Why do the best teams report more mistakes?

A team of nurses spent six months tracking drug errors across hospital wards, reviewing patient charts several times a week. In the first month, a doctoral student named Amy Edmondson handed out a survey on teamwork in the same wards. She expected the well-run teams to make fewer errors.

The correlation came back statistically significant and pointing the wrong way. Teams that scored better on the survey had higher detected error rates. Edmondson has written about staring at that result and dreading the conversation with her supervisor. Then she asked a different question. What if the better teams simply reported more?

Two pieces of evidence

She had added one item to the survey: "If you make a mistake in this unit, it won't be held against you." That item also correlated with detected errors. The more people believed a mistake would not be held against them, the higher the count in their unit.

She then had a research assistant, Andy Molinsky, observe each unit without telling him its error rate or her new idea. He found units that felt like different workplaces. In some, nurses said a non-punitive environment was essential to good care. In others, a mistake meant you got put "on trial". When he ranked the units from most to least open, his list was nearly perfectly correlated with the detected error rates.

What a manager should take from it

The error count was measuring two things at once: the mistakes people made and their willingness to admit them. Edmondson later used the term psychological safety for that difference in climate. The lesson for anyone reading a dashboard is uncomfortable. A falling incident number can mean the team got better. It can also mean the team went quiet.

The same trap shows up in corporate innovation and in AI rollouts. A leader celebrates a quarter with no failed experiments and no reported model errors. That quarter may simply be the one where people stopped telling them.

Four things I would do with this

1. Read any error or incident count next to a measure of climate. Edmondson's survey item is a good start, and you can ask it in a team meeting.
2. Thank the first person who raises a mistake in a meeting, in front of the team, before you ask how it happened.
3. Visit the team with the suspiciously clean record and spend time on its floor before you praise the number.
4. Watch how the most senior person responds to bad news, because the rest of the team is watching too.

Our psychological safety workshop works with intact teams on the reply a leader gives when someone raises a problem. If a model is now part of the work, what psychological safety asks once a model is in the work covers the newer version of the same risk, which is saying out loud that the model got it wrong.