VoxDev Development Economics podcast

S7 Ep59: What drives civil war in poor countries?

0:00
30:46
Rewind 15 seconds
Fast Forward 15 seconds

There are more wars now than at any time since 1946. Most are civil wars, and they cluster in poor countries, in places where the surveys stop and researchers rarely go. So how do you find out what starts a war, or what it leaves behind?

Oeindrila Dube (University of Chicago) and Michael Callen (LSE) tell Tim Phillips about how the answers to questions such as do richer countries fight less need careful measurement, because the answer might depend on who exactly is getting the money. The good news: we have many new ways to collect the data we need to make better policy decisions. 

The research behind this episode:

Dube, Oeindrila, and Michael Callen. 2026. "Conflict and Fragility." Lecture 8, BREAD-IGC virtual PhD-level course on political economy, 27 October 2026. Lectures: "Economic Shocks and Conflict: What Better Data Let Us See" (Dube) and "Fragility, Conflict, and State Building" (Callen).

Dube, Oeindrila, and Juan F. Vargas. 2013. "Commodity Price Shocks and Civil Conflict: Evidence from Colombia." Review of Economic Studies 80 (4): 1384-1421.

Blumenstock, Joshua E., Michael Callen, Anastasiia Faikina, Stefano Fiorin, and Tarek Ghani. 2023. "Strengthening Fragile States: Evidence from Mobile Salary Payments in Afghanistan." CEPR Discussion Paper 18254. Centre for Economic Policy Research (gated).

Bhusal, Bhishma, Michael Callen, Rohini Pande, Soledad Artiz Prillaman, Deepak Singhania, and Apurva Subedi. 2026. "From People's War to People's Rule: Rebel Governance and the Foundations of Inclusive Democracy." NBER Working Paper 35241.

Dube, Oeindrila, Joshua Blumenstock, and Michael Callen. Forthcoming. "Measuring Religion from Behavior: Violence, Climate Shocks and Religious Adherence in Afghanistan." Journal of Political Economy.

To cite this episode:

Phillips, Tim, Oeindrila Dube, and Michael Callen. 2026. "What drives civil war in poor countries?" VoxDev Talks (podcast). 

About the guests

Oeindrila Dube is the Philip K. Pearson Professor of Global Conflict Studies at the University of Chicago Harris School of Public Policy. Her research spans the causes and consequences of conflict, crime and violence, from commodity price shocks and civil war in Colombia to post-conflict reconciliation in Sierra Leone and religious extremism. She co-chairs the Crime, Violence, and Conflict sector at J-PAL and is a co-editor of the Journal of Development Economics.

Michael Callen is an Associate Professor of Economics at the London School of Economics and a research programme director for the State programme at the International Growth Centre. His research spans political economy, development and behavioural economics, using field experiments to tackle failures of accountability and service delivery in the public sector, mainly in Afghanistan, Nepal and Pakistan. He serves on the editorial board of VoxDev.

Research cited in this episode

Uppsala Conflict Data Program (UCDP). The dataset most conflict researchers use. In 2025 it recorded 65 state-based conflicts, the highest number since its records begin in 1946. A state-based conflict has a government on at least one side; eight of the 65 were fought between states, and the rest were inside them.

OECD, States of Fragility 2025. The OECD rates 177 countries and territories on 56 indicators of risk and resilience. The 2025 report finds 61 contexts with high or extreme fragility. They hold 25% of the world's population but 72% of its extreme poor, a share the OECD projects could reach 92% by 2040. Callen cites the 18 rated extremely fragile, including Afghanistan, Sudan and the Democratic Republic of the Congo.

The research gap. Callen estimates that less than 1% of the trials in the AEA registry of randomised controlled trials take place in the 18 extremely fragile countries, and less than half a percent of articles in the top five economics journals study them. His point is the mismatch; that is where a large share of the world's poorest people live.

Cross-country studies of civil war. The first generation of evidence used one observation per country per year. Paul Collier and Anke Hoeffler (2004) and James Fearon and David Laitin (2003) showed that countries with higher GDP per capita are less likely to suffer civil war. The data could not say why, or who in the country got the income.

Opportunity cost and rapacity. Two theories pull in opposite directions. The opportunity cost argument adapts Gary Becker's (1968) economics of crime to rebellion; when wages rise, recruits cost armed groups more, so fighting falls. The rapacity argument treats conflict as violent theft; when there is more to steal, fighting pays more, so it rises. Both are triggered by the same thing, a rise in income.

Coffee and oil in Colombia. Dube and Juan Vargas (2013) separate the two effects using Colombia's two largest exports. Coffee is grown on small, labour-intensive farms, so its price feeds mainly into wages. Oil is extracted with capital, so its price feeds mainly into revenue that can be stolen. Comparing municipalities that produce more and less of each as world prices moved, they find the coffee price collapse of the 1990s cut wages and raised violence in coffee areas. Rising oil prices increased attacks in oil areas, along with municipal royalties and kidnappings of politicians.

Grid cells across Africa. Later studies cut sub-Saharan Africa into cells roughly 50 kilometres square. Nicolas Berman and Mathieu Couttenier (2015) find that falls in agricultural prices increase conflict in cells that grow those crops, the coffee result across a continent. Berman, Couttenier, Dominic Rohner and Mathias Thoenig (2017) find that rising mineral prices increase fighting near mines, the oil result generalised.

Meta-analysis of natural experiments. Graeme Blair, Darin Christensen and Aaron Rudkin (2021) pool 46 natural experiments, within countries and across them. Higher agricultural prices reduce conflict; higher prices for oil, gas and lootable artisanal minerals increase it.

Satellites and machine learning. Dube argues the next step is turning masses of raw data into measures. Hannes Mueller and co-authors (2021) trained a neural network to detect destroyed buildings in satellite images of Syrian cities, producing a city-by-month record of the war. In Gaza, Daniele Rinaldo and co-authors (2026) use radar, which sees through cloud and smoke, together with night-time lights to estimate the economic cost of the destruction.

Measuring religion with phone records. Talking invalidates prayer in Islam, so calls fall during the Maghrib prayer window at sunset. Dube, Joshua Blumenstock and Callen use the size of that fall, across roughly 23 billion anonymised calls, as a daily, district-level measure of religious adherence in Afghanistan. Adherence declines after violent attacks by Islamist insurgents and rises after droughts in agricultural regions.

State capacity as investment. In Timothy Besley and Torsten Persson's (2009) framework, the ability to tax, pay officials and enforce the law is built through costly investments that pay off only over time. If nobody knows whether today's institutions will survive, the incentive to invest weakens. Callen argues that fragile states combine low capacity with exactly that uncertainty.

Mobile Salary Payments in Afghanistan. In a reform run with the Ministry of Education, Blumenstock, Callen and co-authors evaluated the biometric registration of teachers and the payment of their salaries by mobile money, in a randomised trial covering about 30,000 employees in more than 1,000 schools across three provinces. At baseline, only 41% of employees were in school during unannounced visits, 30% of those present held a government ID, and 60% reported salary delays averaging three and a half months. Registration was higher in districts where citizens were more confident that the Afghan army would defeat the Taliban. The authors present this as an association, not a causal estimate.

The cost of the Afghan intervention. Callen puts combined military and development spending at around $3 trillion and the death toll at about 170,000. For comparison, Brown University's Costs of War project estimates US spending alone at $2.313 trillion and about 243,000 direct deaths in Afghanistan and Pakistan. His point is that the effort ended broadly where it began, with the Taliban in control.

Rebel governance in Nepal. During the People's War of 1996 to 2006, the Maoists set up "People's Governments" that ran schools and courts and mobilised historically excluded Janajati (Indigenous) groups, who had been slotted into a caste hierarchy codified by the monarchy. Bhusal, Callen and co-authors (2026) compare places either side of district borders, where only one side fell under rebel rule. Janajati citizens there know more about the post-war constitution and take a bigger part in politics and civil society. Party gatekeepers are more likely to come from excluded groups, and parties across the spectrum, not only the Maoists, select more candidates from them.

More VoxDev Talks episodes

Why democracy alone does not build effective states. Tim Besley on his lecture on state capacity in the same BREAD-IGC series, and why some states can tax and deliver services while others stay stuck in conflict.

Can we use experiments to understand institutions? Michael Callen and Jonathan Weigel on how field experiments can study institutional change as it happens.

Related reading on VoxDev.org

Building state capacity in fragile states: Evidence from Afghanistan, a VoxDev article by Blumenstock, Callen and co-authors on the Mobile Salary Payment reform and what it reveals about why fragility holds back state building.

How rebel governance helped build an inclusive democracy in Nepal, a VoxDev article on how the Maoists' People's Governments left lasting gains in political participation and candidate selection for historically excluded groups.

More episodes from "VoxDev Development Economics"