The Deterrent Effects of the International Criminal Court: Evidence from Libya

ABSTRACT

The International Criminal Court (ICC) was designed to try the worst war criminals for crimes against humanity, genocide, and other instances of mass human suffering. By providing a permanent, international mechanism to hold perpetrators of mass human rights abuse accountable, the ICC is also meant to be a deterrent—to prevent potential genocidaires from committing systematic human rights abuses in the first place. But what if the effect is actually quite the opposite? While advocates of international justice have made conjectures about the effect of the ICC on stopping human rights abuses, the existing scholarship does not empirically test assumptions about the relationship between international criminal justice and violence. This article outlines the causal mechanisms by which the ICC could affect ongoing violence and tests these assumptions using event count models of the relationship between the ICC and the level of violence against civilians in Libya during the 2011 crisis. These analyses suggest that the ICC’s involvement in conflict does have a dampening effect on the level of mass atrocities committed. The results also call for a broad and sustained research agenda on the effect of international accountability efforts on ongoing violence.

KEYWORDS:

• Deterrence
• ICC
• international law
• Libya

Acknowledgments

I would like to thank Carrie Booth Walling, Ross Miller, Nam Kim, Christian Davenport, Will Moore, Erik Voeten, Chris Fariss, Amanda Murdie, and Jim Meernik for their very constructive comments on previous versions of this paper. I am also grateful to Jake Wobig and Rula Jabor for their research assistance and Reed Wood, Anita Ghodes, Yuri Zhukov, and Matthew Baum for directing me toward/sharing data. All errors are mine alone.

Supplemental Material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

¹ This work follows the model set out by Meernik (2005).

² Web appendix A outlines lists the crises in which the ICC got involved while the conflict was ongoing and the mode of referral of the situations.

³ With so many different spellings of Qadaffi, I chose to follow the New York Times’ spelling of his name.

⁴ Figure 1 presents the natural log of fatalities for ease of presentation and interpretation given the large spikes in fatalities.

⁵ These numbers and descriptions are taken from the ACLED data set, described in more detail subsequently.

⁶ All of this information is available from the ICC’s Weekly Updates, which can be found at http://www.icc-cpi.int/Menus/ICC/Press+and+Media/ICC+Weekly+Update/.

⁷ See (The Prosecutor v. Thomas Lubanga Dyilo (2012); The Prosecutor v. Germain Katanga (2014); The Prosecutor v. Mathieu Ngudjolo Chui (2012).

⁸ See, for example, The Prosecutor V. Uhurua Muigai Kenyatta, Confirmation of Charges (Situation in Kenya) (2012); The Prosecutor V. William Samoei Ruto and Joshua Arap Sang, Confirmation of Charges (Situation in Kenya) (2012).

⁹ This same logic is embodied in the principle of “double jeopardy,” which states that you cannot be punished for the same crime(s) more than once.

¹⁰ The ACLED data are conservative in their counting of civilian casualties, but this does not imply that it is a biased undercount. Rather, the ACLED data triangulate accounts of civilian deaths from a number of sources, thereby mitigating potential over- or undercounting biases that might result from only relying on major newspapers or local NGOs. The fact that ACLED is conservative in its counting of civilian casualities should make it even more likely to see an effect of the ICC on lowering the civilian casualty count.

¹¹ For an exhaustive discussion of measuring causalities during conflict, see Seybolt, Aronson, and Fischhoff (2013).

¹² The mean for the Government to Rebels variable is –1.37. The mode and median are both zero, and the standard deviation is 4.49. The mean for the Rebel to Government variable is –1.24, with a standard deviation of 4.37. Both the mean and mode are 0. The mean of the NATO to Government variable is –1.80, the mean and mode are 0, and the standard deviation is 6.37. See also Web appendix C.

¹³ In the robustness checks in Web appendix D, I estimate a zero-inflated negative binomial. I use the three variables that account for the level of conflict at any given day—Government to Rebels, Rebels to Government, and NATO to Government—for the inflation stage of the model, as the level of conflict on any given day should affect whether or not there was violence against civilians on that day.

¹⁴ More information is available here: http://web.ku.edu/~keds/papers.dir/automated.html.