"Examining nuclear war near misses using engineering reliability assessment methods, Barrett, Baum, and Hostetler estimate that there was a 2% chance per year of an accidental U.S.-Soviet nuclear war during the Cold War. Since President Kennedy's estimate of the probability of nuclear war arising from the Cuban Missile Crisis alone puts the probability of accidental nuclear war at between 1.4% and 2.4% per year over the course of the Cold War (1960 to 1989) without even considering the Petrov incident, the NORAD training tape incident, and other such incidents, this 2% per year probability seems very plausible.

Does this Cold War estimate apply to North Korea? Examining the factors that contribute to the probability of accidental nuclear war, I think that the probability of accidental nuclear war with North Korea is if anything higher (and probably much higher) than was the probability of an accidental U.S.-Soviet nuclear war.

To err in favor of the deterrence option, then, I estimate the cost of gambling on deterrence under two cases. In the first, I assume that the probability of an accidental U.S.-North Korean nuclear war is 2% per year. In the second I assume that there is one incident over the next 30 years that creates a 25% chance of an accidental nuclear war."