I like this approach, but the problem is that I'm measuring bounce rate for a registration page, so a single page visit is likely in this case. Maybe I can add on to your calculation with other contraints, where scroll events don't exist, or where time on page is low.