Online comment systems reveal multiple layers of social bias

quote:

---

In the latest issue of Science, a group of researchers decided to see whether people's tendency to follow the crowd extended into online ratings. They tackled this question by carrying out a large-scale experiment in which they rigged the comment ratings on a website. The website they used is a news aggregating website (like reddit) that lets users rate comments and contributions by either “upvoting” or “downvoting” them. The comment’s score, which can be seen by all the website’s users, is simply the raw difference in the number of positive and negative ratings.

---

quote:

---

Comments that were randomly selected to get an upvote were much more likely to also be upvoted by subsequent users; the first user to see a positively manipulated comment was 32 percent more likely to upvote it themselves as well, compared to a comment in the control group. By the end of the 5-month study, the comments in the positive treatment group had ratings 25 percent higher than those in the control group. The researchers refer to this phenomenon as “positive herding behavior,” which essentially means that users tended to jump on the upvoting bandwagon, helping these comments accumulate positive ratings.

---

quote:

---

In the negative treatment group, the results were a bit different. Just as in the positively manipulated group, the comments that the researchers downvoted were also more likely to be downvoted by subsequent users. However, these downvotes were offset by what’s called a “correction effect”—the negatively manipulated comments were also more likely to receive positive ratings down the line.

---

quote:

---

Overall, the researchers think there are three main phenomena that contribute to these findings. First, both experimental conditions increased turnout; in other words, a comment that is rated once is more likely to be rated again, no matter the type of vote it has received. Second, seeing an upvote does help users develop a positive opinion, especially if the user doesn’t have much prior experience with the comment’s author. Finally, users on the site had a natural tendency to upvote rather than downvote.

---