Psychological safety: meaning and application

Your team has a new tool and everyone has been trained, yet nobody asks the question everyone has and mistakes are fixed quietly. This article explains psychological safety: people's estimate of what it costs to reveal ignorance or a mistake. You learn where the concept comes from, what it is not and how to study it carefully.

Psychological safety is the shared belief in a team that you can take interpersonal risks there, such as asking a question, reporting a mistake or raising a dissenting idea, without being rejected, embarrassed or punished for it.

The concept describes a climate at team level, not a personal state of mind. It differs from trust between two people and does not mean that a team sets low standards. Research shows robust associations with learning behaviour and information sharing, but most studies are correlational and based on self-report.

Precursors: change in groups and personal engagement

The concept emerged in the 1960s in research on organisational change. Schein and Bennis described psychological safety as a condition for being able to learn and change (Schein & Bennis, 1965); Frazier and colleagues place that contribution in the 'unfreezing' that precedes change (Frazier et al., 2017). In 1990, William Kahn described psychological safety as feeling able to show and employ one's self without fear of negative consequences to self-image, status or career (Kahn, 1990, cited via Frazier et al., 2017). For Kahn, it is a condition experienced by the individual.

Edmondson: a shared belief in teams

Amy Edmondson shifted the focus to the team. In a field study of 51 teams at an office furniture manufacturer, she defined team psychological safety as a shared belief held by members of a team that the team is safe for interpersonal risk taking (Edmondson, 1999). According to her, that belief is mostly tacit. Teams with more psychological safety showed more learning behaviour, such as seeking feedback and discussing mistakes, and that learning behaviour was associated with performance as rated by managers outside the team. Edmondson herself noted that her cross-sectional design does not provide causal evidence.

Later, Edmondson and colleagues followed sixteen hospitals introducing a new cardiac surgery technique. Successful implementers organised practice sessions that fostered psychological safety and reflected together on procedures (Edmondson, Bohmer & Pisano, 2001). It is relevant to any adoption of new ways of working.

Reviews and meta-analysis

Edmondson and Lei describe psychological safety as people's perceptions of the consequences of taking interpersonal risks in a particular context, studied at individual, team and organisational level (Edmondson & Lei, 2014). Newman, Donohue and Eva systematically mapped the empirical literature and called for better theory on how psychological safety develops (Newman, Donohue & Eva, 2017).

The meta-analysis by Frazier and colleagues combined 136 independent samples with more than 22,000 individuals and nearly 5,000 groups (Frazier et al., 2017). Psychological safety was associated with, among other things, information sharing, learning behaviour, engagement and task performance, and with support from leaders and colleagues, role clarity and interdependence. Associations were stronger when the same person reported both sides, and performance associations were larger for subjective ratings than for objective measures.

On its re:Work website, Google described internal research into 180 teams (Project Aristotle) in which psychological safety was named as the most important of five team characteristics (Google re:Work). That is the company's own description, not peer-reviewed and without public data, so it is not independent evidence.

Not trust between two people. According to Edmondson (1999), psychological safety involves trust but describes a team climate. Frazier and colleagues summarise her later distinction: trust concerns your willingness to give another person the benefit of the doubt, psychological safety your expectation that others will give you the benefit of the doubt when you take a risk.

Not comfort or low standards. Edmondson (1999) stresses that it is neither a careless sense of permissiveness nor unrelentingly positive feelings, but confidence that the team will not embarrass, reject or punish someone for speaking up. In a later book she sets psychological safety against the level of performance standards (Edmondson, 2019): high safety with low standards she calls a comfort zone, high safety with high standards the environment in which learning thrives best. That is a conceptual framework, not a measured typology.

Not group cohesion. A close-knit team is not automatically psychologically safe; cohesion can even reduce the willingness to disagree (Edmondson, 1999).

Not individual wellbeing or clinical safety. The concept says something about the expected social consequences of behaviour in a group, not about an individual's stress, mental health or physical safety. Those topics require their own measurement; see also Stress and brain measurements.

Edmondson developed a short questionnaire with positively and negatively worded statements, such as 'If you make a mistake on this team, it is often held against you' (reverse-scored) and 'It is safe to take a risk on this team' (Edmondson, 1999). According to Frazier and colleagues, that scale or a version of it is by far the most widely used.

Team level requires justified aggregation. Because the concept describes a shared belief, you only combine answers into a team score when team members agree sufficiently and teams differ. Edmondson tested this with measures of within-team agreement and an intraclass correlation. A high average with a wide spread means something different from the same average on which the team agrees.

Self-report has limits. People who feel unsafe do not always say so in a questionnaire. Guarantee confidentiality, avoid reporting on very small groups and combine the questionnaire with behavioural data, such as incident reports, and interviews.

Individual and team are different levels. Frazier and colleagues found comparable associations at both levels, but there are far fewer group-level studies. A statement about a team requires team data.

There is no brain measure of psychological safety. Physiological or EEG measurements can say something about tension or attention in a situation, not about a team's shared belief.

In experience research, the concept raises an often overlooked question: do people dare to say what they experience? In employee research or evaluations of a new workplace, the climate in a group partly determines which answers you get. Workplace wellbeing and neurodesign covers the physical work environment; psychological safety adds the social layer.

In change and adoption, the link is direct. Anyone trying something new makes mistakes at first and asks questions that may sound awkward. If that feels risky, quietly sticking to the old way is the safest choice. Likewise, a promise of an open culture in employer branding needs to be backed by how teams actually experience their work.

1. Define the decision you want the results to support.
2. Set the level: team, department or individual, before collecting data.
3. Choose a validated scale and document translations and adaptations.
4. Guarantee confidentiality and a minimum group size for reporting.
5. Test agreement within teams before reporting team averages.
6. Combine sources: questionnaire, behavioural data and interviews.
7. Report associations, unless your design allows more, such as measurements over time with a comparison group.

This example is fictional.

Situation. For six months, a software team at a mid-sized service provider has been reviewing incidents without assigning blame: the discussion focuses on what happened and why the system allowed it, not on who did it. Yet near misses are rarely reported.

Decision question. Should the review format change, or does the bottleneck lie in the expected consequences of reporting?

Available information. Monthly numbers of reported incidents and near misses, minutes of reviews and informal signals that developers 'don't want the hassle'.

Suitable approach. Give all team members a short psychological safety scale, confidentially and reported only at team level, and check agreement. Hold a few interviews about what people expect to happen when they report their own mistake. Compare this with reporting behaviour over the same period.

Possible interpretation. If experiences are strongly divided, the bottleneck may lie in particular roles or meeting formats. If the score is low and shared, the format has apparently not yet changed the expected consequences of reporting. Both are hypotheses; no results are known in this example.

Next step. Change one element, for example who leads the review, and measure again after a few months, with reporting behaviour as a second outcome measure.

• Assuming causality. Most studies measure safety and outcomes at the same time. Better performance can also increase perceived safety.
• Overrating self-report. Associations are higher when one person reports both sides.
• Confusing safety with ease. The concept is about room to speak up, not about the absence of discomfort or standards.
• Ignoring context. A team can be safe in daily meetings and unsafe in an escalation to management. Measure in the context you want to say something about.
• Ignoring culture. Frazier and colleagues found that associations varied with uncertainty avoidance as a cultural characteristic. A translated scale does not automatically measure the same thing.
• Attributing a team score to individuals. An average says nothing about one specific employee.

Psychological safety describes whether people in a group expect that questions, mistakes and dissenting ideas can be raised without social punishment. The concept is well embedded in research on team learning and change and helps explain why people do not try something new. Use it as a team construct, measure it with a validated scale and a check on agreement, and present associations as associations as long as your design does not allow a causal conclusion.