In recent years, donors and governments have increasingly targeted development aid to conflict-affected areas, often in the hope that aid will reduce conflict by “winning the hearts and minds” of the population. The idea is that by implementing development projects, for example by building roads, schools and hospitals, or by extending technology, cash transfers or insurance to poor people, we can increase popular support for the government and reduce support for insurgent movements. Facing a more hostile population, insurgents will find it harder to recruit fighters, acquire supplies and carry out attacks, leading to an overall reduction in violence. This idea—that development aid can be used to win hearts and minds—is widespread and forms the basis of much of the U.S. Armed Forces’ counterinsurgency strategy. Yet, there is no conclusive empirical evidence that development projects reduce violence. In fact, there is anecdotal evidence for the opposite.A recent report on civil counterinsurgency strategies by the RAND Corporation warns that “insurgents strategically target government efforts to win over the population. Indeed, the frequency with which insurgents attack schools, government offices, courthouses, pipelines, electric grids, and the like is evidence that civil [counterinsurgency] threatens them.”In a recent working paper, my coauthor Patrick Johnston and I offer a simple but frequently overlooked explanation for why the strategy of winning hearts and minds often backfires: if insurgents know that successful development projects will weaken their position,container raspberries they will try to derail them, which may exacerbate conflict. To help us think more clearly about this mechanism, we developed a simple theoretical model of bargaining and conflict around development projects.
The model’s premise is that the government tries to implement a development project while the insurgents threaten to use force to derail it—perhaps by attacking government staff or infrastructure, or by intimidating the population into not participating in the project. The model assumes that the government and insurgents engage in negotiations, during which the government can pay off the insurgents in return for allowing the project’s peaceful implementation. However, the insurgents know that a successful project will win the hearts and minds of the population and will make it harder for insurgents to achieve their political aims in the future. Thus, if the government wants to convince the insurgents to leave the project in peace, it has to compensate them for the shift in power that a successful project will bring about.Previous theoretical work on the causes of conflict has shown that a large shift in power between two parties can cause bargaining to break down. Our model shows that if a project causes a shift in power that is large enough, the government may not be willing or able to compensate the insurgents and conflict will occur. While this theoretical modeling exercise may seem somewhat abstract, it allows us to predict the conditions under which development projects are most likely to cause conflict. First, conflict is more likely if a successful project causes a large shift in the balance of power between insurgents and the government. Second, conflict is more likely if insurgents have a strong military capacity that they can use to effectively derail the project . I will come back to these insights at the end of this article and discuss what they can tell us about the best way to implement development projects in areas affected by conflict.To test the predictions of our theoretical model, we estimate the causal effect of a large development program—the Philippines’ KALAHI-CIDSS program— on casualties in armed civil conflict. From 2003 until 2008, KALAHICIDSS was the Philippines’ flagship anti-poverty program with a budget of $180 million, financed through a loan from the World Bank.
The program distributed grants for small infrastructure projects to the poorest 25% of municipalities in the 40 poorest provinces of the Philippines. In doing so, it followed a community-driven development framework that allowed the population to propose projects and decide which projects to fund through a participatory democratic process. We estimate the effect of this program on the ongoing conflict between the government of the Philippines and the country’s two largest organizations: the communist New People’s Army and the Muslim-separatist Moro Islamic Liberation Front . The New People’s Army is the armed wing of the outlawed Communist Party of the Philippines, a class-based movement that seeks to replace the Philippine government with a communist system. Since taking up arms in 1969, the NPA has relied on guerilla tactics rather than conventional battlefield confrontations against government armed forces. Its current strength is estimated at 8,000 armed insurgents who operate in rural areas all over the Philippines. The Moro Islamic Liberation Front is a separatist movement fighting for an independent Muslim state in the Bangsamoro region of the southern Philippines. It was formed in 1981, when the group’s founders defected from the Moro National Liberation Front, another long-standing southern Philippines insurgent movement. The MILF’s core grievances stem from disputes over lands considered by the southern Muslim population to be part of their ancestral homeland. With an estimated 10,500 fighters under arms, the MILF is larger than the NPA. However, the MILF has a more narrow geographic focus and only operates in parts of the southern island of Mindanao. Overall, conflict with these two groups has been ongoing for over four decades, caused more than 120,000 deaths, and cost the country an estimated $2–3 billion. We had access to information on all conflict incidents that involved units of the Armed Forces of the Philippines between 2001 and 2008.
These data were originally collected for the AFP’s own intelligence purposes, but a declassified version has recently been made available to researchers.Estimating the causal effect of development projects is difficult under any circumstances, and particularly so in conflict-affected areas. To cleanly identify the causal effect of development aid on conflict, one would optimally like to compare two places that are exactly identical in all characteristics, except that one of them received aid while the other did not.Since this is not possible in the real world, researchers usually use regression analysis to “control” for differences in observed variables. By controlling for a variable in a regression, we can “hold its effect constant,” which allows us to compare places that differ in the variable as if they did not. If we were able to measure all the differences between places that receive aid and places that do not, we could control for them in a regression and filter out the pure effect of development aid on conflict. Unfortunately, this is virtually impossible in the real world since many important variables are hard or impossible to measure. We may, for example,draining pots be able to measure and control for differences in demographics, poverty and access to infrastructure, but crucial variables like the strength and militancy of local insurgents and their level of support in the population are nearly impossible to measure. If these unmeasured variables differ systematically between places that receive aid and places that do not, we run the risk of misinterpreting these differences as the causal effect of aid, which would lead us to the wrong conclusions. For example, suppose an aid agency is worried about the safety of its staff and therefore targets aid to places with little or no insurgent presence. In this case, we would most likely find that the places that receive aid from this agency experience less conflict than the places that do not receive aid. However, this does not mean that aid caused a reduction in conflict, but merely that the agency targeted aid towards places that had a low propensity for conflict to begin with. The key to estimating causal effects is therefore to ensure that one is comparing like with like—i.e., that the places one is comparing do not differ in unobserved variables. To overcome this challenge and cleanly identify the causal effect of the KALAHI-CIDSS program on violent conflict, we employ a statistical method called Regression Discontinuity Design . This approach ensures that one is comparing like with like by exploiting arbitrary thresholds in the targeting of interventions. In our case, eligibility for the KALAHI-CIDSS program was restricted to the poorest 25% of municipalities. Thus, municipalities just below the 25th percentile of poverty were eligible and municipalities just above the 25th percentile were not. The basic idea of the RDD approach is that—since the location of the threshold is basically arbitrary—municipalities just above and just below the threshold should not differ systematically in any unobserved variables that determine conflict. We can therefore estimate the program’s causal effect by comparing the intensity of conflict in municipalities just below and just above the eligibility threshold.
The main results of our econometric analysis are summed up in the graph in Figures 1 and 2. The graphs compare the intensity of conflict—measured respectively by the number of casualties and the probability of having at least one casualty in a given month—in municipalities that were barely eligible for the KALAHI-CIDSS program and municipalities that were barely ineligible. Monthly averages of conflict in barely eligible and barely ineligible municipalities are denoted by solid and hollow circles, respectively. Smoothed time trends are plotted as solid lines for eligible municipalities and dashed lines for ineligible ones. The dashed vertical lines mark important dates in the project’s timeline. The first line at t = 0 marks the beginning of preparations for the project in eligible municipalities; the second line marks the start of the project’s implementation six months later. The third vertical line marks the project’s scheduled end after three years. The graphs show that both eligible and ineligible municipalities experienced similar levels of conflict in the period before the project. However, at the start of the project preparations, conflict increased sharply in eligible municipalities but remained virtually unchanged in ineligible municipalities. The difference in the intensity of conflict then became smaller over time and virtually disappeared as the project ended. Overall, the graphs suggest that the KALAHI-CIDSS program caused a large increase in the intensity of conflict over the three years of its duration. The regression results that correspond to these graphs, which are presented in detail in our paper , suggest that the KALAHI-CIDSS program caused a 70– 90% increase in the number of conflict casualties in eligible municipalities. In aggregate, we estimate that the program caused approximately 500 excess casualties over the three years of its duration. Our regression analysis also shows that eligible and ineligible municipalities did not significantly differ in pre-program or post-program levels of conflict, which supports our claim that the observed differences are really due to a causal effect of the program and not due to systematic differences in unobserved variables. Additional results show that the majority of casualties were suffered by insurgents and government troops, while civilians appear to have suffered less. We further find that the program caused similar increases in insurgent-initiated and government-initiated violence, suggesting that the effect is not the result of a one-sided offensive by either party.Our research makes two contributions to the study of civil conflict in developing countries. First, it provides empirical evidence that development projects can cause violent conflict. This evidence is particularly strong because our method of analysis is able to overcome the central problem of causal inference: that places that do and places that do not receive aid differ systematically in important unobserved variables such as the strength of local insurgents. By exploiting a discontinuity in the targeting of aid—the fact that only the poorest 25% of municipalities were eligible—we were able to compare municipalities that were barely eligible for aid with municipalities that were barely ineligible. Since the threshold at the 25th percentile was chosen arbitrarily, barely eligible and barely ineligible municipalities should not differ in unobserved variables, so that the difference in conflict between them reflects the causal effect of the development project. Of course, even though our results show that development aid can cause conflict, they do not suggest that we should stop giving aid to conflict affected areas. Many of the world’s poorest and most vulnerable households live in areas affected by conflict and cutting them off from aid would be throwing out the baby with the bath water. However, we believe that our theoretical model allows us to draw some conclusions about how to implement development projects while avoiding conflict.