The large empirical conflict literature has established that there is a strong negative link between economic variables and the onset of an armed civil conflict. However, it has been difficult to demonstrate a clear causality between poor economic performance and increased risk of conflict because of potential endogeneity issues, especially for large country samples. Most existing studies that analyse the causal links focus on the effects of economic growth on conflict, even though conventional conflict studies find the strongest relationship for income levels. In this article, we use three new exogenous instruments for income per capita, based on historical data for mailing times, telegram charges and urbanization rates. Using instrumental variables methods and global panel data for the period 1946–2014, we show that the negative effect of income per capita on the probability of conflict onset is consistently strong and larger than in conventional estimations using pooled ordinary least square regressions.