Likes breed likes: Study finds 'herding' behavior among commenters

Livefyre comments
Many comment systems have built-in promoting and demoting tools; the one shown is from Livefyre.

The comments sections of websites are home to a herd mentality — but rather than the stream of spam and abuse one associates with Internet comments, there seems to be a bias toward positivity, according to a new study. These "positivity bubbles" could offer insight on other bubbles — like those in investment and real estate.

The study, by researchers at MIT, New York University and the Hebrew University of Jerusalem, monitored and subtly manipulated the comment section on "a major news-aggregation website" for five months, though the authors coyly declined to identify the site.

The comments on the site researchers used were shown in chronological order, with a tally of up votes and down votes from other readers and commenters. What MIT's Sinan Aral and his colleagues did was to give each comment a slight nudge in one direction right after it was posted — a single vote in favor, or a single vote against — and see if that changed how the comment was received.

They found that the single positive vote often created a snowball effect, leading to increased visibility and an overall greater amount of up votes — 25 percent on average, to be exact. Meanwhile, comments given a bad start also had an amplified response — both negative and positive. These "corrective" up votes kept such comments from being totally buried.

"One intuitive explanation is that we tend to go along with positive social influence but be skeptical of negative social signals of value and quality," wrote Aral, in an email to NBC News. "Positive social influence creates herding behavior and positive ratings bubbles, while negative social influence does not have the same effects."

In other words, commenters trust others to promote things, but not to demote them — perhaps thinking someone's opinion shouldn't be smothered because of one or two people who disagree.

But positive behavior doesn't always have positive results, as Aral notes.

"These positive ratings also represent bias and inflation," he says in an MIT press release describing the study. "Stock bubbles represent a positive herding, and they can be dramatically bad in the wrong context."

Could this be a strike against the much-vaunted "wisdom of crowds," by which the best answer supposedly emerges as if by magic from a group of people working together on submitting and sifting information? After all, it might just be the one that got a few votes early by chance.

Aral certainly thinks so, telling MIT that "it’s hard to distinguish the effect of high quality from the effect of social influence bias."

That said, "our message is not that we should do away with crowd-based opinion aggregation." Indeed, whether it's talking about the housing market or an online comment section, all this information is surely still immensely valuable, even if it takes a bit of interpretation or must be taken with a grain of salt. We just need more data to put it all together, he told NBC News.

"Understanding how these things work at population scale is essential to effectively harnessing the wisdom of crowds and taking advantage of collective intelligence," he said.

The study, "Social Influence Bias: A Randomized Experiment," appears in the Aug. 9 edition of the journal Science.

Devin Coldewey is a contributing writer for NBC News Digital. His personal website is