Baron, that is why I directed you to those pages in that book. It specifically addresses how studies have been done to figure out the percentage of drunk drivers on the road and how the proportion of drunk drivers in accidents compares.
Since you do not seem to want to bestir yourself, what it essentially boils down to is, in a number of times and places organisations like the National Highway Safety Administration worked with the local police and simply stopped either every car or something like 1 out of every 4 cars to pass a checkpoint (depending on the study) to breathalyse the driver. (The deal for these was that drivers were given the right to refuse and 3% did, so as a worst-case scenario you could simply assume that those 3% were all over the limit; of those who did not refuse, if they failed at that checkpoint they weren't [necessarily] arrested, they just had to leave the car and find another way to get where they were going.) Anyway, using this method of sampling, they were able to determine that 17-20% of drivers on the roads on weekend nights between 10pm-3am had been drinking.
If drinking made no difference to likelihood of being involved in an accident, then you would have expected that 17-20% of reportable accidents from that exact same time period and city would also involve drivers who had been drinking. This was not the case; looking at the accidents, although the drinking drivers made up max. 20% of drivers on the road, they were involved in 78% of fatality accidents and (searching FARS, because the book doesn't report on non-fatality accidents) they were involved in ~32% of non-fatality accidents.
The thing this last bit tells
me, is that not only are drivers who have been drinking more likely to be in accidents, those accidents are FAR more likely to involve a fatality.
In the UK and the Netherlands, the various traffic regulatory bodies give figures of drinking drivers being roughly 4x more likely to be involved in a fatality accident than a non-drinking driver, derived using much the same methods.
Answer your question?
Oh, and, you may not kill the internets. If you know where to look, it's just too bloody useful. And if you aren't finding the real useful bits, it's your own fault for not knowing where or how to look.