NOTES
1 I am grateful for comments from the editors, two anonymous reviewers, as well as participants at the workshop on this handbook at the Institute of Applied Systems Analysis, Laxenburg, Austria 30 June - 2 July 2007.
This research was supported by grants from the US National Science Foundation (SES-0351670), the Carnegie Corporation, and the Research Council of Norway through its support for the Centre for the Study of Civil War. I am also grateful for travel support from the British Academy.2 To be clear, by civil war I here mean a violent conflict over some incompatibility between at least two organized groups, of which one is a government and one is not a state, that generates some casualties as a direct result of fighting. Note that this excludes one-sided violence, where the victims are not an organized group, inter-communal conflicts that do not involve a government, as well as non-violent forms of conflict such as demonstrations and strikes. For further discussion on different attempts to define civil war, see e.g. Sambanis, (2004).
3 For a systematic review of trends in conflict and the distribution of types of conflict in the Uppsala Armed Conflict Data (ACD), see Gled- itsch et al. (2002) The ACD data are available at www.prio.no/cwp/armedconflict/. Pillar (1983) compares patterns of war termination by type of conflict, and shows that civil wars are far less likely to end in negotiated settlements. Lacina and Gleditsch (2005) examine trends in battle deaths, based on supplementary information for the ACD events available at www.prio.no/cscw/cross/battledeaths.
4 Most research on conflict follows the correlates of War project's distinction between interstate wars where both antagonists are independent nation states and intrastate or extra-systemic conflicts where nation states fight actors that are not states (e.g. Sarkees 2000; Small and Singer 1982).
By these criteria, civil wars can become transformed into a new interstate dispute when foreign states intervene on the side of the opposition and directly confront the government. However, there can be considerable ambiguity over what is meant by states intervening on the side of the opposition and whether a given conflict should be considered interstate or intrastate, and the examples of transnational characteristics cited above suggest that imposing mutually exclusive categories may often generate misleading consequences. I will return to these issues later.5 The problem that outcomes believed to reflect similar mechanisms operating within each unit may stem from diffusion between units was first noted by Galton (1889), hence the term “Galton's problem”. However, the problem of inference runs the other way as well; studies of diffusion face an “inverse Galton's problem” in that similarity in outcomes could be due to the similarity of the units, rather than diffusion between units (see Gleditsch and Ward 2006).
6 Christin and Hug (2004) suggest that the MAR data suffer from selection biases in conflict studies, by only sampling groups on criteria related to conflict, and present empirical evidence suggesting that selection biases plague analyses based on the MAR data. Oberg (2002) supplements the MAR data with additional data on minorities not at risk, and reaches a less pessimistic conclusion regarding the potential selection biases.
7 Indeed, Taylor was in 2003 indicted by the UN Special Court for Sierra Leone for crimes against humanity for his involvement in the conflict in Sierra Leone (see Vines 2003).
8 Somalia is an often-cited example of the link between failed states and terrorism (e.g. International Crisis Group 2002). However, although the absence of an effective central government after the fall of Siad Barre has enabled Islamist groups to operate freely, most of their activities have been limited to Somalia, with the exception of one group (al-Itihaad al-Islami) that has attacked targets in the Ogaden region of Ethiopia.