The report is misleading because AP tends to be used far more on controlled access roads which have fewer crashes per mile. There are other biases too but this alone could easily explain more than the observed 1/3 risk reduction.