1 Violence and Conflict Negotiation: Exploring the Roles of Geography and Culture Sanjeev Kumar 1 University of Memphis Abstract Given the levels of poverty and economic development, India has not experienced a level of interpersonal and inter-group violence that have marked others comparable countries. However, there are large regional variations in the incidence of violence in India. In this paper, the main variable of interest is the percentage of a district area under rice cultivation. The labor intensive rice cultivation is more skill-oriented than the non-rice crops like wheat and corn. In the absence of the modern mechanized agricultural methods, it requires more cooperation from neighbors. I hypothesize that these features of the rice-growing areas would lead to a lower level of violence. In a linear regression framework, after factoring in the issues pertaining to selection on both observables as well as unobservables, the empirical results do support the hypothesis of the moderating influence of the rice-area variable. JEL: K00, N30, O17, O21, P5, Q00, Z10 Keywords: Murder, Violence, Identity, Rice, Geography, Biology, Social Capital, Riot, Violence against Women 1 I am especially grateful to Daniel Millmet, Isaac Mbiti, Thomas Osang, and Karla Hoff for their valuable critiques. I am also grateful to Dilip Mukherjee for sharing his field-level experience working in the rice-growing areas in India. I also wish to thank Mehtabul Azam and Vikrant Bhakta for helpful comments. Any remaining errors remain mine. Correspondence address: Department of Economics, University of Memphis, Memphis, TN 38152-3370 Email: [email protected].
42
Embed
Violence and Conflict Negotiation: Exploring the Roles of - UMdrive
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Violence and Conflict Negotiation: Exploring the Roles of Geography and Culture
Sanjeev Kumar1
University of Memphis
Abstract
Given the levels of poverty and economic development, India has not experienced a level of
interpersonal and inter-group violence that have marked others comparable countries. However,
there are large regional variations in the incidence of violence in India. In this paper, the main
variable of interest is the percentage of a district area under rice cultivation. The labor intensive
rice cultivation is more skill-oriented than the non-rice crops like wheat and corn. In the absence
of the modern mechanized agricultural methods, it requires more cooperation from neighbors. I
hypothesize that these features of the rice-growing areas would lead to a lower level of violence.
In a linear regression framework, after factoring in the issues pertaining to selection on both
observables as well as unobservables, the empirical results do support the hypothesis of the
The neglected dimension of the quality of life ---Freedom from Violence---has become a
matter of central concern the world over, especially with the spread of terrorism. In the aftermath
of `The Great Recession', politics is looking to redefine societal goals (Gertner 2010).
Understandably, a systematic thinking about the central problem of violence is quite relevant in
such endeavor. Violence remains a significant stumbling block in the road to a better future.
Having a better understanding of its predicates would help in containing the ubiquitous threat
that it poses. It certainly would also be of help in achieving a better allocation of the scarce
resources towards the violence-mitigation public policies.
Given the levels of poverty and economic development, India has a low level of violence.
However, there are large regional variations in the incidence of violence. In this paper, I
investigate the causal effect of one particular variable, the percentage area of a district under
rice-cultivation, on violence. The differential nature of institutions necessitated by the
geographical and climatic features of the rice-growing areas is investigated to come in terms with
the lower incidence of violence found there. As discussed by Dixit (2009), a state-led policy to
provide public good, like Freedom from Violence, may unravel prosocial preferences that might
in its absence be a leading cause for a low level violence. Thus, it is relevant to have a better
understanding of the causes and correlates of the existence of public goods in a region like---
Freedom from Violence---in order to contain the cost of its provision. In particular, by various
estimates rice is the most labor and skill intensive crop. That may have played role in creating
norms and behaviors that reduce the level of interpersonal and inter-group violence.
I examine various mediating factors that can explain the low level of violence found in the
3
rice-growing areas. In a linear regression framework, after factoring in the issues pertaining to
selection on both observables as well as unobservables, the empirical results are unambiguous in
their support of this hypothesis. The historical and continued dependence on rice cultivation,
thus, can be treated as a causal factor that, historically as well as at the present, may have helped
in the management of conflict negotiation in India, at both individual-level and community-level.
The main hypothesis of this paper is that the area traditionally under rice cultivation is an
independent variable of principal interest. I will present several possible explanations for the
basic mechanism at work. An individual living in a rice growing area may evolutionarily be
better at restraining violent behaviors.2 Given the recent findings about the role of self-control in
determining educational outcomes, one can posit that the ability to restrain violent tendencies
reflect the presence of a general ability for self-control. Controlling for literacy, contrary to one's
expectation, does not take away the independent effect of the rice-area variable on violence. The
finding becomes starker with the observation that individuals are usually of shorter stature and
weigh less in a rice growing area.3 A recent working paper by Bodenhorn et al. (2010) suggests
that a shorter stature makes people more prone to crime. An additional channel might be the
prevalence of marriage at an early age. Studies suggest that adolescents are more prone to
violence. Moderate to chronic offenders demonstrate a tendency to enter into criminal activity
between ages of 15 and 19 (Bodenhorn et al. 2010). Given that marriage at an early age is more
common in the rice-growing areas, one can argue that this norm might be putting a break on
youths' propensity for violence. However, controlling for this channel does not take away the
independent effect of the rice area variable.
2 It would be interesting to investigate if one finds more neural connections the prefrontal cortex for people living in
the rice growing area for many generation; thus, making it easier for them control their impulses. 3 National Family Health Survey (2005-2006) is used to get this information.
4
Additionally, the labor intensive nature of rice cultivation, historically, has given rise to the
institution of labor-exchange systems in many regions of the world. Such institution may have
facilitated the building of more stable prosocial preferences, thus creating a context for a lower
propensity for violence. I examine if the presence of more social capital contributes in lowering
the incidences of violence in the rice-growing areas.
The remainder of the paper is organized as follows: Section 2 talks about previous literature.
Section 3 talks about the differential nature of rice cultivation and institutions and norms that
have accompanied it. Section 4 presents the identification strategies. Section 5 describes data
used for the analyses. Section 6 presents the results, while section 7 does sensitivity analyses.
Section 8 discusses the results with some concluding thoughts.
2 Literature Review
In their critical evaluation of what economists have learned about the determination of
crime in the last 40 years, Dills et al. (2008, p. 1) write, "...[Economists] know little about the
empirically relevant determinants of crime". They find that factors that find supports in the US
data deemed not very relevant when taken to the international data. Much of the empirical
researches examine deterrence---an idea that policies can reduce crime by raising the expected
cost of committing crime. In the economics literature, we are not used to thinking about violence
or crime in terms of culture, identity, and prosocial preferences. In their recent work of Fehr and
Hoff (2011) and Akerlof and Kranton (2010) explore the role that culture and identity play in
influencing taste and preferences. However, there is not much research talking about the effect of
something like identity and prosocial preferences---and both being determined by culture and
geography---on capital offence like homicide.
5
Research on violent crime in India is a rarefied field of inquiry. Dreze and Khera (2000),
similar in nature to the analyses presented in this paper, discuss the causal role of gender-bias,
captured by an adverse sex-ratio at the district level in influencing violence in India. Their results
find supports in the analysis carried out in this paper too: an adverse sex-ratio makes a district
more violent. Banerjee et al. (2001) talk about the impact that the social heterogeneity has on the
crime rate in India. They find that a higher level heterogeneity is associated with a higher crime
rate. Banerjee and Iyer (2005) talk about the role of the history, which they capture by the land-
tenurial system in India in shaping the nature of conflict. They suggest that the areas under a
feudal-like land tenurial system, which principally are the rice-growing areas, are more
conducive for insurgencies. Such observations make it more intriguing that I find a lower level of
violence in the rice-growing area.
One can create a Prisoner's Dilemma-like game with recourse to violence being a
dominant strategy. With repeated interactions and in the presence of individually internalized
social norms of cooperation or non-recourse to violence, a more efficient of the equilibria---
involving non-recourse to violence---can be sustained. Akerlof and Kranton (2010) indeed talk
about limits placed by society on people's identity can be crucial determinants in people's lives.
Internalized norms and identity of being a more restrained individual ensure self-enforcing
conducts, and if this is situated `historically' in a way that individuals in the rice-growing may be
expected to have internalized such conduct, it becomes part of culture of a community. Then it
provides with loci upon which individuals response to violence inducing circumstances are put to
work (Das Gupta 2009). Such an explanation is similar in spirit to the broken window theory4 of
4 Existence of more broken windows creates an impression that the behaviors leading to it are acceptable, and hence
creates an atmosphere and incentive for similar behaviors.
6
criminal behaviors.
3 Rice Cultivation and Its Accompaniments
Rice-cultivation in India goes back at least 4,000 years. And it continues to be the most
important crop in India (Mbiti 2008). For the states in India around 3rd
B.C., rice-cultivation was
a source of revenue during the time of scarcity. The state would release its stock of rice to pull
down prices and in its wake create revenue for the state exchequer (Basham 1957).
In Sanskrit, there is a word for a ‘rich man’ that is ‘Bahuvrihi’: one who has much rice.
Basham (1954, p. 194) mentions an interesting anecdote, "The Greek travelers were most
impressed by the fertility of India's soil and the energy and ability of her cultivators. The Greeks
found it a great source of wonder that India produced two crops a year. In the wetter parts of the
land the two crops might even grow without irrigation, while in the plains a summer crop of rice
would grow during the monsoon and a second irrigation crop in the dry season." This makes it
clear the year round requirement of labor in rice cultivation that is continuing for millennia.
Working in a rice-field is ten to twenty times more labor intensive than working in an
equivalent size corn or wheat field. The annual work load of a wet rice farmer in Asia is around
3,000 hours a year (Gladwell 2009).5 It is critical to understand the cultivation process of rice, in
contrast to other crops, to elucidate the process giving rise to the emergence of prosocial
preferences, and the ability for better self-control or restraints. Bardhan (1974) and Mbiti (2008)
discuss extensively about the difference between the rice and wheat cultivation process. An
International Rice Research Institute reports that some processes particular to the cultivation of
rice require on average 30 person-days of labor per hectare, thus leading to the general difficulty
5 Talking about the culture built around rice-cultivation in the case of China, anthropologist Goncalo Santos
says, "Rice is life. If you want to be anyone in this part of China, you would have to have rice. It made the
world go around" (see in Gladwell, 2009).
7
faced by farmers in hiring enough labor during this critical phase of cultivation (IRRI, 2003).
Such scarcity of labor has given rise to labor exchange system like chuai kan labor exchange
system in Thailand (Hara 2002).6 Furthermore, historically in contrast to the communities
dependent on animal husbandry, the farmers in the rice-growing areas would never worry about
losing their crops to thieves at night. Furthermore, given the labor requirement, their survival
was crucially dependent on the cooperation of others in the community. These factors provide
the rationale for investing in the creation of prosocial preferences through development of social
and cultural narratives. It is not surprising, therefore, to find stories and narratives putting a
heavy emphasis on the need to maintain a non-violent outlook in the traditional rice-growing
areas in India.
Figure 1: Labor-intensive rice cultivation in the Mekong Delta region, Vietnam7; Figure 2: Rice cultivation
has always been based on community work (an advertisement on the Golden Rice Network)8; Figure 3: Rice
cultivation in Niger Delta9
In contrast to the group-based rice-cultivation captured in the photographs displayed here
from different part of the rice-growing areas in the world, most of the photographs for other
crops that one can find on the internet rarely display a group of labor together at work in the
The major Rice producing areas in India had for a long time been under the landlord
tenurial system until it was abolished immediately after India became independent (Banerjee and
Iyer 2003). The main explanatory variable under consideration: the percentage of a district area
under rice cultivation has not changed significantly over years. Two factors can be argued to
have played significant roles in changing the area under rice cultivation in India, (i) The Green
Revolution, which largely was in the wheat-growing area in the western and north India, and it
mostly worked at enhancing the yield, and (ii) Urbanization. Most of the rice growing area is still
rain-fed (Mbiti 2008). Urbanization is taking place at a snail's pace in India, and is largely
happening around the same cities increasing their densities. India has one of the lowest levels of
urbanization among countries of its size.
Another aspect about the rice-growing area, mentioned by Deaton and Dreze (2008), is
that the average reported hunger in the rice-belt of India is higher than the national average.
Adult men are of shorter stature and less heavy in the states with the rice growing area than those
from the non-rice growing areas. Keeping these accompaniments in mind regarding the rice-
growing areas, the potential interpretation of the main finding of this paper that individuals seem
to have a much better restraint on their violent behaviors becomes quite dramatic.
4 Identification
4.1 Empirics
The economic model of crime can be adapted to illustrate the economics of violence. The
basic economic model of crime was originally developed by Becker (1968) and extended by
9
Issac Ehrlich.10
The empirical approach adopted in this paper is to use the area under rice
cultivation as a ‘treatment’ meted out to individuals that lowers their propensity for violent
behavior. The ideal way would be to compare a rice-growing area with the same area without
rice cultivation. In the presence of missing counterfactuals, a number of methods are used to
create a meaningful comparison district (Millimet 2009; Jha 2008).
At first, a base model is estimated with a parsimonious set of covariates. This approach,
though, it suffers from obvious deficiencies, can still shed considerable light on the empirically
relevant covariates. Any instrumented or multifactorial analysis has to be particularly strong to
overcome the raw correlation (Dills et al. 2008). Thereafter, a full specification model with an
expanded set of covariates is put in operation. The sensitivity of the estimates to the inclusion of
observables would hint to a probable and significant role that unobservables might also be
playing in driving the uncovered relationship (Altonji et al. 2005). These specifications then are
compared to Dreze and Khera (2000) specification for the illustration purpose.11
The following econometric specification is used here:12
ds2ds1dsdsds Z+X TY (7)
dsY is the number of violent acts per million committed in district d nested in state s.13
Since
districts are clustered at the state level, and given that the law and order issues largely fall under
10
see online version for more elaborate discussion on this 11
Using the same model as the one used in Dreze and Khera (2000) would amount to specification bias. Many
factors, that they found insignificant, may overtime have found ways to influence violent behaviors. Their model
also does not control for any possibility of heteroskedasticity, or the state fixed effects. 12
The estimates from the box-cox model reject the logarithmic specification of the dependent variable used in the
literature. In the information based test, linktest, the estimated value of the test statistic is significant and small (see,
Cameron and Trevedi 2009). 13
We also carried out estimation with the absolute number of murder to bring the saliency of murder to the fore, and
the broad conclusions continue to hold.
10
the state jurisdiction, I also control for district invariant state fixed effects through the following
specification:
)(Z+X TY ds2ds1dsdsds s (8)
Xds consists of the variables which includes the district level average of individual
characteristics: literacy, standard of living, proportion of schedule castes and tribes, religion, and
unemployment level. The variables included as characteristics of districts (Zds) are sex-ratio,
population, indicator for left-wing insurgency, measure for inequality, urbanization,
undernutrition, diversity, presence of government infrastructure, and social custom14
to marry
girls below 18 years of age. Our main coefficient of interest is β for variable Tds, which is the
percentage of a district area under rice cultivation. It can be considered as a pre-determined
variable owing to the fact that intensive rice cultivation has been going on in India for last at
least 4,000 years, largely driven by the geographical features suited for rice cultivation.
Estimates of β might not reflect the true impact of idiosyncratic features of rice-growing areas
for the following reasons: (i) There could be measurement error in the variable, Tds; and (ii) there
could be omitted variables correlated both with the district area under rice-cultivation and with
the outcome variable. I have included the variables whose omission can potentially bias the
estimates. To tackle the measurement error problem, I also check whether the results remain
robust with only a binary measure for district area under rice cultivation.15
14
This is a remnant of slowly receding tradition of child-marriage system in India. 15
In future work, I will use a method suggested by Lewbel (1997) as a means to address both measurement error
and endogeneity issue. This method generates instruments using higher moments of the observable data.
11
5 Data
Crime data, reporting various crime statistics for all Indian districts, comes from the
Government of India publication, Crime in India by National Crime Records Bureau (NCRB).
Data on the rice area comes from the Rice Research Institute, Patna, Bihar, India. The variable
used in the analysis, is the average of rice areas for four years from 2001 to 2004. I use data on
437 districts in the 17 states constituting 95% of India's population. The same set of states that
are included in the Dreze and Khera (2000) paper. Data on the volumes (in tonnes) for cash crops
and for fruits and vegetables produce, which is used for conducting the falsification tests, comes
from The Ministry of Agriculture (1998-1999).
I use data on Murder, Rape, and Riot to capture violence. Murder data is a more reliable
among these three. The two other dependent variables are used for the exploratory purpose to see
if the violence restraining feature of the rice-growing areas also holds for (i) violence against
women, and (ii) for group-based violence like riots. The dependent variables are the number of
homicides, reported cases of rape, and registered riot cases per million in the year 1998. The size
of population is also used as a covariate to capture the effect of population growth on the
violence.16
To capture the effect of inequality, the Gini Coefficient is calculated using the Consumer
Expenditure Survey (1998) from the 55th round of the National Sample Survey (NSS).17
To
capture the effects of non-criminal income opportunity, unemployment data at the district level is
used from the NSS 55th
round of the Employment and Unemployment Survey (1998).
Unemployment is defined as number of people in the age group (15-59) actively looking for
employment. Data on the Female-Male ratio comes from the Census of India (2001).
16
India adds population approximately equal to population of Australia each year. It is important to see what kind of
stress it putting on the society, and if it gets reflected in a higher or lower incidence of violence there. 17
‘ineqdeco’ command available through users created command.
12
Undernutrition defined as the proportion of children in the age-group 0-72 months being 3-
standard deviation below the internationally accepted benchmark for the weight-for-age ratio;
data for which comes from the Reproductive and Child Health (RCH) survey conducted by the
World Bank in year 2002-2004. Data on female literacy and total literacy come from the Census
of India (2001). Diversity is defined as the ELF measure:
r
2
iri s-1 d
sir is the proportion of individuals belonging to religion r. I again have used the Census of India
(2001) to create this district level index.18
The 5th wave of the World Value Survey (2005) is used to see if one can rationalize the
interpretation of the rice-growing area variable facilitating more prosocial behaviors at the
country level.19
This survey is carried out in 57 countries in the world. Data for only 52 countries
are used with 8 largest rice producing countries. In the WVS survey, a question regarding the
level of trust is used to see if indeed one can find some difference between the traditional rice
growing countries versus non-rice growing countries.
To evaluate how rice-growing areas fare on the various standard measures of social capital, I
use a nationally representative data set, the India Human Development Survey (IHDS), 2005
collected by the University of Maryland and the National Council of Applied Economic
Research. This data is compiled after surveying 40,000 households across India. The variables of
interest were, (i) percentage of households in a district reporting being part of various social,
political, and economic organizations, (ii) their confidence in different social, political, and
18
Loosely defined, ELF measures the probability that two randomly selected persons in a country will not belong to
the same ethno-linguistic group. Mauro (1995) first introduced ELF to economics literature as a tool for measuring
ethnic diversity. 19
http://www.wvsevsdb.com/wvs/WVSAnalizeStudy.jsp
13
economic institutions, and (iii) three measures of trust, capturing general and specific context for
social trust.
437 districts in the 17 states constitute 95% of India's population. Murder rate ranges from 2.8
per million to 131 per million. Table 1 gives state-level means of the independent variables. The
state level means of the murder rates for India's major states range from 15 in Kerala to 53 in
Jharkhand.
The district area under rice cultivation (RiceArea) ranges from almost negligent (0.06%) in the
case of Rajasthan to 17.3 percent in the case of Punjab. In the traditional rice producing states of
Bihar, West Bengal, Orissa, Uttar Pradesh, and Tamil Nadu, the percentage area under rice
cultivation at the district level is positive in almost all of their districts, see Table 2. These five
states continue to be the major suppliers of rice even today.
In Table 3, the observable characteristics of the traditional and high rice producing districts are
compared to the low rice producing districts. It is clear that the districts with a significant area
under rice cultivation are more densely populated; are largely in the northern part of India; are
poorer, less urban, and are more religiously diverse. Rice dependent districts also have a higher
share of Muslims in their population: 12% of population is classified as Muslims in the high rice
producing districts, while only 10% are in other districts.20
An extreme form of malnutrition is also less prevalent in the rice-growing area. Also, there is a
negative correlation between the rice-area variable and the proportion of households with low
standard of living, a variable which captures the extent of poverty (see Table 3). The rice-
growing areas also have their girls marry below the age of 18; 33% of all marriages in the sample
have brides less than 18 years of age, and the proportion is highest among the traditional rice-belt
20
It possibly is capturing more than three century of Muslim rule in India, which primarily was dependent on the
tax-revenue from the rice-producing areas of India.
14
(52% in Bihar and 48% in West Bengal). One can surmise that the tradition of an early marriage
is a contributory factor in the low level of violence in rice-growing area. It seems that bearing
responsibility at an early age does make individuals less prone to violence.21
Furthermore, a
lower level of publically-funded infrastructure in the rice-growing areas suggests a general
absence of the modern governance infrastructure, thus leaving space for a continued and
entrenched role of the traditional institutions.
I use many strategies to control and check for the potential sources of endogeneity. Ideally, one
should use a district level panel data to control for any time-fixed effects, or have some exclusion
restrictions; but it is hard to find the variables appropriate for the exclusion restriction. In the
absence of these options, the following specifications are used, and the error correlation across
districts within a state is controlled for in all specifications:22
i) District invariant state-level fixed effects specification is used.
ii) Exclusion of those states affected by the Green Revolution, to minimize other district level
characteristics affecting the rice area under cultivation.
iii) To check the robustness of the specifications, other variables capturing violent behaviors
like `attempt-to-murder' and `kidnapping' are used as the right hand side variables to provide
further controls for the unobservables.
iv) Biprobit model is used to check the robustness of the effect to a different level of selection
on unobservables (Altonji et al. 2005).23
21
According to UNICEF's "State of the World's Children-2009" report, 47% of India's women aged 20-24 were
married before the legal age of 18, with 56% in rural areas. The report also showed that 40% of the world's child
marriages occur in India. 22
Coefficients were also estimated using data just for the biggest state in India, Uttar Pradesh (UP), to minimize the
impact of state-level heterogeneity. 23
For the variable capturing violence, a dummy variable is created for all the districts above the 50th percentile
taking value 1. For the rice-area variable, a dummy variable takes the value 1 for all the districts above the 75th
percentile.
15
However, given a low level of mechanization and with the predominance of rain-fed agriculture
in the rice producing area, the selection on unobservables should play a small role.
6. Results
6.1 Baseline
The baseline results, using the full sample, are presented in Table 4. The specifications,
displayed in Column (1-3) in Panel I, include the estimates of the effects of the rice-area variable
on all three measures of violence without controlling for any other covariates. Columns (4-6)
capture the effects of the sex-ratio in a district on all three measures of violence. In these
parsimonious specifications, the rice-area variable has the expected negative effect; for both the
murder rate and the rape rate though the effects are significant only at the 10% level of
significance. The estimate has the expected sign for the riot rate, though it is statistically
insignificant.
The sex-ratio in a district has the expected effects on the murder rate---a finding in
consonance with Dreze and Khera (2000). However, it has an insignificant effect on the other
two variables capturing the incidence of violence. In the case of the riot rate, a more favorable
sex-ratio, rather than reducing ethnic violence, seems to have a negative effect.
In Panel II, where other covariates are controlled for, the estimated coefficients on the
rice-area variable are qualitatively similar to the ones shown in Panel I, for both the murder rate
and riot rate. The estimates, however, are less pronounced in the case of violence against women.
In Panel III, where no covariates are included but the effects of the state level unobservables are
controlled for, the rice-area variable has a more pronounced effect in the case of the murder rate-
--suggesting that the state-level unobservables play a significant mediating role. In the case of
16
violence against women, controlling for the state level fixed effects makes the effect of the rice-
area variable smaller. The effect of the sex-ratio variable, after controlling for the state-fixed
effect, has the expected sign and is also statistically significant in most specifications.
In some specifications, a dummy variable is created, which switches on for the states
suffering from the left-wing extremist violence directed against both the regional and the federal
state. This variable also captures the resource swamping effect in which prevalence of the Maoist
Insurgency dilutes the justice system resources. Table 6 shows the estimates of this variable to
illustrate the effect of insurgency on interpersonal, gender, and on the inter-group violence.
Naxalite affected areas have higher murder rate; but have a positive effect on violence against
women; and seems to restrain the propensity for the group-based violence too, but the last two
effects are statistically not significant.
6.2. Controlling for State-level Unobservables
The violence restraining effect of the rice-area variable becomes more pronounced after
controlling for the state-level district invariant unobservables along with all other covariates. The
estimated effect increases from -0.533 in the case of the murder rate, to -0.918 and remains
statistically significant at the 1% level of significance. In the case of the other two variables, the
estimates are not very encouraging. In the case of violence against women, the estimate becomes
positive and insignificant, suggesting a crucial role of unobservables in driving the result found
in the parsimonious specifications. In the case of ethnic or inter-group violence, the estimated
coefficient continues to be of the expected sign but now smaller and statistically significant at the
5% level of significance. It is illustrative to compare the effects of the sex-ratio variable with
those from Dreze and Khera (2000): the estimated coefficient (-0.11) is quite close to the ones
17
reported in their paper. However, the effect of a favorable sex-ratio is much smaller (see, Col (1)
in Table 5) in comparison with the effect of the rice-area variable, (-0.918). It is reassuring to
find consistency in the estimated effects of sex-ratio spanning almost a decade and half. Such
consistency supports the optimism of both primatologists and some social scientists about a
positive role that a more balanced gender relation may have on the social health.
One of the stark findings is the positive effect of the institution or norm of early marriage
on violence. The estimated coefficient is -0.237, and it is significant at the 5% level of
significance.24
It seems that the institution of early marriage is helping society to restrain the
young adults' homicidal tendencies. However, it is not effective when it comes to the rape and
riot rate in a district. Though the coefficient continues to be negative in the case of violence
against women, it is positive when it comes to its effect on riot, though both coefficients are not
statistically significant.
The deterrence capacity of the state at a district level is proxied by the availability of
public infrastructure, which is defined by more than 60% of villages in a district having a
sufficient health infrastructure. This variable has the expected negative and significant coefficient
in the case of both murder rate and riot rate, but a positive and close to zero and statistically
insignificant estimated coefficient in the case of the rape rate.
7. Robustness Check
7.1 Controlling for Measurement Error
To control for the possible error in measurement in the main variable of interest, the rice-
24
Data is not available on the proportion of boys getting married at an early age. However, it would not be
farfetched to assume that an early marriage of a girl can act as a proxy for an early marriage of a male-child in a
district.
18
area variable; a dummy variable, RiceDummy75, is defined, which switches on for the value
above 75th quantile. These are districts which have historically and predominantly been the rice-
growing districts of the North East and of the South Indian state of Tamil Nadu.
The results are reported in Col.(2), Col.(4), and Col.(6) of Table 5. The estimates of the
coefficient on the rice area variable are very large and statistically significant for both the murder
rate and riot rate. For violence against women, the effect is of the expected sign but not
statistically significant at the conventional levels.
7.2 Non Green Revolution States
One can argue that the estimates of the effect of the rice-area variable would be corrupted
by the introduction of the Green Revolution in some of the Indian states. The states where the
Green Revolution was introduced: Punjab, Haryana, Andhra Pradesh, and Tamil Nadu, are
excluded from the sample, and the estimates are reported in Col.(1) Col.(3), and Col.(5) of Table
6. The results are qualitatively similar in all specifications, including the ones where a dummy
for the rice area variable is used to control of the potential measurement error.
7.3 Bivariate Probit Model25
Biprobit model is used to check the nature of selection of districts into rice cultivation (Altonji
et al. 2005). For the murder rate, a dummy variable is created which switches on for all the
districts above the median of the distribution. For the rice-area variable, a dummy variable is
created which switches on for all the districts above the 75th quantile. Given the low level of
25
The optimization exercise went through smoothly for the main dependent variable of interest in this paper, the
murder rate. For other measures of violence, the rape rate and the riot rate, the maximum likelihood did not
converge. Thus, this method could not be used for those two cases
19
mechanization and the dependence on the rain-fed agriculture, one would expect that the
selection on unobservables would play a small role in the case of the rice-area variable.
If the unobservables that make a district more likely to continue with its dependence on rice-
cultivation, also makes it less prone to violence, then controlling for a negative selection by
constraining ρ to a negative value, the coefficient on the rice area dummy, though continue to
show negative effect, loses its statistical significance with ρ<-0.1. Thus, allowing for a modest
amount of negative selection explains away a statistically significant effect of the rice-area
variable on violence (see Table 7). However, if the experience with the naxalite violence directed
to the state and the federal government machineries is considered, it is quite pervasive in the rice
producing area, and thus, one can argue for a positive selection into the treatment regime.26
When I fix ρ at positive level to control for positive selection, the effect of the rice-area variable
gets more pronounced.
7.4 Controls for Unobservables27
In order to control for the effects of unobserved variables, in two separate specifications I
have included: Attempt-to-Murder rate and Kidnapping rate variables. The results are reported in
Table 8. This provides another round of robustness checks to see if the estimates are driven by
unobservables. The unobservables that affect the three variables that capture violence would
26
The Naxalites or left-wing insurgents in India can be found in the rice-cultivation area. There stated goal is to
throw the state by resorting to violence. I am in the process of investigating the role that the presence of cooperative
infrastructure might have played in bringing down the transaction cost of mobilizing people to stage violent protest
against the state. 27
To rationalize the use of this estimation strategy to control for the unobservables, a simulation study was
conducted by regressing the outcome variables on an endogeneous regressor along with a variable correlated with
the outcome variable on the right hand side. The bias on the endogenous regressor showed monotonically (inverse)
relation with the extent of correlation between the outcome variables and the variable correlated with the outcome
variable.
20
affect these two variables too, thus putting these variables on the right hand side can potentially
control for the district specific unobservables that could not be controlled via the state level fixed
effects. The estimated coefficients of the rice-area variable continue to be qualitatively similar
for the murder rate and riot rate, but flip sign in the case of the rape rate. The estimates in the
case of the murder rate changes from -0.918 to -0.748, continues to be significant at 1% level of
significance, in the model with the Attempt-to-Murder rate on the right hand side. The estimate
on the rice-area variable changes to -.884 with marginally better precision; when the Kidnapping
rate is used as a variable on the right hand side.
The estimates for the others variables, discussed in the baseline section, do not get
changed in any qualitatively significant manner.
7.5 Falsification Test
It would be reassuring to show that the crops other than rice do not have the moderating
influence on violence. First, data from IHDS (2005)28
is used to create variables on the district
areas under both wheat and rice cultivation (data on only 237 districts could be used for this
analysis). When the new rice area variable is used at the place of the old rice-area variable, I find
that the effect on the murder rate continues to be negative, but not statistically significant.
However, when the wheat-area variable is used the effect on the murder rate is positive, but again
not statistically significant at the conventional level of significance (see in online appendix Table
no. 1, 2, and 3).29
28
IHDS collected data on the cultivated area under various crops for each household. From this, a district level
measure of total cultivated area is calculated. Then, the total area under rice and wheat crops is calculated for each
district to get the percentage district-area under both of these crops. The correlation coefficient between the rice-area
variable used earlier, which was available for 437 districts, and this new variable for 237 districts is 0.42. 29
The correlation coefficient between the new rice-area and wheat-area variable is -0.30: indicating somewhat lack
of overlap between the rice area and wheat area.
21
When district level dummy variables30
for the volumes of cash-crops31
and fruits and
vegetables are separately used at the place of the rice area variable, they do show positive effects
on the murder rate, but the estimates are not statistically significant (see online Table 4 and Table
5). This shows that there indeed is something special about the rice-growing areas that make
individuals there less prone to violence.
8 Discussions
8.1 Is Social Capital the Main Mediating Channel?
Theoretically, one of the most convincing accounts for a lower level of interpersonal and
inter-group violence in the rice growing areas is the presence of more social capital. Data on the
variable capturing information on the level of trust from the World Value Survey 2005-2008
(WVS), when categorized on the basis on rice-growing versus non-rice growing countries
provides an interesting pattern. Rice growing countries are found to be quite different from the
non-rice growing areas for the question such as:32
"I now want to ask you how much you trust various groups of people: Using the
responses on this card, could you tell me how much you trust... your neighborhood". With these
options as a possible answer: (1) Trust completely; (2) Trust a little; (3) Not trust very much; (4)
Not trust at all. Lower percentages of the respondents from the traditional rice growing areas
chose (3) and (4) as their response, see Table 9.
Additionally, I control for various measures of social capital used in the literature using
IHDS data set from India. In most cases, the rice area variable continues to exert an independent 30 Both dummy variables switch on for the districts above the 75th quantile of their respective distribution. 31
Major cash-crops in India are: Jute & Cotton, Coconut & Arecanut, Sugarcane, Tea, Coffee, Tobacco,
Spices and Fruits & Potato. 32
http://www.wvsevsdb.com/wvs/WVSAnalizeIndex.jsp
22
moderating influence on violence, although the effect is somewhat less pronounced. This
suggests that the lower incidence of violence in the rice-growing area indeed is mediated through
social capital variables. Table 11 and Table 12 present the results from the fixed-effect
regressions with all the measures of social capital, including three variables measuring different
aspects of trust. These trust variables measure the subjective attitudes toward informal contacts
rather than specific institutions. Table 12 contains the results without these three indicators of
trust-level in a community, as there has been found some inconsistencies in these measures by
some researchers (see, Vanneman et al. 2006).33
Many measures of social capital have the expected moderating influence on violence,
though the variable capturing the associational social capital is more important than the variables
capturing confidence in social, political, and economic institutions.34
Thus, it seems that the available social capital indicators may not be the channel through
which the rice-area variable affects violence. Or, maybe they are not able to capture a more
internalized facet of prosocial preferences, or the non-cognitive ability, like the ability for a
better self-control.
8.3 Biology: A Potential Mediating Factor
Might it be the case that, after all, it is the biology that is one of the crucial mediating
factors making individuals in the rice growing area less prone to violence? In spite of the
discredited theory of genetic disposition toward crime and violence, the literature is coming back
to evaluate the role that human biology and physiology play in creating incentive for committing
Note: Panel I uses no covariates except for the rice area variable. Panel II controls for all the covariates except the state fixed effects. Panel III uses the rice variables along with the state fixed effects. Cluster-Robust standard errors in parentheses; * p<0.01, † p<0.05, ‡ p<0.1.
Note: Cluster-Robust standard errors in parentheses; * p<0.01, † p<0.05, ‡ p<0.1; the state fixed effects are controlled for in all the columns. Other district level variables have done good jobs of capturing the effects of the state-level unobservables.
36
Table 6: Fixed Effects Models without Green Revolution States (1) (2) (3) (4) (5) (6) VARIABLES Murder Murder Rape Rape Riot Riot
Note: Cluster-Robust standard errors in parentheses; * p<0.01, † p<0.05, ‡ p<0.1; The Green-Revolution states (Punjab, Haryana, Andhra Pradesh, and Tamil Nadu) may have biased the estimates. Estimates are very similar to the estimates from Table 15.
37
Table 7: Sensitivity Analysis: Bivariate Probit Results with Different
Assumptions Concerning Correlation Among the Disturbances
* p<0.01, † p<0.05, ‡ p<0.1; ρ=0 denotes an univariate probit regression; ρhat shows the estimated correlation between the unobservables in the violence equation (ε) and the unobservables affecting the probability of the percentage of a district area under rice cultivation (ν).
38
Table 8: Fixed Effects Models with Proxy Variables to Control for
Note: Traditional Rice-growing countries: South Korea, India, China, Taiwan, Thailand, Indonesia, Vietnam, Malaysia; Non-Rice Growing Countries: Rest of the 57 countries covered in the World Value Survey 2005-2008, some countries relevant data were not available.
Table 10: Height and Weight data from the National Family Health Survey-
III (2005)
Summary Statistics: Mean
Rice Height (cm) Weight (kg)
1 163.8 (6.405) 56.7 (6.54)
0 164.8 (11.43) 58.1 (11.65) Note: UP, Bihar, Assam, WB, Orissa, and TN are the traditional rice-growing states.
40
Table 11: Fixed Effects Models with Social Capital Variables with Trust-
Note: Cluster-Robust standard errors in parentheses; * p<0.01, † p<0.05, ‡ p<0.1; Estimates are very similar to the estimates from Table 6 and Table 8. The econometric specifications are similar to the ones in Table 8 with the Attempt-to-Murder rate variable on the right hand side.
41
Table 12: Fixed Effects Models With Social Capital (without Trust Variables) (1) (2) (3) (4) (5) (6) VARIABLES Murder Murder Rape Rape Riot Riot
Note: Cluster-Robust standard errors in parentheses; * p<0.01, † p<0.05, ‡ p<0.1; Estimates are very similar to the estimates from Table 6 and Table 8. The econometric specifications are similar to the ones in Table 8 with the Attempt-to-Murder rate variable on the right hand side.
42
Table 13: Expectations of Violence: Experience versus Reality
Country Percentage of respondents who think it likely they will become victims of violence in the next year
Percentage of respondents with a household member who had been violently attacked or threatened in the last five years
Brazil 75 27
Thailand 50 7
South Africa 48 24
France 33 16
Turkey 30 9
United States 17 12
Canada 16 14
Japan 14 4
Russia 13 16
India 10 1
All countries 22 12 Note: Data comes from a global survey, which polled more than 6000 people worldwide, commissioned by Human Security Centre to Ipsos-Reid in year 2005, for 11 countries around the world: Brazil, Canada, France, India, Japan, Russia South Africa, Thailand, Turkey, the UK and the US (Human Security Report (2005), Human Security Centre, University of British Columbia, Canada