Sober analysis on drunk drivers!

Suppose you had only three sets of data.

x fatal crashes between two drunk drivers.
y fatal crashes between two sober drivers.
z fatal crashes between a drunk and a sober driver.

Can you find out how risky is a drunk driver, when it comes to a fatal crash?

I am impressed by the power of empirical analysis to give us suggestive estimates regarding this. Read Levitt and Porters' paper on How dangerous are drunk drivers?