Poverty, Social Divisions, and Conflict in Nepal

The authors conduct an econometric analysis of the economic and social factors which contributed to the spread of violent conflict in Nepal. They find that conflict intensity is significantly higher in places with greater poverty and lower levels of economic development. Violence is higher in locations that favor insurgents, such as mountains and forests. The authors find weaker evidence that caste divisions in society are correlated with the intensity of civil conflict, while linguistic diversity has little impact.


Introduction
Since 1945, more than 70 civil wars have occurred around the world, resulting in approximately 20 million deaths and displacing more than 67 million people. 1 Understanding the causes of the onset and continuation of such conflicts is therefore of great importance, and the analysis of the determinants of civil conflict has been the subject of a recent and growing literature in economics. 2 There have been several cross-country studies of the proximate correlates of civil war: Collier and Hoeffler (2004) and Fearon and Laitin (2003) both find that poorer countries face a greater risk of civil conflict. Miguel et al. (2004) use instrumental variable analysis to corroborate this view that lower economic growth leads to more conflict. However, this finding is subject to varying interpretations: while Collier and Hoeffler (2004) interpret poverty as proxying for the opportunity cost of rebel recruitment, Fearon and Laitin (2003) emphasize the lack of state capacity in poor countries. These studies also differ in their findings on the impact of other factors: Collier and Hoeffler find that civil conflicts are more likely in countries which are dominated by one ethnic group and which have a larger stock of easily expropriated primary commodities, while the other two studies find no impact of ethnic divisions or primary commodities on the likelihood of conflict.
The results from cross-country studies are subject to the caveat that civil conflict in one country might have very different causes and characteristics than conflict in another. 3 Further, the data on civil conflicts and other variables may not be strictly comparable across countries (Sambanis, 2004). There have been a few within-country studies of civil conflict. Barron et al. (2004) examine village-level conflicts in Indonesia, while Deininger (2004) examines the causes and consequences of civil strife in Uganda. However, both these studies focus on a local level of conflict and construct measures of conflict based on subjective interviews.
They do not study organized conflict like insurgencies or civil wars, and hence are not strictly comparable to the cross-country studies. The set of case studies in Collier and Sambanis (2005)  1 provide richer and more detailed within-country qualitative studies to complement the crosscountry large-sample studies; however, they are not strictly comparable to the cross-country regression-based analyses.
Our study examines violent conflict within one country, Nepal, and applies the crosscountry regression methodology to a within-country context. This approach mitigates many of the concerns regarding cross-country analysis: first, since we are able to study the spread of the same conflict to different parts of the country, several variables which differ across countries such as the existing political system, the aims of the rebel movement or the involvement of other countries, are held constant in such a study. One implication of such approach is that we are not able to examine the impact of these factors, which can be quite important in many contexts. However, these variables are typically not controlled for in many cross-country analyses of conflict. 4 Second, we are able to construct finer measures of conflict intensity based on the number of casualties and other human rights abuses, rather than just using a dummy variable for whether an area experiences conflict or not. Third, we are also able to separately analyze conflict-related deaths caused by activist forces and by the state, and study the evolution of the conflict intensity over time (cross-country studies usually do not go further than dating the onset and end of civil wars). Fourth, we go beyond existing empirical analyses of the conflict in Nepal (Murshed and Gates, 2005;Bohara et al., 2006;Acharya, 2007) by constructing detailed measures of social divisions, based both on caste and linguistic diversity in the country, and by focusing on the role of poverty.
Starting in 1996, Nepal was affected by the "People's War" started by Maoist activists, whose main aim was to establish a republic and change the Constitution. The activists used guerilla warfare techniques to attack government facilities and officials. This conflict spread to all parts of the country, and resulted in a death toll of more than 13,000 people nationwide. Large parts of rural Nepal were under the control of the activists, though the district headquarters remained under the control of the state. In 2006, the Maoists began peace talks with the government. 4 An exception is Sambanis (2001), which examines the role of neighboring country characteristics on civil war incidence.

2
In our empirical analysis, we consider a wide range of economic and social variables which are hypothesized to affect the likelihood of violent conflict, including geographic characteristics, economic development, social diversity and government investment in infrastructure. We find that geographic conditions (presence of mountains and forests) explain a quarter of the cross-district variation in conflict intensity. Controlling for geography, the lack of economic opportunities, measured by higher poverty or lower literacy rates, is significantly and robustly associated with a higher intensity of violent conflict: a 10 percentage point increase in poverty is associated with 23-25 additional conflict-related deaths. Poverty is significantly associated both with the onset of the conflict, as well as its continuation. However, this relationship is not constant over time: both state forces and Maoist activists inflict the greatest casualties in the poorest areas in 2002, when the conflict escalated. As the activists gained control of the poorest areas over time, the highest levels of conflict-related deaths were experienced in the somewhat better-off areas. There is some evidence that greater social polarization (measured by the caste diversity of Nepalese society) is associated with higher levels of conflict. However, this relationship is much less robust than the relationship with poverty; in fact, part of this relationship appears to be due to the fact that social polarization contributes to greater poverty, perhaps by limiting access to economic opportunities for the lower castes. Measures of linguistic polarization or fractionalization have no significant impact on conflict intensity.
The paper is structured as follows: Section 2 provides a brief description of the "People's War" in Nepal, section 3 describes the main hypotheses we examine and the related literature on the Nepalese conflict. Section 4 describes our data and empirical strategy, section 5 summarizes our findings and section 6 concludes.

The Maoist "People's War" in Nepal
Nepal is a poor country with a population of 27 million and a per capita GDP of $270 in 2005. As of 2004, 31% of the population lived under the poverty line. The country is located between India and China, is land-locked and has few natural resources. Agriculture is the 3 major driver of the economy, contributing 40% of GDP in 2005, and employing two-thirds of the workforce. 5,6 A large number of Nepalis have migrated to other countries in search of economic opportunities and remittances constituted 12% of GDP in 2004. Politically, Nepal was a monarchy till 1990, when widespread protests led to the establishment of a multi-party democracy. The first parliamentary elections were held in 1991, and two further general elections were conducted in 1994 and 1999. Even though the Nepalese Congress obtained a majority in the elections of 1991 and 1999, Nepal has faced considerable political instability and governments have been short-lived. Since 1991, there have been 12 governments in as many years.
The Maoist insurgency was started on February 13, 1996, by members of the Communist Party of Nepal-Maoist (CPN-M), with an attack on a police post at Halleri in Rolpa district of Western Nepal. The genesis of the insurgency in the districts of Rolpa and Rukum has been attributed to several factors, including the poverty and general underdevelopment of the area, long-standing grievances against the government (for banning the cultivation of hashish in the 1970s, and the crackdown on CPN-M activists during 1994, among other issues), and a long-standing presence of Communist activists in the area (Gersony, 2003). Since 1996, this "People's War" has spread to most other districts of Nepal, with 73 out of 75 districts reporting casualties in 2004. The Maoists have carried out several attacks on army barracks, police posts, village council buildings and banks.
The chief objectives of the activist movement were to establish a people's republic and set up a constituent assembly to draft a new constitution. In particular, this would mean curtailing some or all of the existing powers of the monarchy. The 75 point manifesto released by the Maoists in November 2001 listed several other aims, including the distribution of land to poor and landless people, equal treatment for all castes and languages, and equal rights for women. Despite being named after Chairman Mao, the activists' activities have been condemned by the Chinese government, and there is no evidence that the activists enjoyed any support from China (Raj, 2004

4
[4] Weatherhead Center for International Affairs similar activist groups in India; the 75 point manifesto promised to support all groups fighting against "Indian expansionism". Estimates of the size of the activist forces varied widely from 7,000-10,000 (Bray et al, 2003) to 10,000-15,000 (BBC report). They were mainly financed by armed raids on banks, as well as taxes on the local population and remittances from abroad.
The Maoists were reported to have set up local governments and local courts in some of the areas under their control.
In the first few years of the insurgency, the response from the government was to use the existing law and order framework to address the problem; more than 1000 people were reported to have been arrested and tortured by the government in 1999 (INSEC, 1999). In several areas, the state attempted to suppress activist activity: the graph in Figure 1 shows that in 1998 and 1999, more people were killed by the state than by the Maoist activists. The political situation changed dramatically in 2001. In June 2001, Crown Prince Dipendra allegedly killed his father King Birendra and most members of his immediate family, including himself. His uncle, Gyanendra, succeeded to the throne. The then Prime Minister Sher Bahadur Deuba declared a truce with the Maoists and a first round of talks was held. However, the Maoists unilaterally broke the cease-fire in November 2001, when they simultaneously attacked army barracks in the mid-western region (Dang) as well as the eastern region (Salleri). Deuba then imposed a state of emergency, declared the Maoists to be a terrorist group, and mobilized the Royal Nepal Army to counter the insurgency. The intensity of the conflict escalated sharply after this, with more than 3,000 people being killed by the state forces in the next year (see The total number of casualties claimed by the "People's War" amounts to more than 13,000. There is considerable variation in the intensity of conflict in different regions of Nepal.
The number of casualties is highest in the mid-Western region where the conflict began, with more than 4,000 deaths between February 1996 and December 2004. In contrast, the Eastern Development region has suffered less than 1,500 casualties in the same period. The map in Figure 2 shows this variation in the intensity of the conflict, while Figure 3 shows the spread of conflict over time. The next section considers some hypotheses about which areas might be more prone to civil conflict.

Possible causes of civil conflict
There is some diversity of views regarding the causes of the conflict in Nepal. Some reports and studies have suggested that the Maoist conflict has found support from the oppressed lower castes (Bray et al., 2003), portraying the insurgency as stemming from "rage against a 6 long legacy of oppression based on caste and ethnicity" (Sengupta, 2005). Deraniyagala (2005) and Murshed and Gates (2005) focus on landlessness and relative deprivation as proximate correlates of the conflict, while Macours (2006) documents the relationship between increasing inequality and the number of abductions by the Maoists. 7 However, other accounts such as Thapa and Sijapati (2004) attribute the conflict mostly to poverty and underdevelopment of the country. In an interview-based approach, Gersony (2003) finds that caste and ethnic divisions are not a major contributor to the conflict. There is also the possibility that government repression might have generated further grievances which led to greater support for the Maoist activists. Examples include the crackdown against CPN-M political activists as part of a highly contentious police operation (Operation Romeo) in the districts of Rolpa and Rukum in 1994. Collier and Hoeffler (1998) suggest an economic framework for the general causes of civil conflict: in their view, conflict is most likely where economic gains are high, and opportunity costs are low. This framework suggests the following hypotheses: first, a higher per capita income or better job prospects should reduce the risk of insurgency, since these increase the opportunity costs of joining the activists, and hence the wages activists need to pay to recruit people to their ranks. Second, conflict is higher where there are easily expropriable resources such as low-weight high-value primary commodities (metals, drugs, precious stones) or where rents from immovable assets (e.g. oil resources) can be easily extracted. They find some support for their framework in cross-country regression analysis; however, the pure "opportunity cost" view of natural resources has been challenged and further refined both by case studies (Collier and Sambanis, 2005) and by more detailed regression analysis (Humphreys, 2005). Fearon (2005) presents an alternative model of insurgency and state response, in which the link with poverty happens through much more indirect channels, since a lower opportunity cost also means lower gains from winning the conflict. Fearon and Laitin (2003) focus on 7 Deraniyagala (2005) is a descriptive study without systematic empirical analysis at the district level. Murshed and Gates (2005) documents the differences in income across various caste categories, but does not investigate the impact of these social divisions on conflict intensity. The estimates in this study also suffer from multicollinearity, since the authors include schooling, life expectancy and the Human Development Index in the same regression. 7 the conditions which specifically favor insurgency. Since insurgents are usually numerically weak compared to the governments they are fighting against, they must be able to hide from government forces and be able to garner local support for their activities. This suggests a number of testable hypotheses: first, the presence of mountainous or forested terrain, poorly served by roads, should increase insurgency. Second, insurgency should be more common in rural areas with low migration, where the rebels can easily punish people who do not support them. Third, the need to form a cohesive fighting force implies that insurgencies should be less common in areas of greater linguistic diversity and fewer means of communication.
Several cross-country studies have suggested that diverse societies are more prone to conflict (Easterly and Levine, 1997;Horowitz, 1985). Collier and Hoeffler (2004) also lay out a "grievance" framework, where the presence of different ethnic groups is likely to lead to the presence of significant animosity between groups; this is especially likely when one ethnic group tends to dominate over the others. Lee et al. (2004) find that the impact of democratization on human rights is limited by the presence of ethnic minorities. Several studies including Alesina, Baqir and Easterly (1999) and Miguel and Gugerty (2005) have also documented the inability of diverse societies to work together to obtain public goods.
Nepal has a very diverse society in several dimensions. Although the majority of the population is Hindu, there are deep caste divisions in the Nepalese society. Discrimination and human rights abuses against the lower castes are not uncommon (Human Rights Watch, 2004).
Hence, in the Nepalese context, the grievance hypothesis suggests that areas with greater caste diversity or a greater dominance of high castes should have more conflict. Nepal also has considerable linguistic diversity, with only 60% of the population speaking Nepali. While other empirical analyses of the Nepal conflict (Bohara et al, 2006;Acharya, 2007) focus on the proportion of people who speak Nepali as a measure of social diversity, they do not construct polarization or fractionalization measures similar to the cross-country studies of civil war. 8 These prior studies also do not examine the role of caste divisions at all.
We conduct our analysis using district-level data. Nepal has 5 major regions and 75 districts.
A district can be thought of as being comparable to a county in the US in the sense that it is administratively below the region level. The average area of a district was 1948 square Profiles based on official data. Details of data sources and variable construction are provided in the Data Appendix. As far as possible, these measures have been constructed using data collected before the violent conflict began in 1996. This helps to partially mitigate the concerns regarding backward causality, such as the possibility that the conflict itself has made some areas poorer or reduced the availability of public goods.
As suggested by the opportunity cost hypothesis, we construct district-level measures of poverty and literacy to proxy for the opportunity cost of recruiting rebel forces. We also have measures of district-level infrastructure such as the number of post offices, schools, banks and health posts. Nepal has very few natural resources which can be easily expropriated (such as oil or diamonds); the only natural resource which the Maoists have expressed an interest in controlling is land. To capture geographic characteristics which make insurgency easier, we use the altitude (elevation) of the district to indicate the presence of hilly territory, as well as the proportion of district area which is forested. We construct a measure of the presence of roads as the total kilometers of road length, normalized by the area of the district. We also construct measures of government investment by looking at the change in the provision of roads and post offices in the 1990s.
To measure the extent of social divisions, we use information on the caste and language divisions in Nepalese society. We collected data on the number of people in different caste groups, as well as speaking different languages from the census of 2001. 10 We consider all castes which constitute more than 1% of the population. This gives us 76 categories, which we use to construct a diversity index. The index we use is the measure of polarization, proposed by Esteban and Ray (1994). Montalvo and Reynal-Querol (2005) find that ethnic polarization is highly correlated with the incidence of civil war in cross-country data. 11 This index attains a maximum value of 1, when the society is divided into two equal-sized groups, thus maximizing the chances of conflict. The index attains a minimum value of zero, either when the society is completely homogenous, or when every single person belongs to a different group. We also compute a measure of caste fractionalization, similar to the ethno-linguistic fractionalization measures used in several cross-country studies. This fractionalization measure captures the probability that two randomly selected persons would belong to different caste groups. We also use two alternative measures for the dominance of upper castes (Brahmins, Chhetris, Thakuris and Newars) in the district: one is simply the proportion of these castes in the population, and the second is a dummy which equals one if the proportion of these castes is between 45% and 90%. 10 We were unable to get this at the district level from the pre-conflict 1991 census; however, the nationwide proportions have changed very little between 1991 and 2001.
11 See the Data Appendix for the details of the construction of the index.

10
[10] Weatherhead Center for International Affairs Around 120 different languages are spoken in Nepal, some by very small numbers of people.
We construct a polarization index as well as a fractionalization index of linguistic diversity (similar to the ones computed for caste divisions), using all languages which are spoken by more than 1% of the population. Since the national language Nepali is the single most spoken language, it might be easier to easier to mobilize forces in Nepali-speaking areas. We thus include the proportion speaking Nepali as another measure of the ease of mobilization.
We should note that caste polarization is highly correlated with the fraction of people who speak Nepali, since the high castes tend to be Nepali-speaking. On the other hand, caste polarization is negatively correlated with linguistic polarization, suggesting that these are two different dimensions of diversity. Table 1 shows the mean of our variables. We note the large variation across districts both in the measures of conflict intensity, as well as in the potential explanatory variables. More than half of all districts have experienced 100 or more conflict-related deaths during 1996-2004, and one-third of all districts have experienced more than 150 deaths. There are also a large number of abductions by the Maoists, which probably reflects forced conscription into their ranks. We also note that, on average, 42% of the population was below the poverty line in Nepal at the time the conflict began. The next section uses regressions to investigate these correlations further.

Empirical analysis of conflict
We investigate the proximate correlates of the Maoist conflict in Nepal by running regressions of the following form:

Main results
Our main results are summarized in Table 3. Geographical factors such as elevation and the presence of forested area explain 25% of the variation in the intensity of conflict across districts (column (1)). 12 Column (2) shows that the pre-conflict poverty level of the district is 12 We also ran specifications including district rainfall and the square of the elevation; these variables are never significant. Re-running these regressions with literacy rates instead of poverty (a measure of current as well as future earnings potential) yields results confirming that economic backwardness is associated with higher levels of conflict (columns (6) and (7)). Areas with higher literacy rates are less prone to conflict; however, now we also find caste polarization to be significantly correlated with the incidence of conflict, while the proportion of advantaged castes is marginally significant. An 11 percentage point (one standard deviation) increase in literacy rates is associated with an decrease of 65 conflict-related deaths, while a one standard-deviation increase in caste polarization is associated with an increase of 34 conflict-related deaths (column (6) estimates). Replacing poverty by a measure of infrastructure (road length per sq. km) also yields a similar result: presence of infrastructure is associated with lower conflict (columns (8) and (9)). It is interesting to note that elevation becomes insignificant when we include the road length variable, suggesting that part of the correlation with elevation arises because of the difficulty of building roads in hilly areas. 14 We also show results with all these variables as regressors (though these are highly correlated with each other) in column (10).
Why do poorer areas witness a higher level of conflict-related deaths? There can be several potential reasons for this: first, consistent with the opportunity-cost view, the rebels 13 For instance, using the estimates in column (2) (4) and (5) are 24, 24 and 23 respectively. 14 Elevation and road length are significantly negatively correlated.

13
[13] Do & Iyer find it easier to recruit people to their cause in poorer areas. Second, it could be the case that the government spends more resources to prevent activist activity in richer areas (through information-gathering and methods other than killing a lot of suspects), so that the number of deaths is lower in these areas. Alternatively, it could be that the state has a lower opportunity cost of violent repression in poorer areas. Third, for the same level of Maoist and state activity, poorer people are more likely to get killed since they have fewer means to protect themselves or to move away from the scene of the conflict (the "collateral damage" view).
We are at this stage unable to distinguish fully between these alternatives. Looking at the number of abductions by the Maoists (presumably to forcibly recruit people for their side), we find the evidence to be ambiguous: poverty is not significant in explaining the number of abductions, while literacy is (Table 5, columns (7) and (8)).

Robustness checks
Appendix Table A1 runs the specification of Table 3,  Appendix Table A2 checks the robustness of our results to different measures of caste and language divisions: caste and linguistic fractionalization measures, the proportion speaking Nepali and an advantaged caste dominance dummy (which equals 1 if the proportion of advantaged castes is between 45% and 90%). 16,17 All these measures are statistically insignificant in predicting the intensity of conflict, except for the proportion speaking Nepali when we use the literacy rate as a measure of economic opportunities.
Appendix Table A3 adds additional control variables: the male-female literacy difference to capture gender differences in society (columns (1) and (2)), 18 the average distance of the district from Rolpa and Rukum where the conflict started (columns (3) and (4)), and the average distance of the district headquarters from the capital Kathmandu, which might be a proxy both for remoteness of the district, as well as the strength of the state (columns (5) and (6)). The results are also robust to dropping the four most urbanized districts from our sample (columns (7) and (8)). 19 Columns (9) and (10) correct our estimates for potential spatial correlation of errors, since there might be spillover effects from one district to another. The results are very similar to those obtained without such correction. We also ran specifications controlling for region fixed effects (the coefficient on poverty becomes statistically insignificant, but literacy remains 15 We also tried normalizing the number of schools, health posts, post offices and banks by the area of the district (rather than the population); in this case, all the infrastructure variables are highly significant predictors of the conflict. The results are available upon request. 16 Most of the cross-country literature uses a similar measure of ethno-linguistic fractionalization to capture social diversity, though the polarization measure is preferred on theoretical and predictive grounds (Esteban and Ray (2004), Montalvo and Reynal-Querol (2005) 15 strongly significant), as well as adjusting the standard errors for region-level clustering (both poverty and literacy remain significant at the 10% level of significance). 20 It is interesting that, in almost all our specifications, the relationship with caste polarization disappears when we control for poverty measures, but not when we include literacy. One possibility is that the effect of social divisions works through poverty i.e. all else being equal, an area with greater social divisions has greater poverty. This could be, for instance, because of economic opportunities being denied to people of lower castes in a more polarized society, or because the distribution of assets or incomes is also more unequal in a more polarized society (resulting in a few rich people and a lot of poor people). Regressing the poverty measure on the caste polarization measure (Appendix Table A4) reveals a positive relationship, even after controlling for geography and literacy rates, suggesting that part of the relationship of caste polarization with conflict can be explained by its impact on poverty.

Onset versus intensity of conflict
In this section, we separate out the impact of poverty and social diversity on the onset of conflict, and the intensity of conflict, once it has begun. Esteban and Ray (2007) suggest that the relationship between polarization, fractionalization and conflict is likely to be highly non-linear; in particular, the relationship with the onset of conflict is likely to be different from the relationship with the intensity of conflict, once it has begun. Looking at Table 5, columns (1) and (2) show results from probit regressions where the dependent variable is a dummy for whether the district witnessed more than 100 conflict-related deaths. In addition to a robustness check on our conflict measure, this can also be interpreted as a measure of the onset of conflict in that district. Columns (3) and (4) show similar regressions for a dummy which equals one if the district had more than 150 conflict-related deaths. Columns (5) and (6) regress the log of the number of people killed on the log of total population as well as other covariates. In all cases, we find that poverty and literacy measures are strongly and significantly correlated with the incidence of conflict in the direction we expect. 20 These results are available upon request.

16
[16] Weatherhead Center for International Affairs The coefficient on caste polarization is negative and significant for the more-than-100-deaths dummy, suggesting that very high levels of polarization in fact raise the costs of conflict.
Columns (7) and (8)  Here we find that the coefficient on poverty is positive but not significant, while the coefficient on literacy is significant; caste polarization is positive and significantly associated with the extent of abductions.
Columns (9)-(12) examine the impact of social and economic factors on the intensity of conflict, once it has started in a given region. We do this by running tobit regressions for the number of deaths per 1000 population, as well as the log number of deaths, provided these variables are above a certain threshold. 21 We find that poorer areas are not just more likely to experience conflict: they also experience a higher intensity of conflict, once conflict has started. We find that the relationship between caste polarization and conflict is non-linear: polarization is negatively related to the incidence of conflict,but is not significantly associated with the intensity of the conflict, once it has started.

Relationship between poverty and conflict over time
Since we have annual data on conflict related deaths for 1999, 2002, 2003 and 2004, we can document the relationship between poverty and conflict intensity over time, and we find that this does not remain constant. Table 6 regresses year-wise conflict deaths on dummies for the bottom three quartiles of poverty (i.e. the omitted category is the top quartile or least poor people), as well as the top three quartiles of literacy (the omitted category is the least literate people). We find that poverty or literacy rates are not significantly associated with Maoist- 21 The threshold for the number of deaths per 1000 population is 0.25, which corresponds to the lowest quartile of its distribution.The threshold for the number of deaths is 100, which corresponds to the "onset" dummy in columns (1) and (2).

17
[17] Do & Iyer caused deaths before 2002 (columns (1) and (2)), while they are significantly associated with state-caused deaths in 1999 itself (columns (9) and (10)). In 2002, when the conflict escalated, deaths caused by the Maoists and by the state are highest in the poorest quartile of districts (columns (3) and (11)); similarly, deaths caused by both sides are highest lower in the most literate districts, compared to the least literate. This might indicate a timeline in which the state was repressing Maoist activity more strongly in the poorest areas, which in turn could have led to greater support for the Maoist movement in those areas, resulting in a very high number of deaths in 2002, when the conflict intensified strongly.
In 2003, the intensity of conflict is not higher in poorer areas (though high literacy areas still show significantly lower conflict). It is interesting that by 2004, when the activists had captured a good deal of territory, we see that the districts in the second quartile of poverty show slightly higher levels of conflict (columns (7) and (15) further implies that what we observe is not just "collateral damage" i.e. poor people getting killed in conflict simply because they cannot protect themselves.

Conclusion
We conduct a within-country empirical analysis of the correlates of conflict intensity in Nepal, analogous to cross-country analyses of civil wars. Our within-country approach enables us to examine the spread of a single conflict-across different parts of the country and over time.
We find that conflict intensity is higher in places with greater poverty, and in places with