I love it that these guys are experimenting with content and analytics but the stats look contradictory. Bounce rate is 1- n_visits_with_more_than_one event/n

I love it that these guys are experimenting with content and analytics but the stats look contradictory. Bounce rate is 1- n_visits_with_more_than_one event/n_visits. The default GA setup doesn't register any events except page views so "more than one event" means more than one page view. However the charts in this article showed that while the bounce rate fell, the average pages per visit stayed constant. For bounce rate to drop you'd expect pages per visit to rise... unless there's also a new event being fired. My guess (as suggested by some commenters here before me) is that they shipped some type of event which fires for a fraction of visitors - scrolling a certain distance down the page perhaps. If this event fires, the visit won't (by default) count as a bounce. it will also effect time-on-site. Time on site is normally zero unless you visit another page or fire an event. In this case it's determined by the time of the last page view or event that google registers. If they did ship an "on 400px scroll" event (for instance) then for each visitor for whom the event fires for, the "time on site" will be registered as the time of the event. For my money, I think that bounce rates for content-driven sites should actually be set by a threshold time-on-site. Most visitors don't read more than one page so it's a very weak indicator of a "bounce". Someone could read for 10 minutes and still count as a bounce which isn't a fair appraisal of a very successful article. The raw "bounce metric" makes sense for a transactional site where onepage view == no Revenue but makes little sense for a site where one ten minute read of an article is exactly what the author hopes for.