I find this table on page 14 to be very interesting:
Similar typology was used in the FHWA study by Hunter et al. (1996). In a six-state study of
3,000 bicycle crashes taken from hard copy police reports, the most frequent bicycle/motor
vehicle crash types were as follows:
Crossing Path Crashes % of All Crashes
• Motorist failed to yield to bicyclist (includes drive out/through 21.7
at intersections and at Midblock/driveway locations)
• Bicyclist failed to yield to motorist at an intersection 16.8
• Bicyclist failed to yield to motorist, midblock 11.8
• Other crossing path crashes 7.2
Parallel Path Crashes
• Motorist turned or merged into bicyclist’s path 12.2
• Motorist overtaking bicyclist 8.6
• Bicyclist turned or merged into motorist’s path 7.3
• Other parallel path crashes 7.4
From these numbers, we can do a very crude assignment of fault. If we assume that whoever fails to yield, or merges into someone else's path is at fault, and the overtaking vehicle is at fault in a collision, we get:
Motorist fault: 21.7+ 12.2 +8.6 = 42.5%
Cyclist fault: 16.8 +11.8+7.3= 35.9%
So at a rough level, the motorist is at fault about 20% more of the time than the cyclist.
Keep in mind that these determinations come from the police reports, and police are notorious for having a car-centric bias when it comes to determining fault. Also, in many car-bike collisions the driver is the only witness, as the cyclist is much more likely to be incapacitated or killed. However, in many car-bike collisions the driver is not a reliable witness because he never saw the cyclist.