I couldn't find the article without passing through a paywall. But this article [1] summarizes the methodology decently:

"We used a site very similar to Reddit," Aral says. Here's how it worked: The researchers convinced an anonymous website–under the promise it would remain nameless–to allow them to test slight manipulations in online voting. Over six months, the research team took a small percentage of more than 100,000 written comments and slightly altered their vote rating as they were submitted. Four percent of the comments were given a positive edge (a single upvote) while another 2 percent were struck with an immediate disadvantage (a lonely downvote).

[1] http://www.popularmechanics.com/science/health/nueroscience/...