Could also be that they are only looking at the subset of collisions involving bicycles where the cyclist was determined to be at fault, which isn't data available in the analysis given here.