A Unified Statistical Model of Conflict Onset and Escalation

In this article I argue that much of the inconsistency in the scholarly literature on conflict escalation can be attributed to selection bias. To control this bias statistically, I use a censored probit model to evaluate hypotheses about conflict escalation over the time period of 1950-1985. I report four results. (1) Power parity and economic development are found to influence conflict nonmonotonically. (2) Although joint democracy and joint satisfaction with the status quo are found to have robust pacifying effects on the onset of conflict, the results suggest that they are unrelated to the escalation of disputes to war. (3) Allied pairs of states appear less likely to escalate their disputes. (4) Finally, the unified model suggests that it is essential for researchers interested in the escalation behavior of states to consider first how states become involved in disputes. Conflict onset and escalation appear to be related processes. any scholars have sought to isolate the causes of escalation, but discrepancies between the theoretical expectations and the empirical results persist. I attribute these discrepancies to a selection effect. Pairs of states do not become entangled in hostilities randomly. They instead select or are selected into disputes by a strategic process. Following previous empirical research focusing on selection effects (Achen 1986; Smith 1996a, 1996b, 1998, 1999; Gartner and Siverson 1996; Leeds and Davis 1997; Signorino 1999), I employ a statistical model to control for the interdependent relationship between the onset of disputes and escalation to war over the time period of 1950-1985. The results suggest that controlling for the selection effect statistically stipulates a link between the formal and empirical studies of escalation. The scholarly literature on what causes escalation is mixed. Although some suggest that joint democracy inhibits escalation (Rousseau et al. 1996), others find that joint democracy actually makes dyads more escalatory (Senese 1997). There is also some disagreement about the effect of military capabilities on conflict onset compared to escalation (Morgan 1984, 1990, 1994; Morrow 1989; Fearon 1994b; Bueno de Mesquita, Morrow, and Zorick 1997). It seems reasonable to suspect that the factors that influence conflict onset may also affect escalation directly and/or indirectly. If this is the case, it is important to consider how the factors that influence onset and escalation may be related to each other. Constructing a unified statistical model offers a first cut at modeling the process of conflict. Similar to the formal literature (Morrow 1989; Bueno de Mesquita and Lalman 1992; Fearon 1994a, 1994b; Smith 1998, 1999) that describes explicitly the selection process by which states get into and escalate disputes, I address the same issue from an empirical perspective (Smith 1996b, 1998, 1999; Signorino 1999). Modeling such selection empirically manages sources of bias and allows one to make truer inferences about the conflict generating process.