Upvotes and Downvotes on Social Media Sites Not Created Equal

PHOTO: A new study shows that upvotes and downvotes affect social media users differently.
Share
Copy

The social media site Reddit has risen in popularity as an Internet news hub, even grabbing the attention of President Obama.

One of site's distinguishing features is the way it delivers content: Users score Reddit stories with an upvote or downvote, which affects where the stories appear on the website.

But is there intrinsic power to the upvotes and downvotes themselves? A new study published this week in Science says yes. Sinan Aral, an associate professor of IT and marketing at MIT, saw that a quick upvote might have more sway than a quick downvote.

Instead of creating a social network from scratch, Aral and his colleagues observed how users of a Reddit-style social network reacted to more than 100,000 different comments. When a comment was initially created, Aral would immediately give it an upvote, a downvote or leave it alone. He then scored the comment, calculated as the number of upvotes minus the number of downvotes.

"We did the experiment on the comments and not the articles, because an upvoted article might get higher placement on the website," Aral told ABC News. However, comments on the social network were sorted only chronologically, regardless of the score. Aral's initial vote didn't affect where the comment appeared on the site, but it did affect how others received it.

Overall, upvoted comments received significantly higher scores than comments that were downvoted or left alone. In addition, comments that were downvoted ended up with the same scores as those that were left alone.

Aral found that users were effectively "neutralizing" the downvote. "We were surprised to see this correction effect," he said.

Social Media Makes for Better Student Writing, Not Worse, Teachers Say

A similar analysis on Amazon.com product reviews was conducted by Cristian Danescu, a researcher at the Max Planck Institute for Software Systems in Germany. He was also surprised by the move to correct, or neturalize, a downvote but wondered whether such reactions were universal.

"It would be interesting to [see] what extent this is dependent on the cultural background of the users," he said. "Would it be reversed in a cultural setting where negative feedback is highly valued?"

Although the study did not vote on the actual articles themselves, Aral did find different voting effects depending on the article.

"Stories about politics and culture had a strong social influence bias," he said. "But in other topics, it's nonexistent."

The findings in the new study carry over to other websites that rely on user feedback, such as Amazon and Yelp. Though those websites have spotted companies trying to buy five-star reviews, the effects observed in this study may be a back door approach for a business to raise its review score. "To what extent [does] the rating of the first reviewer influence that of subsequent reviewers and the success of the product?" asked Danescu. "The answer to this question can guide the design of more truthful review communities."

These observed effects might transfer from news sites into review sites, but Felix Reed-Tsochas, a professor of social sciences at the University of Oxford, said that the two types of sites were distinct from one another. "We are embedded in multiple social networks with many different types of ties," he wrote in an email. "It is not necessarily easy to understand how effects that are observed in one environment can be translated to another.

Regardless of the extent that Aral's findings might have on user reviews, any influence from other users can have major consequences. "User ratings are the second most important criteria to judge the quality of a product or service, after referrals from friends and family," he said. "I just had a baby and relied on online ratings to narrow down the doctors. That's a significant decision for which ratings matter."

Join the Discussion
blog comments powered by Disqus
 
You Might Also Like...