Community Playbook #3: One Feature That Doubled User Visits (A 6-Month Experiment)

In Part 2, we covered two psychological forces that build community attachment: group identity and interpersonal bonds. The theory sounds solid. But theories are cheap. The question is: does it actually work when you build it into a real product?

A research team from the University of Minnesota and Carnegie Mellon decided to find out. They ran a six-month field experiment on MovieLens, a movie recommendation site with over 100,000 users. The results were striking.

The Setup

The experiment ran from January to July 2007. The team recruited 4,818 MovieLens members and randomly assigned them to one of three groups:

Control group (1,544 people): Same old MovieLens. No new features.

Identity group (1,625 people): Got features designed to strengthen group identity. They were assigned to one of 10 "movie groups" with animal names (Tiger, Eagle, Polar Bear, etc.). Each group was built using a clustering algorithm so members within the same group actually had similar movie tastes.

Bond group (1,649 people): Got features designed to strengthen personal connections. Instead of group pages, they got individual profiles, personal activity feeds, and tools to communicate one-on-one with other members.

What They Actually Built

Each condition got three features, designed as mirrors of each other:

Feature 1: Profile pages. The identity group got a group profile. It showed the group's name, icon, a fun description ("Bears love to watch sci-fi and fantasy blockbusters while not hibernating"), movies the group collectively liked, and rankings against other groups. The bond group got individual profiles with personal info, photos, and a comparison of movie tastes between the viewer and the profile owner.

Feature 2: Activity feeds. The identity group saw recent ratings and posts organized by movie group, with group names and icons. 80% of the content came from their own group. The bond group saw the same type of content but organized by individual members, showing names and personal photos.

Feature 3: Communication. The identity group could leave comments on their group's profile page, visible only to group members. The bond group could leave comments on individual profiles, visible to anyone.

The Results

After six months, the differences were clear.

Self-reported attachment. Compared to the control group, people in the identity condition reported 76% stronger attachment to their movie group (p < .001). They also reported 7% stronger attachment to MovieLens as a whole. The bond group reported 27% stronger attachment to their movie group (p = .004), but no significant increase in attachment to MovieLens overall.

Visit frequency. This is where it gets interesting. People in the identity group visited MovieLens an average of 7.15 times over the experiment. The bond group visited 5.52 times. The control group visited 4.96 times. That's a 44% increase for the identity group.

But the real outlier was a specific combination. Members in the identity condition who had access to both group profiles and the group activity feed visited an average of 11.6 times. Everyone else averaged 5.7. That's roughly double.

Forum engagement. The identity group viewed 36% more forum posts than the control group. The bond group showed no significant difference.

Newcomers vs. Old-timers: A Surprising Split

When the researchers split the data by experience level, a pattern emerged that matters a lot for anyone building a community.

Both types of features worked well for newcomers. Compared to newcomers in the control group (5.0 logins), newcomers with identity features logged in 7.8 times (56% more). Newcomers with bond features logged in 6.0 times (20% more).

For old-timers, the story was different. Identity features still helped a bit: 5.5 logins vs 4.8 in control (a modest 10% increase). But bond features actually backfired. Old-timers with bond features logged in only 4.2 times, which is 16% fewer than the control group.

Why? The researchers think it's because established members who came to MovieLens for movie recommendations didn't want social features pushed on them. As one member put it: "I do enjoy ratings, predictions, graphs and classifications. The social aspect of it doesn't mean anything to me."

Why Identity Beat Bonds

The researchers offered two explanations for why group identity features consistently outperformed bond-based features.

First, group identity is just easier to establish. Give a group a name and a logo, and people start identifying with it almost immediately. Tajfel showed this in 1971 with random labels. Bonds, on the other hand, take time. They require repeated one-on-one interactions and self-disclosure, which is hard in a community where most people visit once a month.

Second, MovieLens wasn't the kind of place where people came looking for friends. They came for movie recommendations. Group features felt like a natural extension of that experience. Personal profiles and one-on-one messaging felt like a different product entirely.

The One Thing That Didn't Work

Despite all the positive effects on engagement, neither set of features increased long-term retention. People visited more often and engaged more deeply during the experiment, but they weren't more likely to stick around after it ended.

The researchers suggest that online communities compete so directly for attention that retention depends more on whether a better alternative exists than on how attached someone feels. Someone who loves your movie community can still leave if a shinier one shows up. Engagement and retention, it turns out, are different problems that need different solutions.

In the final post of this series, we'll pull together everything from the research and turn it into concrete tactics you can apply to your own community.

---

Community Playbook series: ← Part 2 · Part 4: Steal These 5 Tactics →