On sites like Reddit, Imgur, and Genius, the cream always rises to the top — or gets pushed there anyway. By virtue of voting aggregation, the most upvoted posts are the ones most likely to be read. This is all by design, but, even so, cream isn’t actually good for anyone; it just tastes good. And, like our taste buds, internet denizens may struggle to sort the good from the bad, much less the great from the good.

The psychology of upvoting is mostly unexplored, even though we use sites powered by voting aggregation all the time. The sub-subfield’s father is Dr. Sean J. Taylor, a computational social scientist at Facebook who helped pioneer the field by co-authoring a research paper on the science of upvoting in 2013. Using a real site very similar to Reddit (it had to remain anonymous), Taylor’s team added either a single upvote or downvote to a fraction of 100,000 total posts as they were posted. Adding a single upvote early in a post’s life caused subsequent upvotes to snowball, resulting in a 32 percent higher chance that it would receive additional positive votes.

“The presence of upvotes tends to cause more upvoting,” Taylor told Inverse. “Comments with either upvotes or downvotes receive more attention, and more subsequent voting.” This snowball effect is a phenomenon known as “positive herding.” There’s no question that it happens IRL — the Reddit hivemind is very much a thing — but why it happens is still unclear.

Taylor has his theories. People use simple cues to make value judgments, he says, and on sites like these, an upvote is a pretty convenient indication that something is worthwhile, especially when users aren’t exactly sure what to think. It could be because a highly rated post is more likely to draw attention and, therefore, more votes. Or, like so many of our internet tendencies, positive herding might simply point toward our desire for conformity.

The content of posts does matter, of course, but perhaps not as much as a reader would expect. And it’s hard to quantify the importance of actual sentiment; Taylor’s study was designed to estimate the average effect of voting manipulation over many posts and it’s hard to control for quality without making subjective judgments. The theory that people use simple clues to help them make their decisions, however, implies that positive herding has an even stronger effect when the people doing the voting aren’t sure about the post’s quality in the first place. Our collective upvoting habits, it seems, have somewhat defeated the point of voting aggregation.

“I think that they do discount the validity,” says Taylor. “For those using ratings for decision making, at least two types of mistakes are possible — the items with the best ratings may not be the best, and some great items may be unlucky enough to have not received many positive ratings at all.”

Still, opinion aggregation isn’t completely worthless. With an upvote-downvote ratio of 6 to 1, according to Taylor, the hivemind tends to enable more ideas than it impedes. His team’s study merely serves as a gentle reminder that, like all things on the internet, boosterism should be met with skepticism.