Forming student teams

Should teammates be alike or different? Three of our studies of student software teams give a more careful answer than the usual advice. Below is what they found, what they did not find, and a tool for building teams from personality or other scores.

What the studies found

Similar teams started better, and the advantage faded

In two earlier studies, teams whose members had similar Big Five profiles (N = 129, 34 teams) or similar cultural values (N = 141, 39 teams) reported more cohesion and did better early in the course. In the culture study, mixed teams did better on the last project instead.

Mixed teams felt slightly worse at the time, but students preferred them looking back

In the latest study (N = 219), students rated mixed teams slightly lower on cohesion (d = −.23), satisfaction (−.19), wanting to work together again (−.18) and belonging (−.15). Psychological safety and learning did not differ. But when asked to compare the two teams directly, students favored the mixed one (d = .50), most of all for “challenged me to think differently” (d = .72).

Feeling personally different mattered more than a varied team

A student who felt different from their teammates reported worse outcomes than one on a team they saw as varied (−.62 vs −.35). That is why the builder below flags a student who sits far from everyone else on their team.

Fair shares of the work predicted wanting to work together again

How evenly the work was shared was the strongest predictor of wanting to work with the same team again (β = .28). This was measured at the same time as the outcome, so it shows an association, not a cause.

What they did not find

That similar-then-mixed is the right order. Every student had the similar team first, so the order was never varied. Similar-first remains a hypothesis.
That conscientiousness is the trait to match on. An earlier draft reported that a wider spread of conscientiousness predicted worse teams (r = −.64, 15 teams). With 52 teams it did not hold up (r = −.17, p = .23). Our workshop materials weighted conscientiousness 2.0; the builder starts every trait at 1.
That belonging protects mixed teams. There was a hint of it (b = .42, p = .022) but not enough to rely on.
Anything about AI or remote teams. These were in-person student teams working without AI tools.

What to do with it

Mixed teams are a reasonable default: they cost a little comfort and students value them afterwards.
Make shares of the work visible: assign roles, hold short check-ins, and use peer evaluation that asks about contribution.
Watch students who are the only one of their kind on a team. Move them, or check in with them early.
Have each new team agree on deadlines, how they will communicate, and what counts as done before they start.

A hypothesis about AI, not a finding

The engagement model suggests a link these studies did not test: teammates who use AI in different modes could stretch each other the way mixed personalities did, and a student whose AI habits differ from everyone else's could feel like the odd one out. Treat that as something to watch, not a rule for forming teams.

Team builder

Names stay on your computer

This tool runs in your browser. Your file is read here, teams are built here, and nothing on this page sends it anywhere. You can use pseudonyms from class access codes instead of names. The same file, settings and seed always give the same teams.

1. Your class file

A CSV with one row per student: a name or pseudonym column, then one column of numbers per trait (for example the five IPIP scale averages). Columns that are not numbers are ignored.

Sources

Anderson, G., Keith, M. J., Francisco, J., & Fox, S. (2018). The effect of software team personality composition on learning and performance. Proceedings of the 51st Hawaii International Conference on System Sciences, 3476–3485.
Anderson, G., Keith, M., Albrecht, C., Sprill, A., & Pettit, C. (2019). Optimizing software team performance with cultural values. Proceedings of the 52nd Hawaii International Conference on System Sciences, 3–11.
Keith, M., & Anderson, G. (2026). Personality-composition transitions: Concurrent ratings, comparative appraisals, and felt dissimilarity [Unpublished manuscript]. Brigham Young University.