Syndicate content

Do local development projects during civil conflict increase or decrease violence?

Jed Friedman's picture

A “hearts and minds” model of conflict posits that development aid, by bringing tangible benefits, will increase population support for the government. This increased support in turn can lead to a decrease in violence, partly through a rise in population cooperation and information sharing with the government. At least one previous observational study in Iraq found that development aid is indeed associated with a decrease in conflict. A very recent working paper approaches the same question through an evaluation of community development programs in the Philippines and explores whether these local aid activities increase or decrease insurgent violence. Interestingly, the study identifies a significant increase in violence during the implementation period of the aid program.

The study, by Benjamin Crost (University of Colorado Denver), Joseph Felter (Stanford University), and Patrick Johnston (RAND Corporation), covers the first six years of a widespread community development program known by the acronym KALAHI-CIDSS. In part financed by the World Bank, the program began in 2003 and by 2009 had operated in 4000 villages located in 184 municipalities across 40 provinces in both conflict and non-conflict areas. KALAHI supports local infrastructure and institution-building activities through block grants given to municipalities. Eligible municipalities were determined by a municipal level poverty score where the lowest 25% of municipalities in each province were invited to participate in the program. The transfers were substantial – grant size totaled 15% of the average municipality’s annual budget.

The Philippines is a setting where relatively low-levels of insurgent violence have simmered for quite some time. The two largest insurgent groups in the country over the study period were the New People’s Army, a militant wing of the Communist Party, and the Moro Islamic Liberation Front. Various smaller groups were also somewhat active. A real strength of this study is the use of micro data on the incidence and extent of insurgent conflict. Most conflict data exist only in aggregate form, such as battle deaths per country-year, and thus cannot speak to the impacts of micro-level interventions. In this case the authors compiled data from all official reports of conflict incidents that involved the Philippines Armed Forces over the period 2001 – 2008.

The authors leverage program assignment by poverty score to identify program impact on local violence through a Regression Discontinuity Design. Familiar to many of our readers, this method assumes that municipalities on either side of the eligibility threshold do not differ with respect to any observed or unobserved factor that may cause conflict. (Initial data checks do confirm that the municipalities in the vicinity of the cut-off score do not differ across observable dimensions.)

What did the authors find? It turns out that KALAHI municipalities experienced a statistically significant increase in violence compared to non-eligible areas. Moreover the timing of the violence is rather remarkable – while equal levels of violence are seen in treatment and control areas before the project start date, violence ticks up soon after the start of local project preparation. The level of violence remains elevated until the project ends whereupon violence declines to the level measured in non-project municipalities. The estimates suggest that, over the three year lifecycle of a project, treated municipalities experienced an additional 3 casualties due to insurgency related violence.

Because conflict is not in every municipality, in fact it occurs in only a minority of municipalities, the impact estimates are identified off of comparatively few observations. Nevertheless the results are precisely estimated and are robust to alternative conflict measures. For example a KALAHI project increases the likelihood of having any conflict related casualties (a measure less sensitive to outliers than the casualty count) by 13 percentage points, which is quite a large increase since the mean likelihood of any casualty related violence is 29%.

To explain these seemingly counter-intuitive results, the authors sketch a theoretical framework where development projects may induce conflict if (1) the completion of a successful project will shift population sentiment towards the government and (2) violent action by insurgents has some likelihood of derailing the project.  The observed timing of violence is exactly as Crost, Felter, and Johnston predict since it is only during the project implementation period when insurgents can hinder the likelihood of successful implementation.

Aid programs take on various shapes and sizes. Might all aid projects in conflict settings bring an increased chance of violence? Not necessarily. The authors predict that development projects would only cause violence if insurgents have the ability to derail them. It may be that local infrastructure investments are relatively easy to disrupt through violence or the threat of violence (at the very least these projects present a tangible target). Many other common development programs, such as agricultural input subsidies or conditional cash transfers, are much more costly to disrupt since the beneficiaries are widespread as well as harder to identify.  Consequently we would expect a temporary increase in violence around only some types of local development projects. In the coming years I am sure that other evidence will emerge to test this proposition.

 

Comments

Submitted by Gary Milante on
Interesting results and very useful summary, Jed. It is not surprising that there is more conflict during the project, we're inserting more wealth and "raising the stakes" of success in the local community and the model proposed by the authors sounds reasonable. It is a bit surprising that there is not less conflict after the program ends - I wonder what the definition of "after" is here - months or years? Also, did the authors look at success rates of the projects? Were the insurgents successful in interrupting development? Looking forward to reading the paper.

Gary: Thank you for your interest in my paper with Ben Crost and Joe Felter, and for your comment above. A few thoughts in response to your comment and question: "It is a bit surprising that there is not less conflict after the program ends - I wonder what the definition of "after" is here - months or years?" One thing to bear in mind is that the mission of KALAHI-CIDSS and most CDD programs isn't conflict resolution--at least not explicitly. So perhaps it shouldn't be a surprise that KALAHI isn't associated with reduced conflict. Yet many believe development programs should reduce conflict regardless, because they can increase income, help build institutions, and improve governance--all of which are generally unfavorable to insurgency. But as we illustrate in the paper, what this conventional wisdom fails to capture is that programs risk exacerbating conflict for these very reasons--militants have an incentive to prevent programs from achieving these outcomes in order to avoid being made worse-off. So once a program ends, this incentive diminishes or goes away altogether, and so we shouldn't expect higher conflict in program areas. This is consistent with what we see empirically--after KALAHI ended in 2005 or 2006 in the initial batches of municipalities where it was rolled out, the difference in conflict between eligible and ineligible municipalities in subsequent years (through 2008) was close to zero and wasn't statistically significant (see Table 5 on page 40 of the paper). This does not, of course, imply that socioeconomic benefits of KALAHI couldn't have a violence-reducing effect over the longer term, but it is consistent with the short-term incentives story laid out in the paper.

Submitted by Homira Nassery on
From work we did in Helmand province of Afghanistan, we found that much of the violence occurring during development project implementation had to do with sharing the spoils of the project's revenue in terms of employment for family/tribal members, rent-seeking behavior on the part of grassroots predators/strongmen, and other factors not related to insurgent activity. Sarah Chayes' book "The Punishment of Virtue: Reconstruction in Post-Taliban Afghanistan" conducts a good analysis of this behavior using the Kandahar Air Force base as an example - the contract for the entire circumference of the base's perimeter was given to one tribe, and that one tribe controlled everything going in and out into the base, which resulted in huge revenues for this one tribe, naturally others felt excluded from the goodies and attacks ensued. This is not related to the positive effects of the project/construction itself, but rather to the lack of inclusiveness and equity (as well as pre-construction political economy analysis) in the bidding process. In addition, we found our tractors and other heavy equipment in road projects burned or destroyed in Helmand - many times because we used an outside contractor. Wolfgang Koehling from INTSC can shed more light on these phenomenon.

Submitted by Scott Guggenheim on
Interesting stuff. It will be useful to see the paper itself. Two quick questions come to mind. Both the NPA and Moro conflicts are relatively confined compared to the nationwide scope of Kalahi. So in their areas of operation, the impact of Kalahi on exacerbating conflict must have been very, very high to be detected amidst a nationwide sample. I'm sure this is accounted for in the paper, but it'll also be important to make sure that what we're seeing isn't just stepped up conflict overall. Since Kalahi was supposed to be targeted on the poorest municipalities and because the insurgents are disproportionately concentrated in remote, hillside, and marginal areas, without proper statistical controls this could otherwise just be covariance. However, it does feel a bit silly of me to comment (at all!) without actually having the paper!

... the paper is linked in the original post, and you are right that the results are essentially estimated off of comparatively few Kalahi districts (since most districts are not in broader conflict areas). The authors are aware of this and pursue some tests of robustness... in terms of a general rise in conflict, the comparison districts are assessed at the same time (since it uses RDD to indentify results) so that doesn't appear to be driving the results. Might there be a displacement of conflict towards the Kalahi districts? I'd have to review the paper again as well to investigate that...

Submitted by Cliff Burkley on
Is there a significant difference between traditionally Moro and NPA areas? KALAHI did not operate in most of the Moro areas but I think it penetrated many NPA areas. And disruptions are common in community development and infrastructure projects but I think less where targets are households and individuals. I could be wrong. Would be interesting to see how the project coped, if it did at all, compared with other projects. Might be that development projects in those areas (KALAHI or not) has the same impact? Would love to read the paper fully.

Submitted by Anne Knight on
I concur with Homira's points. This excerpt is key: "KALAHI supports local infrastructure and institution-building activities through block grants given to municipalities....The transfers were substantial - grant size totaled 15% of the average municipality's annual budget." What were the accountability mechanisms for these grants to municipal governments? Monitoring and evaluation, tracking of grant money to ensure it wasn't just pocketed or otherwise disposed of in a manner that would have the opposite effect of reinforcing local confidence in the government -- such as providing to family members or members of the same tribe/ethnic group -- is key.

Submitted by MattJ on
In my opinion, I feel that violence will always increase when aid is given to municipals in civil conflict. I think it causes more of an uproar when free money is brought in. For example, it can cause much corruption and unequal treatment, which then leads to acts of violence. The lower class, or any class for the matter that doesn’t benefit, will rise up against the benefitted class. You would expect the people to turn to the government for help once the aid is given, but instead, the people try to take matters into their own hands.

Submitted by Sean Bradley on
Jed, Some thoughts/observations on the Crost et.al. study from the World Bank team working on the KALAHI-CIDSS program with the Government of the Philippines. *Data *- The authors use data from the Armed Forces of the Philippines and there's some debate in the Philippines regarding how reliable the data are, especially as the army is not an unbiased party to the conflict. The data only cover incidents between the army and insurgent groups and do not cover incidents involving non-insurgent groups (i.e., paramilitary, vigilante, goons of local politicians, etc.) which could also be affected by the project. In addition, the army has the tendency to report civilian casualties as casualties suffered by the insurgents, to (i) increase the number of insurgents killed in combat and (ii) deflect any accusation of human rights abuses committed against civilians. The increase in presence of government staff in project areas could affect the way the army report incidents (even without any changes in the level of conflict) but the authors do not discuss if and how this might affect their results. We also note that the authors have been less than forthcoming in providing access to their data, despite repeated requests. * **Nature of the conflict* - The Philippines is characterized by the presence of two rebel groups with sharp ideological differences and modes of operation. This also affects the Army's strategies with the 2 groups. One group (the NPA) is mostly concerned about raising money to finance their operations while the other (MILF)appear to be concerned about grievances regarding the economic situation of Muslims in Mindanao and their exclusion from the political process. One would expect different impacts depending on whether the NPA or the MILF is present in the area. *Conflict localisation* - The authors use a rather narrow definition of conflict by only focusing on conflict incidents occuring in a given municipality. This is problematic for a number of reasons: (i) attacks in a given municipality will likely have an impact on welfare in neighboring municipalities, (ii) it fails to account for potential displacement of conflict between project and non-project areas and, (iii) patterns of attacks will depend on geographic conditions and thus there might not be a one-to-one relationship between location of government actions and of violent attacks. It is important to note that those three factors are likely to be especially important in the Philippines as municipalities tend to be small, with an average land area of 200 km2. Also, you might be interested in another paper looking at the KALAHI-CIDSS impacts on conflict (here ). In this paper, Arcand et.al. use newspaper reports of conflict incidence between the Armed Forces of the Philippines (AFP) and either the New People’s Army (NPA) or the Moro Islamic Liberation Front (MILF) and find that the project led to a decline in MILF-related events, but to an increase in NPA-related events. Both studies rely on nationwide conflict data and estimate project impacts using difference-in-differences and regression discontinuity techniques. The differences between the two sets of results could come from the variation in data sources but also from the different definitions of “conflict event” used.

Submitted by Ben Crost on
Dear Sean, thank you for your thoughtful comments on our working paper. We especially appreciate feedback and insights from those with experience in the Philippines. We hope to have sanitized replication data approved for release soon and look forward to making it available to other scholars. Your other points are well taken. We agree that there is potential for reporting bias in the AFP's data. But we'd like to point out that our main results do not disaggregate insurgent and civilian casualties, so that they are not affected by a possible misclassification of civilian casualties as insurgents. Either way, it will be interesting to see how our findings compare to ones based on newspaper reports. One should however keep in mind that newspapers have their own biases and have incentives to please certain stakeholders in the Kalahi program. The issue of MILF versus NPA conflict is somewhat difficult to address, since our data shows that only about 10% of incidents in Kalahi municipalities involve the MILF. Our results are therefore mainly driven by an increase in incidents involving the NPA, though we cannot reject the null hypothesis that the effect of Kalahi is the same for both insurgencies. We also agree that spillovers are likely to be an important issue in this context. Preliminary evidence (not shown in the current version of the paper) suggests that conflict increases in provinces in which Kalahi is active, even in municipalities that are not eligible for the program. We hope to be able to provide further evidence on spillovers in a future revision of the paper.