freejung - 5:02 pm on Feb 28, 2011 (gmt 0)

But why did they return and click on another one? Did the first one not answer the question? ...

I would think that with a large volume of data you wouldn't want to try to work out these kinds of questions on a case-by-case basis. Rather, you would want to develop several independent quality metrics, and then test them against each other and against known datasets to extract patterns, then apply those patterns to new datasets and test again.

User behavior will vary a lot from user to user, from site to site, and from one niche to another, but on a large scale patterns will emerge.