After a project fails, a team holds a long retrospective. Everyone talks, a list of lessons appears on the wall, and the next project repeats the same mistake. Another team has only a brief conversation but changes its checklist and catches the error next time. Which team learned? The number of words spoken is observable; learning behavior is a more demanding process of seeking, interpreting, testing, and using information. A useful measure must distinguish those steps from the ceremony that resembles them.
Measuring team learning behavior asks what a group does to improve its shared understanding and action while working together. Feedback seeking, error discussion, experimentation, reflection, information sharing, and adjustment can all matter, but they are not automatically equivalent. This guide compares team-member surveys, observation, task traces, and outcome measures. It treats psychological safety as a possible condition for learning behavior, not the behavior itself, and it distinguishes a reported lesson from evidence that a team’s practice changed.
Locate the behavior in a learning cycle
Edmondson’s 1999 study described team learning behavior as activities including asking questions, seeking feedback, experimenting, reflecting on results, and discussing mistakes. These activities expose the team to information it may otherwise miss. The original field study linked team psychological safety to learning behavior across 51 manufacturing teams. It did not show that every question is useful or that a team with a high activity count necessarily improves performance. The measure should identify which part of the learning cycle is observed.
Input, interpretation, action, and consequence can be separated. A team might receive customer complaints but never discuss them. It might discuss the complaints but reject their implications. It might agree on a change but fail to implement it. Or it might change a process and discover that the new version works worse. If a study measures only the first stage, call it feedback exposure or seeking rather than completed learning. If it measures the latter stages, preserve enough temporal detail to connect the feedback to the response.
Team learning is also more than the sum of individual training. One member may learn a new technique while the team’s routines remain unchanged. Conversely, a team may build a shared checklist that helps all members without each person explaining the theory behind it. Define the unit of analysis. A team-level claim needs evidence that information was exchanged, incorporated into coordination, or embedded in a shared practice rather than merely acquired privately.
Ellis and colleagues examined team learning in a task setting and showed how collectively connecting information relates to team performance. Their experimental work illustrates the importance of distributed information and collective integration. It does not make one laboratory task a universal model of learning in every workplace. In a practical assessment, ask what information was uniquely held by different members, whether it entered discussion, and whether the team used it in a decision.
Survey behavior without turning attitude into action
A team-member survey can ask how often the group seeks feedback, discusses errors, experiments with new approaches, or reflects on its work. Edmondson’s learning-behavior items were designed to target these activities separately from psychological safety. Keep that distinction in the questionnaire. “We can admit mistakes without embarrassment” is a safety belief; “We discuss mistakes to improve our work” is a reported behavior. If both appear in one total score, the study cannot easily test whether safety predicts learning.
Savelsbergh and colleagues developed and validated a multidimensional instrument for team learning behaviors. Its existence matters because a single general item can conceal patterns: a group may share information readily but avoid challenging assumptions, or may reflect often but not experiment. Choose subscales that fit the question and report their scores rather than using a broad total by default. If a different setting requires rewritten items, test whether those items still have the intended factor structure and meaning.
Frequency items need a time frame and an opportunity denominator. “We often ask for feedback” may mean once a month in a team with one major decision, or daily in a team with many customer interactions. A count per meeting, per completed task, or per feedback opportunity may be more interpretable. If the team has no external customer during the sampled period, a low frequency of customer feedback requests is not necessarily reluctance to learn. Describe what opportunities existed.
Self-report can be influenced by social desirability. Teams may recognize that talking about mistakes sounds virtuous and overreport it. A survey also asks members to summarize weeks of interactions from memory, which favors salient meetings and recent events. Protect confidentiality, use clear behavioral wording, and where possible compare responses with meeting observations or task records. Convergence strengthens a claim, but disagreement can reveal which behaviors are invisible to different members.
Aggregation requires care. A team average can be informative when several members describe a shared pattern, but a leader’s answer is not the whole team’s behavior. Report response rates and variation within teams. If junior members say questions are discouraged while senior members report frequent discussion, the contrast is analytically important. A high overall score may reflect whose behaviors are most visible or whose voice dominates a meeting.
Observe interaction, not just volume
Observation can capture an error being raised, a request for evidence, a test of an assumption, a proposal for an experiment, or a revision to a plan. It can also show how others respond. A meeting with many exchanges is not necessarily a learning meeting; participants may repeat positions without integrating anything. Code the function of each exchange. Did it introduce information, invite clarification, challenge an interpretation, compare alternatives, or decide a next test?
Edmondson’s study of operating-room teams adopting a new cardiac-surgery technology used qualitative and quantitative evidence to examine speaking up and coordination. The setting makes the measurement problem vivid: an observation of voice is meaningful when linked to what the team is trying to learn and how a leader responds. It cannot by itself establish that a new routine was retained. Observational protocols should follow the work through implementation when the research question concerns sustained learning.
Video or audio analysis can preserve sequences of interaction but introduces consent, privacy, and selection issues. Teams may behave differently when recorded, and sensitive errors may occur outside formal meetings. Sample multiple moments in the workflow, state what was not recorded, and blind coders to intervention status where feasible. Inter-rater agreement should be reported for categories such as question, feedback request, reflection, and action decision. A coding scheme that cannot be applied consistently will not support a fine-grained learning claim.
Sequence matters more than a raw count. A concern may be voiced, ignored, revived by another member, investigated, and finally embedded in a procedure. A count of one “error discussion” misses the difference between dismissal and productive inquiry. Event-based coding can track the link from trigger to interpretation to action. If analysis stops at meeting end, the strongest defensible claim may be that the team engaged in learning-oriented interaction, not that it learned the intended lesson.
Silence has several interpretations. A team that identifies no problems may be effective, unaware of problems, or reluctant to expose them. A high rate of reported errors may indicate a safe reporting culture rather than poor work. Avoid scoring the raw number of problems as a simple inverse measure of learning. Combine it with opportunities, severity, response, and subsequent change. The same count can mean different things in different risk environments.
Trace experiments and changes in work
Behavioral traces can show whether a team requested outside feedback, changed a protocol, ran a trial, reviewed results, and retained or reversed a change. These records are valuable because they reduce dependence on retrospective opinion. Yet they can miss informal learning, and a logged change can be cosmetic. Define what counts as an experiment: a proposed variation, a prediction, a trial, observation of consequences, and an explicit decision about what follows. Not every change is an experiment merely because it is new.
A structured debrief can make reflection observable. Eddy and colleagues compared guided and unguided team-led debriefing methods and found differences in subsequent team processes under their design. The study supports the idea that the form of reflection can matter. It does not justify treating a debrief’s mere occurrence as a learning outcome. Measure whether members identify specific evidence, consider alternatives, allocate follow-up actions, and revisit results later.
Performance outcomes can help test whether learning behavior mattered, but performance is affected by resources, task difficulty, and luck. A team may engage in excellent learning behavior while facing a harder project; another may perform well through an easy task or inherited expertise. If performance is used, select an outcome linked to the learning target and account for baseline differences. The correlation between a learning survey and success is not itself proof of the process by which success occurred.
Van den Bossche and colleagues examined social and cognitive factors in collaborative learning environments. Their work helps distinguish constructive interaction and shared understanding from individual endorsement of teamwork. In a group task, the relevant evidence may include whether members build on each other’s explanations and reach a usable shared account. A team can agree quickly without learning if its initial view was never tested. Conversely, disagreement can be productive when it makes assumptions visible and leads to inquiry.
Account for context and power
The ability to seek feedback depends on who controls information. A team with no access to customers or outcome data cannot demonstrate frequent external feedback seeking in the same way as a team with a live dashboard. Organizational constraints should be documented. A low observed behavior rate may reflect missing opportunities or authority rather than a team disposition. An intervention to improve learning should address those structural barriers as well as meeting habits.
Power can shape whose information is treated as credible. A senior member may be thanked for an observation that a junior colleague raised earlier without response. A measure that counts only whether an issue was eventually discussed can miss this pattern. Code who introduced information and who received uptake. Protect individuals in small-team reporting, because publishing that one person repeatedly voiced ignored concerns can expose them. Good measurement should not increase the cost of learning behavior.
Team membership may change during a project. New members bring information but may not know current routines. If the sample is a temporary action team, a survey of “usual behavior” may be less appropriate than event-based observation. If the team is long-lived, repeated measurements can show whether feedback seeking, reflection, and experimentation become routine. Do not compare a newly assembled group with a stable group without considering the opportunities each had to develop shared practices.
Task interdependence matters. People working independently under one label may not need collective learning for every decision. A highly coupled surgical or product team may need rapid shared updating. Define the task that the team performs jointly. A team learning score without a shared task can become a vague measure of collegiality. The methods should fit the way information must move for this particular work to improve.
Distinguish process from outcome
Learning behavior is an activity; learning outcome is a change in knowledge, coordination, or performance that can be attributed, at least partly, to that activity. The two should be measured separately. A reflective conversation can be sincere yet mistaken. An experiment can produce a null result and still teach the team something useful. Conversely, improved performance can occur without a measured learning process. Report the observed process and the observed outcome, then state how confidently they are linked.
Temporal ordering is essential for stronger claims. Measure the team’s baseline understanding or routine, observe the feedback and response, and assess later understanding or behavior. If all variables are collected in one retrospective survey, members may report learning because they feel successful. A longitudinal or experimental design can better separate that story from change. Even then, alternative events between waves should be recorded rather than assuming a single meeting caused the improvement.
What the evidence does not show
The cited research does not show that more discussion is always better, that every mistake disclosure leads to learning, or that a high survey score proves a team updated its work. Measures of psychological safety, voice, reflection, and team learning behavior overlap in practice but answer different questions. A team can feel safe and fail to learn; it can learn under pressure; it can report many learning activities while acting on none of them. The instruments should preserve those distinctions.
The primary studies come from particular work and learning settings. Their scales, coded interactions, and interventions cannot be imported unchanged into every culture, profession, or team structure. The Institute’s synthesis is to trace a specific information path: what was noticed, who brought it forward, how the group examined it, what changed, and what later evidence showed. When one link is missing, describe the part measured instead of calling the entire chain “team learning.”
Report the chain of evidence
A transparent report identifies the shared task, team boundary, measurement period, survey items, respondents, observation sample, coding rules, and behavioral traces. It distinguishes opportunity from action and process from outcome. It reports uncertainty in a team-level score and documents disagreements among members instead of hiding them in an average. If a debriefing intervention was used, it gives the format and follow-up interval.
The retrospective in the opening example would count as a learning opportunity, not as proof of learning. Its value becomes visible only when one can follow the lesson into the next decision and assess whether the team adapted. A shorter conversation that changes a routine may provide stronger evidence. Careful measurement rewards the work of inquiry and revision, not the appearance of reflection.
Research references
- Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44, 350–383. View source
- Savelsbergh, C. M. J. H., van der Heijden, B. I. J. M., & Poell, R. F. (2009). The development and empirical validation of a multidimensional measurement instrument for team learning behaviors. Small Group Research, 40, 578–607. View source
- Ellis, A. P. J., et al. (2003). Team learning: Collectively connecting the dots. Journal of Applied Psychology, 88, 821–835. View source
- Edmondson, A. C. (2003). Speaking up in the operating room: How team leaders promote learning in interdisciplinary action teams. Journal of Management Studies, 40, 1419–1452. View source
- Eddy, E. R., Tannenbaum, S. I., & Mathieu, J. E. (2013). Helping teams to help themselves: Comparing two team-led debriefing methods. Personnel Psychology, 66, 975–1008. View source
- Van den Bossche, P., Gijselaers, W. H., Segers, M., & Kirschner, P. A. (2006). Social and cognitive factors driving teamwork in collaborative learning environments. Small Group Research, 37, 490–521. View source