It is difficult to do meaningful statistical studies of human behavior. With crime statistics, about the only statistic that can be trusted is the number of murders - bodies have a way of showing up. But, most folks don't call the police when a $100 bike is stolen - they know reporting the crime will not lead to a police investigation. So, bike thefts are greatly underreported.
But, it turns out, even murder statistics can be misunderstood. In Houston, the number of murders using firearms dropped about 20% between 1989 and 2000. The politicians took the credit, and said that their "get tough" policies and "long sentences", and handing out the death penalty willy-nilly deserved the credit.
Then, some folks looked a bit closer. They discovered that the number of shootings in Houston was as high in 2000 as in 1989. The difference in "murders" was the result of the doctors at emergency rooms saving a far higher percentage of gunshot victims due to growing levels of experience and new techniques. In fact, the doctors bragged that they had saved some guys three or four times...regular customers.
Likewise, if the number of cyclists killed by motor vehicles in 2004 is lower than in 1994, there can be many explanations, including a higher quality of care at emergency rooms. But, in my city, two explanations that can be ruled out are "Better Drivers" and "Better Roads".