As shown in Exhibit 5-17, at British roundabouts bicyclists fare worse in terms of crashes at roundabouts than at signalized intersections.

A French study (7) compared the crashes in 1988 in 15 towns in the west of France at both signalized intersections and roundabouts, as shown in Exhibit 5-18. The conclusions from the analysis were:

All European countries report that a more careful design is necessary to enhance bicyclists’ safety. The type of bicycle crashes depends on the bicycle facilities provided at the roundabout. If there are no bicycle facilities, or if there is a bike lane on the outer area of the circulatory roadway, crashes typically occur between entering cars and circulating bicyclists as well as between cars heading into an exit and circulating bicyclists. Improperly placed signs on the splitter island may also be a contributing factor.
As a result, most European countries have the following policies:
Speed is a fundamental risk factor in the safety of bicyclists and pedestrians. Typical bicyclist speeds are in the range of 20 to 25 km/h (12 to 15 mph), and designs that constrain the speeds of vehicles to similar values will minimize the relative speeds and thereby improve safety. Design features that slow traffic such as tightening entry curvature and entry width, and radial alignment of the legs of a roundabout, such as with the urban compact design, are considered safe treatments for bicyclists (17).
In the Netherlands, a 90 percent decrease in injury crashes was experienced with separate bicycle paths around roundabouts where bicyclists do not have right-ofway at the crossings (17).
A bicycle crash prediction model from Sweden has been validated against data for Swedish, Danish, and Dutch roundabouts (18). The model provides reasonable results for roundabouts with up to 12,000 vehicles per day and 4,000 bicycles per day. The model tends to over-predict crashes (i.e., is conservative) for roundabouts carrying more than 12,000 vehicles per day that are also designed with separate bicycle paths with crossings on the approach legs. It is calibrated for crossroad intersections as well as roundabouts. To obtain the expected cycling crashes per year at roundabouts, the value derived from the general junction model is factored by 0.71, implying that bicycle crashes at roundabouts are 71 percent less frequent than at junctions in general. However, the reader is cautioned when extrapolating European bicycling experience to the U.S., as drivers in Europe are more accustomed to interacting with bicyclists.