011 min
The problem psychological safety names
Psychological safety is easiest to see in a team that lacks it. Picture a team of engineers in a release review. A backend engineer has noticed that a database migration may drop rows when it runs on production data. They have not tested it, so they are not sure. Two weeks earlier, the tech lead answered a similar late concern with a sigh and a reminder that the plan was already agreed.
The engineer does the sum that most people would do. If they speak and are wrong, they look careless. If they speak and are right, they delay a release that the lead cares about. If they stay quiet, nobody will ever know that they could have spoken. The backend engineer says nothing and the release goes out.
Nothing in that meeting looked unusual. Nobody argued, the review ended early and everyone agreed. The cost appeared a day later, as missing customer data that the team could not recover.
The belief that shaped the engineer's choice is a property of the team, not of the engineer. Another engineer on another team, with a lead who had thanked the last late concern, might have spoken within seconds. The rest of this article follows that belief: where researchers found it, how it works, what it did in real organisations, and what a person who leads others can do about it.
022 min
Where psychological safety comes from
Researchers noticed this pattern long before anyone measured it. The idea goes back to 1965, when the organisational researchers Edgar Schein and Warren Bennis wrote that people need psychological safety before they can feel secure enough to change.
The link to teams came from a surprise in a hospital. In 1996 Edmondson compared eight units at a university hospital and found that the units with stronger managers and better relationships had recorded more medication errors.
Why better units reported more errors
The result looks backwards, and Edmondson's explanation is the key to the whole idea. The error counts measured the errors that people were willing to report, not the errors that happened. In a unit where a nurse manager answered a mistake with blame, nurses had a reason to hide it. In a unit where managers were approachable, mistakes were reported and discussed, so they were counted. A low error count could therefore mean a safe unit or a silent one.
Three years later: a name and a measure
In 1999 Edmondson surveyed 51 work teams in one manufacturing company and found that psychological safety was linked to learning behavior, while team efficacy was not, once safety was taken into account. Learning behavior means actions such as asking for feedback, trying a change and talking openly about errors. Team efficacy is a team's confidence that it can do its job well. A confident team that does not feel safe, in other words, still does not learn.
In that paper Edmondson defined team psychological safety as a belief held by the whole team that it is safe to take interpersonal risks. That one sentence is the definition used in most research since.
032 min
How psychological safety works
Psychological safety works by changing what it costs a person to speak. An interpersonal risk is any move that could lower a person's standing with the people around them. Asking a question risks looking uninformed. Disagreeing risks looking difficult. Admitting a mistake risks looking incompetent. Offering a rough idea risks looking foolish.
Each of these moves also helps the team. The question finds a gap, the disagreement tests a plan, the admitted mistake can be fixed and the rough idea can be improved. People weigh the personal cost of speaking against that benefit, quickly and often without noticing. When the team has treated earlier risks kindly, the cost is low and speaking wins. When it has not, silence wins, even if the person is sure of the facts.
Where the belief comes from
Edmondson found that team members rarely argue for these beliefs. They hold them as tacit, taken-for-granted views of how things are around here. They form from what the team has actually seen happen, and the leader's reactions count most. A leader who is supportive, coaches and answers questions without getting defensive tells members that the team is safe. A leader who acts in an authoritarian or punishing way makes members reluctant to take the risks that learning needs.
In the release review, the tech lead's sigh two weeks earlier was exactly this kind of event. One reaction cost the lead almost nothing, but it taught every engineer in the room what a late concern would cost them. Because the lesson is shared, the belief belongs to the team and not to any one person. Two teams in one company can differ, because each learns its own rules from its own leader and its own history.
Safety is not an effect that switches on by itself
Edmondson's 1999 study also listed what makes a team safer: a clear goal, enough resources and information, and a leader who coaches. Safety is therefore something a team builds, one reaction at a time, and it can be lost the same way.
042 min
Psychological safety at Google: Project Aristotle
Psychological safety also showed up when Google went looking for what makes teams effective. The company's research group called the project Project Aristotle. It studied 180 teams: 115 project teams in engineering and 65 sales pods, with some strong teams and some weak ones.
The researchers collected hundreds of variables, from personality and skills to how it felt to work with teammates, and ran more than 35 statistical models on them. Who was on the team mattered less than how the team worked together. Of the five dynamics that stood out, psychological safety ranked first. The other four were dependability, structure and clarity, meaning, and impact.
Google measured safety by asking members how far they agreed that, on their team, a mistake is not held against them. Teams that agreed more were more likely to be in the group of effective teams.
What Aristotle did not show
Headlines turned this into a claim that safety causes performance. Google's own wording is more careful: it describes a strong team culture that was correlated with each member's view of the risk of speaking. A correlation does not say which comes first. A team that performs well may feel safer because it is winning. The hospital studies above point the other way, from safety to behaviour, but they are different studies in a different field. Treat Aristotle as a strong hint from a large company, not as proof.
052 min
Psychological safety in an operating room
Psychological safety matters most where the people who know something rank below the person who decides. Cardiac surgery is such a place. In a study published in 2001, Edmondson, Richard Bohmer and Gary Pisano followed 16 hospitals in the United States as they adopted a new way to operate on the heart through small openings.
The new method changed who knew what. A surgeon who works through small openings cannot see everything directly. Other team members read the monitors and tell the surgeon what they show. A method that depends on nurses and technicians speaking to a surgeon runs into a field where each role has a narrow, fixed speaking part.
The hospitals differed in how far their teams moved past that. At one successful hospital, a nurse said that nurses, surgeons and everyone else took part in the conversation about how communication would matter. The researchers concluded that psychological safety made it easier for team members to share information and to change behavior, both of which the new method needed.
When the setting works against safety
The same team members also said that safety depended on the surgeon. The researchers wrote that it could vary greatly between teams, and that the deep habits of cardiac surgery do not encourage the kind of coaching they saw in the successful teams. A strict chain of command makes every speaking turn a decision about rank. Safety in such a team has to be created on purpose, by the person at the top, because the structure does not create it.
061 min
What missing psychological safety cost NASA with Columbia
Psychological safety shows its value most clearly when its absence has large stakes. The space shuttle Columbia broke up on its return to Earth in 2003. The Columbia Accident Investigation Board found that the cause lay in NASA's history and culture as well as in the damage the shuttle took at launch.
The Board found that structures at NASA kept important safety information from moving and discouraged engineers and managers from disagreeing. Engineers made three requests for images to assess the damage, but sent them outside the normal channels. A manager who heard informally that a request existed checked with three colleagues, and none of them knew of a requirement for images. The concern was real, and the route to the people who could act on it was not one the engineers could safely use.
The Board did not use the term psychological safety. It wrote in the vocabulary of safety culture. Reading its findings with this idea is our reading, and a reasonable one: the same pattern as in the release review, at the scale of a national space programme, where a person with a concern met a system in which raising it formally and openly was harder than staying quiet. This case also fits the broader point about groupthink: a group that values agreement can lose the information that would change its mind.
072 min
Psychological safety vs. nearby concepts
Psychological safety is easy to mistake for ideas that sound close to it. The test that separates them is whether the idea is about what happens to a person who speaks up.
| Concept | What it is about | The question that separates it |
|---|---|---|
| Psychological safety | What happens to a person who speaks up | Will I be punished or embarrassed if I say this? |
| Group cohesion | How close and loyal the members are | Do the members want to stay together? |
| Comfort or niceness | How pleasant the team feels | Is it pleasant to be here? |
| Accountability | Whether people answer for results | Do people answer for what they promised? |
Cohesion is the confusion with the highest stakes. Edmondson pointed out that a close team is not automatically a safe one, because cohesion can make people less willing to disagree, as in groupthink. A close team may protect its agreement by keeping its doubts to itself.
Comfort and standards are separate from safety too. Safety says nothing about how high the standards are. A team can be safe and demanding, where people can say that a standard is not being met. A team can also be pleasant and careless, where nobody mentions it. Trust is a related idea but a different one: trust is one person's belief about another person, while psychological safety describes what a whole team believes about speaking up. Healthy disagreement at work needs safety, but safety does not mean there is no conflict.
082 min
What psychological safety means for your work
Psychological safety changes everyday decisions for anyone whom others depend on. The decisions look small, and they add up to what the team learns.
An engineer on call decides whether to report their own mistaken change at once. They are more likely to do it if the last incident review looked at the sequence of events rather than at the person who caused it. Blaming the person is a common attribution error: it explains a mistake by the person's character and overlooks the situation that made the mistake easy.
A product manager decides how to run a roadmap review. Asking each person for the strongest reason the plan could fail, before asking anyone to agree, makes disagreement part of the task. A missed estimate treated as news about the plan, not as a failing of the person, makes the next estimate more honest.
A designer decides how to run a critique. Rules that keep the comments about the work and not the maker let a junior designer say that they do not understand the problem. That statement is the most useful one a critique can get.
A founder with five employees is the whole environment. Every reaction to bad news is seen by everyone, and the first one sets the pattern. A founder who is certain about everything teaches the team that doubt is unwelcome, which is the opposite of what the founder needs to hear. Giving people real empowerment to stop work when they see a problem only works if the team believes that stopping it will not be held against them.
092 min
How to build psychological safety
Psychological safety is built by what a leader does in the first few seconds after bad news. These moves are small and can start in tomorrow's meeting.
- Say what you do not know, early. A leader who admits uncertainty makes it normal for others to do the same.
- Ask for the specific risk, not for questions in general. People rarely answer "any questions?", and they often answer a precise request.
- Answer the first piece of bad news with thanks and a question, not a defence.
- Treat a mistake as information about the process. Ask what made the error easy to make.
I might be wrong about the migration. What am I missing? Thank you for telling me. What would you need to check it?
Each reaction is also a lesson to everyone who watched. The same cycle that creates silence creates speech: how the leader reacts decides what the next person does, and that decides the next reaction.
Signals that it is working
- Concerns arrive before a deadline, not after it.
- Questions and disagreements come from several people, not only the most senior.
- Mistakes are described to the team by the person who made them.
Signals that it is not
- Meetings end quickly in full agreement, and the concerns appear later in private messages.
- Only senior people ask questions.
- Problems come up only once they can no longer be hidden.
To measure it, ask team members how far they agree with statements such as "If you make a mistake on this team, it is often held against you." Edmondson used items like this in 1999, and Google used them later. Read the result team by team, because the average for a company hides the team where a member stays quiet.
102 min
Where the evidence on psychological safety is limited
Psychological safety has a strong research base, but the base has limits. The founding studies came from single organisations:
| Study | Setting | Size |
|---|---|---|
| Edmondson 1996 | One university hospital | 8 units |
| Edmondson 1999 | One manufacturing company | 51 teams |
| Edmondson, Bohmer and Pisano 2001 | Cardiac surgery in the United States | 16 hospitals |
| Google, Project Aristotle | One technology company | 180 teams |
A larger check came in 2017. M. Lance Frazier and colleagues published a meta-analysis, a study that combines the results of many earlier studies, of 136 independent samples, drawing on more than 22,000 people in nearly 5,000 groups. One of their aims was to test whether study design and national culture change what psychological safety predicts. That is an open question for a practitioner: a result from a US factory or a large technology company may not describe a ten-person agency, or a team in a culture where disagreeing with a manager is rare.
There is also a limit on what a survey can show. Safety is measured by asking members what they believe, and the best evidence of an effect comes from studies that look at teams at one point in time. Such studies show that safety and good outcomes go together, and they cannot show which causes the other.
Where does that leave you? Use psychological safety as a strong explanation for why a team speaks up or stays silent, and check it on your own team by watching who speaks and when. Do not treat it as a guaranteed way to raise performance.
?5 questions
Questions people ask
Is psychological safety the same as trust?
Does psychological safety mean lower standards or no conflict?
Who coined the term psychological safety?
Does psychological safety work for small or remote teams?
What is an example of psychological safety?
§6 sources
Sources
Edmondson, A. (1999). Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350-383. A course reading copy of the paper.
Agency for Healthcare Research and Quality, PSNet. A summary of Edmondson, A. (1996), Learning from Mistakes Is Easier Said Than Done, Journal of Applied Behavioral Science, 32(1).
Google re:Work. Understand team effectiveness (originally published 2015), describing Project Aristotle.
Harvard Business School Working Knowledge. Inside the OR: Disrupted Routines and New Technologies, a summary of Edmondson, A., Bohmer, R., and Pisano, G. (2001), Administrative Science Quarterly.
Show all 6 sourcesShow fewer sources
Congressional Research Service (2003). NASA's Space Shuttle Columbia: Synopsis of the Report of the Columbia Accident Investigation Board, RS21606.
Frazier, M. L., Fainshmidt, S., Klinger, R. L., Pezeshkan, A., and Vracheva, V. (2017). Psychological Safety: A Meta-Analytic Review and Extension. Personnel Psychology, 70(1), 113-165. Repository record with abstract.