Apologies if this was posted before...but I didn't find it in a search.
If you google 'Proof Google is Using Behavioral Data in Rankings', the first article is article I'm referring to.
In the first test they had 100 participates try to manipulate search rankings...by just clicking the link from the SERP's... (for a competitive term). Then they did the same for a non-competitive ranked site with 65 participants. It didn't work for the competitive term, but it did for the uncompetitive term.
The second test was also quite interesting... They had 100 users from google SERP's click into a test site and 'spend' 30 seconds to 60 seconds on the site. The bounce rate decreased in GA and the average time on site increased dramatically. The ranking jumped from 80 to 33...but when they stopped their test...the rankings (and the GA results) returned to normal.
This is pretty convincing evidence at the least that google is using behavior data and the question is now to what extent it affects SERP's and what other factors are they measuring besides click-throughs, bounces and time on site? I assume this didn't work for the competitive term merely because the competitive term probably required a bigger sample size. Very scary to think of this happening on a bigger scale.
I first kind of stumbled on this idea when I was going over the stats for one of my personal sites. It has extremely good 'time on site averages' but competes for a lot of obscure long tail phrases. What I would notice is after somebody found my site and spent a decent amount of time there...I received a huge placement bump for these obscure terms...yet I did no changes to the site. After a while my 'reward' for netting that long lasting visitor wore off and my placements went crashing down. Again the correlation was seemed to be when people clicked on my link from the SERP's and spent time on my site position for the term they used improved...and vice versa when nobody clicked on that term. I then did some googling and found the above article.