Psychological safety — Organizational Development

Amy Edmondson's definition: a shared belief held by members of a team that the team is safe for interpersonal risk-taking. In a psychologically safe team, people believe they can speak up with questions, concerns, mistakes, and half-formed ideas without being humiliated, punished, or written off.

This is the best-evidenced construct in this collection, with a large empirical literature linking it to learning behavior, error reporting, innovation, and performance — particularly in knowledge work with high interdependence.

Evidence: Strong. The best-evidenced construct in this collection. A large literature spanning three decades links it to learning behavior, error reporting, information sharing, innovation, and performance, with meta-analytic support and effects that hold across sectors including health care and software. Three honest caveats: most measurement is self-report; the relationship with performance is strongest where work is interdependent and uncertain, which is precisely this guide's audience; and the measures have multiplied faster than the construct has been pinned down, with recent systematic review work finding no consensus on what belongs inside psychological safety and what merely correlates with it. Use Edmondson's original seven items rather than a vendor's variant, and do not compare your score against one produced by a different instrument.

Use when: the work requires people to speak up, admit not knowing, or challenge someone more senior. That is nearly all knowledge work.

Do not use when: as a synonym for comfort. Confusing the two produces conflict-avoidant teams and gives the concept a bad name it does not deserve.

What it is not

Not niceness. Safe teams often have more open disagreement, not less.
Not comfort. People are still accountable and still stretched.
Not lowered standards. Edmondson's 2x2 pairs safety with accountability:
- Low safety, low accountability → apathy
- Low safety, high accountability → anxiety
- High safety, low accountability → comfort
- High safety, high accountability → learning and high performance The goal is the fourth quadrant. Safety without standards is the failure mode critics attack; standards without safety is the failure mode most organizations actually have.
Not an individual trait. It is a property of a group, and it varies sharply between teams in the same organization — which is good news, because a local leader can change it without waiting for the enterprise.

The counterintuitive finding

Edmondson's original hospital study expected better teams to report fewer errors. They reported more — because they were willing to. Reporting rate measures willingness to disclose, not underlying error rate. Any organization improving safety culture must brief its board on this before the numbers move, or the improvement will be read as deterioration.

Measuring it

Edmondson's seven-item scale, rated 1–7 (items 1, 3, 5 are reverse-scored):

If you make a mistake on this team, it is often held against you.
Members of this team are able to bring up problems and tough issues.
People on this team sometimes reject others for being different.
It is safe to take a risk on this team.
It is difficult to ask other members of this team for help.
No one on this team would deliberately act in a way that undermines my efforts.
Working with members of this team, my unique skills and talents are valued and utilized.

Report at team level only, with a minimum response threshold. Individual-level reporting destroys the thing being measured. See od-ethics.

Behavioral indicators, which are harder to fake than a survey:

Who speaks in meetings, and in what proportion. Measure it once; the result is usually startling.
Whether anyone asks a basic question in front of senior people.
Time between a problem being known by someone and being known by the team.
Whether "I don't know" and "I was wrong" are ever said by senior people.

Building it

Edmondson's leader toolkit:

Frame the work. Name the work as learning or as complex, not as execution of a known routine. If people believe the answer is known, not knowing becomes shameful.
Acknowledge your own fallibility. "I may miss things — I need to hear from you." Specific, not ritual.
Model curiosity. Ask real questions and wait. Silence after a question is the leader's tool; filling it is the leader's most common mistake.
Respond productively. The response to the first person who raises something uncomfortable sets the norm for everyone watching. Thank them specifically and visibly, and act.

Structural supports: round-robins, written-first meeting formats, pre-mortems, blameless reviews, anonymous channels as a starting ramp rather than a destination.

Examples by audience

Health care: a nurse or junior doctor questioning a senior clinician's drug dose. The evidence that this changes outcomes is strong, and the hierarchy gradient that suppresses it is the target. Structural aids — read-back, graded assertiveness, checklists with a designated challenge point — work even where the culture has not yet changed. See crew-resource-management.
Software: an engineer saying "I don't understand this design" in review, or reporting that they caused an outage within minutes rather than hours. The measurable proxy is time-to-disclosure during incidents.
Executive: the hardest room. The executive who says "here's what I got wrong last quarter" buys more safety than any program. The executive who interrupts, or who visibly punishes one bad-news bearer, removes it for everyone in the room and everyone they tell afterwards.

Limitations

Cannot survive a leader who behaves badly under pressure, regardless of program investment.
Team-level construct; enterprise-wide averages hide the variance that matters.
Popular usage has drifted toward "comfort", which invites backlash. Use Edmondson's definition explicitly, every time.

Questionnaire

A paired, copy-and-use instrument for this model: psychological-safety-questionnaire. It carries Edmondson's seven-item scale, the safety-and-accountability grid, and a set of behavioral indicators that are harder to fake than a survey.