The Political Economy of the Maoist Conflict in India: An
Empirical Analysis
Preliminary Draft
March 2011
Joseph Flavian Gomes
Department of Economics
Universidad Carlos III de Madrid
Office: 11.2.32
Phone: 91 624 9320
Email: jgomes@eco.uc3m.es
Abstract
Papers using cross country aggregate data abound in the literature on civil conflicts. However, such cross country studies have some inherent limitations: data might not be comparable across countries; reasons for conflict might vary from country to country; conflicts are
often localized and depend on unequal spatial distribution of resources within the country.
Using aggregate data is also problematic, because conflicts might be due to individual-level
or group-level characteristics. This paper contributes to a small but burgeoning literature
that uses sub-national micro data to identify the causes of civil conflicts. In particular, we
study the Maoist/Naxalite conflict in India by constructing a comprehensive district level
database using conflict data from four different terrorism databases and combining to it socioeconomic and geography data from myriad sources. In addition to exploiting the within
country regional heterogeneity, we use the micro structure of the data to construct grouplevel characteristics. Using data on 362 districts for 3 time periods, we find evidence on how
land inequality is extremely important for the Maoist conflict. We also show how historical
property rights institutions from colonial times that go back centuries can affect present
day conflict outcomes through their impact on economic outcomes, social relations and the
political environment in the district. Finally, making use of the micro structure of the data
we are able to ask whether exclusion of the low castes and tribal people from the growth
story of India is important. At best we get mixed evidence on this. While incomes of lower
castes and tribal communities impact the number of Maoist incidents, they do not affect the
probability of conflict.
1
1
Introduction
In recent years the relation between economic performance and civil conflicts has generated a
considerable amount of interest among economists. Not surprisingly within the span of a few
years a lot has been written on the subject 1 . This paper contributes to two different strands
of the literature. The first strand is the research using sub-national-micro data exploiting the
spatial heterogeneity within a country and the micro characteristics of the data to pin down the
causes of civil conflict (in contrast to the literature that uses cross country regressions for the
same). The second strand falls in a broad class of literature that traces divergences in current
economic outcomes to differences in historical institutions in a country 2 . The focus is on one
specific country - India and one specific conflict - The Maoist conflict (aka the Naxalite conflict).
We use district level conflict data for the period 1979-2009 along with socio-economic and geography data from multiple NSS sample surveys, Censuses etc. to build a district level dataset.
Finally using this dataset on 362 districts (for the 16 main states) for three time periods this
paper provides evidence on how land inequality is one of the key determinants of the Maoist
conflict controlling for all other factors that the literature has found to be important. Moreover,
there is evidence on how class antagonism driven by land institutions that have lingered for
centuries also has a significant impact on the conflict.
The Maoist conflict in India has existed for more than 40 years (since 1967). However, it has
seen a terrifying increase in proportions only in the last decade 3 . In the period 2004-2010 there
have been more than 5000 lives lost (by official estimates). Including the number of wounded
and displaced the figure would be many times higher. In fact, it has been identified as ”the
single biggest security challenge to the Indian state” by Dr. Manmohan Singh, the Prime Minister of India. Indeed such a conflict if not tackled on a timely and efficient manner, could have
enormous negative socio-economic consequences. Such violence often leads to the destruction
of existing infrastructure and discourages investment, apart from the loss in human capital. A
good part of the civil conflict literature finds underdevelopment as a primary determinant of civil
conflict. If the conflict indeed is mostly located in the more impoverished regions of the country,
the existence of such a conflict would give a further adverse shock to the economy of the region.
Moreover, if the conflict intensity rises unabated (like it has in the past few years) for too long a
period, then it could eventually negatively affect the growth that India is currently experiencing.
And this would be a consequence over and above the tremendous direct human suffering that
such conflicts inevitably cause. All this makes the Maoist conflict a serious issue and calls for
an in-depth analysis identifying its causes and suggesting potential policy interventions. We
believe that there is no work (to the best of our knowledge) in the existing literature that does
an in-depth empirical analysis of the Maoist conflict in India using a district level panel. 4 . In
fact, even some of the studies that look specifically on India have focussed more on the several
communal riots between the Hindus and the Muslims that the country has experienced over the
1
See Blattman and Miguel (2010) for a recent survey of the existing literature.
2
A very novel concept in the conflict literature: Jha (2008) for example, associates the Hindu-Muslim conflicts
in India to historical institutions.
3
see figure in appendix
4
Barooah (2008) studies the Naxalite conflict, but that study relies on a simple cross section analysis
2
years. Fortunately, such conflicts have seen a general decline in India in the recent years. 5 This
makes the analysis of the Maoist conflict even more interesting since one would not be tempted
to wrongly conclude that perhaps the law and order situation in the country has deteriorated
leading to more violence and conflict.
The existing literature has witnessed several different approaches to empirically identify the
causes of civil conflicts. There are two clear directions in which this literature needs progress.
The first direction is using sub-national micro data in order to overcome the short comings of the
cross country analyses. Cross country studies on civil conflicts abound in the literature. Collier
and Hoeffler (2004), Fearon and Laitin (2003) for example, do cross-country regressions trying
to identify the variables that impact conflict probabilities. Miguel et al. (2004) and Ciccone
(2010) on the other hand, look at variation in conflict intensity due to economic shocks instrumented by weather shocks. However, such cross country studies usually suffer from some serious
shortcomings. Do and Iyer (2009) point out two caveats that apply to the use of cross country
data: (1) Data might not be comparable across countries. (2) Reasons for the conflict might
vary from country to country. 6 In fact, cross country studies do not allow us to control for
all the factors that are constant within the country viz. the macroeconomic variables and they
ignore the within country heterogeneity by treating the country as a unit of observation. This is
a serious shortcoming since conflicts are often localized and depend on the unequal spatial distribution of resources within the country. For example, if we look at the Maoist conflict in India,
in West Bengal which is one of the severely affected states (located in the east of the country),
the conflict is very pronounced in the Midnapore and Puruliya districts while it is completely
absent in districts like Howrah, North & South 24 Parganas. 7 This is the kind of heterogeneity
that one cannot take into account using even states as units of analysis. Thus, using countries
as the units of analysis could be pretty unsatisfactory. In addition, ”the country level of analysis
has [other] inherent limitations. Individual and group level conflict factors such as poverty and
ethnic hostility are imperfectly tested at the national level” (Blattman and Miguel (2010) from
Sambanis (2004)). Indeed, tribal and low caste participation in the Maoist conflict has been
much talked about, suggesting that perhaps some sections of society are losing behind in the
growth story of India leading to more tension. 8 Moreover, the sudden rise in the conflict levels
from 2005 onwards is even more surprising given that the country has been experiencing above
8% growth rates in the same period. The use of micro data could be particularly helpful in this
regard in order to identify if there is regional and group level heterogeneity in the growth rates
and development that might explain this rise. Thus, as Blattman and Miguel (2010) puts it ”...
the most promising avenue for new empirical research is on the sub-national scale, analysing
conflict causes, conduct, and consequences at the level of armed groups, communities and individuals”.
5
see Iyer (2009) for a study of all conflicts in South Asia
6
Further, even within a country focussing on one specific conflict might give us more interesting insights than
looking at overall violence.
7
Again, if one looks at Maharashtra (which is a state located in the West of the country), almost all the Maoist
incidents have been concentrated solely in the Gadchiroli district (formerly part of the Chandrapur district).
Again, while the conflict affects virtually all the districts of Bihar, it is almost completely absent in the states of
Rajasthan, Punjab and Haryana.
8
Chattisgarh for example, a predominatly tribal state is one of the hotbeds of the conflict.
3
Part of the lack of sub-national level studies has been of course due to the unavailability of
reliable sub-national data and/or the tedious task of building such datasets. However, recent
times have witnessed a small but burgeoning literature that seeks the causes of civil conflicts
using sub national micro data with some very interesting insights 9 . Do and Iyer (2009) for
example study the Maoist Conflict in Nepal using a district level dataset comprising economic,
social and geography variables and show how the lack of economic opportunities trigger such
conflicts while suitable geographic conditions help perpetuate it once the conflict has started.
However, ethnic and caste polarization, land inequality and political participation don’t affect
violence. Jha (2008) does a town level study of India focussing on the Hindu-Muslim communal
conflicts. His primary conclusion is that historical institutions determine peaceful coexistence or
conflict between Hindus and Muslims. Bohlken and Sergenti (2009) do a state level study of the
same Hindu-Muslim riots and find local economic conditions as a significant determinant of such
communal riots in India. Iyer (2009) looks at all terrorist incidents in the Indian subcontinent
and infers that lagging regions are more prone to conflicts. For India and Nepal she runs district
level regressions. She points out that on the whole terrorist incidents have been on the rise
over the years in the Indian subcontinent with the lagging regions more prone to it. Moreover,
making use of the district level regressions she shows that it is the lagging districts in India and
Nepal that are more prone to conflicts than others. Looking only at the aggregate country level
for example would have deprived us of this insight. Thus from the above interesting conclusions
it is evident how micro level studies are a clear advance over the cross country studies.
The other critical issue in this literature is establishing a causal relation between conflict levels
and its determinants. This is due to two main problems 10 (1) There might exist unmeasured
factors that affect both conflict intensity and pre conflict characteristics. (2) Districts that
are experiencing more violence might also be districts that have experienced high past conflict.
Recent studies have tried to address this issue using an Instrumental variable approach when
clearly exogenous instruments have been available. Miguel et al. (2004) and Ciccone (2010)
have used weather shocks. Dube and Vargas (2008) have used exogenous shocks to agricultural
and resource prices. When clear instruments have not been available authors have tried to use
data on covariates from the pre conflict levels (Do and Iyer (2009); Mitra and Ray (2010)), in
order to prevent endogeneity arising out of reverse causality. Also, all potential covariates are
controlled for in order to reduce endogeneity arising out of omitted variables. Following in the
same vein in this paper we (i) Use data from the pre conflict period; (ii) Control for all possible
variables (that the literature suggests are important) that might play a role subject to the data
availability (iii) Always control for the presence of past conflict (iv) also at least try to run some
IV regressions where possible with previous period income as IV for present income.
Apart from the above mentioned points, this paper has another important contribution. It traces
the origins of civil conflict to a new channel over and above the channels like underdevelopment
and geography already identified in the literature. We see how historical colonial institutions
from centuries ago can impact conflict outcomes in the present times. In this context Jha (2008)
has a very interesting insight in his paper. He does a town level study of India focussing on the
Hindu-Muslim communal conflicts. His primary conclusion is that historical institutions deter9
10
e.g. Dube and Vargas (2008), Do and Iyer (2009), Jha (2008), Bohlken and Sergenti (2009)
from Do and Iyer (2009)
4
mine peaceful coexistence or conflict between Hindus and Muslims. In the presence of a historical
complementarily between the two groups the probability of conflict is lower. Besley and Persson
(2008) on the other hand highlight the importance of political institutions in explaining conflict
outcomes. While their main conclusion is that higher market prices of exported and imported
primary commodities have a positively significant effect on the incidence of civil conflict, they
find that those effects are heterogeneous depending on the quality of political institutions in
the country 11 . There is indeed a huge amount of literature showing how historical Institutions
have a persistent effect on current economic performance. Acemoglu et al. (2001, 2002), show
how institutions developed centuries ago determine the development or underdevelopment of
countries in the present times. La Porta et al. (1998, 1999, 2000) also trace the differences
in the economic performance of a country to its colonial origin. Their conclusion is that the
legal system of a country is determined by whether the country was colonized by the British or
other colonial powers and this difference in legal systems affects the economic performance of
the country. Engerman and Sokoloff (1997, 2002) also trace the divergence in the growth rates
of Brazil and the USA to the history induced divergence in institutions. Following in the same
vein, Banerjee and Iyer (2005) show that the colonial land revenue institutions set up by the
British in India, lead to sustained differences in economic outcomes in the post independence
period. Areas in which proprietary rights in land were historically given to landlords have significantly lower agricultural investments and productivity in the post-independence period than
areas in which these rights were given to the cultivators. These areas also have significantly
lower investments in health and education. In fact, these land revenue institutions created some
social divisions that continued to undermine economic progress way into the future since, ”...
the memory real or imagined, of having been exploited, can create a divide that can continue
to hurt the economy many years into the future”. The allocation of land revenue collection
responsibility to the landlords (which translates into a de facto property right over the land)
gave birth to a reason for perpetual conflict between the peasants and landlords. ”Elsewhere,
the colonial state directly collected the land revenue from the cultivator, thereby avoiding this
particular source of internecine conflict” (Banerjee and Somanathan (2007)).
These historical institutions could affect the conflict both, through their effects on the economic
outcomes, or directly through their effects on the social relations among people. Economic underdevelopment has been associated with higher conflict in the literature. As Banerjee and Iyer
(2005) have shown, the landlord areas 12 are also the areas with worse economic outcomes and
thus they could be experiencing more conflict via the underdevelopment channel. Moreover,
landlords areas might still have a more unequal land distribution which might potentially incite
social tensions. 13 ”However, districts with worse land distribution historically have also seen
more land reforms in the post independence period. This makes it unlikely that the persistence
of the landlord effect is mainly through its effect on contemporaneous land distribution.” ”However, as late as 1990, 64% of all landholdings in the landlord areas were classified as ”marginal”
(less than 1 hectare) which is about 1 % point higher than the NL [non-landlord] districts. Further 48 % of all holdings are small to midsized (1-10 hectares) in individual based areas, but
11
i.e. whether the country is a parliamentary democracy, or has a system of strong checks and balances
12
from now on we will refer to the areas where historically the revenue collection rights were given to landlords
as the landlord areas
13
see figures in appendix for differences in land inequality between Maoist and non Maoist districts.
5
only 35% in landlord areas. However, there is no significant difference in proportion of extremely
large holdings, probably due to land ceiling laws passed post independence.” Banerjee and Iyer
(2005)
The historical institutions could also have a direct impact on the conflict. ”The class based
antagonism that [the land revenue institutions] created within the communities in these areas
has persisted well into the post independence period” Banerjee and Iyer (2005). Thus, in effect
the class antagonism might have survived even after land reforms might have reduced land inequality. ”Areas most associated with Maoist uprisings are WB, Bihar & Srikakulam district of
AP- all landlord areas. Paul R. Brass (1994 pg. 326-327) argues explicitly that these peasant
movement have their roots in the history of exploitation and oppression of peasants by landlords.” (Banerjee and Iyer (2005))
Thus, there are reasons to believe that the Land institutions have an effect on the conflict over
and above the effect through the contemporaneous land inequality and underdevelopment.In this
paper we thus test for the direct effect of the Landlord institutions on the Conflict outcomes. If
we do not include the direct effect of the Land Institutions variable, we will have problems due
to the omitted variable since as we will see later the Land Institutions variable indeed comes
out to be significant.
Thus, to summarize, this paper makes the following contributions to the existing literature. (1)
It adds to a small but burgeoning literature that uses regional level micro data to study civil
conflict, which is a clear progress over existing cross country literature. It is also one of the very
first attempts to put together district level data from a variety of sources in order to analyze
civil conflicts rigorously. Moreover, exploiting the micro characteristics of the data we are able
to ask some interesting questions about whether exclusion from growth of certain sections has
an impact on the conflict. (2) It is the first to analyze rigorously the Maoist conflict in India
which has turned out to be a very serious problem. Moreover, India is a really interesting case
since it is a fast growing economy and the success of its economic ambitions determine the fate
of one third of the world’s poor that it is home to. (3) It shows that in addition economic
underdevelopment, how Land relations and historical institutions within a country could lead
to conflicts.
In the next section we add a small discussion on the Maoist Conflict in India, in section 3 we
list the main hypotheses of the study. Section 4 discusses the data, section 5 the econometric
results and section 6 concludes.
2
The Maoist/Naxalite Conflict
Overall there are five major sources of conflicts/violence in India
14 :
• Maoist/Naxalite/left wing extremism ,
• Hindu- Muslim communal conflicts,
• Separatist movements in the North Eastern states,
• Islamic fundamentalist terrorism,
14
see Iyer (2009) for a discussion on the different sources of violence in India and the rest of South Asia
6
• The Kashmir conflict.
Only the first three of the above five can be termed as ”Civil” conflict in the true sense of the
term. Pure acts of terrorism like the Mumbai terrorist attacks originating from Islamic fundamentalist groups often coming from across the border is now an international problem. The
Kashmir problem has been more of a territorial dispute. Thus, both these issues would require
a very different analysis from that of conflicts or unrests arising from within the country. The
Separatist movements in the North-East are important and interesting but they also require
a separate analysis since they are more secessionist in nature (unlike the Maoist conflict). 15
Moreover, over the last decades progress has been made in solving some of the disputes in the
North-East and violence has reduced significantly. The Hindu-Muslim communal conflicts have
already been studied by various authors in considerable detail and have also been on a declining
trend over the years 16 . The Maoist conflict has on the other hand seen a huge increase in
intensity over the last decade and is the focus of this paper.
The start of the conflict is marked by a peasant uprising in the year 1967 in Naxalbari (whence
the term Naxalite), a small village in West Bengal. ”A tribal youth having obtained a judicial
order went to plough his land on 2 March 1967. The local landlords attacked him with the help
of their goons. Tribal [peasants] of that area retaliated and started forcefully recapturing their
land” (Kujur (2008)). The rebellion left nine tribal people and one police personnel dead and the
Naxalite movement in India was born. The rebellion itself was contained by government forces
within 72 days with the use of force, but had already gathered huge visibility from Communist
revolutionaries from across the country. After West Bengal the movement spread to the state
of Andhra Pradesh where the formation of the People’s War Group (PWG) in 1980 marks the
revival of the movement post the Naxalbari uprising. It has since then spread across various
states in India including Bihar, Jharkhand, Madhya Pradesh, Orissa, Chhattisgarh, Maharashtra and Karnataka across many districts and has existed in varying degrees across the country
17 . However, it was the 2004 merger of the PWG with the Maoist Communist Center (MCC)
that lead to the formation of the Communist Party of India-Maoist (CPI-Maoist) that marks the
modern revival of the movement and followed a huge rise in insurgency and violence thereafter.
While the term ”Naxalite” comes from the place of birth of the movement the term ”Maoist” is
used due to the Maoist ideologies that many of these rebel groups adhere too. The CPI-Maoists
for example claim to be committed to a ”democratic revolution” through ”a protracted peoples war with the armed seizure of power remaining as its central and principal task” (SATP;
Iyer (2009)). While the conflict as a whole has been termed as a Maoist/Naxalite conflict, the
movement itself is hardly homogenous. It has in fact, always had a fragmented structure with
multiple groups operating without a centralized Movement organization. Different issues like
control of common property resources, better wages and housing, protection of land ownership
etc, have also been taken up on a timely basis in different places.
While in some small pockets of India like the Dantewada district in Chattisgarh the movement has reached a war like situation with constant terror and violence on a daily basis, the
entire movement has not always nor everywhere been violent. ”Common forms of nonviolent
15
Also, there is severe lack of good quality data for the North East India.
16
Bohlken and Sergenti (2009), Mitra and Ray (2010), Jha (2008) are some of the more recent studies
17
see map in appendix
7
action include sabha (meeting), bandh (closure), aarthik nakebandi (economic boycott), samajik
bahishkar (social boycott), jan adalat (people’s court), dharnas (e.g., the 14-day dharna organised by Liberation in Ara against Ranbeer Sena in 1995), gheraos (e.g., the famous gherao of the
state assembly after the Arwal massacre in 1986), rallies (including silent marches, torch processions, and more), chakkajaam (road blocks), putla dahan (effigy burning), and of course strikes.
Even hunger strikes have figured in this rainbow of agitations ...... Cultural media (sanskritik
madhyam) such as songs and plays have an important role in mobilisation, especially since a
large majority of the people in central Bihar are illiterate. Often the songs are made by people
themselves and convey their existing reality with great poignancy” (Bhatia (2005)). However,
figures suggest that at least in the last five years the violent part of the movement has taken
precedence, making the analysis of the conflict ever more crucial. In 2011 alone, by March, 13
already 116 people had been killed due to the conflict (SATP).
There are various estimates of the severity of the conflict from a variety of official and unofficial
sources. 18 Even according to the official source of the Government of India there have been
more than 5000 lives lost (civilians, rebels and security personnel) and more than 12000 incidents of violence in the period of 2004-2010 due to the Naxalite conflict. The number of people
displaced in Chhattisgarh alone was more than 43000, end of 2006. Moreover, the geographical
spread of the conflict has seen a phenomenal increase from 55 districts in eight states in 2003 to
194 districts spread across 18 states in 2007 (SATP; Iyer (2009)). ”While it is difficult to put an
exact figure to the number of rebels, there is an estimated 10000 to 20000 full time fighters with
countless thousands of village militias controlling particularly remote jungle areas where the
state is hardly present. On April 6th, 2010 several hundred Maoist guerrillas attacked a convoy
in a forest in eastern Chhattisgarh state, killing 76 armed policemen. This was reckoned to be
the worst loss in the stuttering, four-decade-long conflict” (The Economist). While the death
and injured toll looks grim, the other crucial issue is the divergence of development funds to the
Maoists. ”Nitish Kumar (the chief minister of Bihar) is among the few who actually put a figure
to it; he estimates that the Maoists in his state pull in Rs.150 crore a year-not counting the
cost of outright looting of money and snatching of arms and ammunition from security forces.....
Raman Singh (the Chief Minister of Chattisgarh) ....estimates Rs.100 crore from Bastar, the
Dandakaranya state within his state.” (Chakravarti (2008))
Surprisingly however, there is no in-depth empirical analysis of the Maoist conflict. The conflict is perhaps best understood through the various case studies that exist on the issue. Many
of these studies are concentrated on particular states identifying the nature and causes of the
conflict in that state through field studies, interviews of local people, and talks with Maoist
leaders and cadres etc. Bhatia (2005) for example provides an analysis of the nature, causes and
outcomes of the conflict in the context of Central Bihar. While her analysis focuses exclusively
on Central Bihar the key insights she brings out could well be true for the rest of the country as
well. She identifies three distinct albeit inter-related classes of reasons behind the entire Naxalite
movement viz. (1) Economic rights (2) Social rights (3) Political rights. The most important
economic issues that have been taken up by the movement as identified by Bhatia are as follows:
(1) land rights; (2) minimum wages; (3) common property resources; and (4) housing. The
Government of India in its own study recognizes the following socio economic factors behind
the discontent of people and support to Naxalism: Land related factors, Displacement & Forced
18
see appendix for some of the estimates.
8
evictions, Livelihood, Social Oppression, (apart from the socio economic reasons the GOI also
mentions issues arising out of bad governance and policing)(Government of India (2008).
Land rights are indeed one of the most important issues taken up by the Naxalites: Khet par
adhikar ke liye ladho, desh mai janwad ke liye badho (Fight for land rights, march towards
democracy in the country - Liberation (a Naxalite group) slogan) (Bhatia, 2005). ”[Indeed the]
focus of the Naxalite movement is on trying to provide land, whether the land of landlords or government land, to the landless. In fact, as pointed above the ’Naxalite’ term is itself derived from
the peasant uprising arising out of land disputes in Naxalbari. However, there are instances
of such struggle even later and elsewhere in the country. ”[For example in Bhojpur, Bihar,]
Jagdish Mahato, a local teacher who had forged links with Naxalite leaders from West Bengal,
led a protracted struggle against exploitative landlords” (Bhatia (2005)). While the inequality
in land distribution and the strife between big landowners and landless peasants seems to very
important in Bihar and West Bengal, in Chhattisgarh the strife between poor tribal landowners
and the state seems to be more important. As Bahree (2010) puts it ”There is a proxy war
underway in India’s interior- a bloody conflict raging over that rare and valuable commodity
in this too crowded country: land. On one side, powerful rebel groups claim to be fighting for
the poor-farmers and small agrarian tribes in particular. On the other side, the government is
locking up land and the resources buried beneath it (particularly coal and bauxite) for some
of India’s biggest private companies”. Some areas where the conflict is really intense like Bastar while are economically impoverished regions, are very rich in mineral resources” (Bahree
(2010)). ”Campaigners say that the reason why the government has opened a new front in
this battle lies beneath Chhattisgarh’s fertile soil, which contains some of the country’s richest
reserves of iron ore, coal, limestone and bauxite. Above live some of India’s most impoverished
people: semi-literate tribes who exist in near destitution. India’s biggest companies have moved
stealthily into the forest areas, buying up land and acquiring the rights to extract the buried
wealth” (The Guardian 2006). ”People are fighting against land acquisitions and the government
is labelling them Naxals and then using that to suppress them. .... If the Naxals were not there,
the government would be able to acquire more land for the private sector” (Sundar (2008)). In
fact, regardless of the ideologies of the different Maoist factions in practice land redistribution
has remained one of the main goals of the movement. This is as evident from the failed peace
talks between the Andhra Pradesh government and the PWG in 2004 where this was one of the
main issues (Iyer, 2009).
Underdevelopment has also been seen as one of the key reasons behind the rebellion. ”Even if
there was no truth behind the accusations that poor tribes are being exploited and are being
illegally disposed of their lands, it is fairly certain that the Naxalites are feeding on the festering
discontent of the impoverished and marginalized tribal communities. According to the 2001
census for example, about three quarters of Dantewada’s 1,220 villages are almost wholly tribal;
1,161 have no medical facilities; 214 have no primary school; the literacy rate is 29% for men
and 14% for women” (The Economist, 2006).
As far as the participation is concerned, ”...the social base of the movement ...... consists overwhelmingly of the landless, small peasants with marginal landholdings, and to a lesser extent
middle peasants. In caste terms, the base of the movement consists of lower and intermediate
castes” (Bhatia, 2005). The fight against the social oppression that the dalits [lower castes] and
the lower among the OBCs [other backward castes] have been regularly subjected to is perhaps
9
the most significant among the issues used by the Naxalite movement” (GOI, 2008).
Given the above discussion on the Naxalite conflict we now come to the main hypotheses of the
study.
3
The Main Hypotheses
The literature speaks of a variety of different factors that might lead to civil conflicts. ”Civil Wars
are more likely to occur in countries that are poor, are subject to negative income shocks, have
weak state institutions, have sparsely populated peripheral regions and possess mountainous
terrain” (Blattman and Miguel (2010)). Some of the other factors that have been mentioned in
the literature are, Ethnic and religious diversity and fragmentation, lack of democracy and civil
liberties, linguistic and religious discrimination, inequality, new states and political instability,
geographical factors like mountains and non-contiguous territory, population pressure, colonial
occupation, weak state institutions etc. 19 In this study we try to identify which factors might be
more relevant in the context of the Naxalite conflict in India. Thus, combining the findings from
the previous literature and our understanding of the conflict we test for different hypotheses.
Hypothesis 1: Land inequality increases conflict.
Following up from the discussion in the previous section we think that land issues are crucial.
Land inequality and exploitation of the landless/poor farmers by the wealthy landlords has been
attributed to be one of the most important causes of the Naxalite movement. In fact, that’s
how it started in the first place. ”Land is a very strategic socio-economic asset, particularly
in poor societies where wealth and survival are measured by control of, and access to, land”
(USAID (2005)). Moreover, land often has valuable natural resources buried beneath it. In a
predominantly agrarian economy, the importance of land cannot be overemphasized and survival
of the poor often depends on their access to land. There is in fact, empirical evidence on the
importance of land on conflicts from elsewhere in the world as well. Andre and Platteau (1996)
study a highly densely populated area in the Northwest of Rwanda during the period 1988-93 and
find how the land distribution had become increasingly unequal and land dispossession rampant.
They further show that pervasive incidence of land disputes and the threat of landlessness had
led to rising tensions in social relations. They relate these adverse conditions in land distribution
to the civil conflict that broke out in 1994. Verwimp (2003) offers an empirical analysis of the
peasant participation in the Rwandan genocide. He uses data from pre-genocide agricultural
survey sample and builds a panel dataset. He finds that the genocide perpetrators comprised
mostly of poor wage workers and land renters, while the victims were primarily from the landlord
class.
There are several potential ways in which the land distribution could affect the conflict. A highly
skewed land distribution also reflects higher disparities in the social and economic lives of the
people and thus a higher potential for grievance. Moreover, if the distribution is too unequal and
dominated by a few large landlords while the vast majority being small and landless then there
is an additional source of problem. In a growing India the government is trying to acquire land
for a variety of purposes including industrialization, mining or building dams. In case of land
acquisition by the government while the entire community is adversely affected the compensation
19
See Fearon and Laitin (2003) for a discussion.
10
mostly or fully goes to the big landlords who to start with enjoy a higher socio-economic status,
while the vast majority of the affected are left uncompensated. Moreover, it is also common
that tribal people in remote places have been living on a certain plot land for generations but
do not have any formal title deeds to the land resulting in no compensation whatsoever in case
of acquisition by the government. Thus, the Naxalites are fighting for land issues against both
big landlords and the state 20 . Indeed, the land distribution is crucial not only for the Maoist
conflict but for the overall health of the rural economy. The Land reforms in India in the post
independence period thus sought to directly improve the access to land of the poor households.
Besley and Burgess (2000), in state level analysis of the Indian land reforms find that indeed the
land reforms have been associated with significant poverty reduction. However, implementation
was hardly homogeneous across states. These heterogeneities exist even within states which
have witnessed widespread implementation. ”The 3 extreme Maoist districts in West Bengal are
West Midnapur, Puruliya, Bankura-where land reforms on the scale affected by the CPI (M)
in other parts of the state haven’t taken place.” (Chakravarti, (2008))” Thus, we test to see if
indeed more land inequality is associated with more conflict.
Hypothesis 2: Underdevelopment leads to more conflict.
In other words, districts with higher per capita incomes should experience less conflict. The
theoretical idea goes back to Becker (1968) where he argues that rise in returns to crime induces
more workers to the criminal sector. In fact, the relation between income and conflict is one of
the most robust relations in the empirical literature on conflicts. Both Collier and Hoeffler (1998,
2004) and Fearon and Laitin (2003) for example suggest that lower per capita incomes lead to
higher probabilities of conflict. Miguel et al. also demonstrate how negative income shocks
lead to higher levels of conflict. Collier and Hoeffler (2004) argues that the opportunity cost of
fighting is lower when incomes are low and thus making it easier to recruit rebels. Dube and
Vargas (2008) identify two distinct channels through which income shocks affect armed conflict.
They argue that higher wages reduce conflict (lower international coffee prices lead to a negative
shock to certain regions of Colombia leading to higher conflict levels).This is the opportunity
cost channel. If on the other hand the price of natural resources viz. oil goes up it increases
the contestable income and thus an increase in the conflict levels. This has been termed as the
rapacity effect. In the Indian context the rapacity channel seems to be unlikely since the fighting
is not (at least not yet) for a share from resource revenues 21 . Fearon and Laitin (2003) on the
other hand argues that lower incomes reflect limited state capacity to put the rebellion down.
In other words, low per capita income is an indicator of weak and corrupt governments who
might be unable to devise effective counter insurgency policies. Moreover, guerrilla warfare is
usually concentrated in remote rural areas where the rebels get suitable conditions to pursue such
insurgencies. The favourable conditions that such regions provide include superior knowledge of
local conditions and territory by rebels than by government forces, rough mountainous territory
or forests providing suitable hideouts to insurgents etc. Even for nations with strong military
20
In a recent hostage incident when a government collector was kidnapped, two of the primary demands of the
Maoists were as follows: grant of land rights to tribal people in scheduled areas; minimum displacement of tribals
while making space for industries and mining (The Hindu, Feb 22, 2011: Orissa accepts eight Maoist demands:
http://www.hindu.com/2011/02/22/stories/2011022264251300.htm).
21
However, indeed the government has boosted up spending on development efforts in the Naxalite affected
districts and some of this spending is being siphoned off to Naxalites. But that is a consequence rather than a
cause of the conflict.
11
capabilities such conditions could turn out to be pretty daunting (e.g. the US in Vietnam, British
in the initial days in Northern Ireland) 22 , for poorer nations with weak military capabilities
and corrupt bureaucratic set ups it could be hopeless. Thus, we try to test if lower per capita
incomes indeed imply higher conflict levels. 23
Hypothesis 3: Historical land institutions directly impact the conflict.
As already discussed in section 1, the allocation of the responsibility of collecting the land revenue
to the landlords (which translates into a de facto property right over the land) gave birth to a
reason for perpetual conflict between the peasants and landlords. ”Elsewhere, the colonial state
directly collected the land revenue from the cultivator, thereby avoiding this particular source
of internecine conflict” Banerjee and Somanathan (2007). Moreover, land acquisition for mining
purposes, building dams or for private industry is an important issue pursued by the Maoists.
Such land acquisition often leads to large scale displacement and loss of livelihood of people.
Given adequate compensation people might be less discontent with the displacement induced
by land acquisition. In this context, Duflo and Pande (2007) show how the landlord districts
do worse than the non-landlord districts as far as effects of dams are concerned. They argue
that since the social relation in the landlord districts somehow renders collective action difficult
it leads to inadequate compensation. Following the same argument if land is taken away in the
landlord districts for industrial, mining or the purposes of building dams, historically people
in such districts have less potential for collective action leading to inadequate compensation,
which in turn leads to more grievances. This makes for easier Maoist recruits. Thus, we try to
test whether the historical property rights institutions have any effect over and above its effects
through underdevelopment and land distribution.
Hypothesis 4: The presence of disadvantaged castes Scheduled Castes (SC) and
Scheduled Tribes (ST)) leads to more conflict.
As discussed in the previous section ”the main support for the Naxalite movement comes from
dalits (SC) and adivasis (ST) (Government of India (2008)).” In fact, the predominantly tribal
areas like Chhattisgarh have experienced higher conflict levels. While the SCs and STs have
enjoyed affirmative action in the post independence period, they might still not be economically
and socially at par with others. ”There are over 2,00,000 pending cases of atrocities against lower
castes in India, and the conviction rate is just a little over 2 percent. ”There are an estimated
162 million untouchables in rural India, according to the National federation of Dalit Land
rights movement. 70% of them don’t own land” (Chakravarti (2008)). Thus Caste and tribal
identities are important issues that could exacerbate the conflict and we directly test to verify if
a higher SC & ST percentage indeed lead to higher conflict. Moreover, if the benefits of the high
growth that the country is experiencing are not homogeneous across groups, there is further
potential of grievances arising out of feelings of exclusion. ”Some case studies [Sambanis 2005,
Frances Stewart 2001] suggest that ’horizontal’ inequality-inequality that coincides with ethnic
or other politically salient cleavages-is a particular important driver of civil conflict” Blattman
and Miguel (2010).” Thus we test to see how the incomes of the different groups impact the
conflict.
22
examples from Fearon and Laitin (2003)
23
In the geographical controls we also include proportion of barren and rocky land which also indicates presence/lack of economic opportunity.
12
Apart from the variables mentioned in the above hypotheses we control for several other variables
that the literature suggests to be important. For example, Geographical factors have been found
to be important in such a context. Geography determines the cost, time and tactics necessary
to curtail and check such activity. Fearon and Laitin (2003) for example find that conditions
favouring insurgency like rough terrain increases the likelihood of conflict. We thus control for
the geographical factors like percentage of forests, sloping land, sandy land, barren rocky land
etc. in the district. We also control for income inequality and size variables like population
density, area etc. 24
4
Data
Districts are the smallest administrative units in India for which coherent reliable data is available. The idea of the study is to exploit micro data and thus all regressions are at the district
level. We use data from the 16 main states in India which is about 90% of the Indian population.
The data is pooled from diverse sources. In this section we describe the original sources of the
data and also try to justify the use of the different variables.
Naxalite/Maoist incidents:
The data on the Maoist incidents comes from four different sources:
• Global Terrorism Database (GTD) I: 1970-1997 & II: 1998-2004: In fact now there is one
consolidated GTD which has data till 2007.
• Rand-MIPT Terrorism Incident database (1998- present)
• Worlwide Incidents tracking system (WITS), National Counter Terrorism Centre (20042007)
• South Asia Terrorism Portal (SATP) maintains an excellent web portal tracking all terrorism incidents in South Asia. It has most detailed accounts of the major Naxalite incidents
from 2005 onwards.
All the above mentioned sources track data on terrorist/violent incidents from different sources
like newspapers, official reports etc. 25 . The data is across countries often with the village/city/district name of the place where each particular incident took place specified. We
have filtered out all the data pertaining to India and then kept only the incidents that were
clearly identifiable as Maoist/Naxalite in nature. To do so we used the name of perpetrator
unless it was clearly mentioned as being a Maoist/Naxalite incident. Then except for the cases
where the district is clearly mentioned the town or village mentioned was placed in its corresponding district. More often than not we were successful in doing so. However, we had to leave
out the cases where it was not possible to do identify the districts. Finally, we have consolidated all the above data sources to construct a comprehensive consolidated district level Maoist
24
A higher total population might by itself indicate more potential recruits. Moreover, when coupled with a
higher proportion of barren rocky land it could imply a less per capita availability of usable land and thus a
population pressure on land.
25
The first three sources have been used by Iyer (2009)
13
incidents database. Thus, there is information on presence, incidents and the number of deaths
and injuries at the district level from 1979-2009.
Consumption & Poverty: District-specific indicators of income or expenditure are not available for India. Instead, per capita consumption expenditure is calculated from the National
Sample Survey (hereafter NSS) data. The NSS conducts every 5 years the Consumption expenditure survey (among other surveys) whence district specific per capita consumption expenditures are calculated. The last four NSS thick rounds 26 were undertaken in 1987-88 (43rd),
1993-94 (50th), 1999-00 (55th), 2004-05 (61st). Given that district specific data is used, the
50th round is unusable since there are no district identifiers. For the purposes of this paper we
use the other three rounds. The per capita consumption expenditure is calculated for all the
districts using the Marginal per capita consumption (MPC hereafter), using the 30 day recall
period since the 30 day recall period is the only one that is common across rounds. The MPC is
used as a proxy for per capita income and the MPC figures are also used to calculate the district
specific poverty rates and Gini coefficients of inequality within the districts.
Demography: Apart from the Sample surveys by the NSS, India conducts a country wide
census every 10 years. The data on demography and public goods are stored in the so called
village directory data and is available for each census year since 1961. The 2001 census is used,
whence we have data on total population, Schedule tribe and Caste population, population by
religious groups. All the above information is available at the village level which have been used
to construct the corresponding district level numbers.
Geography: We control for geographical terrain by the fraction of the districts uncultivated
area that is barren rocky, sandy and steep sloping. We also control for remoteness by the log
of the distance from the state capital. Such data is made available in the Wasteland atlas of
India, Department of Land Resources (Ministry of Rural development) in collaboration with the
National Remote Sensing agency, (Department of Space). These variables by definition do not
vary over time. We also have data on forest cover in the different districts. The data comes
from the various ”State of forest cover” reports of the Forest Survey of India (FSI) and gives
the percentage of forest cover in each district. The 2005 data is used.
Land distribution: The data on operational holdings of agricultural land comes from the
Agricultural census of India which collects such data every 5 years. The data gives the number
of of operated landholdings in the different size classes (viz. Small, semi medium, medium and,
large). Both the 1991 and the 2001 censuses are available. The gini coeeficient for land inequality is calculated using this data. The computed Gini coefficient varies from 0.12 to 0.78 and 0.14
to 0.79 in 1991 and 2001 respectively. The average inequality has gone up from 0.47 in 1991 to
0.5 in the year 2001 (data on Bihar not available in 2001). The land inequalities across the 2
years are highly correlated with Correlation coefficient of 0.97. We use the 1991 data since that
way we have data prior to the high conflict years and also no major state is missing.
Colonial Land Institutions: This data comes from Banerjee and Iyer (2005). However, out of
the 362 districts we use in the analysis we have data on Land Institutions for only 233 districts.
27
26
only the thick rounds can be used since the thin rounds have too few observations
27
This data is available for districts which were directly under British control.
14
5
Empirical Analysis
We have annual data on the conflict variables for the period 1979-2009. However, since we have
only three rounds of NSS data viz. 1987-88 (43rd), 1999-00 (55th), 2004-05 (61st), we club the
conflict data to match these three different NSS periods. The district names are first mapped
to the districts that existed in 1987. Then all the conflict data from the period 1988- 1999 are
collapsed and clubbed together and matched to the 1987-88 NSS data, all the conflict data from
the period 1999- 2004 are clubbed together and matched to the 1999-2000 NSS data, all the
conflict data from the period 2005- 2009 are clubbed together and matched to the 1987-88 NSS
data. Thus we have data on 362 districts for three time periods.
There are primarily two variables of interest that we try to explain viz. ’Probability of Conflict’
and ’Intensity of Conflict’. In order to explain the probability of conflict we create a 0-1 binary
outcome variable (called Maoist) that takes the value ’1’ if a district has seen any Maoist activity at all in the relevant period while it takes the value ’0’ otherwise. While it is extremely
important to understand which factors increase conflict probabilities it is equally important to
understand what factors lead to higher intensity of conflict. Intensity is a latent variable that
is measured using two variables viz. ”Total number of incidents” in a particular district and
”Total number of deaths + wounded” in the district in the relevant periods. The econometric
specification is very simple and is as follows:
(Conf lict)j,t = α(Conf lict)j,t−1 + βXj,t−1 + γGj + αs + δt + j,t
The (Conf lict)j,t variable gives the conflict (presence or intensity) in district ”j” in round ”t”.
The explanatory variables Xj,t include economic variables like MPC and measures of income inequality. ”G” includes all variables that do not change over time: land inequality, demographic
variables, presence of marginalized sections like schedule tribes and castes and geographic variables like barren and rocky land , steep sloping land, percentage forests etc. ”s” is the state
dummy, while ”t” is the time dummy.
For the 0-1 probability of conflict variable we use Probit specifications. The intensity variables
are both Count Variables by nature taking integer values from zero upwards. The Poisson model
is the standard model used in such cases. However, the Poisson model assumes the mean and
variance to be equal. In the presence of many zeros in the data like in our case there is overdispersion and thus the equidispersion assumption of the Poisson model does not hold true. The
number of zero counts in my data is more than 70%. The standard parametric model to account
for overdispersion is the negative binomial (Cameron and Trivedi (2005)). Thus, we use the
Negative Binomial model 28 for explaining intensity. 29
Main Results:
In the first set of regressions we try to identify the potential deteminants of the presence and
intensity of conflict, primarily focussing on land inequality and MPC . All the specifications
control for Time dummies and geography controls. We add controls for income inequality, percentages of Scheduled Castes & Tribes, size variables, and the proportion of the district that was
28
Poisson is a special case of it.
29
All the regressions use cluster robust standard errors, clustered at the state level
15
non landlord. We also control for conflict presence in the previous period in all the specifications.
Table 1, 5, 8: Pooled Probit regressions explaining presence of Maoist Conflict in districts.
Table 2: IV Probit regressions explaining presence of Maoist Conflict in districts.
Table 3, 6, 9: Pooled Negative Binomial regressions explaining no. of Maoist incidents in districts.
Table 4, 7, 10: Pooled Negative Binomial regressions explaining no. of dead & wounded in
districts.
Column 1 includes MPC, land inequality and time and geography controls. Column 2 adds
controls for ethnic group sizes, population density and income inequality. Column 3 controls for
the initial period income while Column 4 controls also for proportion non-landlord. Column 5
adds state controls (but not the initial period income nor proportion non-landlord). Column 6
adds proportion non-landlord to Column 5. In order to address potential endogeneity concerns,
in Table 2 we use lagged consumption as an instrument for previous consumption. 30
Income, Land inequality and Conflict
A lower per capita income (as proxied by the mean per capita consumption expenditure) significantly increases the risk of conflict in three of the six specifications in the Pooled Probit
regressions (Table 1) explaining presence of Conflict and four and three specifications respectively in the Nbreg regressions (Tables 3 & 4) explaining the number of incidents and the number
of people dead and wounded due to the conflict. However, in the IV Probit regressions (Table 2),
where we instrument MPC with its lagged values, MPC is a significant and robust predictor of
presence of Conflict in all the specifications. Thus, we indeed have evidence that poorer regions
experience more conflict. This is pretty much in line with the previous literature and is in fact
one of the most robust results in the conflict literature. In terms of Average marginal effects, a
1 SD increase in the log of MPC increases the probability of conflict in the next period by 13
% (highest), and results in 4 more incidents and 28 more dead and wounded people more in the
next 5 years.
On the other hand, while income inequality does not seem to play any role in the context of
the Maoist conflict, Land inequality is a highly significant and robust predictor of conflict in
all the specifications. Land inequality also significantly increases the intensity of conflict and
this is robust to all sorts of controls. This gives support to the the greivance arising out of land
inequities being a key reason behind the conflict. In terms of Average marginal effects, a 1 SD
increase in land inequality increases the probability of conflict in the next period by at least 9
% , and results in 5 more incidents and 35 more dead and wounded people more in the next 5
years.
While a higher SC population does not indicate higher presence of conflict, tribal areas do tend
to experience more conflict, at least in some of the specifications. However, this is not robust
in all the specifications. In the next set of regressions we will use the incomes of two groups
separately (Tables 5 to 7).
The land Institutions variable, p nland i.e. the proportion of districts that was not under land30
Thus we use data from the last two available rounds. In any case, conflict occurrence is highly concentrated
during this period).
16
lord control is highly significant and negative in the specifications where it has been included.
We will analyse its robustness in the tables 8 to 10.
As far as the geography variables are concerned, barren & rocky areas experince more conflict.
This is in agreement to our expectations. In a primarily agrarian economy, a higher proportion
of barren & rocky land indicates lack of economic opportunity. This supports the opportunity
cost story. Percentage steep sloping land is significant in almost all the specifications and always
has a negative coefficient. This could sound a bit counter intuitive since most of the previous
literature mentions mountainous land as being more suitable for conflict. However, given that
the Naxalite conflict is almost exclusively concentrated in the plains of India this result makes
absolute sense. If we could have used even finer dummies for regions than what we have used,
perhaps the results would have been different. More interestingly, the percentage of forest cover
comes out to be highly significantly increasing the probability of conflict in almost all the specifications (in fact of all the specifications in Tables 2, 3 & 4). There could be several plausible
explanations for this. For one thing, large patches of forest cover provides the perfect hiding
and fighting conditions for rebels and makes it quite difficult for security forces to keep up with
them. While such an explanation is perfectly plausible there could be a deeper meaning to the
forest cover variable turning out to be positively significant even when we account for income,
inequality and regional dummies. What this might be showing is the confrontation of modernization and development with previously isolated communities who are mostly concentrated in
remote areas with higher forest cover. As discussed in section 2, there seems to be some relation
between the land acquisitions from tribal peasants in previously untapped resource rich areas
by the state with the conflict.
Accounting for differences in income of Lower castes/tribes and others: Exclusion :
In the tables 5 to 7, we check specifically for the exclusion story. We run the same set of regressions as above but instead of the overall income we use the incomes clubbed in two groups:
Scheduled Castes & tribes vs General Castes. The better thing to do would have been to put
the incomes of all the three groups separately but we are unfortunately unable to do so since
the number of Scheduled tribal people surveyed is just too low in many instances and thus we
would lose a lot of observations. Instead we club the the Scheduled Castes and Tribes together.
Overall, the incomes of both the groups i.e the General Castes & the SCSTs matter and they
both reduce the probabilty of conflict (at least when they matter significantly). Interestingly in
the specifications where both the variables are significant, the SCST consumption has a higher
magnitude. But the relationship is not robust in all the specifications when we are trying to
explain the presence of conflict. However, we find that SC/ST income has a negatively significant and robust impact on the number of incidents in a particular district, ceteris paribus. In
other words, districts where the SCs & STs have a lower income experience significantly more
Maoist incidents than other districts. The same result does not hold for the incomes of the
general castes. If we look at the number of people dead and wounded the results are pretty
much similar. Moreover, in almost all the specifications we have percentage of Scheduled Castes
and Tribes in the district, significantly increases the conflict presence and intensity. While this
is not conclusive but it does give some confirmation of the participation from the lower castes
and tribes. In terms of AME, we have a 1 SD increase in the log(MPC) of the SC/STs resulting
in around 4-6 more Maoist incidents while the similar increase in the incomes of the General
Castes result in around 1-3 more incidents.
17
Institutions and Conflict:
Finally, we directly test for how far the colonial institutions matter (Tables 8 to 10). While,
following the earlier discussion we know that the land institutions could matter either indirectly
through its effect on underdevelopment and/or land inequality or directly through the environment of class antagonism it has created. But we finally test for the direct effect. We do this
since if indeed the land institutions have a direct effect on the conflict we cannot use it as an
instrument for income or land inequality. However, using the proportion non-landlord variable
directly, we can get the remaining effect that the institutions have on the conflict outcome, once
we control for income and land inequality. We note that the proportion of the district i.e. not
landlord is robust to the various specifications and has significant negative effect on the presence
and intensity of conflict. We also note that land inequality and the geography variables still
continue to be significant in explaining presence of conflict. In terms of Average marginal effect,
an increase of 1 SD in the proportion of the district that was not controlled by the landlords
reduces the probability of Maoist conflict in this district by around 8%, results in 2 less incidents
and around 10-12 less dead & wounded people.
Growth and Conflict:
In the next set of tables we try to identify what impacts growth might have on the conflict
(Table 11). The evidence in favour of low growth leading to higher conflict is not too strong.
While indeed in 3 of the 5 specifications in Table 11, growth significantly reduces conflict levels,
the result is not robust once we account for historical land institutions. Land inequality on the
other hand continues to be significant. Moreover, the Average Marginal effects are too small in
magnitude. We also try to identify if the growth of income of the different subgroups i.e. SCs
& STs versus the general castes have any significantly different impact on the conflict (Tables
12-14). The incomes of both the groups come out to be significant in some of the cases. However,
this result is also not robust in all specifications. However, we find that the magnitudes of the
growth rates of incomes of the disadvantaged castes/tribes are always higher than that of the
general castes. Moreover, we find that the SCST income growth rate has a significant effect in
many cases at least as far as the intensity of conflict is concerned. Even in this case the average
marginal effects are too small.
6
Conclusion
This paper is the very first attempt to study the political economy of the Maoist conflict in India
in-depth using a district level panel. It contributes to the civil conflict literature by adding
to a small but growing literature that uses sub-national micro data to study civil conflicts.
Making use of a newly constructed district level conflict database the paper provides some very
interesting insights on the causes of the Maoist conflict. The evidence on how lower income
leads to more conflict is pretty much in line with the existing literature and in addition the
significance of the proportion of non agricultural land that is barren rocky in turn gives support
to the grievance and opportunity cost story. Land inequality on the other hand comes out to
be very robust in explaining the Maoist conflict and thus has immediate policy implications in
terms of land reforms. The other important contribution of this paper is to the Institutions
literature. This paper shows how centuries old historical institutions could have a significant
impact on present day conflict outcomes. The significance of the Historical Land institutions
18
variable is also interesting since it shows that while the landlord districts might have indeed
experienced more land reforms, the land reforms per se haven’t been able to address the class
based antagonism and embittered social relations that these districts continue to experience.
The Government of India has indeed recognized that apart from an appropriate police strategy
to handle the conflict there is need for a non-military approach involving development of the
relevant regions and solving disputes arising out of land. 31
Also there is evidence that while the income of the general castes are not that important the
lower incomes of the SC/ST lead to higher conflict incidents (once we control for incomes of
the general castes). This could be evidence in favour of the exclusion story, i.e. the income rise
in India is not homogeneous and some groups are falling behind. While this evidence is not
conclusive we still have been able to ask this interesting question due to the Micro nature of the
data.
Future research would be in the direction of controlling for land reforms at the district level and
verifying its effects on the conflict outcomes 32 . The other direction for future research would
be to find adequate instruments for the income variables that would allow us to further ensure
that our results are not affected by endogeneity. Moreover, a lot of work remains to be done in
terms of data collection from the households of the perpetrators and victims in order to further
pin down both the causes and consequences of the Maoist conflict at the household/individual
level.
31
The Ministery of Home Affairs, Government of India(http://mha.nic.in/uniquepage.asp?Id Pk=540) points
out over and above the policing issues the other aspects related to Naxalism are:
• fair and firm revenue (land) administration, with attention on elements like proper maintenance of land
records, expeditious recording of mutations and fair disposal of land disputes, without undue delay.
• appropriate mechanisms for grievance redressal, public contact and public awareness, for creating an overall
positive environment and confidence of the people in the State administrative machinery.
32
See Besley and Burgess (2000) for a state level study of the land reforms.
19
References
Daron Acemoglu, Simon Johnson, and James A. Robinson.
The colonial origins of comparative development: An empirical investigation.
American Economic Review, 91 (5):1369–1401, 2001.
Daron Acemoglu, Simon Johnson, and James A. Robinson.
Reversal of fortune: Geography and institutions in the making of the modern world income
distribution.
Quarterly Journal of Economics, 117 (4):1231–94, 2002.
Catherine Andre and Jean-Philippe Platteau.
Land tenure under unendurable stress: Rwanda caught in the malthusian trap.
Cahiers de la Faculte des Sciences Econommiques et Sociales de Namur, Serie Recherche, 164
- 1996/7, 1996.
Megha Bahree.
The forever war: Inside india’s maoist conflict.
World Policy Journal, Summer, 27.2:83–89, 2010.
Abhijit Banerjee and Laxmi Iyer.
History, institutions and economic performance: The legacy of colonial land tenure systmes
in india.
American Economic Review, 2005.
Abhijit Banerjee and Rohini Somanathan.
The political economy of public goods: Some evidence from india.
Journal of Development Economics, Elsevier, 82(2):287–314, 2007.
Vani K. Barooah.
Deprivation, violence and conflict: An analysis of naxalite activity in the districts of india.
International Journal of Conflict and Violence, 2 (2):317–333, 2008.
Gary S Becker.
Crime and punishment: An economic approach.
Journal of Political Economy, 76(2):169–217, 1968.
Timothy Besley and Robin Burgess.
Land reform, poverty reduction, and growth: Evidence from india.
Quaterly Journal of Economics, 115 (2):389–430, 2000.
Timothy J. Besley and Torsten Persson.
The incidence of civil war: Theory and evidence.
NBER Working Paper Series, Working Paper 14585, 2008.
Bela Bhatia.
The naxalite movement in central bihar.
Economic and Political Weekly, 40 (15) Apr 9-15, 2005.
20
C. Blattman and E. Miguel.
Civil war.
The Journal of Economic Literature, 48(1):3–57(55), 2010.
Anjali Thomas Bohlken and Ernest Sergenti.
Economic growth and ethnic violence: An empirical investigation of the hindu-muslim riots
in india.
2009.
A. Colin Cameron and Pravin K. Trivedi.
Microeconometrics:methods and applications.
Cambridge University Press, 2005.
Sudeep Chakravarti.
Red sun-travels in naxalite country.
Penguin Global, 2008.
Antonio Ciccone.
Transitory economic shocks and civil conflict.
unpublished, 82(2):287–314, 2010.
P. Collier and A. Hoeffler.
Greed and grievance in civil war.
Oxford Economic Papers, 56(4):563–596, 2004.
Quy-Toan Do and Lakshmi Iyer.
Geography, poverty and conflict in nepal.
HBS Working Paper, 07-065, 2009.
Oeindrila Dube and Juan F. Vargas.
Commodity price shocks and civil confict:evidence from colombia.
CID Graduate Student and Postdoctoral Fellow Working Paper No. 14, 2008.
Esther Duflo and Rohini Pande.
Dams.
The Quarterly Journal of Economics, 122(2):601–646, 05, 2007.
Stanley L. Engerman and Kenneth L. Sokoloff.
Factor endowments, institutions, and differential paths of growth among new world economies:
A view from economic historians of the united states.
in Steven Haber, ed., How Latin America fell behind: Essays on the economic histories of
Brazil and Mexico, 18001914. Stanford: Stanford University Press, 1997.
Stanley L. Engerman and Kenneth L. Sokoloff.
Factor endowments, inequality, and paths of development among new world economies.
Economia: Journal of the Latin American and Caribbean Economic Association, 3 (1):41–88,
2002.
21
J. Fearon and D. Laitin.
Ethnicity, insurgency and civil war.
American Political Science Review, 97 (1):75–90, 2003.
Lakshmi Iyer.
The bloody millennium: Internal conflict in south asia.
Harvard Business School, Working Paper, 09-086, 2009.
Saumitra Jha.
Complementarities and religious tolerance: evidence from india.
Stanford GSB Working Paper No. 2004, 2008.
Rajat Kujur.
Naxal movement in india: A profile.
Institute of Peace and Conflict Studies, New Delhi, 2008.
Rafael La Porta, Florencio Lopez de Silanes, Andrei Shleifer, and Robert Vishny.
Law and finance.
Journal of Political Economy, 106 (6):113–55, 1998.
Rafael La Porta, Florencio Lopez de Silanes, Andrei Shleifer, and Robert Vishny.
The quality of government.
Journal of Law, Economics, and Organization, 15 (1):222–79, 1999.
Rafael La Porta, Florencio Lopez de Silanes, Andrei Shleifer, and Robert Vishny.
Investor protection and corporate governance.
Journal of Financial Economics, 58 (1-2):3–27, 2000.
E. Miguel, S. Satyanath, and E. Sergenti.
Economic shocks and civil conflict: An instrumental variables approach.
Journal of Political Economy, 112 (41):725–753, 2004.
Anirban Mitra and Debraj Ray.
Implications of an economic theory of conflict:hindu-muslim violence in india.
unpublished, 2010.
Government of India.
Development challenges in extremist affected areas.
Report of an Expert Group to the Planning Commission, 2008.
Nandini Sundar.
Subalterns and sovereigns, an anthropological history of bastar (1854-2006).
Oxford University Press, 2008.
USAID.
Land and conflict: A toolkit for intervention.
US Agency for International Development (USAID) Office of Conflict Management and Mitigation, 2005.
22
Philip Verwimp.
Micro-level evidence from rwanda.
HiCN Working Papers 08, Households in Conflict Network, 2003.
23
7
Appendix
24
maoist 1
Table 1: Dep Var: Maoist; Probit regressions
(1)
(2)
(3)
(4)
probit1
probit2
probit3
probit4
1.912*** 1.875*** 2.111*** 1.453***
(5)
probit5
0.928***
(6)
probit6
0.626***
prop sandy
-2.497
1.074
5.001
23.67***
24.40**
31.53**
log stcapdistance
-0.0847
-0.0841
0.0488
0.247
-0.0181
0.192
prop barrenrocky
7.877***
12.37***
17.32***
34.39***
19.24***
33.93***
prop steepsloping
-68.93***
-59.95***
-81.18**
-61.78**
-132.7***
-106.9***
proportion forest cover
1.145**
1.100
1.863**
1.572**
1.961***
2.040***
log area
0.558**
0.671***
0.724***
0.902***
0.732***
0.759***
consumption pc
-1.621***
-1.413***
-1.805***
-0.828
-1.097
-0.588
land inequality
3.337***
3.434***
3.441***
2.992***
4.374***
5.324***
%Scheduled Castes
-0.531
0.116
-0.953
2.282
3.958**
%Scheduled Tribes
0.849*
1.106*
0.269
1.457***
1.256**
population density
-59.74
-150.6**
-113.9***
19.63
2.098
income inequality
-0.178
-3.348*
-5.958***
1.166
-1.320
-0.263
0.387
initial consumption pc
Prop. Non landlord
-1.392***
State Dummies
Time Dummies
Observations
Yes
996
Yes
996
Yes
656
* p < 0.10, ** p < 0.05, *** p < 0.01
25
Yes
431
-1.256**
Yes
Yes
Yes
767
Yes
548
consumption pc
Table 2: Dep Var: Maoist; IVProbit regressions
(1)
(2)
(3)
(4)
ivprobit1 ivprobit2 ivprobit3 ivprobit4
-3.078*** -2.808*** -3.427*** -3.215**
maoist 1
(5)
ivprobit5
-3.551***
(6)
ivprobit6
-3.536*
1.954***
2.007***
1.919***
1.308***
0.954***
0.639**
prop sandy
-0.907
5.844
5.729
26.31***
35.48
43.08*
log stcapdistance
-0.0245
0.00487
0.0325
0.192
0.0776
0.226
prop barrenrocky
10.43***
17.33***
18.22***
32.26***
25.49***
34.56***
prop steepsloping
-56.51**
-45.69*
-75.42**
-56.67*
-82.55**
-49.45
proportion forest cover
1.278*
1.866**
1.818**
1.529*
2.758***
3.018***
log area
0.338
0.575**
0.553**
0.661**
0.708***
0.723***
3.610***
3.215***
3.039**
2.363**
3.530***
3.884***
%Scheduled Castes
-0.118
-0.353
-1.095
2.030
3.950**
%Scheduled Tribes
0.657
0.548
-0.841
0.955*
-0.505
population density
-132.3**
-135.1**
-90.23***
-72.44
-79.00
income inequality
-1.333
-0.190
-0.763
3.478
1.564
0.716
1.338
land inequality
initial consumption pc
Prop. Non landlord
-0.953**
State Dummies
Time Dummies
Observations
Yes
640
Yes
640
Yes
638
* p < 0.10, ** p < 0.05, *** p < 0.01
26
Yes
420
-1.314***
Yes
Yes
Yes
520
Yes
359
maoist 1
Table 3: Dep Var: No. of Incidents
(1)
(2)
(3)
(4)
nbreg1
nbreg2
nbreg3
nbreg4
2.891*** 2.731*** 2.935*** 2.165***
(5)
nbreg5
1.201***
(6)
nbreg6
0.888**
prop sandy
-17.30
-12.67
-13.25
1.164
8.291
18.12
log stcapdistance
-0.287*
-0.302
-0.0125
0.159
-0.104
0.0306
prop barrenrocky
19.24**
30.17***
36.71***
39.24***
21.59***
27.99***
prop steepsloping
-112.9***
-101.3**
-135.1***
-152.3**
-136.2**
-124.1*
proportion forest cover
2.385**
2.200*
2.767**
2.991**
2.960***
3.762***
log area
1.570***
1.865***
1.891***
2.434***
1.536***
1.955***
consumption pc
-2.887***
-2.004**
-1.744*
-1.376
-1.718**
-1.637
land inequality
7.590***
8.066***
7.977***
6.160***
6.499***
6.511***
%Scheduled Castes
-0.867
-0.0752
-1.311
2.821
5.415*
%Scheduled Tribes
2.198***
2.302***
-0.579
2.358***
1.188
population density
-107.7***
-137.6***
-132.0**
85.04
50.03
income inequality
-3.240
-9.450**
-14.12***
0.246
-1.734
-0.931
0.0657
initial consumption pc
Prop. Non landlord
-1.279***
State Dummies
Time Dummies
Observations
Yes
996
Yes
996
Yes
656
* p < 0.10, ** p < 0.05, *** p < 0.01
27
Yes
431
-1.183**
Yes
Yes
Yes
996
Yes
655
maoist 1
Table 4: Dep Var: No. of dead wounded
(1)
(2)
(3)
(4)
nbreg1
nbreg2
nbreg3
nbreg4
2.925*** 2.755*** 2.941*** 2.298***
prop sandy
-50.69***
-39.58**
-48.38***
-40.40***
11.45
36.52
log stcapdistance
-0.497***
-0.636***
-0.151
-0.127
-0.251
-0.189
prop barrenrocky
21.44***
30.25***
40.17***
40.48***
34.47***
42.84***
prop steepsloping
-191.3***
-187.0***
-234.0***
-255.9***
-306.5***
-323.7***
proportion forest cover
4.119***
3.691**
4.248**
5.890***
4.949***
6.060***
log area
1.940***
2.409***
2.317***
3.038***
2.144***
2.691***
consumption pc
-4.398***
-2.845**
-1.674
-1.499
-3.329*
-2.612
land inequality
11.44***
12.34***
11.14***
8.581***
13.13***
14.11***
%Scheduled Castes
-3.005
-3.845
-5.684
5.467
9.513***
%Scheduled Tribes
2.819**
1.972*
-3.349***
2.764
1.037
population density
-88.03**
-119.2***
-128.2
149.5
124.7
income inequality
-5.387
-14.42***
-21.21***
-1.728
-5.001
-2.611**
-1.829
initial consumption pc
Prop. Non landlord
(6)
nbreg6
0.639
-1.186***
State Dummies
Time Dummies
Observations
(5)
nbreg5
0.943**
Yes
996
Yes
996
Yes
656
* p < 0.10, ** p < 0.05, *** p < 0.01
28
Yes
431
-1.335*
Yes
Yes
Yes
996
Yes
655
maoist 1
Table 5: Dep Var: Maoist
(1)
(2)
(3)
probit1
probit2
probit3
1.916*** 1.901*** 2.174***
(4)
probit4
1.468***
(5)
probit5
0.926***
(6)
probit6
0.632***
prop sandy
-2.343
1.654
4.163
24.90***
24.08***
28.40**
log stcapdistance
-0.0853
-0.0573
0.0525
0.270
-0.0195
0.134
prop barrenrocky
8.558***
12.59***
16.43***
34.48***
18.93***
33.41***
prop steepsloping
-74.89***
-60.84***
-83.24**
-63.82**
-134.0***
-108.2***
proportion forest cover
1.188**
1.295*
1.740**
1.601*
1.770***
1.600**
log area
0.572**
0.643***
0.703**
0.896***
0.753***
0.773***
log gen mpc
-0.570
-0.513*
-0.635***
-0.236
-0.340***
-0.122
log scst mpc
-1.154***
-1.042***
-1.070**
-1.009
-0.614
-0.578
land inequality
3.277***
3.196***
3.350***
2.765***
4.436***
5.535***
%Scheduled Castes & Tribes
1.077*
1.804**
-0.000150
1.760***
1.861***
population density
-63.02
-136.8**
-108.9***
21.24
4.844
income inequality
-0.731
-4.405**
-6.291***
0.802
-1.346
-0.976*
0.345
0.515
0.174
General MPC
SCST MPC
Prop. Non landlord
-1.396***
State Dummies
Time Dummies
Observations
Yes
993
Yes
993
* p < 0.10, ** p < 0.05, *** p < 0.01
29
Yes
651
Yes
426
-1.278**
Yes
Yes
Yes
765
Yes
546
maoist 1
Table 6: Dep Var: No. of Incidents
(1)
(2)
(3)
2.869*** 2.777*** 3.034***
(4)
2.301***
(5)
1.228***
prop sandy
-16.40
-11.72
-12.20
4.016
8.007
log stcapdistance
-0.305*
-0.244
0.0183
0.141
-0.104
prop barrenrocky
20.33***
31.37***
36.70***
34.73***
21.40***
prop steepsloping
-119.1***
-97.48***
-129.9***
-147.1***
-133.1**
proportion forest cover
2.594**
2.511**
2.743**
2.559*
2.847***
log area
1.589***
1.778***
1.841***
2.486***
1.558***
log gen mpc
-1.201***
-0.932
-0.683
-0.306
-0.700***
log scst mpc
-2.079***
-1.710***
-1.669***
-2.161***
-1.161*
land inequality
7.425***
7.703***
7.502***
5.505**
6.465***
%Scheduled Castes & Tribes
2.611***
3.297***
-0.852
2.614***
population density
-107.2***
-142.2***
-132.2**
75.03
income inequality
-3.356
-9.596***
-13.84***
-0.549
-1.733***
-1.312
0.956
1.725
General MPC
SCST MPC
Prop. Non landlord
-1.140***
State Dummies
Time Dummies
Observations
Yes
Yes
993
Yes
993
* p < 0.10, ** p < 0.05, *** p < 0.01
30
Yes
651
Yes
426
Yes
993
maoist 1
Table 7: Dep Var: No. of dead wounded
(1)
(2)
(3)
nbreg1
nbreg2
nbreg3
2.880*** 2.771*** 3.072***
(4)
nbreg4
2.562***
(5)
nbreg5
1.076*
prop sandy
-46.41**
-39.40*
-39.59*
-37.42**
8.840
log stcapdistance
-0.521***
-0.486***
-0.0354
-0.176
-0.273
prop barrenrocky
22.31***
31.30***
39.94***
34.49***
32.08***
prop steepsloping
-191.8***
-174.3***
-207.9***
-253.5***
-278.3***
proportion forest cover
4.461***
3.712**
3.910**
5.464***
4.613***
log area
2.116***
2.315***
2.168***
3.131***
2.271***
log gen mpc
-2.113***
-1.844***
-1.106
-0.0971
-1.374**
log scst mpc
-2.848***
-2.169***
-1.994*
-2.717**
-2.063
land inequality
11.39***
12.06***
10.58***
6.911**
12.74***
%Scheduled Castes & Tribes
3.841***
4.145***
-3.230*
3.332**
population density
-88.55**
-128.7***
-147.3*
118.8
income inequality
-4.222
-12.39***
-21.41***
-2.793
-2.876***
-3.084**
0.835
1.522
General MPC
SCST MPC
Prop. Non landlord
-1.161**
State Dummies
Time Dummies
Observations
Yes
Yes
993
Yes
993
* p < 0.10, ** p < 0.05, *** p < 0.01
31
Yes
651
Yes
426
Yes
993
maoist 1
Table 8: Dep Var: Maoist
(1)
(2)
(3)
(4)
1.904*** 1.707*** 1.504*** 1.503***
(5)
1.357***
(6)
1.364***
(7)
0.611***
Prop. Non landlord
-1.047***
-1.266***
-1.526***
-1.268***
-1.436***
-1.234***
-1.230**
16.62***
18.69***
19.84***
20.33***
20.42***
31.42**
log stcapdistance
0.0838
0.151
0.164
0.188
0.192
0.193
prop barrenrocky
9.442**
25.87***
24.55***
28.18***
27.63***
33.81***
prop steepsloping
-57.69*
-27.05
-22.45
-39.45
-33.89
-107.7***
proportion forest cover
0.784
1.559**
1.674**
1.260**
1.296*
2.091***
log area
0.336
0.587***
0.579***
0.761***
0.725***
0.734***
population density
-120.6***
-125.4***
-85.26***
-78.17***
4.929
%Scheduled Castes
-1.269
-1.069
-1.498
-1.394
3.966**
%Scheduled Tribes
-0.302
-0.901
0.765
0.211
1.229**
-0.948**
-0.724
2.973***
5.391***
prop sandy
consumption pc
-1.087***
land inequality
3.014***
State Dummies
Time Dummies
Observations
Yes
Yes
698
Yes
668
Yes
668
32
Yes
655
Yes
668
Yes
655
Yes
548
maoist 1
Prop. Non landlord
Table 9: Dep Var: No. of incidents
(1)
(2)
(3)
(4)
3.378*** 2.307*** 1.991*** 2.150***
-1.089*
(5)
1.805***
(6)
1.904***
(7)
0.864**
-1.986***
-2.186***
-1.589***
-1.883***
-1.381***
-1.145**
prop sandy
-0.110
2.987
7.379
6.717
8.738
18.12
log stcapdistance
-0.251
-0.104
-0.0130
-0.0156
0.0753
0.0388
prop barrenrocky
13.46
32.29***
24.36***
45.14***
39.17***
28.30***
prop steepsloping
-77.67*
-51.75
-45.80
-69.91
-62.04
-126.9*
0.507
2.533**
2.759**
2.277*
2.257*
3.788***
1.487***
1.831***
1.817***
2.171***
2.073***
1.906***
population density
-186.3***
-192.7***
-143.1***
-113.3***
52.93
%Scheduled Castes
-0.868
-1.097
-0.906
-1.061
5.609*
%Scheduled Tribes
-0.646
-2.461***
1.200
-0.489
1.175
-2.218***
-1.842
6.632***
6.681***
proportion forest cover
log area
consumption pc
-2.365***
land inequality
6.593***
State Dummies
Time Dummies
Observations
Yes
Yes
698
Yes
668
Yes
668
33
Yes
655
Yes
668
Yes
655
Yes
655
maoist 1
Table 10: Dep Var: No. of dead wounded
(1)
(2)
(3)
(4)
nbreg1
nbreg2
nbreg3
nbreg4
3.026*** 2.104*** 1.875*** 2.012***
(5)
nbreg5
1.782***
(6)
nbreg6
1.813***
(7)
nbreg7
0.549
Prop. Non landlord
-1.911***
-2.595***
-2.732***
-2.135***
-2.534***
-2.053***
-1.207
prop sandy
-24.70
-3.391
-1.021
-0.578
-2.990
38.57
log stcapdistance
-0.585
-0.566
-0.309
-0.343
-0.172
-0.126
prop barrenrocky
17.62
43.17***
37.41**
59.70***
54.96***
46.07***
prop steepsloping
-133.4***
-114.6*
-106.4*
-148.4**
-143.4**
-337.2***
1.735
4.455***
4.649***
4.797***
4.857***
6.277***
1.728***
2.482***
2.285***
2.951***
2.722***
2.483***
population density
-281.5***
-279.4***
-181.7***
-148.0***
130.5
%Scheduled Castes
-4.150
-4.325
-3.072
-3.013
10.10***
%Scheduled Tribes
-1.441
-3.643***
1.312
-0.489
1.336
-2.431*
-3.207
10.87***
14.79***
proportion forest cover
log area
consumption pc
-2.852**
land inequality
10.86***
State Dummies
Time Dummies
Observations
Yes
Yes
698
Yes
668
Yes
668
* p < 0.10, ** p < 0.05, *** p < 0.01
34
Yes
655
Yes
668
Yes
655
Yes
655
maoist 1
Table 11: Dep Var: Maoist (Growth)
(1)
(2)
(3)
2.375***
2.109***
2.161***
(4)
1.468***
(5)
1.069***
prop sandy
-1.552
3.260
3.244
25.81***
43.60
log stcapdistance
-0.0501
0.000879
0.00972
0.200
0.150
prop barrenrocky
3.563
15.00***
17.17***
34.31***
27.98***
prop steepsloping
-54.89***
-71.01***
-89.14***
-69.43*
-110.3***
proportion forest cover
0.958
0.779
1.483*
1.687**
2.119**
log area
0.0728
0.581*
0.542*
0.745**
0.942***
-0.0361***
-0.0509***
-0.0582***
-0.0344
-0.00706
%Scheduled Castes
0.399
0.350
-1.227
2.205
%Scheduled Tribes
2.355***
1.599**
0.975
2.174***
population density
-82.71
-83.13
-103.0***
-6.836
gini growth
-0.00143
-0.00375
-0.0159
-0.0226*
land inequality
3.984***
3.776***
3.207***
4.158***
-1.612**
-0.172
mpc growth
initial consumption pc
Prop. Non landlord
-1.539***
State Dummies
Time Dummies
Observations
Yes
Yes
640
Yes
639
35
Yes
637
Yes
419
Yes
500
maoist 1
Table 12: Dep Var: Maoist (Growth by subgroups)
(1)
(2)
(3)
(4)
2.441***
2.187***
2.319***
1.584***
(5)
1.110***
(6)
0.732***
prop sandy
-1.552
2.655
4.493
24.91***
43.48**
47.46**
log stcapdistance
-0.0524
0.0323
0.0445
0.262
0.149
0.242
prop barrenrocky
3.282
13.64**
15.28***
32.78***
27.69***
37.51***
prop steepsloping
-58.38***
-68.27***
-84.31**
-61.28*
-112.0***
-84.32*
proportion forest cover
1.098*
1.014
1.710**
1.949**
2.173***
2.368***
log area
0.0821
0.517
0.521
0.691**
0.935***
0.827***
General MPC growth
-0.00709**
-0.00884*
-0.0168*
-0.0128
-0.00247
-0.00749
SCST MPC growth
-0.0275**
-0.0314**
-0.0294**
-0.0229
-0.00616
-0.00923
2.208***
2.089***
0.250
2.129***
2.347***
population density
-67.85
-110.7
-95.94***
-4.027
-15.14
gini growth
-0.0123
-0.0276***
-0.0305***
-0.0233*
-0.0185
3.653***
3.086**
2.668***
4.240***
5.059***
General MPC
-1.935***
-0.176
SCST MPC
-0.00814
-0.309
%Scheduled Castes & Tribes
land inequality
Prop. Non landlord
-1.574***
State Dummies
Time Dummies
Observations
Yes
636
Yes
636
* p < 0.10, ** p < 0.05, *** p < 0.01
36
Yes
633
Yes
415
-1.504***
Yes
Yes
Yes
497
Yes
355
maoist 1
Table 13: Dep Var: No. of incidents
(1)
(2)
(3)
nbreg1
nbreg2
nbreg3
3.512*** 3.066*** 3.180***
(4)
nbreg4
2.264***
prop sandy
-13.70
-7.699
-13.36
-2.861
log stcapdistance
-0.289
-0.114
-0.00654
0.0991
prop barrenrocky
2.695
32.27***
37.39***
40.74***
prop steepsloping
-82.08**
-78.88***
-126.5***
-157.3*
proportion forest cover
2.242
2.099
2.322*
2.010
log area
0.674
1.458***
1.578***
2.134***
General MPC growth
-0.00500
-0.0186
-0.0244
-0.0255
SCST MPC growth
-0.0537**
-0.0436**
-0.0298
-0.0443**
%Scheduled Castes & Tribes
4.307***
3.755***
0.358
population density
-123.2***
-137.6***
-122.2***
-0.0209
-0.0402**
-0.0588**
7.808***
7.227***
6.118***
-2.887***
-1.946*
0.130
0.505
gini growth
land inequality
General MPC
SCST MPC
Prop. Non landlord
Time Dummy
Observations
-1.410***
Yes
636
* p < 0.10, ** p < 0.05, *** p < 0.01
37
Yes
636
Yes
633
Yes
415
maoist 1
Table 14: Dep Var: no of dead + wounded
(1)
(2)
(3)
nbreg1
nbreg2
nbreg3
3.518*** 3.257*** 3.299***
prop sandy
(4)
nbreg4
2.337***
-33.48
-29.17
-36.71
-47.97**
log stcapdistance
-0.532**
-0.246
-0.0300
-0.0571
prop barrenrocky
-1.995
27.94***
39.11***
43.96***
prop steepsloping
-122.2***
-127.0***
-180.8***
-255.5**
proportion forest cover
3.174
2.405
3.108*
3.952**
log area
0.616
1.599***
1.867***
2.627***
-0.0114
-0.0401**
-0.0445**
-0.0354
-0.0765**
-0.0492*
-0.0319
-0.0537*
5.593***
4.784***
-0.918
population density
-102.0*
-130.2***
-118.8*
gini growth
-0.0438
-0.0628**
-0.115***
11.05***
10.31***
8.629***
-4.507***
-3.808***
-0.486
-0.231
General MPC growth
SCST MPC growth
%Scheduled Castes & Tribes
land inequality
General MPC
SCST MPC
Prop. Non landlord
Time Dummy
Observations
-1.643***
Yes
636
* p < 0.10, ** p < 0.05, *** p < 0.01
38
Yes
636
Yes
633
Yes
415
39
Table 15: Districts affected
Round 1
Conflict t-1 Conflict t+1
Not Affected
358
338
Affected
3
23
Round 2
Not Affected
339
293
Affected
23
69
Round 3
Not Affected
293
269
Affected
69
93
40
41
42
43
44
45
46
47
48
49
50
Variable
maoist
deadwounded
nincidents
maoist 1
gini
mpc 87
General MPC 87
SCST MPC 87
landineq91
p forest
pop density
log mpc
log gen mpc
log scst mpc
log area
prop sandy
prop barrenrocky
prop steepaloping
log stcapdistance
SCST percent
log mpc lag
log gen mpc lag
log scst mpc lag
p nland
Table 16: Summary: All 3 rounds
Obs
Mean Std. Dev.
Min
1085 0.170507
0.376251
0
1085 6.322581
59.3128
0
1085
2.41106
15.79039
0
1085 0.087558
0.282781
0
1046 0.266298
0.060809 0.103525
1030 5.052897
0.244983 4.410083
1030 5.132978
0.249794 4.498148
1028 4.870675
0.250028 4.180412
1055
0.47359
0.178855 0.120449
1082 0.167094
0.18037 0.000699
1082 0.006383
0.022805 0.000692
1046 5.849775
0.628196 4.410083
1043 6.002583
0.696857 4.498148
1045 5.674889
0.628981 4.180412
1082 8.721257
0.696394 6.475433
1040 0.006336
0.040049
0
1040
0.00799
0.020191
0
1040 0.002556
0.010052
0
1031 5.485296
0.811263
0
1082 0.292082
0.15533 0.016113
686 5.608404
0.620172 4.410083
685 5.727645
0.666179 4.498148
685 5.440889
0.634257 4.180412
698 0.522766
0.429809
0
51
Max
1
1837
428
1
0.525915
5.777959
6.014436
5.941787
0.788384
0.832942
0.416596
7.352093
8.501844
7.374076
10.7288
0.688315
0.265584
0.129534
6.899219
0.946497
6.848523
7.336274
7.374076
1
Variable
maoist
deadwounded
nincidents
maoist 1
gini
mpc 87
General MPC 87
SCST MPC 87
landineq91
p forest
pop density
log mpc
log gen mpc
log scst mpc
log area
prop sandy
prop barrenrocky
prop steepaloping
log stcapdistance
SCST percent
p nland
Table
Obs
361
361
361
361
350
350
350
350
351
360
360
350
350
350
360
346
346
346
343
360
232
17: Summary: Round
Mean Std. Dev.
0.063712
0.244578
1.227147
7.815471
0.171745
0.874566
0.00831
0.090907
0.268654
0.049943
5.053116
0.24615
5.134089
0.252915
4.871676
0.250661
0.473529
0.17919
0.166137
0.179325
0.00637
0.022845
5.053116
0.24615
5.134089
0.252915
4.871676
0.250661
8.72076
0.69759
0.006344
0.040126
0.007996
0.020229
0.002559
0.010071
5.485555
0.812822
0.291302
0.154554
0.52343
0.430869
52
1
Min
0
0
0
0
0.14319
4.410083
4.498148
4.180412
0.120449
0.000699
0.000692
4.410083
4.498148
4.180412
6.475433
0
0
0
0
0.016113
0
Max
1
114
9
1
0.517323
5.777959
6.014436
5.941787
0.788384
0.832942
0.416596
5.777959
6.014436
5.941787
10.7288
0.688315
0.265584
0.129534
6.899219
0.946497
1
Variable
maoist
deadwounded
nincidents
maoist 1
gini
mpc 87
General MPC 87
SCST MPC 87
landineq91
p forest
pop density
log mpc
log gen mpc
log scst mpc
log area
prop sandy
prop barrenrocky
prop steepaloping
log stcapdistance
SCST percent
log mpc lag
log gen mpc lag
log scst mpc lag
p nland
Table
Obs
362
362
362
362
335
330
330
329
352
361
361
335
334
334
361
347
347
347
344
361
351
351
351
233
18: Summary: Round
Mean Std. Dev.
0.190608
0.393324
3.743094
15.56642
1.743094
6.166011
0.063536
0.244262
0.222355
0.044069
5.050845
0.242649
5.129295
0.243505
4.867642
0.249575
0.47362
0.178942
0.167571
0.181136
0.006389
0.022816
6.18721
0.256414
6.347735
0.295582
6.03601
0.253107
8.721505
0.696764
0.006332
0.040069
0.007987
0.020201
0.002554
0.010057
5.485167
0.811668
0.292471
0.155929
5.055982
0.251596
5.137588
0.260924
4.874592
0.256194
0.522436
0.430207
53
2
Min
0
0
0
0
0.103525
4.410083
4.498148
4.180412
0.120449
0.000699
0.000692
5.413915
5.419044
5.395524
6.475433
0
0
0
0
0.016113
4.410083
4.498148
4.180412
0
Max
1
122
61
1
0.409687
5.777959
5.831726
5.941787
0.788384
0.832942
0.416596
6.848523
7.336274
7.374076
10.7288
0.688315
0.265584
0.129534
6.899219
0.946497
6.059236
6.362353
5.941787
1
Variable
maoist
deadwounded
nincidents
maoist 1
gini
mpc 87
General M 87
SCST MPC 87
landineq91
p forest
pop density
log mpc
log gen mpc
log scst mpc
log area
prop sandy
prop barre y
prop steep g
log stcapd e
SCST percent
log mpc lag
log gen mpc lag
log scst mpc lag
p nland
Table 19: Summary: Round
Obs
Mean Std. Dev.
362 0.256906
0.437532
362 13.98343
100.8417
362 5.312155
26.38211
362 0.190608
0.393324
361 0.304792
0.056793
350 5.054611
0.246686
350
5.13534
0.253162
349 4.872531
0.25051
352
0.47362
0.178942
361 0.167571
0.181136
361 0.006389
0.022816
361 6.309026
0.302151
359 6.528189
0.379725
361 6.119516
0.284002
361 8.721505
0.696764
347 0.006332
0.040069
347 0.007987
0.020201
347 0.002554
0.010057
344 5.485167
0.811668
361 0.292471
0.155929
335
6.18721
0.256414
334 6.347735
0.295582
334
6.03601
0.253107
233 0.522436
0.430207
54
3
Min
0
0
0
0
0.158474
4.410083
4.498148
4.180412
0.120449
0.000699
0.000692
5.641609
5.802069
5.494266
6.475433
0
0
0
0
0.016113
5.413915
5.419044
5.395524
0
Max
1
1837
428
1
0.525915
5.777959
6.014436
5.941787
0.788384
0.832942
0.416596
7.352093
8.501844
7.244851
10.7288
0.688315
0.265584
0.129534
6.899219
0.946497
6.848523
7.336274
7.374076
1