When Google set out to find what made its best teams great, it expected the answer to be talent: the smartest people, the most experienced, the right mix of skills. It found something else entirely. The single strongest predictor of team performance was psychological safety, whether members felt safe to speak up, admit mistakes, and ask questions without being punished for it. Amy Edmondson’s research had reached the same conclusion years earlier, and from a counterintuitive starting point. Here is what psychological safety actually is, why it drives performance through learning rather than comfort, and how to build it.
The Hospital Study That Found the Opposite
Edmondson’s research origin is counterintuitive. As a doctoral candidate at Harvard, she designed a study to test the hypothesis that better-performing hospital nursing teams made fewer medication errors. She had every reason to expect a clean positive correlation: better teams, fewer mistakes.
The data showed the reverse. The better-performing nursing teams reported more medication errors than the lower-performing teams. The initial reaction, that her hypothesis was simply wrong, dissolved on closer inspection.
The better teams weren’t making more errors. They were more willing to report them. The lower-performing teams were hiding errors, minimizing them, or simply not surfacing them in contexts where they might reflect poorly on individual nurses. The difference was not mistake frequency. It was willingness to speak.
This observation shaped Edmondson’s subsequent research program. The question became: what is the organizational condition that determines whether team members are willing to speak up, report problems, ask questions, and admit uncertainty?
The 1999 Administrative Science Quarterly Paper
Edmondson formalized the framework in a 1999 paper published in Administrative Science Quarterly: “Psychological Safety and Learning Behavior in Work Teams.” The paper studied 51 work teams in a manufacturing company, not the hospital setting that inspired the question.
The core finding: psychological safety predicted learning behavior, and learning behavior predicted performance. The relationship between safety and performance was not direct. It ran through a mediating variable. Teams that felt safe to speak up engaged in more learning behaviors (seeking feedback, experimenting, discussing errors, asking for help). Those learning behaviors were what drove superior performance outcomes.
The causal chain matters for intervention design. You cannot shortcut from “we want better performance” to “let’s declare ourselves psychologically safe.” The path runs through the learning behaviors that safety enables. If those behaviors are absent, the performance benefit does not materialize regardless of the stated culture.
What Psychological Safety Is Not
Edmondson is precise about this. Psychological safety is not:
- Comfort. Psychologically safe teams can have high standards, difficult feedback, and demanding expectations. Safety is not the absence of pressure.
- Consensus. Teams with high psychological safety disagree openly, which is the point. Conflict of ideas is a sign of safety functioning correctly; its absence often signals suppressed dissent.
- Nice culture. Team members can be blunt and direct within a psychologically safe environment. Safety describes willingness to take interpersonal risk, not warmth of interaction.
- Trust in individuals. Psychological safety is a property of the team context, a shared belief about what is safe to do here with this group, rather than a bilateral trust relationship between specific individuals.
Google’s Project Aristotle
In 2012, Google launched an internal research initiative called Project Aristotle to identify what separates high-performing teams from low-performing ones. The project analyzed 180 Google teams across multiple years, coding for team composition, skills, individual talent levels, interpersonal styles, and dozens of other variables.
The strongest single predictor of team effectiveness was psychological safety, more predictive than team composition, collective IQ, technical skills, or experience levels. Google’s independently derived conclusion matched Edmondson’s: whether team members felt safe to take interpersonal risks determined how well the team performed, more than who was on the team.
The convergence of two independent research programs (one from academic study of manufacturing teams, one from internal analysis of tech teams) using different methodologies and data sources strengthens the underlying finding considerably.
The Counterintuitive Diagnostic
One practical implication of Edmondson’s research: the absence of reported problems is not evidence of health. A team that never surfaces conflicts, errors, or concerns is not necessarily performing well. It may have low psychological safety. The silence reflects suppression, not absence of problems.
The manager who interprets smooth meetings and harmonious team dynamics as signs of excellence should ask whether the team is psychologically safe enough to tell them what is actually wrong. The answer often reveals that the apparent consensus is produced by the same dynamic that hid medication errors in Edmondson’s hospital study.