Longer suspensions on platforms like Roblox could help curb bad behavior, new research finds

Social platforms are constantly trying to strike a balance when it comes to managing bad behavior. How do you crack down on harassment and cyberbullying without slipping into censorship that drives people off your platform?

A new study finds one potential solution: longer suspensions for misbehavior.

The research, conducted by a Northeastern University researcher and Roblox, the massive online gaming platform, found that longer suspensions for users with reported bad behavior on Roblox were effective in two ways. It not only reduced the likelihood that users would repeat their bad behavior — it also didn’t have much impact on how much time people spent on the platform.

“That’s what you hope for, that a reprimand or consequence is a targeted intervention where players reduce that kind of behavior but are not reducing more positive interactions with other players,” says Jeffrey Gleason, a doctoral student at Northeastern who worked as a data science researcher on Roblox’s study.

With an average 85.3 million daily active users, Roblox is one of the largest online gaming platforms in the world. It allows users to design and create their own games and then share those games with other people on the platform.

Although most online platforms have community standards policies, Gleason says there has been surprisingly little evidence for how effective specific kinds of consequences are.

“This is especially true when you start asking questions about the frequency of a consequence or its duration or the sequencing of consequences,” Gleason says. “That was really the motivation [of this study]: to understand the impact of this one parameter, the length of a temporary suspension, and also internally to improve Roblox’s policies around dealing with violations of community standards.”

The researchers ran two large-scale experiments involving close to 800,000 Roblox users. The first experiment compared the impact of one-hour suspensions and one-day suspensions on first-time violators. The second experiment focused on one-day versus three-day suspensions for users who were on their second violation.

Both experiments analyzed players 13 years or older who had violated Roblox’s community standards policies around harassment, bullying and scamming, which includes asking for personal identifiable information.

While there were some key differences in the findings of both experiments, Gleason explains that they told a largely similar story.

“In both experiments, the longer suspension reduced consequence-related metrics, reduced reoffense rate, number of consequences, number of user reports against the violating user,” Gleason says.

Longer suspensions in both experiments didn’t push people away either. There was very little reduction in how much time people spent in Roblox. Across both experiments, the time a user spent playing Roblox decreased by at most 2%.

The researchers did find a difference that longer suspensions curbed future bad behavior much more effectively in first-time violators than in people with multiple violations. In the first experiment alone, longer suspensions reduced the reoffense rate for people with one violation by 13%. Frequent violators only experienced a 4% decrease in their reoffense rate.

“We think that can help us potentially look at differentiating consequences even more, taking different approaches, which is something … the industry is exploring more broadly,” Gleason says.

The study is also a reminder that one strategy on its own is not a silver bullet. The researchers observed that longer suspensions helped to curb bad behavior in users for at least three weeks but the impact faded over time. Gleason says it’s further motivation to look at using other tools in concert with longer suspensions, including “more proactive or educational [strategies] that could provide more durable effects on behavior.”

The findings from this study have already informed Roblox’s own suspension policies, but Gleason says the hope is for other platforms to learn from this research as well.

“We’re hoping that other platforms look at the targeted effectiveness of consequences here and the tradeoff with engagement and can maybe use that to inform their own policies,” Gleason says. “There’s not a lot of research on the effectiveness of suspension on social platforms even though most platforms are grappling with that question.”