Teams and individuals

Psychological safety

Amy Edmondson's definition: a shared belief held by members of a team that the team is safe for interpersonal risk-taking. People believe they can speak up with questions, concerns, mistakes, and half-formed ideas without being humiliated, punished, or written off.

This is the best-evidenced construct in this guide.

What it is not:

• Not niceness. Safe teams often have more open disagreement, not less.
• Not lowered standards. Edmondson's 2×2 pairs safety with accountability:

Safety without standards is the failure mode critics attack; standards without safety is the failure mode most organizations actually have.

• Not an individual trait. It is a property of a group and varies sharply between teams in the same organization — which is good news, because a local leader can change it without waiting for the enterprise.

The counterintuitive finding. Edmondson's original hospital study expected better teams to report fewer errors. They reported more, because they were willing to. Reporting rate measures willingness to disclose, not underlying error rate. See the reporting-rate paradox.

Measuring it. Edmondson's seven-item scale, 1–7 (items 1, 3, 5 reversed):

1. If you make a mistake on this team, it is often held against you.
2. Members of this team are able to bring up problems and tough issues.
3. People on this team sometimes reject others for being different.
4. It is safe to take a risk on this team.
5. It is difficult to ask other members of this team for help.
6. No one on this team would deliberately act in a way that undermines my efforts.
7. Working with members of this team, my unique skills and talents are valued and utilized.

Behavioral indicators are harder to fake than a survey: who speaks in meetings and in what proportion (measure it once — the result is usually startling); whether anyone asks a basic question in front of senior people; time between a problem being known by one person and known by the team; whether "I don't know" and "I was wrong" are ever said by senior people.

Building it — Edmondson's leader toolkit:

1. Frame the work as learning or as complex, not as execution of a known routine. If people believe the answer is known, not knowing becomes shameful.
2. Acknowledge your own fallibility, specifically rather than ritually.
3. Model curiosity. Ask real questions and wait. Silence after a question is the leader's tool; filling it is the leader's most common mistake.
4. Respond productively. The response to the first person who raises something uncomfortable sets the norm for everyone watching.

Structural supports: round-robins, written-first meeting formats, pre-mortems, blameless reviews, and anonymous channels as a starting ramp rather than a destination.

Health care: a nurse or junior doctor questioning a senior clinician's drug dose. Structural aids — read-back, graded assertiveness, checklists with a designated challenge point — work even where the culture has not yet changed. Software: an engineer saying "I don't understand this design" in review, or reporting an outage they caused within minutes rather than hours; the measurable proxy is time-to-disclosure during incidents. Executive: the hardest room. An executive who says "here's what I got wrong last quarter" buys more safety than any program; one who visibly punishes a bad-news bearer removes it for everyone in the room and everyone they tell afterwards.

See topics/psychological-safety.

What makes teams effective

Google's Project Aristotle studied 180 teams over two years. Its most useful contribution is the null result.

Did not predict effectiveness: individual ability or seniority; personality mix or introvert/extrovert balance; colocation; tenure, team size, or team age; whether members socialized outside work.

A great deal of team intervention money is spent on factors that showed no effect.

The five factors that did, in descending importance:

1. Psychological safety — by a wide margin.
2. Dependability — can we count on each other to do high-quality work on time? The most neglected in favor of the first. Safety without dependability produces comfort, not performance.
3. Structure and clarity — clear goals, roles, and execution plans.
4. Meaning — is the work personally important to each member?
5. Impact — do we believe the work matters and creates change?

Limitations: one company, one sector; the null result on individual ability may reflect restricted range, since everyone hired was already strong; and popular retellings compress it to "psychological safety is all that matters", dropping four of the five factors.

Health care: structure and clarity is frequently the binding constraint in multidisciplinary teams, where role boundaries are contested and nobody wants to name it. Meaning and impact are usually strong, which is why erosion of the others is so demoralizing. Software: hiring optimizes hard for individual ability, which the study found irrelevant to team effectiveness — the implication is reallocating some hiring investment into team norms. Executive: dependability is the executive-team factor most often broken — commitments made in the room and not delivered, with no consequence.

See topics/google-project-aristotle.

Team development models

Tuckman: forming, storming, norming, performing, adjourning. The most widely taught team model and one of the least supported — derived from a literature review of therapy groups, and later research does not find that teams reliably proceed in this order.

Its one genuinely useful message: early conflict in a new team is normal and is not evidence of a wrong hire or a bad team. That prevents real damage and justifies the model's continued use. Beyond that, use something better.

A practical alternative: charter instead of stage-watch. In week one, agree:

• Purpose — why does this team exist, and what happens if it does not?
• Success — how will we know, by when?
• Roles — who decides what, and who must be consulted?
• Norms — meetings, response times, disagreement, escalation.
• Review — when do we revisit this? Put the date in the calendar now.

Drexler-Sibbet is more useful as a diagnostic, because each of its seven stages names both the resolved and unresolved state:

Two informative patterns: when a team disagrees about which stage is unresolved, the disagreement is itself the finding; and a late-stage symptom very often has an early-stage cause — missed deadlines (5) are usually unclear goals (3) or absent commitment (4), and more project management will not fix that.

Lencioni's five dysfunctions — absence of trust, fear of conflict, lack of commitment, avoidance of accountability, inattention to results — is a business fable rather than research, and remains popular because executives recognize themselves in it. Work bottom up: attacking inattention to results with dashboards fails when the cause is three levels below. Its best single question for a CEO: "which team is your first team — the one you lead, or the one you're a member of?" Executives who answer "the one I lead" are describing the fifth dysfunction, and the answer is usually honest.

Health care: rotating junior staff mean many clinical teams re-form every few months. The remedy is fast explicit onboarding of norms — a written charter, an introduction ritual, and a named person to ask. Software: a squad arguing about branching strategy in week two is fine; one still arguing in month six has a decision-rights problem, not a stage problem. Executive: a new executive team's storming is expensive and public; accelerate it deliberately with a facilitated chartering session rather than waiting for it to emerge in front of the organization.

See topics/tuckman-stages-of-group-development, topics/drexler-sibbet-team-performance-model, and topics/lencioni-five-dysfunctions-of-a-team.

Conflict

Thomas-Kilmann maps five conflict modes on assertiveness and cooperativeness:

Two observations carry most of the value. Collaborating is not always right — it is expensive, and a team that collaborates on trivia exhausts itself. And compromising is not collaborating: compromise splits the difference and settles positions; collaboration digs beneath positions to interests and can produce an option neither party brought. The tell is whether a new option appeared.

Interests, not positions (Fisher and Ury): separate the people from the problem; focus on interests, not positions; invent options for mutual gain before deciding; insist on objective criteria. Two units fighting over headcount (positions) may have interests in delivery certainty and career progression, both satisfiable without the headcount.

Health care: a surgeon and anaesthetist disagreeing about proceeding should not compromise — this needs collaboration if time permits, or a pre-agreed decision rule if it does not. Splitting the difference on clinical risk is the worst available option. Software: architecture disagreements are avoided until they become expensive; use a decision record with an explicit deadline to force the conflict into the open while it is cheap. Executive: executives overuse compromising in resource allocation — proportional cuts, nobody happy, strategy unserved. Strategy requires choices, which means someone loses. An executive team that has never had a genuine loser has not made a strategic choice.

Structural aid: crew resource management. Aviation's response to accidents in which technically airworthy aircraft crashed because junior crew could not effectively challenge a captain who was wrong. The transferable tool is graded assertiveness — a scripted escalation ladder:

1. Probe — "I'm not sure that's right; can you help me understand?"
2. Alert — "I'm concerned about the dose."
3. Challenge — "I need you to stop and re-check."
4. Emergency — "Stop. I am not going to continue."

Scripting works because it converts an interpersonal risk into a professional routine: the junior person does not have to invent the words or judge whether the moment warrants confrontation. This makes it effective even where the culture has not yet changed, which distinguishes it from culture-first approaches. Related: CUS ("I'm Concerned, I'm Uncomfortable, this is a Safety issue"), closed-loop communication with read-back, and structured handover formats such as SBAR and I-PASS.

See topics/thomas-kilmann-conflict-modes and topics/crew-resource-management.

Motivation

Self-determination theory is the best-supported account available. Three innate psychological needs:

1. Autonomy — actions experienced as self-endorsed. Not independence; a person can be autonomous while following a rule they endorse.
2. Competence — experiencing oneself as effective and growing.
3. Relatedness — feeling connected to and cared about by others.

The motivation continuum replaces the intrinsic/extrinsic binary: amotivation → external regulation (reward/punishment) → introjected (guilt, ego) → identified ("this matters to me") → integrated ("this is who I am") → intrinsic (inherent enjoyment). The goal is not to make work enjoyable, which is often impossible; it is to move regulation toward identified and integrated — by explaining rationale, acknowledging that a task is unpleasant, and offering choice about how it is done.

The undermining effect. Tangible, expected, contingent rewards reduce intrinsic motivation for tasks that were already interesting, and narrow attention to the rewarded metric. Nuances that matter: the effect applies to interesting work, not to dull algorithmic work; unexpected rewards and informational praise do not undermine; and underpayment is a powerful demotivator. Deci's practical advice — pay people well and fairly, then take money off the table as a lever.

Designing for the needs. Autonomy: give rationale for unwelcome requests, offer real choice, acknowledge negative feelings, minimize controlling language. Competence: optimal challenge, fast specific feedback, visible progress, and removal of obstacles that make good work impossible. Relatedness: stable teams — churn is a relatedness tax — and managers who show genuine regard.

Herzberg's hygiene/motivator distinction survives as a useful heuristic even though the two-factor structure is likely a method artifact. Hygiene factors (pay, policy, conditions, security) remove dissatisfaction; motivators (achievement, recognition, the work itself, responsibility, growth) create satisfaction. Fixing hygiene moves people from dissatisfied to neutral. His prescription was job enrichment — a vertical addition of responsibility — as opposed to job enlargement, which adds more tasks of the same kind and achieves nothing.

Maslow's hierarchy is weakly supported as a strict ordering, and Alderfer's ERG (existence, relatedness, growth) is a better formulation — three categories, no strict prepotency, and a frustration-regression mechanism: when growth is blocked, people intensify demands on relatedness and existence, which is why pay and conditions disputes rise when development is frozen.

One durable implication survives: you cannot run higher-order interventions on people whose safety needs are active. During restructuring, redundancy consultation, or unsafe staffing, vision workshops and values exercises are received as insulting, and survey scores fall regardless of intervention quality. The right intervention is information — "we don't know yet, and we'll tell you on the 14th" outperforms optimistic vagueness.

Health care: burnout research maps closely onto SDT. Clinicians rarely lack meaning; they lack autonomy (protocols and targets imposed without rationale or consultation) and competence satisfaction (prevented from delivering the standard of care they were trained for). Pizza and resilience training address neither. Software: autonomy over how is the strongest engineering retention factor in most surveys, and competence is damaged by long feedback loops — a two-hour build denies the sense of effectiveness dozens of times a day. Executive: individual bonus schemes tied to narrow metrics predictably produce narrow behavior. If the work is complex and interdependent, contingent individual reward is working against the design of the organization.

Goal setting theory (Locke and Latham) is the other strongly supported piece here, and the one most degraded in practice. Specific and difficult goals beat "do your best" — provided the person is committed to the goal and gets feedback on progress. Both conditions are necessary, and both are routinely dropped.

Two refinements matter in knowledge work. First, learning goals versus performance goals: a performance goal ("reduce lead time to three days") works when the method is known; a learning goal ("identify and test three causes of the delay") works when it is not. Setting a performance goal on a task whose method is unknown reliably degrades performance, because attention goes to the outcome rather than to the strategy search the task requires. Second, the documented side effects — narrowed focus, unethical reporting, distorted risk taking, and eroded cooperation — all follow from the same mechanism that makes goals work: they direct attention. A goal that directs attention at a proxy directs behavior at the proxy. Pair every goal with a balancing measure, and avoid attaching individual financial consequence to a single number in interdependent work.

Burnout is the failure state of all of the above, and it is a property of the job rather than of the person. Maslach's three dimensions are exhaustion, cynicism, and reduced professional efficacy; the actionable frame is the six areas of worklife — workload, control, reward, community, fairness, and values. Values mismatch is the strongest predictor in much of the research and the one wellbeing programmes never address. Reviews consistently find that organization-directed interventions outperform person-directed ones: if the cause is a job condition, teaching the person to tolerate it does not remove it, and offering tolerance training in place of a fix is itself a fairness and values mismatch. That is why wellbeing programmes are sometimes received with anger by the people they were meant to help.

See topics/self-determination-theory, topics/goal-setting-theory, topics/burnout, topics/herzberg-two-factor-theory, and topics/maslow-hierarchy-of-needs.

Work design

The job characteristics model (Hackman and Oldham) is strongly supported and targets the design of the work itself, which is more durable than motivating the person.

Five core characteristics:

1. Skill variety
2. Task identity — completing a whole, identifiable piece of work with a visible outcome
3. Task significance — the work substantially affects other people
4. Autonomy — freedom in scheduling and determining procedures
5. Feedback from the job itself — direct information about performance from doing the work, not from a manager. The most commonly missed characteristic.

Variety, identity, and significance produce experienced meaningfulness; autonomy produces experienced responsibility; feedback produces knowledge of results. All three states are required.

MPS = ((variety + identity + significance) / 3) × autonomy × feedback

The multiplication is the point: autonomy or feedback near zero collapses the score regardless of the other three. A meaningful job with no autonomy and no feedback is not motivating, however important it is.

Implementing principles: combine tasks (reverse fragmentation); form natural work units (give a team a coherent domain); establish client relationships (connect the worker directly to the person they serve — powerful for both significance and feedback); vertical loading (give planning and control that was the manager's); open feedback channels.

Health care: task identity is damaged when care is fragmented across handoffs and no clinician sees the whole journey or outcome. Named-nurse and continuity-of-care models are direct applications, as is simply telling a team what happened to the patients they treated. Feedback from the work is frequently absent in acute settings, and its absence is demoralizing in a way staff struggle to articulate. Software: "you build it, you run it" raises task identity, autonomy, and feedback simultaneously. A team that writes to someone else's specification, hands off to QA, and never sees production has MPS near zero whatever the salary. Fast pipelines and observability are feedback mechanisms, not just conveniences. Executive: when automating or outsourcing, examine what the remaining job contains. Removing the interesting fragments and leaving exception handling destroys skill variety and task identity, and the resulting attrition is usually attributed to something else.

See topics/job-characteristics-model.

Learning

Single-loop vs. double-loop learning (Argyris and Schön). Single-loop corrects the action, leaving goals and assumptions unchanged: "we missed the target; work harder." Double-loop questions the governing variables: "why is that the target?"

Most organizations are competent at single-loop and structurally incapable of double-loop, because it requires questioning things senior people have committed to publicly.

Espoused theory vs. theory-in-use. What people say they believe, versus what their behavior implies they believe. The gap is invisible to the person and visible to everyone else. Do not ask people about their values; observe behavior and infer the theory-in-use. Espoused: "we want challenge and debate." Theory-in-use: "challenge is welcome from peers, not juniors, and never in front of the board."

The ladder of inference — observable data → selected data → added meanings → assumptions → conclusions → beliefs → actions, with a reflexive loop from beliefs back to what data we select. This is why two competent people can observe the same meeting and disagree completely about what happened. Facilitation moves: "what did you see or hear that led you to that?"; "here is my data and the meaning I put on it — what do you see differently?"

Organizational defensive routines prevent embarrassment while preventing the organization from identifying their causes: the reorganization that avoids naming an underperforming executive; the "communication problem" diagnosis that avoids naming the disagreement; the consultant hired to say what employees have said for two years; the meeting after the meeting.

Immunity to change (Kegan and Lahey) explains why people fail at changes they sincerely want. The failure is a competing commitment protected by an untested big assumption. The four-column map: (1) improvement goal; (2) what I am doing or not doing that works against it; (3) the hidden competing commitment; (4) the big assumption that makes column 3 feel necessary. Then design a small, safe, reversible experiment to test the assumption.

At group level this is the highest-value use in OD, because it converts vague cultural complaints into falsifiable claims. An executive team's column 4 might be "we assume a visible mistake ends a career here" — which is a testable claim about the organization, and frequently a true one, in which case the intervention is not psychological. It is changing what actually happens to people who are wrong.

The learning organization (Senge): personal mastery, mental models, shared vision, team learning, and systems thinking. The concept fails when it stays abstract; the operational mechanisms are blameless post-incident review with published findings, decision records capturing context and expected consequences, explicit predictions checked afterwards, rotation and boundary-spanning, protected slack, and consequence-free channels for bad news.

Senge's "learning disabilities" are more useful than the disciplines: "I am my position"; "the enemy is out there"; fixation on events while blind to slow patterns; the delusion of learning from experience, since the most consequential decisions have consequences beyond the horizon in which we can observe them; and the myth of the management team — a group that presents unity outwardly, suppresses disagreement inwardly, and is therefore "skilled at keeping itself from learning."

Health care: strong individual learning cultures — CPD, audit, M&M meetings — coexist with weak organizational learning, because findings do not travel between wards. The fix is structural and boring: a route by which one ward's learning reaches the other forty, with someone accountable for it. Software: post-incident reviews are the strongest learning mechanism most engineering organizations have, and publishing them internally in full is the differentiator. Executive: ask when the executive team last publicly changed its mind. If the answer is never, the team is managing appearances.

See topics/double-loop-learning, topics/immunity-to-change, and topics/learning-organization.

Leading individuals

Situational leadership (Hersey and Blanchard) proposes matching leadership style to the follower's development level for a specific task.

The two classic errors: over-delegating to D1 and D2 — a new person given autonomy and no structure in the name of empowerment, who flounders and is read as a hiring mistake; and over-directing D3 and D4 — an expert supervised as though new, which is the fastest way to lose senior people while believing you are being helpful.

The critical nuance: development level is per task, not per person. A senior clinician is D4 in clinic and D1 on a new digital system. A staff engineer is D4 in their domain and D1 leading their first team.

Evidence: weak. The style-matching prescriptions have repeatedly failed confirmation. Its real value is as a conversation structure giving a manager and a report language for "I need more support here" and "I need you to back off there" — a genuine service, independent of whether the four-quadrant theory is true. Better-evidenced neighbors: goal setting theory (specific, difficult goals with feedback) and self-determination theory's autonomy-support mechanism.

Health care: clinical supervision is formally graded by competence — the gap is on non-clinical tasks, where a consultant asked to lead a service change is often D1 and treated as D4. Software: a promoted tech lead is D4 as an engineer and D1 as a manager, and typically receives S4 delegation because they are senior. This is the standard mechanism by which good engineers become unsupported managers. Executive: newly appointed executives are D1 on the political system of the organization whatever their track record; the first ninety days need S2, and it is rarely offered because asking for it looks like weakness.

See topics/situational-leadership-model.