Long-running deviations become the new normal

What it is

Normalization of deviance (a term from sociologist Diane Vaughan) describes an organizational decision chain: management approves departures from an established safety boundary step by step, exception by exception, without any single operator ever "breaking a rule." Each approval looks locally reasonable at the time. This is distinct from an individual worker turning a convenient workaround into personal habit — normalization happens at the approval layer, and operators are often doing exactly what was formally authorized.

Why it happens

The first exception request usually comes with a specific, defensible reason — a part temporarily unavailable, schedule pressure, a similar case already waved through one level up — and the approver grants it after weighing the tradeoff; nothing bad happens right away. The next time a similar case comes up, the approver no longer reaches back to the original rule's full risk assessment; they reach for the precedent that "a similar exception was approved before and nothing happened," and approval itself becomes the default, with the risk assessment quietly skipped. This repeats up and down the approval chain: each level sees only the precedent set by the level before it, and cites that precedent instead of re-evaluating. After enough rounds, what began as an exception has, in practice, replaced the original procedure as the standard way of working — without a single person or a single meeting ever deciding to abandon the safety rule. The outcome is assembled from a string of locally reasonable exception approvals, not produced by any one decision.

Studying it

The object of longitudinal reconstruction is the approval record itself, not just the operating record: the stated reason for each exception, who approved it, and whether a risk assessment was attached. The share of approvals missing a risk assessment, tracked over time, marks the point where approval shifted from evaluation to citing precedent. Analyze near misses and uneventful operating periods alike — reconstructing only from the last deviation before an accident misses the earlier, seemingly harmless approvals that built the pattern.

Where it stops holding

Not every recurring exception is normalization: when each approval carries its own fresh risk assessment and the new condition is written into an explicit boundary, that is ordinary procedure revision, not normalization. The distinguishing question is whether the approver assessed this case's risk or merely cited last time's approval — the latter is the marker. The absence of an accident does not disprove normalization either; accidents are low-probability events, and not having one yet does not mean the margin has not actually narrowed.

Applying it

• Log every exception approval and audit it retrospectively: pull a sample of already-approved exceptions and check whether the original reasoning and risk assessment still hold, rather than only checking whether current operation matches whatever was last approved.
• Put a counter and expiry date on recurring temporary exceptions; once a count or time limit is reached, trigger a formal review automatically instead of letting the same approval carry on indefinitely.
• Rotate approvers, or bring in a reviewer independent of the production line, so the same people are not repeatedly citing their own past approvals as justification.
• How to check: track the share of approvals in each exception category that carry a complete risk assessment; a share that keeps falling while the approval count keeps rising is a direct signal that normalization is underway.

Related

• Same group: Y7.03.1 Deviations often arise because procedures do not match reality · Y7.03.3 Understand motives for deviation rather than only prohibiting it
• Nearby: Y7.04 Tension between production and safety · Y3.10 Parameter limits and safety interlocks
• Search terms: normalization of deviance · practical drift · risk migration