The effort to explain escalation processes using formal and empirical
models has yielded numerous insights. However, there have been relativ
ely few attempts to integrate these two approaches such that the resul
ts from an empirical test of escalation behavior are interpreted withi
n and based on a well specified theory of escalation. This paper bridg
es that gap by developing a formal theory of escalation processes to d
erive hypotheses that identify the conditions under which states escal
ate in conflict. A player's decision to escalate or back down is based
on an estimate of the opponent's cost tolerance-the maximum costs a p
layer will withstand to achieve a demand in the game. The hypotheses d
erived from the theory are general and applicable in many contexts tha
t involve escalation processes. In this paper, the hypotheses are test
ed empirically in the context of extended deterrence crises. The test
confirms the theory's expectations and reveals strong support for the
hypotheses. The theory offered here, I argue, provides a better explan
ation for the patterns found in the data than do the theories used in
previous studies of extended deterrence.