For HR & Culture Leaders Article · November 2026 ~12 min read

Psychological Safety: Definition, Research, and Limits

Psychological safety is a shared belief held by members of a team that the team is safe for interpersonal risk-taking. Amy Edmondson defined it in 1999 and it has been measured that way ever since. This is the reference version: the definition, the research history, what the construct does not cover, and where a team needs something other than more safety.

By Jeff Morris, "The High Trust Guy" DreamSmart Behavioral Solutions

Most of what circulates about psychological safety is a paraphrase of a paraphrase. The term has been in general management vocabulary for about a decade now, long enough that it gets used to mean comfort, morale, niceness, or whatever the speaker wishes the team had more of. That drift is a problem for anyone who has to defend a training budget to a skeptical executive, because the moment the definition goes soft, the evidence behind it stops counting.

This page is the reference version. The definition as its author wrote it, where the research came from, how the thing is actually measured, what it does not cover, and the point at which a team needs something other than more safety. The last part is a position, and it is labelled as one when it comes. The first half of this page is not opinion. It is other people's work, credited.

What is psychological safety?

Psychological safety is a shared belief held by members of a team that the team is safe for interpersonal risk-taking. That definition belongs to Amy Edmondson, a professor at Harvard Business School, who set it out in 1999 in "Psychological Safety and Learning Behavior in Work Teams" in Administrative Science Quarterly. In practice it means team members believe they will not be embarrassed, rejected or punished for speaking up, asking a question, admitting a mistake or offering an idea that might not work.

Two parts of that definition carry most of the weight and are the first to go missing when the term gets repeated. It is a shared belief, which makes it a property of a group climate rather than a description of any one relationship inside the group. And it is about interpersonal risk, the social cost of speaking, rather than about physical safety, job security or emotional wellbeing. Everything else written on this page depends on those two words holding.

One more distinction belongs in the definition rather than further down the page. Psychological safety is not a personality trait and not something an individual carries between teams. The same person can be candid on one team and silent on another in the same building two floors up, reporting to a different manager, and nobody learns anything useful by treating that as a fact about them. The unit is the team.

Where the concept came from

The route to the 1999 paper runs through a hospital. In earlier research on medication errors in nursing units, published in the Journal of Applied Behavioral Science in 1996, Edmondson found that the better-led units were associated with higher rates of detected and reported error, which is the opposite of what she expected. The reading she took from it was that the units were not making more mistakes. They were surfacing more of them, because the climate made reporting survivable. That finding is what the construct was built to explain.

The 1999 paper is the one that defined it. Edmondson studied 51 work teams inside a single manufacturing company and found that team psychological safety was associated with learning behavior, asking questions, seeking feedback, discussing error, and that learning behavior in turn related to team performance. That chain matters more than it is usually given credit for. Safety was not proposed as a direct cause of performance. It was proposed as the condition under which a team learns, and learning is what moves performance.

The term reached general management use through a different door. Google's Project Aristotle, reported by Charles Duhigg in The New York Times Magazine in February 2016, identified psychological safety as the most significant of the dynamics that distinguished the company's effective teams. That article did more to put the phrase into everyday corporate speech than the preceding seventeen years of academic work, which explains both the reach of the idea and the looseness of how it is now used.

What psychological safety is not

This is the section worth bookmarking, because most confusion about the term traces back to one of these six.

  • It is not niceness. A safe team often argues more than an unsafe one, because disagreement has stopped being expensive.
  • It is not comfort. Being told your analysis has a hole in it is uncomfortable, and a safe team does it anyway.
  • It is not a lower standard. Edmondson's own framing pairs psychological safety with high standards, not in place of them. Safety without standards produces a pleasant team that ships late.
  • It is not consensus. People can be free to speak and still be overruled by a decision they disagree with, as long as the disagreement itself did not cost them anything.
  • It is not the absence of conflict. It is the absence of punishment for conflict, which is a different condition and often looks louder from the outside.
  • It is not the same thing as trust. A shared belief about a room and an observable pattern between two people are different measurements of different things.

The practical consequence is that "we have psychological safety" and "everyone here gets along" are not the same claim, and the second one is often used as evidence for the first.

How it is measured

Psychological safety is measured with a climate survey. Edmondson's original instrument is a short set of agreement statements, answered by every member of the team, with several items reverse-scored so that agreement signals low safety rather than high. The best known of them asks whether a mistake made on this team is often held against you. Individual responses are then aggregated into a team-level score, because the construct was defined at team level and a single person's answer is not the unit of analysis.

That design does what it was built to do, and it has three blind spots a practitioner should know before presenting the results.

The first is that it measures belief, not behavior. It tells you what people expect would happen if they took a risk, which is not the same as a record of what happened the last four times somebody did.

The second is aggregation. A team mean of 4.1 can be produced by twelve confident people and four quiet ones, and the four are usually the ones the score was supposed to find. Look at the distribution before you look at the average, and look at it by tenure.

The third is timing. The same team surveyed two weeks after a restructure announcement returns a different score than it did the month before, and neither score is wrong. Both are the climate on the day. Anyone treating a single administration as a stable trait is reading it incorrectly.

The four stages, and who they belong to

The four stages model is frequently attached to psychological safety without attribution, so to be clear: it belongs to Timothy R. Clark, who set it out in his 2020 book The 4 Stages of Psychological Safety. The stages are inclusion safety, learner safety, contributor safety and challenger safety, each one a level of permission a team grants in exchange for more vulnerability from the person receiving it.

Clark's model is a practitioner framework built on Edmondson's construct rather than a replacement for it or a competing measurement of it. Its usefulness is diagnostic: when a team's survey score lands in the middle and nobody can say what is missing, the stage names give you somewhere specific to look.

Psychological safety and trust

These two get used interchangeably and they are not interchangeable. The clearest way to hold them apart is to ask five questions of each.

QuestionPsychological safetyBehavioral trust
What is itA shared belief that the team is safe for interpersonal risk-takingA pattern of observable behaviors between specific people
What level does it operate atThe group. One climate, one scoreThe relationship. Person to person, and role to role
How is it measuredA climate survey, aggregated to team levelScored behavior across defined dimensions, such as the 12 Cs of Trust in the Team TrustFlow Index
What does it predictLearning behavior, speaking up, error reporting, and through those, performanceFollow-through, reliability under pressure, and team performance directly
What does it missWhether what people say, once they are willing to say it, can be relied onWhether people feel able to speak in the first place, which it assumes

The comparison goes deeper in is psychological safety the same as trust, which is the post to read if you only want that one distinction. The short version for this page: a team can hold a lot of one and very little of the other, and the survey you ran will not tell you which. Patrick Lencioni’s model makes a related point with vulnerability-based trust at the base of its pyramid; see the Five Dysfunctions of a Team, explained.

What the research supports, and where the claims outrun it

The construct is well supported, and I want to be careful to say so before I say anything else. Psychological safety has been studied for more than twenty-five years across healthcare, manufacturing, software and aviation, it is associated with learning behavior, error reporting and speaking up, and the underlying instrument has held up across repeated use. If a leader in the room is treating it as a fashion, that position is not defensible against the literature.

The overreach is in the retelling rather than in the research. Three claims in particular travel further than the evidence carries them. That safety causes performance directly, when Edmondson's own chain runs through learning behavior. That a single team score is a stable property of the team rather than a reading on a day. And that more safety is always better, which nobody with a serious instrument has demonstrated and which ignores the standards half of the pairing entirely.

There is also a limit in the shape of the evidence itself. Most of this work is field research built on self-reported climate measures, which means the accurate verb is usually associated with rather than causes. That is not a weakness peculiar to psychological safety. It is true of most organizational research that cannot randomize people into teams. It is still worth saying out loud in the room, before the one person who knows it says it for you.

Then there is DreamSmart's own position, labelled as such: psychological safety is necessary and insufficient. That is professional judgment from 30 years of executive coaching and organizational consulting, not a head-to-head study, because no published study we know of directly compares safety climate and behavioral trust as predictors of the same team outcomes. What the trust side can point to is the meta-analytic work. Across 112 independent studies and 7,763 teams, intrateam trust showed a positive relationship with team performance (ρ = .30) that held after controlling for trust in the leader and past performance (De Jong, Dirks and Gillespie, 2016, Journal of Applied Psychology). That is a separate line of research about a separate construct, which is exactly the point.

Why safety alone does not finish the job

Psychological safety is the floor. It is not the building. A team needs enough of it that people will take the small interpersonal risks that everything else depends on, because a silent team gives a leader nothing to work with and nothing to correct. Without the floor, nothing else on this page is reachable.

What a floor does not do is determine what gets built on it. Once people are willing to speak, the next question is whether what they say holds: whether the commitment made in the meeting survives the week, whether the person who took the action closes the loop, whether the same standard applies when the deadline moves and nobody is watching. That is behavioral trust, and in the TrustFlow™ methodology it is scored as twelve observable behaviors in four quadrants, Foundations, Essentials, the Work and Results, rather than as one climate score.

The failure mode this produces is specific and quiet. A team reports high safety, everyone speaks freely, the meetings are candid and well attended, and the commitments made in those candid meetings still do not land. Nothing in the safety score predicted that, because nothing in the safety score was ever asking about it. The leader is left managing half a system and reading the other half as noise.

How to build psychological safety at work

A team calculates how safe it is from the leader's reaction to the first mistake, the first disagreement and the first piece of bad news. The offsite does not set it and neither does the open-door policy. The invitation is free to issue. The reaction is what the team writes down. One punished disclosure outweighs several unpunished ones, because the punished one carries information about the ceiling.

So the build is mostly rehearsal, and four conditions do most of the structural work:

  • A predictable response. The team should be able to guess your reaction to bad news before they deliver it, and be right. A leader who is consistently direct is easier to bring problems to than one who is usually kind and occasionally sharp.
  • Blame and analysis as separate steps. Run the mechanism first, in full. Accountability conversations, where they are warranted, happen afterward and usually in private.
  • Visible fallibility. The leader names their own error early and specifically. "I read that forecast wrong and we staffed for it" costs about four seconds and is the cheapest Micro-Trust Activation™ available to a manager.
  • A standing route for bad news. A fixed, scheduled slot where unwelcome information is expected rather than exceptional.

Put them in that order, one at a time, over roughly ninety days, which is about how long a team needs to collect enough new data points to revise its estimate. Running all four at once asks the team to believe four new behaviors simultaneously, and it will discount all of them.

Psychological safety examples you can observe

A survey tells you what people believe about the room. Observation tells you what they do in it. If you only get to watch three things, watch the first disagreement, the first mistake and the first piece of bad news after a team forms or a new leader arrives, because those three set the price of every one that follows. Some paired examples, with what it looks like when safety is present and when it is missing:

  • An analyst says the demand number in the deck is wrong, the VP asks which input is off, and the deck changes in the room. Missing: the analyst sends the correction privately afterward, and one person decides whether it travels.
  • In an incident review, the engineer who pushed the change says "that one was mine, here is what I did not check" in the first two minutes. Missing: forty minutes of timeline reconstruction and nobody uses the word "I".
  • A report tells their manager eleven days out that the date will slip. Missing: the slip appears on the due date with a recovery plan already attached.
  • A new hire asks a basic question in a public channel in week three and gets an answer with no joke attached. Missing: new hires ask in direct messages, and the channel history shows it.

What psychological safety training should include

Most psychological safety training is a session about the concept, and three months later nothing has changed, because the session issued another invitation and did not touch the reaction. Training that holds starts with a baseline of where trust is actually leaking on the team, works on the leader's response behaviors before anyone else's, installs the standing rituals described above, and re-measures. That is how DreamSmart's psychological safety training is built: a Team TrustFlow Index (TTFI) baseline, facilitated workshops mapped to the team's specific gaps, leader coaching, weekly safety rituals, and the TTFI run again every six to twelve months. If you want a quick read first, take the free Trust Score, or book a call to talk through your team.

Frequently asked questions

What is psychological safety?

Psychological safety is a shared belief held by members of a team that the team is safe for interpersonal risk-taking. Amy Edmondson defined it that way in her 1999 paper in Administrative Science Quarterly. It means people believe they will not be embarrassed, rejected or punished for speaking up, asking a question, admitting an error or offering an idea. It is a property of a group climate, measured at team level, rather than a description of how any two people treat each other.

Who created the concept of psychological safety?

Amy Edmondson, a professor at Harvard Business School, defined team psychological safety and built its measure in 1999 in Administrative Science Quarterly, in a paper titled Psychological Safety and Learning Behavior in Work Teams. Earlier researchers had used the phrase, but Edmondson's paper is the one that defined it at team level, built the instrument that measures it, and linked it to learning behavior. Timothy R. Clark's four stages are a later practitioner model built on top of that construct.

Is psychological safety the same as trust?

No. Psychological safety is a shared belief about a group's climate and is measured as one team-level score. Trust is observable behavior between specific people: whether commitments get kept, whether loops get closed, whether the standard holds under pressure. The two usually move together, and a team can hold a lot of one and very little of the other. That gap does not show up on a safety survey because the survey was never asking about it.

How do you measure psychological safety?

With a climate survey. Edmondson's original instrument is a short set of agreement statements answered by every member of a team, several of them reverse-scored, then aggregated to a team-level score. It captures what people believe about the room. It cannot see what any individual does after the survey closes, it cannot separate a confident majority from a quiet subgroup, and it reads differently the week after a layoff than it did the month before.

How do you build psychological safety at work?

By changing how the leader reacts, not by issuing more invitations to speak up. Make your response to bad news predictable, separate the analysis of a mistake from any blame conversation, name your own errors early, and create a standing slot where unwelcome information is expected. Put those in one at a time over about ninety days, because the team is scoring repetition rather than announcements.

What is psychological safety training?

Training that changes the behaviors a team uses to decide whether speaking up is safe, starting with the leader's reactions. Good programs measure a baseline first, work on specific behaviors rather than the concept, put standing rituals in place so the work continues after the session, and re-measure. A session that only explains the concept tends to fade within a quarter.

You have measured
the climate.

The Team TrustFlow Index scores the behavior underneath it, across twelve dimensions, in under ten minutes.