Joining by Number: Military Intervention in Civil Wars

Abstract

Understanding why and when states militarily intervene in civil wars is crucial. Intervention can increase civil wars’ severity and the strategies employed in civil wars are shaped by the possibility of military intervention. This article argues that potential military interveners react to information revealed about warring parties’ intentions and relative power. Without revealed information, potential military interveners are unlikely to reconsider their initial decision to remain out of the war. Revealed information causes non-belligerent states to update their expectations about the trajectory of the civil war causing them, at times, to change their calculus about the benefits of belligerency and thus intervene. This helps explain why civil wars spread and when they do so. This explanation is tested using generalised estimating equations on a new data-set of unexpected events for the civil wars in the Correlates of War Intrastate War and PRIO Armed Conflict data-sets.

Acknowledgements

I would like to thank Henk-Jan Brinkman, John Harden, Michael Lee, Zachary Malitz, Madalene O’Donnell, Gary Uzonyi, and Alex Weisiger for their help.

Notes

1. Unexpected events are defined in the ‘Methodology and Results’ section.

2. Confirming evidence could prompt risk averse states to intervene by increasing the certainty of states’ estimates.

3. Since multiple states often join the same civil war in the same month, there are more instances of intervention in the non-belligerent war-month data-set than in the war-month data-set.

4. Lags also help avoid serial correlation in the error term (Goldstein and Pevehouse 1997).

5. The various levels and types of unexpected events do not sum to the total number of unexpected events as lagging these variables creates overlaps in some months.

6. The asymptotic consistency of the βs estimated by GEE models holds even if the exact nature of the intra-cluster relationship is unknown. Thus, it is not necessary to know precisely how certain observations are causally linked to control for interdependence.

7. The results for the control variables remain essentially unchanged.

8. Military events are only significant at the 0.10 level with shorter lags in Model 6. The lag chosen does not affect any events variables in Model 7.

9. The result holds with shorter, but not longer, lags.

10. The results for Level Two variables hold for shorter lags, but hold only at the 0.10 level for longer lags. The results for Level Three variables are not sensitive to the lag chosen.