(Page 19 of 30) - Balance of Power Dynamics and War Escalation authored by Bennett, Scott.

states were coded as changes – cases with no shift in the balance, or where the shift was because
of data that changed over a year break – had values of 0. I set values to 0 even where the balance
changed slightly because of a shift in the COW data on January 1 to avoid “contaminating” the
balance of power tests concerning intervention and active external balancing with simply
updated information.

Analysis suggests that while the change in favor or against the MID initiator did not

matter to dispute escalation, the overall magnitude of changes (regardless of direction) did. In
particular, cases with a larger shift in the balance in any direction had a significantly lessened
expectation of the time to war, with no change in the expected time to peace. Reversing
direction to speak in terms of probabilities, when there was a major change of the balance of
power in any direction, the probability of peaceful settlement remained about the same, but the
probability of war increased significantly. Intervals within disputes that have major shifts in the
balance of power, in any direction, are more dangerous than those where the balance is relatively
constant. There are of course parallels to this finding in literatures such as the power transition
and other power shifts literature, but here the finding is within MIDs, within an appropriate
research design.

Additional Findings
The new analysis also allows proper assessment of the effect of numbers of disputants on

dispute escalation. The findings on number of disputants are quite unequivocal: having a larger
number of participants in a dispute interval leads to a shorter time to war (that is, an increased
per-day probability of war), as well as a longer time to expected peace (that is, a decreased per-
day probability of peaceful settlement). This is the same substantive result as Petersen et al.
(2004), but the finding here is not contaminated by endogeneity over the full course of a dispute.

Several other findings are very interesting, and the new method applied here reveals

several nuanced variable effects on dispute escalation. Having at least one major power as one
of the dispute originators leads to both a shorter duration to war (higher per-day probability of
war) and a longer duration to peaceful settlement (lower per-day probability). With the new data
set up, contiguity is again revealed to have a strong effect on both war and peace, with effects
paralleling major power status, namely a higher per-day probability of escalation to war and a
lower per-day probability of peaceful settlement. Joint democracy appears to have no effect; as
before, this suggests that the major effect of democracy may be at the level of deterring disputes
rather than aiding their settlement. And the alliance, hostility score, and S score variables reveal
nuanced effects that again illustrate the power of the competing risk setup. In each of these
cases, the variable does not affect the per-day probability of war. However, each affects the per-
day probability of peaceful settlement. So, considering alliance, the presence of an alliance does
not affect the probability of the dispute escalating to war, but it does increase the probability of
peaceful settlement (decreasing the expected time to settlement). Disputes between allies should
thus be somewhat shortened, with a constant probability of war and an increased probability of
peace. S functions in exactly the opposite direction, with higher S scores actually serving to
increase the expected time to peaceful settlement (lowering the per-day probability) while not
affecting the per-day probability of war. Higher values of the Crescenzi and Enterline hostility
measure actually indicate less hostility/more cooperation; those higher values lead to no change

A=90 and side B=10. The nondirected balance would still be 0.90, even though there had been a major shift
between t=1 and t=2. Change in the directed balance would pick this up. And here, side A need not be thought of
as “initiator” but simply as a constant identifier for one of the sides.