Psychological Safety: The precondition for everything else.
Why a concept from a Harvard hospital study became the most important variable in occupational safety.
In 1996, a young Harvard doctoral student named Amy Edmondson set out to prove what everyone already believed about hospital teams. Higher-performing nursing units, she hypothesized, would make fewer medication errors than lower-performing ones. She spent hundreds of hours building the survey, attending biweekly meetings with the physicians and nurses tracking drug errors at two Boston-area hospitals, and bicycling to interview teams within hours of major incidents. The data came back and showed the opposite. Better teams reported more errors, not fewer.
Edmondson, by her own later account, sat "dumbfounded, staring at the computer screen." The finding looked like a failure. It turned out to be a discovery. The high-performing teams were not making more mistakes — they were more willing to talk about them. The low-performing teams were hiding theirs. What separated the two was a quality Edmondson eventually named psychological safety, defining it as "a shared belief held by members of a team that the team is safe for interpersonal risk-taking."
That paper, published in Administrative Science Quarterly in 1999, has been cited more than 20,000 times. Almost two decades later, Google's internal People Operations team ran a multi-year analysis of 180 of its own teams — a project they called Aristotle — looking for the single variable that most predicted team effectiveness. They tested seniority, education, personality mix, team size, every plausible factor. The strongest predictor by a wide margin was psychological safety. The New York Times wrote about Project Aristotle in 2016, and the concept moved from organizational psychology into the everyday vocabulary of leaders.
What took longer was the recognition that the same construct applies, directly and operationally, to occupational safety and health. That recognition is finally arriving, and the data behind it should reorder how every safety practitioner thinks about the work.
Consider the numbers. The National Safety Council's SAFER Trend Survey found that workers who felt psychologically unsafe on the job were 80 percent more likely to report an injury requiring medical attention or missed days of work. Workers who perceived their employer as discouraging reporting were 2.4 times more likely to have experienced a work injury. A separate body of research, summarized in a 2021 review across five high-risk industries, found that roughly 71 percent of workplace accidents go unreported when organizational safety climate is perceived as poor — and that the strongest single moderator of underreporting was psychological safety. The OSHA national reporting system has been estimated to be missing somewhere between 50 and 68 percent of actual workplace accidents, depending on the study. That is not a measurement problem. It is a culture problem with a measurement consequence.
Let me make the implication explicit, because the implication is harder than the data.
If 71 percent of accidents in a low-safety-climate workplace are going unreported, then every recordable rate, every leading indicator dashboard, every quarterly review based on those numbers is reading from a partial deck. The plant that just hit a record low TRIR may have improved. It may also have simply driven its reporting underground. Senior leaders who do not understand this difference will reward the wrong behavior, invest in the wrong programs, and discover the problem only when it surfaces as a serious injury or fatality the data never warned them about. The metric did not fail. The culture was suppressing the inputs to the metric.
This is why I have come, over the last decade, to think of psychological safety not as a soft skill or a wellness initiative but as the precondition for every other safety system. The hierarchy of controls assumes that hazards are identified. Hazard identification assumes that workers will speak up. Workers speak up when they believe speaking up will help them rather than cost them. Strip that belief away, and every layer above it is operating on degraded data. Human and Organizational Performance, drawing on Sidney Dekker and Todd Conklin, makes the same point from a different angle. Conklin has written that workers are not the problem to be controlled; they are the solution. That formulation only works if workers are actually free to speak. In a low-safety-climate organization, they are not.
Edmondson is careful, and I want to be careful too, about what psychological safety is not. It is not niceness. It is not the absence of conflict. It is not lowered standards. In her words, it is not "an unrelentingly positive affect." Genuinely psychologically safe teams disagree often, and sometimes sharply. What separates them from unsafe teams is that the disagreement happens about the work, not at the expense of the people. A welder can tell the supervisor the rigging is wrong and not pay a social cost for it. A line worker can flag a near-miss without being labeled as "that guy." A new hire can ask the question everyone else has been embarrassed to ask, and the answer comes back as information rather than judgment.
If you are a safety leader, here is what I would offer as the practical implication of fifty years of converging research.
First, treat your near-miss reporting rate as a measure of psychological safety, not a measure of hazard prevalence. When reports go up, that is almost never bad news. It usually means workers have decided to trust the system. When reports go down without a corresponding change in operations, that is rarely good news. It usually means trust is leaking. Most senior leadership teams have this exactly backwards and reward the wrong direction of movement.
Second, audit your supervisory layer for the behaviors that destroy psychological safety. Public correction. Disparaging remarks about workers who raise concerns. Treating the first message as a complaint rather than data. The line workers in your organization can tell you, with remarkable precision, which supervisors are safe to speak around and which are not. The supervisors themselves often cannot. This is a knowable problem with knowable interventions, and it is not solved by another round of leadership training. It is solved by holding supervisors accountable for a different set of behaviors than the ones their performance review currently measures.
Third, change the question you ask after an incident. The single most damaging question in our profession is, "Why did the worker do that?" Edmondson's research, and three decades of HOP work building on it, makes the answer almost always the same: because the conditions made it the most reasonable thing to do at the time. The better question is, "What did the system do to put a person in a position where this outcome was possible?" That question can only be asked honestly in organizations where workers feel safe enough to answer it honestly. Psychological safety and learning are the same thing seen from two angles.
Edmondson titled her 2018 book The Fearless Organization. The phrase is more precise than it first sounds. She is not arguing that fear can be eliminated from work. She is arguing that fear, at the level that suppresses honest information, is incompatible with the kind of learning that produces safe operations. Every senior leader who has ever stood in front of a press conference after a serious incident has, in some sense, paid the cost of fear at the wrong moment.
The chain runs in a straight line. Psychological safety produces honest reporting. Honest reporting produces accurate data. Accurate data produces better decisions. Better decisions produce safer work. Cut the chain at the first link, and everything downstream degrades — quietly, predictably, and often invisibly until something terrible makes it visible.
Build the first link. Everything else depends on it.
Sources & Further Reading
• Edmondson, A. C. (1999). Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350–383.
• Edmondson, A. C. (2018). The Fearless Organization. Wiley.
• Duhigg, C. (2016). What Google Learned from Its Quest to Build the Perfect Team. The New York Times Magazine.
• National Safety Council. (2023). SAFER Trend Survey: Psychological Safety Correlates to Physical Safety.
• Probst, T. M. & Estrada, A. X. (2010). Accident under-reporting among employees: Testing the moderating influence of psychological safety climate and supervisor enforcement of safety practices. Accident Analysis & Prevention.
• Conklin, T. (2019). The 5 Principles of Human Performance. Pre-Accident Investigation Media.
• Dekker, S. (2015). Safety Differently: Human Factors for a New Era (2nd ed.). CRC Press.
