It’s also probably looking at scroll speed. So if the people conducting the experiment tended to linger longer examining content they disliked, that could result in getting more of it.
Would need to see a more detailed explanation of the methodology. Ideally the scrolling was done in an automated way, at a consistent speed.