I am more then slightly baffled by the statistical arguments. Is he just using data from crashes with fatalities? This does not seem like the right way to go. Comparing data from accidents in which *someone else* (or even better, the driver) dies should be better.