapse, based on random annexation of Nagpur state due to
death of the ruler.
First stage of IV-2SLS: The first stage regression for the IV strat-
egy is:
princelystatei ¼ a þ binstruN agurLapsei þ dX i þ ei
where princelystatei is the measure of colonial indirect rule for the
i th constituency, as in the OLS regressions above, and Xi stands
for the same constituency characteristics controlled for in the OLS
model. In the first stage of the 2SLS, I use instru_NagpurLapse as
the instrument for the choice of direct vs. indirect rule in the British
49
These indicate the ability of state forces to quickly reach distant rural areas, and
may be better measures of actual state penetration in rural areas, than measures of
per capita income at the sub-district level. Model 7 includes voter turnout from 1985
assembly elections in Chhattisgarh as a possible measure of level of democracy.
Model 8 includes ENPV lagged for the 1998 Chhattisgarh elections, which Wilkinson
(2004) found has an effect on ethnic riots in India.
districts of Chhattisgarh based on the lapse of Nagpur state to Bri-
tish rule.
The results of the first stage of the IV-2SLS are presented in
Table 3A below. Replicating Model 1A, 1B, 1C, 1D, 2A and 2B from
Table 2A of the OLS regressions, the first stage regressions gener-
ally show that Instru_Nagpur Lapse has a strong negative correla-
tion with colonial indirect rule through princely states.50 This is
because these districts switched to direct British rule due to death
of ruler. Bound, Jaeger and Baker (1995) show how the IV-2SLS esti-
mates may be somewhat biased in the same direction as OLS esti-
mates in finite samples if there is weak correlation in the first
stage between the instrument and the endogenous independent
variables. For Model 1A, the first stage regression with princely state
as the dependent variable yields an Angrist-Pischke F statistic of
37.27 (p value of 0.000) for the excluded instrument. For Models
1B, 1C, 1D, the Angrist-Pischke F statistic is 17.44, 12.95, 11.36
respectively. For Model 2A and 2B, which are the longer models,
the Angrist-Pischke F statistic is 9.08 (p value of 0.0082) and 11.63
(P value of 0.0036) respectively.51 The Angrist-Pischke F statistic of
the instrument for princelystate is larger than 10 in all models except
2A, which is the usual F statistic value required to suggest that
instruments are sufficiently strong. Given this information, it is pos-
sible to reject the null hypothesis of no correlation between instru-
ment and endogenous regressors at the 95% confidence level, and
this implies that the instruments are not weak.52
Second stage of IV-2SLS: The second stage regression for the IV
strategy estimates the impact of colonial indirect rule through
50
Estimating first stage regressions for Models 3–8 in Table 4A below will give
similar results, and are not shown.
51
The Kleibergen-Paap Wald statistic is the same as the Angrist-Pischke statistics,
since there is only one endogenous independent variable, namely princelystate. See
Angrist & Pischke (2009) for discussion on instrumental variables.
52
See Stock, Wright and Yogo (2002) for detailed discussion about weak
instruments, and see Baum, Schaffer, and Stillman (2007) for how to check for weak
instruments, and the STATA code to use.
 S. Mukherjee / World Development 111 (2018) 113–129 125

Table 4A
IV-2SLS estimates of impact of colonial institutions on Maoist rebellion.
Dependent Variable: Maoist Control (% Highly Sensitive Polling Stations in each Constituency).

Pre colonial controls Post colonial controls

Model 1A Model 1B Model 1C Model 1D Model 2A Model 2B
Princely State 22.06*** (0.000) 13.92*** (0.000) 20.81*** (0.007) 15.13** (0.011) 16.47*** (0.005) 11.79** (0.037)
Forest Cover 0.256*** (0.005) 0.275** (0.019) 0.290** (0.011) 0.118 (0.158)
Elevation 0.00737(0.744)
0.0109 (0.680)
0.00716 (0.792)
0.0303 (0.144)
Popln Density 0.000179 (0.974) 0.00233 (0.535)
Rural % 0.0267 (0.652) 0.0686 (0.120)
PerCapita Income -0.000324 (0.446) 0.000172 (0.424)
Male Unemployment 0.453 (0.293) 0.846* (0.052)
Scheduled Caste%
0.505** (0.013)
Scheduled Tribe% 0.450*** (0.001)
Electricity 26.95** (0.016)
ST% * Electricity
1.030*** (0.000)
Male Literacy
0.482* (0.061)
Constant 1.341** (0.023)
3.105* (0.085)
1.002 (0.889) 0.00386 (1.000)
30.34 (0.343)
13.42 (0.752)
Observations 90 90 90 90 90 90
R2 0.293 0.429 0.299 0.433 0.441 0.651
Adjusted R2 0.285 0.416 0.283 0.413 0.393 0.597
Endogeneity & Overidentification Test Diagnosis:
Hansen’s J statistic y y y y y y
Hansen’s J (p value) y y y y y y
Anderson-Rubin Wald test na 9.54 5.29 5.33 4.25 2.10
Anderson-Rubin (p-value) na (0.0070) (0.0352) (0.0347) (0.0559) (0.1668)

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
Robust Std. Errors adjusted for 17 clusters by district.
y
Hansen J Statistic (overidentification test of all instruments) is 0.000, since equation is exactly identified because the number of endogenous regressors equals the number of
instruments.

princelystate on Maoist control, and has the same specification as
the OLS model:
Y i ¼ a þ b princelystatei þ dX i þ ui ;
where Yi is the dependent variable of Maoist control for constituency
i used in the OLS regression table, princelystatei is the same measure
of indirect rule, Xi stands for the same set of control variables, and ui
represents the error term. Standard errors are robust to clustering
by district, though district fixed effects cannot be shown because
of lack of within district variation in the independent variable of
princelystate, since current district boundaries follow colonial era
district boundaries, as explained above.
Tables 4A below and 4B show the results of the second stage of
the IV-2SLS regressions, and are the same specifications as OLS
models in Tables 2A and 2B above. It is not possible to do the Han-
sen’s J test for overidentifying restrictions, since the number of
instruments is equal to the number of endogenous regressors,
and the equation is exactly identified, but the instrument should
satisfy exclusion restrictions.53
5.2.4. Analysis and discussion of results
Colonial institutions: In Table 4A, Model 1A, the instrumental
variable regression yields a point estimate of 22.06 on princelystate
which is significant at 99% confidence level. The size of the IV-2SLS
coefficient in Model 1A is slightly larger than the OLS regression
coefficient of 20.09 for princely state in Table 2A, Model 1A. The
effect of being a princely state in Chhattisgarh on the possibility
of Maoist insurgency in the future is quite large—moving a con-
stituency from direct rule under the British to indirect rule through
a princely state increases the % of Maoist affected polling stations
in the constituency by almost 22 percentage points. In Models 1B
to 1D, on including forest cover and elevation, princely state is still
statistically significant and positively correlated with Maoist con-
trol. These results suggest that controlling for pre-colonial determi-
53
See Kurt Schmidheiny, ‘‘A Short Guide to Microeconometrics”, Spring 2012,
Downloaded from internet.
nants of choice of indirect rule like terrain, as well as using an
instrument for indirect rule to address selection effects, princely state
institutions have a separate causal effect on the probability of Maoist
insurgency. Once the institutions of indirect rule were in place, they
had an additional effect on state penetration and development
which persisted through ‘sticky’ path dependent mechanisms and
created political opportunity structures for leftist insurgency.
In Table 4A, Model 2A, 2B, I include the other post-colonial and
more proximate factors, and in Table 4B, I control for the larger set
of post-colonial proximate factors like public goods, economic
development, literacy, employment, alternate state capacity mea-
sures. Princely state continues to be statistically significant predic-
tor of Maoist control. This suggests that colonial indirect rule
institutions have a causal effect beyond the effects of proximate
factors on Maoist insurgency. Colonial indirect rule acts as a deep
rooted structural condition which creates several of these proxi-
mate measures of state capacity, ethnic inequalities, and levels of
literacy and development, and these proximate factors then act
as mechanisms that are exploited by the Maoists.
Proximate post-colonial factors: The IV2SLS models include prox-
imate factors in Models 2A, 2B, and 3–7. Some notable results are
that forest cover is significant and has a positive coefficient in the
shorter models, but in the longer models like 2B, and Table 4B,
Models 3–8 it is not statistically significant. Elevation, which is a
proxy for rough terrain, has a positive but non-significant coeffi-
cient in Model 1C, but in all other models with more control vari-
ables, elevation has a negative coefficient, which could be because
of influential observations.54
In Model 2B, Scheduled Tribe (ST) % has a positive and significant
coefficient, since the Maoists recruit mainly from the poor tribal
populations in Chhattisgarh. The other papers like Gawande et al.
(2017), Chandra and Garcia-Ponce (2014), Gomes (2015) also find
that scheduled tribes have a positive correlation with Maoist
54
Maybe two influential observations of Konta and Bijapur constituencies that score
highest on Maoist control but low on elevation are causing the correlation to become
negative. On excluding them, elevation has a positive coefficient. (See Appendix F for
graph).
126 S. Mukherjee / World Development 111 (2018) 113–129

Table 4B
IV-2SLS estimates of impact of colonial institutions on maoist rebellion.
Dependent Variable: Maoist Control (Measured as % of Highly Sensitive Polling Stations in each Constituency).

Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Princely State 12.15** 11.14** 11.75** 11.28* 12.44*** 12.70**
(0.024) (0.040) (0.041) (0.051) (0.009) (0.018)
Forest Cover 0.130 0.0912 0.105 0.116 0.116 0.108
(0.133) (0.318) (0.165) (0.167) (0.162) (0.181)
Elevation
0.0310
0.0287
0.0290
0.0284
0.0317
0.0264
(0.134) (0.138) (0.161) (0.157) (0.115) (0.191)
Popln Density
0.000337 0.00780 0.00226 0.00774* 0.00276 0.00448
(0.947) (0.272) (0.555) (0.063) (0.439) (0.148)
Rural % 0.0673 0.0809 0.0568 0.0795* 0.0687 0.0850*
(0.121) (0.125) (0.255) (0.077) (0.132) (0.067)
PerCapita Income 0.000178 0.000417*
(0.423) (0.075)
Male Unemployment 0.841** 0.825** 0.776* 0.805* 0.881** 0.558
(0.049) (0.042) (0.075) (0.054) (0.041) (0.129)
Scheduled Caste%
0.431**
0.454**
0.449**
0.557**
0.491**
0.741***
(0.023) (0.014) (0.030) (0.011) (0.020) (0.005)
Scheduled Tribe% 0.462*** 0.494*** 0.393*** 0.409*** 0.461*** 0.314**
(0.001) (0.000) (0.009) (0.003) (0.001) (0.014)
Electricity 26.33** 29.59*** 24.12** 25.24** 25.96*** 25.06***
(0.012) (0.007) (0.010) (0.020) (0.008) (0.008)
ST% * Electricity
1.032***
1.048***
0.960***
0.968***
1.016***
1.002***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Male Literacy
0.479*
0.499**
0.489*
0.529**
0.501*
0.478*
(0.059) (0.049) (0.064) (0.049) (0.083) (0.066)
Paved Road 7.420
(0.214)
Post Office
24.42
(0.239)
Distance from Town 0.0684
(0.488)
Railway Station
171.6*
(0.056)
VoterTurnout 1985 0.0587
(0.684)
ENPV 1998 6.364***
(0.009)
Constant
16.41
11.37
8.404
7.016
16.69
11.28
(0.702) (0.780) (0.846) (0.866) (0.685) (0.779)
Observations 90 90 90 90 90 90
R2 0.653 0.657 0.655 0.655 0.652 0.687
Adjusted R2 0.599 0.604 0.601 0.602 0.592 0.633
Endogeneity & Overidentification Test Diagnosis:
y y y y y y
Hansen’s J statistic
y y y y y y
Hansen’s J (p value)
Anderson-RubinWald test 2.31 2.05 2.03 1.85 2.79 2.35
Anderson-Rubin (p-val) (0.1484) (0.1713) (0.1734) (0.1923) (0.1141) (0.1450)

p-values in parentheses.
Robust Std. Errors adjusted for 17 clusters by district.
*
p < 0.10, **p < 0.05, ***p < 0.01.
y
Hansen J Statistic (overidentification test of all instruments) is 0.000, since equation is exactly identified because the number of endogenous regressors equals the number
of instruments.

insurgency, and this show a common finding in all studies on the reduces Maoist insurgency by providing employment in rural areas
Maoist insurgency, and suggests support for the Cederman, (Gawande et al., 2017; Hoelscher et al., 2012).
Wimmer, and Min (2010) mechanism of ethnic grievances. Sched-
uled Caste (SC) % has a negative and significant coefficient, because
6. Robustness checks
the lower castes in Chhattisgarh live in the central districts where
there is no insurgency.55
I do various robustness checks, and colonial indirect rule mea-
Another interesting result is that Male unemployment has a pos-
sured by princely state continues to remain statistically significant.
itive and significant effect in the longer models, which could be
I briefly mention the tests here, please see Online Appendix G for
support for both opportunity as well as grievance mechanisms,
detailed description.
since unemployment may create lower opportunity costs to joining
rebellion, and also may generate grievance among tribal unem-
ployed males.56 This supports the finding that NREGS program 6.1. Clustering by colonial era boundaries of princely states and British
districts

55
There can be unobservable similarities for current day electoral
In other states of India, like Bihar, the Maoists recruit mainly from the lower
castes, thus indicating multiple causal pathways to insurgency in India.
constituencies which were historically part of the same princely
56
Sambanis (2004) points out how it is better to measure rebel opportunity costs at state, or same British direct rule era district. To address this, I clus-
the sub-national level using fine grained data on male unemployment. ter by colonial era princely state and British district boundaries,
 S. Mukherjee / World Development 111 (2018) 113–129 127

Table 4C
IV-2SLS estimates of impact of colonial institutions on Maoist rebellion.
Dependent Variable: Maoist Control (Measured as % of Highly Sensitive Polling Stations in each Constituency).

Pre colonial controls Post colonial controls

Model 1A Model 1B Model 1C Model 1D Model 2A Model 2B
Princely State 22.06*** (0.000) 13.92*** (0.000) 20.81** (0.029) 15.13*** (0.009) 16.47*** (0.003) 11.79*** (0.003)
Forest Cover 0.256* (0.055) 0.275* (0.094) 0.290* (0.062) 0.118 (0.212)
Elevation 0.00737 (0.751)
0.0109 (0.680)
0.00716 (0.791)
0.0303 (0.162)
Popln Density 0.000179 (0.969) 0.00233 (0.526)
Rural % 0.0267 (0.566) 0.0686* (0.096)
PerCapita Income
0.000324 (0.444) 0.000172 (0.409)
Male Unemployment 0.453* (0.078) 0.846** (0.032)
Scheduled Caste%
0.505** (0.048)
Scheduled Tribe% 0.450*** (0.009)
Electricity 26.95** (0.013)
ST% * Electricity
1.030*** (0.000)
Male Literacy
0.482* (0.067)
Constant 1.341** (0.029)
3.105 (0.183)
1.002 (0.892) 0.00386 (1.000)
30.34 (0.148)
13.42 (0.745)
Observations 90 90 90 90 90 90
R2 0.293 0.429 0.299 0.433 0.441 0.651
Adjusted R2 0.285 0.416 0.283 0.413 0.393 0.597

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
Robust Std. Errors adjusted for 16 clusters by colonial boundaries of princely states and British districts.

Table 4D
IV2SLS (spatial lagged) estimates of colonial institutions on Maoist rebellion.
Dependent Variable: Maoist Control (Measured as % of Highly Sensitive Polling Stations in Each Constituency).

Pre colonial controls Post colonial controls

Model 1A Model 1B Model 1C Model 1D Model 2A Model 2B
Princely State 12.43* (0.081) 10.27* (0.094) 8.331 (0.133) 8.519* (0.094) 7.814* (0.061) 5.882** (0.048)
Forest Cover 0.156** (0.011) 0.148*** (0.007) 0.150*** (0.003) 0.104** (0.020)
Elevation
0.0108 (0.339)
0.0196* (0.083)
0.0136 (0.139)
0.0154* (0.081)
Popln Density 0.00541 (0.352) 0.00566 (0.309)
Rural % 0.0637 (0.283) 0.0814 (0.161)
PerCapita Income
0.000248 (0.554)
0.000107 (0.788)
Male Unemployment 0.578* (0.059) 0.674** (0.027)
Scheduled Caste% 0.0234 (0.918)
Scheduled Tribe% 0.181* (0.076)
Electricity 5.097 (0.561)
ST% * Electricity
0.427* (0.068)
Male Literacy 0.193 (0.124)
Constant
0.0343 (0.992)
2.802 (0.338) 2.331 (0.531) 2.774 (0.454)
40.74* (0.050)
65.79*** (0.007)
Lambda
Constant 0.519 (0.163) 0.404 (0.188) 0.857*** (0.000) 0.701*** (0.001) 0.864*** (0.000) 1.019*** (0.000)
rho
Constant 0.505 (0.217) 0.484 (0.116) 0.191 (0.633) 0.233 (0.521)
0.105 (0.764)
0.532 (0.119)
Observations 90 90 90 90 90 90

p-values in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.
No robust standard errors since spatial ivreg does not allow that option. Spatial Weights matrix used to calculate spatial lag of Maoist control.

and present the IV2SLS results in Table 4C below which show prin-
cely state to be strongly significant.
6.2. Spatial autocorrelation – contagion from neighboring
constituencies
To control for spatial diffusion or contagion of insurgency from
neighboring constituencies, I test for spatial autocorrelation by cre-
ating a spatial weight matrix for Chhattisgarh, and including a spa-
tial lag of Maoist control in the model (Lane (2016: 397), Gleditsch
(2007: 302), Salehyan and Gleditsch (2006). The results in Table 4D
show some evidence for spatial autocorrelation, but princely state
is still significant.57
6.3. Controlling for alternate historical factors
First, to control for effects of pre-colonial conflict on current
conflict (Besley & Reynal-Querol, 2014), I develop and control for
a new measure of pre-colonial agrarian rebellion for the Mughal per-
iod (1556–1707), see Online Appendix G, Table 5. Second, I control
for previous organizational legacies (Jha & Wilkinson, 2012;
Zukerman-Daly, 2012) that could facilitate future conflict, by using
an alternate measure of leftist strength. Finally, I control for prior
subaltern ethnic party success from 1980 elections (Chandra &
Garcia-Ponce, 2014).
6.4. Other measures of dependent variable of Maoist rebel control

57
See Drukker et al. (2013a, 2013b) for the STATA code on how to create spatial I check the robustness of the regression estimates for other
weight matrices and include spatial lag in instrumental variable regression. measures of Maoist control from alternate sources like Ministry of
128 S. Mukherjee / World Development 111 (2018) 113–129
Home Affairs (MHA), police strength in constituencies during elec-
tions, and perception of an Inspector General of Police collected
during field work in Chhattisgarh. See Tables 6A and 6B in Appen-
dix G and Figs. 3A–3F.
6.5. Relationship between Maoist control and Maoist violence
I check whether the relationship between Maoist violence and
Maoist control is non-linear as suggested by Kalyvas (2006). The
results in Table 7 and Fig. 4 in Appendix G show a non-linear rela-
tionship between Maoist control and Maoist violence.
7. External validity of this study
7.1. Broader testing on all India dataset show similar results
As discussed earlier, using an IV2SLS analysis based on a new
instrument on European Wars in an all India dataset, I find that dis-
tricts which had these different forms of indirect rule have a higher
probability of Maoist control (Mukherjee, 2018). This shows
broader generalizability of this argument within India. Thinking
beyond the Maoist insurgency, the other cases of insurgency in
India, like Kashmir and the North Eastern states, also were differ-
ent forms of indirect rule, which future studies should analyze.
Within India, Jafa (n.d.) suggests, ‘‘A number of areas which the Bri-
tish ruled indirectly through native princes or tribal chiefs upto
1947 . . . have been in the forefront of dissent or open armed defi-
ance . . . The Nagas and the Mizos were ruled through the tribal
chieftains and headmen, Manipur and Tripura through their
respective Maharajas. Jammu and Kashmir was ruled indirectly
through the Maharaja. . . .” So there is generalizability of this theory
to the several cases of insurgency in India.
Beyond India: There is potential generalizability of this causal
mechanism of indirect rule, weak state capacity, and ethnic grie-
vances evident in Chhattisgarh to other similar cases in South Asia
like tribal grievances in the Taliban insurgency in the FATA area of
indirect rule in NWFP in Pakistan (Naseemullah, 2014), the Baluchi
insurgency in princely states of Kalat in Pakistan (Khan, 2015), the
various ethnic secessionist insurgencies in ‘frontier areas’ of indi-
rect rule in Burma (Callahan, 2003).58 Outside South Asia, British
indirect rule allowed the Islamic caliphate and sharia law to con-
tinue in the north of Nigeria, thus creating conditions for Islamic
movements and more recently the emergence of the Boko Haram
insurgency in Nigeria’s north (Sampson, 2014: 312-15). This micro-
level analysis of Chhattisgarh sets the agenda for further research on
how colonial institutions create conditions for persistence of conflict
in other cases.
8. Conclusion: broader contributions and policy implications
Historical institutions are potential omitted variables that need
to be taken seriously by studies of civil war and Maoist insurgency
in India. This article exploits sub-national variation within the
Indian state of Chhattisgarh, to show the historical origins of
sub-national variation in state capacity, and tribal grievances,
which were used by Maoist rebels to mobilize successfully. By
focusing on the causal effects of colonial institutions, and using
an instrument for princely state based on the random death of an
Indian king, the paper addresses the issue of endogeneity of
socio-economic and state capacity measures to ongoing conflict.
While emerging econometric studies of the Maoist insurgency in
India have disaggregated to the district level in India, this paper
58
I describe two shadow cases of the Taliban insurgency in Pakistan and the Maoist
insurgency in Nepal in Appendix J (supplementary files).
creates a new assembly constituency (AC) level dataset for
Chhattisgarh.
There are policy implications of my study of Maoist insurgency
in Chhattisgarh, where Maoist violence escalated in 2005 due to
the rise of the vigilante Salwa Judum movement. Since the British
policy of indirect rule was continued by the post-independence
government of Madhya Pradesh (of which Chhattisgarh was a part
till 2000) into the 1980s, it set up the conditions of weak state
capacity, and required the BJP government in Chhattisgarh to indi-
rectly depend on vigilante movement to tackle the Maoists, since
they did not have an efficient police force (Miklian, 2009). Also,
while officially the landlord and princes were abolished following
independence, the landlords and landed tribal elites continued to
have power within society and politics (Mahoney, 2000), and it
was a landed elite named Mahendra Karma, who started the Salwa
Judum vigilante movement to counter the Maoist threat of dis-
tributing land to tribals. Avoiding such policies of indirect rule
and exploitation would have possibly prevented successful Maoist
mobilization, and also not led to the rise of a violent movement like
the Salwa Judum in former princely state areas of Chhattisgarh.
Acknowledgements
I acknowledge financial support for research on this topic from
the USIP dissertation research grant 2008, the NSF doctoral disser-
tation grant 2011-12 while a Ph.D. student at Yale university, and
the Connaught New Scholar Grant 2016-17, at the University of
Toronto.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at https://doi.org/10.1016/j.worlddev.2018.06.
013.
References
Acemoglu, D., Johnson, S., & Robinson, J. (2001). The colonial origins of comparative
development: An empirical investigation. American Economic Review, 91(5),
1369–1401.
Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricist’s
companion. Princeton: Princeton University Press.
Banerjee, A., & Iyer, L. (2005). History, institutions and economic performance: The
legacy of colonial land tenure systems in India. American Economic Review, 95
(4), 1190–1213.
Banerjee, A., & Somanathan, R. (2007). The political economy of public goods: Some
evidence from India. Journal of Development Economics, 82.
Baruah, S. (2005). Durable disorder: Understanding the politics of northeast India. New
Delhi: Oxford University Press.
Baum, C. F., Schaffer, M. E., & Stillman, S. (2007). Enhanced routines for instrumental
variables: Generalized method of moments estimations and testing. The Stata
Journal, 7(4), 465–506.
Besley, T., & Reynal-Querol, M. (2014). The legacy of historical conflict: Evidence
from Africa. American Political Science Review, 108(2), 319–336.
Bhatia, B. (2005). The naxalite movement in central Bihar. Economic and Political
Weekly, 1536–1549.
Boone, C. (1994). States and ruling classes in postcolonial Africa: The enduring
contradictions of power. In Migdal, Kohli and Shue (Ed.), State power and social
forces: Domination and transformation in the third world, pp. 117–118.
Boone, C. (2013). Property and political order in Africa: Land rights and the structure of
politics. New York: Cambridge University Press.
Boone, C. (2017). Sons of the soil conflict in Africa: Institutional determinants of
ethnic conflict over land. World Development, 96, 276–293.
Bound, J., Jaeger, D., & Baker, R. (1995). Problem with instrumental variables
estimation when the correlation between the instruments and the endogenous
explanatory variable is weak. Journal of the American Statistical Association, 90
(430), 443–450.
Callahan, Mary (2003). Making Enemies: War and State Building in Burma. Ithaca,
N.Y.;London: Cornell University Press.
Cederman, L.-E., Wimmer, A., & Min, B. (2010). Why do ethnic groups rebel? New
data and analysis. World Politics, 62(1), 87–119.
Chandra, K. (2004). Why ethnic parties succeed: patronage and ethnic headcounts in
India. Cambridge, UK; New York: Cambridge University Press.
 S. Mukherjee / World Development 111 (2018) 113–129
Chandra, K., & Garcia-Ponce, O. (2014). Why subaltern-led parties crowd out armed
violence: Explaining maoist violence in India. Paper presented at workshop on
India’s Maoist insurgency. Princeton University.
Dasgupta, A., Gawande, K., & Kapur, D. (2017). (When) do anti-poverty programs
reduce violence? India’s rural employment guarantee and Maoist conflict.
International Organization, 71(3), 605–632. https://doi.org/10.1017/
S0020818317000236.
Dell, M. (2010). The persistent effects of Peru’s mining mita. Econometrica, 78,
1863–1903.
Dirks, Nicholas (2001). Castes of the Mind: Colonialism and the Making of Modern
India. Princeton: Princeton University Press.
Drukker, D., Peng, H., Prucha, I., & Raciborski, R. (2013). Creating and managing
spatial-weighting matrices with the spmat command. Stata Journal, 13(2),
242–286.
Drukker, D., Prucha, I., & Raciborski, R. (2013). A command for estimating spatial-
autoregressive models with spatial-autoregressive disturbances and additional
endogenous variables. Stata Journal, 13(2), 287–301.
Fearon, J., & Laitin, D. (2003). Ethnicity, insurgency, and civil war. American Political
Science Review, 97(1), 75–90.
Fisher, M. H. (1991). Indirect rule in India. Residents and the residency system, 1764-
1858. Delhi: Oxford University Press.
Foa, R. (2016). Ancient polities, modern states Dissertation. Harvard University.
Gawande, Kishore, Kapur, Devesh, & Satyanath, Shanker (2017). ‘‘Renewable
Resource Shocks and Conflict Intensity: Findings from India’s Ongoing Maoist
Insurgency”. Journal of Conflict Resolution, 61(1), 140–172. https://doi.org/
10.1177/0022002714567949.
Gerring, J. (2007). Is there a (Viable) crucial-case method? Comparative Political
Studies, 40(3), 231–253.
Gerring, John, Daniel, Ziblatt, Van Gorp, Johan, & Julian, Arevalo (2011). An
Institutional Theory of Direct and Indirect Rule. World Politics, 63(3), 377–433.
Gleditsch, K. (2007). Transnational dimensions of civil war. Journal of Peace Research,
44(3), 293–309.
Gomes, J. F. (2015). The political economy of the maoist conflict in India: An
empirical analysis. World Development, 68, 96–123.
Guardado, J. (2016). ‘‘Forced labor, ethnicity and conflict: Evidence from the Peruvian
Mita”, Manuscript, presented at OSU conference on colonialism, June 2016.
Gujarati, D. (2004). Basic econometrics (4th ed.). McGraw Hill.
Hegre, H., & Sambanis, N. (2006). Sensitivity analysis of empirical results on civil
war onset. Journal of Conflict Resolution, 50(4), 508–535.
Hoelscher, K., Miklian, J., & Vadlamannati, K. (2012). Hearts and mines: A district-
level analysis of the Maoist conflict in India. International Area Studies Review, 15
(2), 141–160.
Iyer, L. (2010). Direct versus indirect colonial rule in India: Long term consequences.
The Review of Economics and Statistics., 92(4), 693–713.
Jafa, V. S. Unknown date. Administrative policies and ethnic disintegration
engineering conflict in India’s north east. Faultlines, volume 2. Downloaded
on August 3, 2015 from http://www.satp.org/satporgtp/publication/faultlines/
volume2/Fault2-JafaF.htm.
Jha, S., & Wilkinson, S. (2012). Does combat experience foster organizational skill?
Evidence from ethnic cleansing during the partition of South Asia. American
Political Science Review, 106, 883–907.
Kalyvas, S. (2006). The logic of violence in civil war. Cambridge; New York:
Cambridge University Press.
Kalyvas, S., & Kocher, M. (2009). The dynamics of violence in Vietnam: An analysis
of the hamlet evaluation system (HES). Journal of Peace Research, 46(3),
335–355.
Kennedy, J., & Purushotham, S. (2012). Beyond naxalbari: A comparative analysis of
maoist insurgency and counterinsurgency in independent India. Comparative
Studies in Society and History, 54(4), 832–862.
Khan, Adeel. (2015). ‘‘Renewed Ethnonationalist Insurgency in Balochistan,
Pakistan”, in Jugdeep Chima, (Ed.), Ethnic Subnationalist Insurgencies in South
Asia, (pp. 125–29).
Kohli, A. (2004). State directed development: Political power and industrialization in
the global periphery. New York: Cambridge University Press.
Lane, M. (2016). The intrastate contagion of ethnic civil War. The Journal of Politics,
78(2), 396–410.
Lange, M. (2009). Lineages of despotism and development: British colonialism and state
power. Chicago: University of Chicago Press.
Lange, M., & Dawson, A. (2009). Dividing and ruling the world? A statistical test of
the effects of colonialism on postcolonial civil violence. Social Forces, 88(2),
785–818.
Lee-Warner, S. W. (1910). The native states of India. London: Macmillan and Co., Ltd.
Lee, Alexander (2017). Redistributive colonialism: The long term legacy of
international conflict in India. Politics and Society, 45(2), 173–224.
Lieberman, E. (2005). Nested analysis as a mixed-method strategy for comparative
research. American Political Science Review, 99(3), 435–452.
Mahoney, J. (2000). Path dependence in historical sociology. Theory and Society, 29.
Mahoney, J. (2010). Colonialism and postcolonial development: Spanish America in
comparative perspective. New York: Cambridge University Press.
Mamdani, M. (1996). Citizen and subject. Princeton: Princeton University Press.
129
Miguel, Edward, Satyanath, Shanker, & Sergenti, Ernest (2004). Economic Shocks
and CivilConflict: An Instrumental Variables Approach. Journal of Political
Economy, 112(4), 725–753.
Miklian, J. (2009). The purification hunt: The Salwa Judum counterinsurgency in
Chhattisgarh, India. Dialectical Anthropology, 33, 441–459.
Mitchell, E. (1968). Inequality and insurgency: A statistical study of South Vietnam.
World Politics, 20(3), 421–438.
Mukherjee, S. (2018). Colonial origins of Maoist insurgency in India: Historical
institutions and civil war. Journal of Conflict Resolution, published online
September, 2017 at http://journals.sagepub.com/doi/full/10.1177/
0022002717727818
Naseemullah, A. (2014). Shades of sovereignty: Explaining political order and
disorder in Pakistan’s northwest. Studies in Comparative International
Development, 49, 501–522.
Ramusack, Barbara N. (2004). The Indian Princes and Their States. The New
Cambridge History of India. III.6. Cambridge University Press.
Robinson, J. (February 2013). Colombia: Another 100 years of solitude? Current
History.
Salehyan, I., & Gleditsch, K. (2006). Refugees and the spread of civil war.
International Organization, 60(2), 335–366.
Sambanis, N. (2004). Using case studies to expand economic models of civil war.
Perspectives on Politics, 2, 259–279.
Sampson, I. T. (2014). Religion and the Nigerian state: Situating the de facto and de
jure frontiers of state-religion relations and its implications for national
security. Oxford Journal of Law and Religion, 3(2), 311–339.
Sartori, Giovanni (1970). Concept Misformation in Comparative Politics. The
American Political Science Review, 64(4), 1033–1053.
Scott, J. (2009). The art of not being governed: An anarchist history of upland Southeast
Asia (p. 2009). New Haven: Yale University Press.
Sinha, S. (1989). Maoists in Andhra Pradesh. New Delhi: Gian Publishing House.
Stock, J. H., Wright, J. H., & Yogo, M. (2002). A survey of weak instruments and weak
identification in generalized method of moments. Journal of Business and
Economic Statistics, 20, 518–529.
Sundar, N. (2007). Subalterns and sovereigns: An anthropological history of Bastar
(1854–2006). Delhi: Oxford University Press.
Teitelbaum, E., & Verghese, A. (2016). Colonialism and armed conflict in the Indian
countryside. Manuscript.
Thachil, T. (2011). Embedded mobilization: Nonstate service provision as electoral
strategy in India. World Politics, 63(3), 434–469.
Tudor, Maya., (2013). ‘‘Explaining Democracy’s Origins: Lessons from South Asia”,
Comparative Politics.
Vanden Eynde, O. (2017). Targets of violence: Evidence from India’s Naxalite
conflict. Manuscript. The Economic Journal, 128(609). https://doi.org/10.1111/
ecoj.12438. Forthcoming in.
Verghese, A. (2016a). British rule and tribal revolts in India: The curious case of
Bastar. Modern Asian Studies., 50(5).
Verghese, A. (2016b). The colonial origins of ethnic violence. Stanford University
Press.
Wilkinson, S. (2017). Looking Back at the Colonial Origins of Communal and Caste
Conflict in India. The Wire, at https://thewire.in/110535/communal-violence-
caste-colonialism/.
Wucherpfennig, J., Hunziker, P., & Cederman, L.-E. (2015). Who inherits the state?
Colonial rule and post-colonial conflict. Forthcoming in American Journal of
Political Science.
Zukerman-Daly, S. (2012). Organizational legacies of violence: Conditions favoring
insurgency onset in Colombia, 1964–1984. Journal of Peace Research, 49(3),
473–491.
Government Documents/Reports:
Baden-Powell, B. H. (1892). The land-systems of British India, Volumes 1–3. Oxford:
Clarendon Press.
The imperial gazetteer of India, Volumes 1–26. 1908. Oxford: Clarendon Press.
Chhattisgarh state assembly election – 2003. Chief Electoral Officer, Government of
Chhattisgarh.
Census of India, 1951, Volume VII, Madhya Pradesh, Part 1-A Report, Nagpur, Govt.
Printing, 1953.
Census of India, 2001, Chhattisgarh Administrative Atlas, published by Directorate
of Census Operations, Madhya Pradesh and Chhattisgarh, p. 15, Map 7:
‘Chhattisgarh Administrative Divisions 2001’.
Maoist and Human Rights Documents:
Dandakaranya Special Zonal Committee, CPI-Maoist. 12 Aug 2005. Jan Jagran Nahi,
Jan Daman Abhiyan.
People’s Union of Civil Liberties (PUCL). July 1989. Bastar: Development and
Democracy, Madhya Pradesh.
Shankar, P. (2006). Yeh Jangal Hamara Hain. New Vitas Publications.