Culture

What culture is

Edgar Schein's definition:

A pattern of shared basic assumptions that the group learned as it solved its problems of external adaptation and internal integration, that has worked well enough to be considered valid and, therefore, to be taught to new members as > the correct way to perceive, think, and feel in relation to those problems.

Two consequences. Culture is learned solutions that worked, which is why it is defended — it is the accumulated answer to real past problems. And culture operates below awareness, which is why asking people to describe their culture returns the wrong layer.

Culture is over-diagnosed, because it is unfalsifiable and blames nobody in particular. Rule it in only after structure, incentives, and workload have been excluded.

Schein three levels

1. Artifacts — visible, easy to observe, hard to interpret. Office layout, dress, rituals, language, tools, meeting behavior, stories. You can catalog artifacts in a day and misread every one.

2. Espoused beliefs and values — stated strategies, goals, philosophies. The values page. What the group says it wants.

3. Basic underlying assumptions — unconscious, taken-for-granted, and non-negotiable in the sense that contradicting them produces confusion or anger rather than debate. This is the culture.

The diagnostic work is inferring level 3 from level 1, and treating the gap between levels 2 and 3 as the finding.

Reading the third level. Never ask directly. Instead:

• What happens when someone makes a mistake here? Watch, do not ask.
• What gets someone promoted? Compare with the values page.
• What can be said in a meeting, and what is only said in the corridor?
• Which rules are broken routinely and without consequence?
• Who is a hero here, and what story is told about them?
• What surprised you in your first month? Ask new joiners in week three.

Health care: espoused "speak up for safety"; underlying assumption "the consultant is not questioned." The revealing artifact: who speaks first on a ward round, and who never speaks. Software: espoused "we value quality"; underlying assumption "shipping on the announced date is what counts." The artifact: what happens in the last week before every release. Executive: Schein's most useful executive observation is the persistent clash between three occupational subcultures — the operator culture (the people doing the work), the engineering culture (designers of systems, who prefer people-free solutions), and the executive culture (financial abstraction, necessarily impersonal at scale). Most enterprise dysfunction is this three-way misunderstanding, not a missing values statement.

See topics/schein-model-of-organizational-culture.

Competing values framework

Cameron and Quinn's two axes — flexibility ↔ control, internal ↔ external — produce four types. The associated OCAI instrument is among the most widely used culture measures.

The "competing" part is the point. These values genuinely conflict. An organization cannot maximize all four. Most executive teams, asked what culture they want, describe all four — and that answer is the finding.

Method: respondents divide 100 points across the four types on six dimensions, twice: now and preferred. The gap is the conversation.

That forced split is also the instrument's main weakness. Because the four scores must total 100, one type can only rise if another falls, so the profile is a relative ranking and not a measurement of how much Clan or Market a unit actually has. Read it as "which of these dominates here", never as a level, and do not compare a unit's raw numbers against a published benchmark. Useful readings: leadership and frontline producing different "now" profiles means leadership does not know its own organization; different units with incompatible profiles means friction at their boundary; everyone wanting more Clan usually signals exhaustion rather than a culture strategy.

Health care: typically Hierarchy-dominant for defensible reasons — safety, regulation, standardization. Ambitions to become "more innovative" without relaxing any control mechanism are incoherent, and staff know it. Software: startups begin Adhocracy, drift to Market at scale, then add Hierarchy at maturity. Founders experience this necessary drift as decline; naming the intended profile at each stage defuses a lot of unproductive nostalgia. Executive: after an acquisition, run OCAI on both organizations before integration planning. An Adhocracy acquired by a Hierarchy will lose the people who created the value being purchased.

See topics/competing-values-framework.

Westrum typology

Ron Westrum classified culture by how information flows, particularly information someone would rather not hear.

"Failure leads to inquiry" is the key row. After a failure, a pathological culture asks who, a bureaucratic culture asks which rule was broken, and a generative culture asks what about our system made this the reasonable thing to do. Only the third produces learning; the first two produce hidden information, which is the precondition for the next, larger failure.

Evidence: good. The DevOps research program found Westrum generative culture statistically predicts both software delivery performance and organizational performance, using a six-item survey scale. This is one of very few culture constructs with a demonstrated link to outcomes.

The six-item measure, rated 1–7: information is actively sought; messengers are not punished for bad news; responsibilities are shared; cross-functional collaboration is encouraged and rewarded; failure causes inquiry; new ideas are welcomed. Short enough to run quarterly. Report at team level and above.

Moving toward generative: respond to bad news with curiosity and visible thanks, every time, including when the news is inconvenient and the messenger is annoying; make failure reviews blameless by design and facilitation, not by slogan; reward the person who surfaces a problem early, publicly, in front of peers; remove the mechanisms that punish disclosure; fund bridging roles.

One senior leader who reacts badly to bad news once can undo a year of this. That asymmetry is the whole problem.

Health care: wards with high near-miss reporting are usually safer, not more dangerous — establish this with the board before reporting rates rise. Software: an incident review ending with "human error" is a pathological or bureaucratic artifact. Executive: measure how long it takes for bad news to reach you and how distorted it is on arrival. Ask your directs what they did not tell you last quarter, and react well to the answer.

See topics/westrum-organizational-culture-typology.

Just culture

An approach to accountability after failure that rejects both blame culture and blame-free culture. Most errors are produced by systems, and individual accountability still exists — but it attaches to behavioral choices, not to outcomes.

Three behaviors and three responses:

1. Human error — inadvertent slip, lapse, or mistake. Console, then fix the system that made the error likely and its detection unlikely. Punishing human error does not reduce error; it reduces reporting.
2. At-risk behavior — a choice that increases risk where the risk is unrecognized or believed justified. Coach. If everyone drifts the same way, the procedure is the problem.
3. Reckless behavior — conscious disregard of a substantial and unjustifiable risk. Discipline. Rare, and genuinely warranted when present.

The substitution test. Would three other similarly trained, similarly experienced peers, in the same situation with the same information and the same pressures, have done the same thing? If yes, the individual is not the cause.

The outcome bias problem. Identical decisions are judged differently depending on how they turned out. The clinician who is interrupted and administers correctly is invisible; the one who errs is investigated. Same system, same behavior, different luck.

Implementation requirements: a published decision algorithm applied consistently, so people can predict how they will be treated — predictability is the mechanism; consistency across seniority, without which the policy is dead and everyone knows it; separation of learning review from disciplinary process, with different people and a clear statement of which is happening; managers trained in the distinctions.

Health care: medication errors are overwhelmingly system-produced — look-alike packaging, interruption-heavy environments, understaffing, poor handover. Suspending the nurse removes one person and leaves the system that will produce the next error, while ensuring it is not reported. Software: an engineer deletes production data. Just culture asks why the command was available, why it required no confirmation, why the backup was untested, and why one person could do it alone. Firing them purchases no safety and loses the person who now understands the failure best. Executive: the test is what happens the first time a serious failure is publicly embarrassing. Every prior statement about learning culture is settled in that moment, and the organization will calibrate on it for years.

See topics/just-culture.

Hofstede cultural dimensions

Six dimensions describing national cultural differences: power distance, individualism/collectivism, motivation toward achievement (originally masculinity/femininity), uncertainty avoidance, long-term orientation, and indulgence/restraint.

Use it to generate hypotheses about cross-border friction. Never to characterize an individual.

The ecological fallacy is the fatal misuse: scores are national averages, and within-country variance exceeds between-country variance on most dimensions. Applying a national score to the person in front of you is stereotyping with a citation attached. Other serious criticisms: single company (IBM), single era, mostly male professional respondents, nations treated as cultures, and scores treated as static.

Useful applications are about interaction design, not about people:

• Power distance mismatch: a manager says "challenge me" and reads silence as disengagement, while team members read the invitation as a test. Remedy: structured input — written first, round-robin, or anonymous.
• Uncertainty avoidance mismatch: one group wants a specification, the other wants to prototype. Both are competent. Remedy: agree explicitly what is fixed and what is open, in writing.

Health care: international recruitment is standard in many health systems, and power distance differences directly affect willingness to challenge a senior clinician — a patient safety issue. The remedy is structural: graded assertiveness protocols and read-backs, which work regardless of background. Software: distributed teams need explicit norms about who speaks, how disagreement is registered, and whether silence means consent. Executive: in cross-border integration, expect the acquirer's practices to be experienced very differently than intended. Use the dimensions to generate questions to ask locally, never as a substitute for asking.

See topics/hofstede-cultural-dimensions.

How to actually change culture

Schein's position: culture cannot be changed directly. It changes when the group learns that a new way solves its problems better.

Levers, in descending order of power:

1. What leaders pay attention to, measure, and control — routinely, not occasionally.
2. How leaders react to critical incidents and crises. One layoff, one handled failure, one tolerated bad actor teaches more than a year of values work.
3. How leaders allocate resources and rewards.
4. Deliberate role modeling and coaching.
5. Criteria for recruitment, promotion, and exit — the strongest long-run lever, because it changes who "we" are.
6. Secondary: structure, systems, rituals, physical space, stories, formal statements.

Formal values statements are the weakest mechanism available, and they are where most organizations start. A values statement that contradicts observed behavior is worse than no statement, because it documents the hypocrisy and gives every cynic a citation.

A practical culture change sequence:

1. Establish what the underlying assumptions actually are, by observation and by asking new joiners.
2. Identify the specific behaviors that would differ if the assumption changed. Behaviors, not attitudes.
3. Find the bright spots — where the desired behavior already happens here — and study the conditions that permit it.
4. Change what is measured, rewarded, and promoted so the new behavior is cheaper than the old.
5. Have leaders visibly do the new behavior under pressure, and visibly handle one uncomfortable case well.
6. Remove the structural obstacles that make the old behavior rational.
7. Only then talk about it — and describe what changed, not what you aspire to.

Timescale: years, not quarters. Anyone promising a culture change in two quarters is selling something.