Is Psychological Safety a Shared Belief? What 1.3 Million Data Points Say

Psychological safety lives in each person’s read of the room, not in the room itself. That sentence is the whole argument, and the rest of this piece is the data behind it and what it changes about how you run a team.

Read it with your own team in mind. Most of it is abstract until there are real names attached, and immediately useful after.

Two definitions, and only one survives the data

Since the second edition of The 4 Stages of Psychological Safety launched, one question has come back again and again, at the launch webinar and in a steady stream on LinkedIn: “How does the 4 Stages model square with the prevailing definition of psychological safety, the shared belief that a team is safe for interpersonal risk-taking?”

We’re grateful for that definition. It carried psychological safety into the mainstream and built the audience this work now reaches. But the two schools are describing different things. The prevailing definition treats psychological safety as a team-level climate, something the group either has or doesn’t. The 4 Stages school defines it as an individual’s perception of the risk attached to acts of vulnerability: different person to person, different moment to moment, developing through four stages.

Only one of those descriptions survives contact with the data.

Same room, same manager, opposite experiences

Build the most similar working experience you could engineer. Same team, same room, same manager, same work, same conversation. Now watch the two people sitting next to each other. One of them speaks up constantly, challenges the plan, asks for what she needs. The other has never asked a question.

Not one, ever.

Now put the question to the prevailing definition: does that team have psychological safety? There’s no good answer, because it’s a trick question. One of them might. The other might not. How shared is that?

If this were a rare arrangement it would be a curiosity. It isn’t rare. In LeaderFactor’s PSindex® dataset, four out of five teams contain a version of it: the same assessment question answered nine or 10 by one member and below six by a teammate.

The same team rates the same environment differently

The analysis was supposed to be routine. The question kept arriving, so the team pulled PSindex® data expecting to confirm the obvious, that some intra-team variance exists, and move on. What came back surprised the people who run the instrument.

The dataset: over 1.3 million data points, from 81,504 assessment responses across 513 organizations, cut down to 10,432 intact teams of three or more people. Every item runs on an 11-point scale, zero to 10.

What we measured	What we found	What it means at your table
Highest vs. lowest rater	Median gap of 2.5 points; the middle half of teams fall between 1.4 and 4.2	The typical team spans a quarter of the scale internally
The wide tail	One team in three spans more than a third of the scale; 17% span five-plus points	On one team in six, two people sit half the scale apart on the same question
Opposite readings	On four of five teams, one member answers 9–10 where a teammate answers below 6	Two people, one room, opposite experiences
Promoter and detractor together	26.8% of teams hold both a blue-zone member (composite 9 or higher) and a red-zone member (6 or lower)	One team in four has someone paying for the average
Intra-class correlation (ICC)	0.16, on a scale of zero to one	Far below what a shared belief would produce
Where the variance lives	16% between teams, 84% inside them	Teams look alike from the outside and differ sharply inside

The ICC is the number to sit with. It measures how much the people inside a team agree with one another, from zero to one. Anything below 0.5 counts as very low agreement, and a construct that genuinely lived at the team level would clear 0.6. This one comes in at 0.16. Even a generous reading, the kind that would concede “largely shared” at 0.55 or 0.6, has nothing to work with here.

If psychological safety were a shared belief, agreement would have been the finding. Disagreement is the finding, at scale, in the instrument’s own data.

Where both definitions are right

Teams are not hallucinating different meetings. On team-level agreement, a statistic called RWG, these same teams run a median of 0.87 to 0.95, comfortably above the 0.70 bar. In plain terms: the people in the room agree about what is happening in it. They can all tell you who got interrupted this morning and how the last dissent was received. What they do not share is what it would cost them, personally, to be the next one to speak.

So both schools are holding a real finding. Psychological safety can be measured at the team level, and LeaderFactor will keep measuring it there: means and benchmarks tell you where a team stands relative to others, and aggregation is what keeps individual answers anonymous. But it is not experienced at the team level. The experience happens one person at a time.

If you’ve watched this argument run in the comments, that’s the resolution to carry back into it. Teams agree about the room. They disagree about the bill. Both findings sit in the same dataset.

A shared understanding was never on the table

The numbers show the understanding isn’t shared. The harder claim is that it never could have been. Ask what a team would have to do to genuinely share an understanding of interpersonal risk.

Walk it through with a real meeting in mind. Eight people would each need to read the same moment the same way: what it meant when the VP checked her phone during the new analyst’s question, whether the silence after the proposal was consideration or verdict, what happened to the last person who pushed back and whether it would happen again. Each person runs that read through a different filter: a different seat in the hierarchy, a different tenure, a different history with the person talking. For the understanding to be shared, eight private calculations would have to land on the same answer at the same moment and hold it there as the meeting moves. Full agreement never arrives, on any team, ever.

The full guide takes the finding the rest of the way: what you lose when you read safety as shared (five things, starting with the person inside the average), a spread-versus-mean grid for placing your own team without an instrument, why disagreement peaks at Challenger Safety, a fill-in table for pricing acts of vulnerability per direct report, six broadcast questions with the named version of each, and a five-day practice that ends with one person and one conversation.