In 2016, Charles Duhigg wrote a fascinating article for The New York Times Sunday Magazine chronicling Google's three-year effort to discover why some teams were highly productive and others flopped. The article: "What Google Learned From its Quest to Build the Perfect Team" revealed that what separated high-performing teams from lackluster ones wasn't a matter of whether a team was stacked with standout leaders or even members who had previously worked on other high-performing teams. What mattered was the climate for risk-taking established within the team, otherwise referred to as psychological safety.

Because Duhigg's writing and Google's research, named Project Aristotle, were so popular, the link between psychological safety and team performance lodged in people's minds and over a decade later, many still assume that the correct way to assess psychological safety is at the team level.

However, psychological safety was first defined at the individual level as a necessary condition for organizational change. In 1965, researchers Edgar Schein and Warren Bennis at MIT explained that for workers to try new roles, potentially fail, and then try again, they needed to feel psychologically safe (Schein & Bennis, 1965 as cited in Frazier et al., 2017).

It wasn't until 1999 that Amy Edmondson defined team psychological safety as a shared belief that the team was safe for interpersonal risk taking. This is what Duhigg's article was about: team psychological safety and for many people, the term became synonymous with team climate. But what's less well known and potentially more valuable in certain situations is that individual-level safety and relational or dyadic safety have their own research traditions.

MEASURING WORKPLACE RELATIONSHIPS

Because the Anchor Relationships assessment explores how people work together in ways that respect each other’s competence and enable each person to raise concerns, dissenting views, or bad news, it contains questions that signal the presence or absence of safety within a workplace relationship. While the assessment does not measure safety, it identifies relational qualities from which safety can be inferred. For this reason, understanding how psychological safety operates at the relational level is important for interpreting the assessment.

Two recent studies shed light on why it's so important to measure safety at the interpersonal level: The first is by Lehmann et al. (2023), and it explores whether humility and psychological safety live in specific working relationships rather than in individuals or in teams.

The second is by Verresen et al. (2026) and it examines psychological safety in relationships and on teams. In both studies, researchers used round-robin designs, in which every team member rates every other member individually, providing a comparison of the team-level experience and the relationship-level experience among the same people at the same time.

Here's what the researchers found:

1. Measuring safety at the team level captures more disagreement than shared belief:

   In Lehmann et al., a study of 133 people spread across 32 different teams, each person answered a short set of questions about every member on the team. There were no questions about the team, every answer was about a named colleague. Ratings from the same team were nearly as varied as ratings from different teams. The team explained roughly 11% of the rater's response.

   In Verresen et al., a study of about 500 employees across 70 teams, each person rated their team (using five items from Edmondson's seven-item scale) in addition to the members of their team (using two items based on Edmondson's with the referent swapped). In this study, the team-level survey produced the most consequential finding. Only 16% of the variation in answers reflected differences between teams. The other 84% was people on the same team disagreeing with each other about how safe their own team is.
   

2. The team you are on has little to do with how safe you feel with a colleague.

   Teams differ in how safe people feel, but the relationships differ much more. Both studies found that ratings of psychological safety in relationships were drawn from a mix of factors. Part reflects the rater, since some people feel safe with almost everyone; part is unique to the pair; and part is attributed to the team (Lehmann et al., 2023; Verresen et al., 2026).

   In Lehmann et al.'s (2023) Main Study, each relationship rating broke down into the following: 32% for the rater and 45% for the pairing, with the remainder reflecting the person rated and error (p. 817). Team membership, tested separately, explained 11% and was not significant (p. 818).

   In Verresen et al. (2026), each relationship rating was broken down as: 38% for the rater; 39% (member-leader groupings) to 49% (peers) for the pair; while 8% was attributed to the team (p. 280).

   Then the Verresen study revealed that disagreement among team members reflected individuals' felt safety in specific relationships. Each person's view of the team tracked how safe that person felt in the single relationship they depended on most. The relationship people depended on least had no measurable bearing on team-level safety (Verresen et al., 2026).

RELATIONSHIPS AS AN INTERVENTION POINT

Both studies found that the largest variation in psychological safety occurred between relationships, showing that efforts to improve safety at the relationship-level might be more effective than changes implemented at the team-level. The same person can feel safe with one colleague and not with another. What these two studies show is that safety isn't primarily a property of teams or individuals. Where it resides most reliably, according to these studies, is in relationships.

By turning the focus from teams to relationships, new ways of understanding valuable organizational practices can be seen. This is the hope and the opportunity inherent in a movement toward greater relational literacy for business leaders, consultants, and individual employees alike. The measurements we use and the very questions we ask about unlocking potential can change what we see and how we see it.

It’s interesting to look back and question why Google’s famed Project Aristotle didn’t look at relationships as a possible answer to their question about high-performing teams– especially considering that Aristotle’s most famous writings extolled flourishing within virtuous relationships as essential for happiness.

Frazier, M. L., Fainshmidt, S., Klinger, R. L., Pezeshkan, A., & Vracheva, V. (2017). Psychological safety: A meta-analytic review and extension. Personnel Psychology, 70(1), 113–165. https://doi.org/10.1111/peps.12183

Lehmann, M., Pery, S., Kluger, A. N., Hekman, D. R., Owens, B. P., & Malloy, T. E. (2023). Relationship-specific (dyadic) humility: How your humility predicts my psychological safety and performance. Journal of Applied Psychology, 108(5), 809–825. https://doi.org/10.1037/apl0001059

Verresen, R., Vangrieken, K., & Loeys, T. (2026). Team psychological safety (de)composition: Connecting lower-level dyadic perceptions to higher-level beliefs. Small Group Research, 57(2), 263–303. https://doi.org/10.1177/10464964251347901