Psychological safety and innovation: what the research supports

In innovation programs, the question we hear most often from executives is why ideas do not show up. The company set up a suggestion channel, ran a hackathon, set aside budget. The proposals that arrive are few and too safe. Almost always the answer lies less in the process and more in what happens in team meetings when someone admits they do not know, disagrees with the boss, or reports that a test failed.

Amy Edmondson named this phenomenon in a 1999 article: psychological safety, defined as a shared belief held by members of a team that the team is safe for interpersonal risk taking.

What the original study showed

Edmondson's research followed 51 work teams in a manufacturing company. The central finding was that psychological safety was associated with team learning behavior: asking for help, seeking feedback, discussing errors, experimenting. Learning behavior, in turn, was associated with performance.

There is an important nuance. In earlier work with hospital teams, Edmondson had observed that the better teams reported more errors, not fewer. The explanation was not that they made more mistakes, but that they felt comfortable reporting them. That inversion is the heart of the concept: environments that seem to have few problems may simply have few people willing to talk about them.

The Google case, with due caution

The topic gained visibility outside academia through Project Aristotle, Google's internal initiative on team effectiveness. According to Charles Duhigg's reporting in The New York Times Magazine (2016) and material the company itself published, psychological safety emerged as the most important of the dynamics identified in the teams studied.

Caution is warranted: this was an internal study, disclosed in summary form, not peer-reviewed published research. It is useful as an illustration and as a way into the conversation with executives. The scientific foundation lies in Edmondson's work and in the review by Edmondson and Lei (2014), which brings together two decades of studies on the construct.

Psychological safety is not comfort

The most frequent misunderstanding is confusing psychological safety with an environment without accountability. Edmondson insists on the opposite. She crosses psychological safety with performance standards and describes four situations: low safety and low standards produce apathy; high safety and low standards produce comfort; low safety and high standards produce anxiety; high safety and high standards create the learning zone.

The implication for innovation is clear. It is not about lowering demands, but about making demands compatible with exposing doubts, hypotheses, and failures. An R&D team with aggressive targets where reporting a failed experiment is normal is in the right zone. A team with the same targets where failure becomes a performance review topic will hide bad results until the last possible moment.

How to measure it

Edmondson's original scale has seven items, answered on an agreement scale. Some are negatively worded, such as the perception that a mistake made on the team is held against you, or that it is difficult to ask other members for help. It is short, has been widely used in later research, and works well as a component of broader innovation culture assessments.

Three practical precautions:

• Measure at the team level. It is a group construct. The company average hides what matters, which is the variation between teams under different leaders.
• Combine it with behavioral indicators. Number of recorded experiments, ideas that reach the test stage, frequency of post-project reviews. If the scale says it is safe to fail and nobody has recorded a failed experiment in a year, something does not add up.
• Share results with the team, not only the manager. A carefully facilitated conversation about the results is already an intervention.

What leaders can do on Monday

Practice	Concrete example
Frame the work as learning	Open the project by stating explicitly what is still unknown and which hypotheses will be tested
Acknowledge your own fallibility	The manager describes a personal mistake and what they learned, before asking the team to do the same
Ask genuine questions	In meetings, ask "what are we missing?" and wait for the answer
Respond well to bad news	Thank whoever raised the problem before discussing the solution
Separate intelligent failure from negligence	A written criterion to distinguish a failed experiment from an avoidable lapse

None of these practices requires budget. All of them require consistency, because psychological safety builds slowly and can be lost in a single meeting. In our experience, the most honest sign that the work is paying off is an increase in problems reported early, not a decrease.