NOTES

1 On the evolution of the field, see Kriesberg (1997).

2 On changing patterns of warfare, see Holsti (1996); Human Security Centre (2005); Marshall and Gurr (2005); and Harbom, Hogbladh, and Wallensteen (2006).

3 For useful reviews, see Galtung (1965), Zartmann (1985,1995), Azar and Burton (1986), Burton (1990), Kriesberg (1992), Bercovich (1996), Carnegie Commission (1997), Fisher (1997), Stern and Druckman (2000), Malone and Hampson (2001), and Wallensteen (2002).

4 Case study methods are a subset of qualitative methods, which include interpretive ethnographic studies, archival analysis, elite interviews, macrohistor- ical analysis, intensive analyses of particular historical episodes, "qualitative comparative analysis” based on Boolean and fuzzy set methods (Ragin 1987, 2000), alternative conceptions of causation (Goertz and Levy, 2007), and a range of other topics. Myfocus here, following most but not all of the expanding literature on case study methodology in the fields of international relations and comparative politics field (e.g., George and Bennett, 2005: 18-19), is on comparative and case study methods that aim to produce causal explanations and to develop a set of cumulative generalizations about the social world. I exclude postmodern narratives and other qualitative work that rejects the possibility of causal explanation, while incorporating other forms of interpretive or discourse analysis that accepts the goal of causal explanation and the possibility of generalization.

5 Verba (1967) distinguished between configura- tive and disciplined configurative analyses. Lijphart (1971: 691) distinguished among atheoretical, inter­pretive, hypothesis-generating, theory-confirming, theory-infirming, and deviant case studies. Eckstein (1975: 96-123) suggested a similar typology: configurative-idiographic, disciplined configurative, heuristic, and crucial-case studies based on most-likely and least-likely designs, and also plausibility probes.

6 Idiographic refers to the aim of inquiry (the explanation of an individual case), and not whether the inquiry is theoretical or not. Theory can be used to structure an idiographic case study (Levy, 2001).

7 The revised hypothesis cannot, however, be tested against the same data that was used to generate the hypothesis to begin with (King, Keohane, and Verba, 1994; George and Bennett, 2005).

8 Critics of case-study research often complain that case studies are so pliable that researchers' can interpret any outcome as consistent with their theoretical argument. One motivation for the growth of qualitative methodology is to eliminate whatever remnants of that research practice still existed. For an example of a case study that begins with a falsifiable interpretation, suggests evidence that would falsify the argument, and in fact uncovers that evidence and concludes that the hypothesized interpretation was false, see Gochal and Levy (2004).

9 After early debates, the literature on case study methodology now equates the comparative method with the analysis of a small number of cases (Collier, 1993: 105).

10 Note that Mill defines agreement or difference in terms of values on the dependent variable, whereas Przeworski and Tuene (1970) define similar and different in terms of explanatory variables. Thus, Mill's method of agreement is equivalent to a most different systems design, and Mill's method of difference is equivalent to a most similar systems design. Scholars often confuse these different terminologies.

11 The logic of inference is much more similar to what philosophers of history call genetic expla­nation (Gallie,1963; Nagel, 1979: 564-68) than to explanations based on covering-laws and deductive- nomological logic (Hempel, 1942).

12 Experimental methods may be superior for testing many of these hypotheses, but it is often difficult to generalize from highly controlled labo­ratory settings that cannot fully replicate the stakes and emotions inherent in the world we are trying to explain.

13 Prospect theory (Kahneman and Tversky, 1979) posits that people “frame” choice problems around a reference point, give more weight to losses than to comparable gains as defined by that reference point, and engage in risk-averse behaviorwith respect to gains and risk-acceptant behavior with respect to losses. It helps to explain why people fight to keep territory and other things they would not have fought to gain in the first place, why threats are more effective in deterring people from improving their positions than in coercing them to accept losses from their reference point, and why they often take enormous risks to eliminate losses - even at the risk of incurring even greater losses. For applications to international relations, see Levy (2000).

14 I treat crucial case studies as a design that serves the hypothesis-testing function of case studies, rather than as a distinct type of cases studies, which is more common in the literature.

15 Another example of a most-likely case design is Ripsman and Levy's (2007) analysis of the absence of a “preventive war” against Germany in the 1930s. If ever conditions were ripe for preventive action against a rapidly rising and threatening adversary, it was in the mid-to-late 1930s with Hitler's Germany. On preventive war, and how it differs from preemption and other sources of better-now-than-later logic, see Levy (2008).

16 Scholars often define crises as situations involv­ing a threat to basic values, a high probability of involvement in military hostilities, and a finite time for response (Brecher, 1980: 1). Crisis management can also be applied to intra-war crises, where it involves an attempt to avoid war or limit or control the escalation of violence. Although the literature on crisis management focuses on interstate confli­cts (Williams, 1976; George, 1991a), it can also be applied to crises involving rebel groups or non-state actors.

17 Thus, I define crisis management in terms of the de-escalation of conflict (Kriesberg, 1992), rather than in terms of the more permanent elimination of the conflict of interests between parties (Burton, 1987: 7-8). See also Stern and Druckman (2000: 44) and Maoz (2004: 17-18).

18 On entrapment, see Brocknerand Rubin (1985). For a summary of recent interpretations suggesting that Germany was in fact not so adverse to a world war, see Lieber (2007). If correct, this would further reinforce the argument that the outbreak of World War I was not a failure of crisis management, but undercut the argument about entrapment in an escalating conflict.

19 George (1991b) contrasted Kennedy's behavior with Truman's behavior in the Korean War-escalating both his objectives (unifying the two Koreas by military force) and the means for achieving them (permitting MacArthur to march north toward the Yalu).

20 George (1991b) acknowledged some brink­manship behavior that violated the “limited means” criterion (such as the anti-submarine warfare activities of the US navy), but noted that it reminded each side of the risks of provoking the other and of an inadvertent escalation, and that it ultimately contributed to de-escalation.

21 George's (1991b) multidimensional conception of risk contrasts with standard treatments in the literature, which generally assume that risk is a uni­dimensional variable and which treats actors in terms of their degree of risk aversion or risk acceptance. For formal decision theorists, risk orientation is simply the shape of the utility function (concave downward for risk aversion, linear for risk neutrality, and convex for risk acceptance). Prospect theorists posit an S-shaped value function with varying risks as a function of losses and gains (Levy, 2000).

22 The “tried to satisfy himself” phrasing sug­gests another hypothesis about the psychology of risk - that risk does not shape motivation but is endogenous to it, and that a highly motivated actor will subconsciously adjust its risk assessment (or perhaps consciously, if it wants to convince other decision-makers) so as to justify an action it wants to take.