Makes sense in theory
The model works in many other areas. The idea is predictive modeling. I'm no baseball fan but this is the same math that helped the Oakland A's and the Boston Red Sox win the world series. Basically, you can analyze all the factors that are collected in baseball stats. You run them through a regression analysis and it will tell you the 3 or four factors that are most significant in predicting the outcome. In baseball, James Ashenfelter did this for player selection to help his teams win the World Series.
CSA is doing the same thing. It's looking for the most significant predictors of accidents. If you can change driver's behaviors BEFORE an accident, that is really what you want to do.
In practise, it's hard to tell whether it will work or not. Your talking about trying to change the behavior of hundreds of thousands of drivers, who will definately change their behaviour when you introduce new rules, however, the change may not be as intended and you could possibly shift the whole dynamic of what the model was based on in the first place.
P.S. - If your a stats geek like me, I recommend reading "Super Crunchers" by Ian Ayres. He goes into more detail about the Baseball stats and many other interesting applications of this method.