Psychological safety operates at the level of the team, not the individual or the organization. It is created by specific, identifiable leader behaviors and destroyed by specific, identifiable leader behaviors. The organizational-level safety score conceals more than it reveals.
Edmondson's (1999) introduction of psychological safety as a team-level climate variable defined it as the shared belief that the team is safe for interpersonal risk-taking, specifically for raising concerns, sharing incomplete ideas, asking for help, and reporting mistakes without fear of negative interpersonal consequences. The construct has since been established as one of the most consistent predictors of team learning and performance in the organizational literature, with Frazier, Fainshmidt, Klinger, Pezeshkan, and Vracheva (2017) finding in a meta-analysis that psychological safety predicted information sharing, learning behavior, creativity, and performance across diverse organizational and cultural contexts. This article reviews the multilevel evidence on psychological safety, examines the specific leader behaviors that create and destroy it at the team level, addresses the distinction between safety and comfort, and considers measurement and intervention approaches appropriate to the team level at which safety operates.
Edmondson (1999) established psychological safety as a team-level climate phenomenon rather than an individual characteristic or an organizational condition. The team-level architecture has important practical implications: the same individual may feel highly safe speaking up in one team context and highly unsafe in another, and both assessments may be accurate. The determinants of safety are primarily team-level, specifically the norms established by the team's leader through repeated behavioral modeling and response to team member communication attempts, rather than individual-level characteristics of team members or organization-wide culture conditions. This means that safety cannot be improved through organizational-level culture initiatives alone, but requires direct investment in the specific behaviors of specific team leaders across the organization.
The within-organization variance in psychological safety is typically much larger than between- organization variance, and is almost entirely attributable to differences in team-level leadership behavior. Nembhard and Edmondson (2006) found that leader inclusiveness, specifically the degree to which leaders actively and explicitly invited input and responded positively to contributions regardless of their status or initial quality, was the most powerful predictor of team psychological safety across diverse organizational contexts. This finding is practically significant because leader inclusiveness is a behaviorally specific, trainable leadership practice rather than a dispositional characteristic that leaders either have or do not have. It is produced through specific repeatable actions, and it can be developed through deliberate practice and structured feedback.
Clark (2020) extended Edmondson's framework into a four-stage typology of psychological safety: inclusion safety, the safety to belong and participate; learner safety, the safety to ask questions and make mistakes without fear; contributor safety, the safety to make meaningful contributions without excessive vetting; and challenger safety, the safety to challenge the status quo and offer alternative views without social cost. The developmental sequence implied by this typology has practical implications for intervention targeting: organizations where challenger safety is absent while inclusion and learner safety are established face different development challenges than those where even inclusion safety has not been established. The correct intervention depends on the specific type of safety that is most limiting the team's performance, not on a generic safety improvement initiative calibrated to an average safety deficit.
The leader behaviors most consistently and rapidly destroying psychological safety are those that demonstrate to team members that speaking up carries social cost. These behaviors fall into two categories: punishing behaviors, which produce direct negative consequences for team members who raise concerns or disagreements, and neglecting behaviors, which produce no visible response to team member communication attempts, effectively teaching through silence that speaking up has no organizational effect. Both categories teach the same lesson through different mechanisms: speaking up does not produce positive organizational consequences and may produce negative ones.
Punishing behaviors include visible frustration or dismissal in response to questions or challenges, public criticism of team members who raise concerns without the complete information to support them, the informal labeling of challenging voices as non-team-players or difficult, and the exclusion of dissenting voices from consequential conversations or decision processes. Each of these behaviors, observed once by any team member, is processed as information about what speaking up costs in this team's social environment, and the lesson generalizes across the team whether or not the observer was the direct target of the punishing response. A single visible punishing response can suppress speaking-up behavior across an entire team for months, as other members update their assessment of the safety of raising concerns in this context.
Neglecting behaviors are subtler and slower in their safety-destroying effects but equally consequential over time. When team members raise concerns, ideas, or questions that the leader does not acknowledge, validate, or respond to substantively, they learn that their communication has no organizational effect. Over time, this pattern produces communication withdrawal: team members stop investing in communication that has been taught through repeated non-response to produce no visible organizational consequence. The cumulative effect on team information quality can be severe, because the information most valuable to the team, the observations and concerns that require interpersonal courage to raise, is precisely the information most likely to be withdrawn when neglect signals that it will not be meaningfully received.
The distinction between safety and comfort is practically important and frequently confused in organizational application of the psychological safety concept. Psychologically safe teams are not necessarily comfortable teams: safety enables and is in fact associated with more candid and sometimes more uncomfortable conversations than low-safety teams, where the avoidance of discomfort is achieved through the suppression of the challenging communications that would produce it. Edmondson and Lei (2014) clarified that psychological safety should not be equated with absence of conflict, excessive agreeableness, or low performance standards. Safe teams are teams where the interpersonal risk of speaking up has been sufficiently reduced that team members are willing to raise the concerns and challenges that improve team performance, which often makes them more, not less, productively contentious than their low-safety counterparts.
Organizational-level psychological safety scores conceal the team-level variance that makes safety measurement organizationally useful. An organization with an average safety score of 3.8 on a 5-point scale may contain teams ranging from 2.1 to 4.9, with the lowest-scoring teams representing the highest organizational risk and the highest intervention priority. Organizational aggregate scores cannot identify these teams or target intervention toward them. Team-level measurement, reported and acted on at the team level, is the diagnostically appropriate unit for both identifying safety deficits and evaluating the effectiveness of safety interventions. Organizations that report safety data only at the organizational or business unit level are measuring safety in a way that prevents them from using the data to improve it.
Frazier et al. (2017) identified the leader behaviors most predictive of team psychological safety across their meta-analytic sample, confirming leader inclusiveness and leader modeling of fallibility as the most consistently significant predictors. These findings suggest specific assessment content for safety-related leader behavior assessment: items assessing how often the leader explicitly invites input, how the leader responds when team members raise concerns or make mistakes, and whether the leader models intellectual humility and genuine updating in response to team member contributions. Assessment of these specific leader behaviors, rather than only assessment of team member safety perceptions, provides the actionable diagnostic information that guides leader behavior development in specific, safety-relevant directions.
The assessment challenge in psychological safety measurement is the response bias produced by the very climate variable being measured: teams with low psychological safety are precisely those where team members are least likely to report accurately about safety conditions, because accurate reporting itself requires the safety that is absent. This methodological problem is not fully solvable within standard survey approaches, but it can be partially mitigated through anonymous administration, item framing that probes specific behavioral observations rather than direct safety assessments, and supplementary methods including structured interviews with former team members and observation of team discussion dynamics. Organizations whose safety measurement approach cannot detect within-team variation are generating data insufficient to guide the specific, team-level interventions that safety improvement requires.
| Dimension | Low psychological safety | High psychological safety |
|---|---|---|
| Conflict level | Low visible conflict (suppressed) | Productive conflict present; minority views surface |
| Error reporting | Errors managed individually to protect appearance | Errors reported rapidly for collective resolution |
| Information sharing | Partial; strategic; filtered for self-protection | Proactive; relevant information shared without prompting |
| Voice behavior | Suppressed; concerns not raised to leadership | Active; concerns raised with expectation of engagement |
| Team feel | Comfortable surface; underlying tension | Sometimes uncomfortable; genuinely productive |
The leader development approaches with the strongest evidence base for improving team psychological safety are those targeting specific leader behaviors rather than general safety awareness or cultural values. Leader inclusiveness training that develops specific behavioral skills, including asking questions before evaluating, explicitly attributing credit to team members whose ideas are used, visibly engaging with challenging input rather than dismissing it, and acknowledging one's own uncertainty and mistakes in front of the team, produces measurable improvements in team safety scores across multiple organizational contexts and leadership levels. Training that targets general awareness of psychological safety without developing specific leader behaviors produces smaller and less consistent effects.
The behaviors that most directly build psychological safety are not the most natural or comfortable behaviors for most organizational leaders. Acknowledging uncertainty and mistakes in front of the team, actively inviting challenge to one's own positions, and publicly updating one's views in response to team member input all require leaders to accept temporary status costs in exchange for the safety benefits these behaviors produce in their teams. Leaders who are most comfortable with their current status and most sensitive to status threat are those most likely to avoid these behaviors, and most likely to produce the low-safety environments that the research consistently associates with lower team performance in learning-intensive contexts.
Organizational structures supporting safety development include regular structured reflection on specific interactions rather than general performance, peer observation in which leaders observe each other's facilitation of team discussions and provide behavioral feedback, and measurement systems that assess team-level safety as a leadership performance outcome alongside operational performance metrics. Organizations investing in these structures produce leaders with stronger psychological safety-building behavioral repertoires than those relying on general leadership development programs that treat psychological safety as one competency among many rather than as the foundational climate condition that determines the effectiveness of every other form of team-level leadership investment.