I'm not sure where the 17% is coming from either. The linked study talks about 300 deaths over a 10 year period. That's 30/year which, relative to the 30K or so automotive fatalities in the US per year is 0.01%--i.e. a almost vanishingly small percentage. It's actually not implausible to me that there would be some tiny effect associated with switching time/schedules even if it were only because of "OMG, I forgot to switch my clock and now I'm late for a meeting!" [EDIT: If you apply the 30 annual number against weekly fatalities you can get up into the few percent range. But you're really in the noise with these kind of numbers.]
