Normalization of Deviance¶

Why It Matters¶

Organizations don't suddenly decide to accept dangerous conditions—they drift into them through incremental normalization. Each anomaly that doesn't produce failure makes the next anomaly seem acceptable. This creates invisible escalation of risk: the organization is objectively in a more dangerous state but subjectively feels safer ("we've done this before without problems"). It explains why organizations violate their own safety rules while genuinely believing they're being careful.

The Key Move¶

Track the history of anomalies over time. For each deviation from standard: when was it first observed? What was the initial reaction? When did it occur again? When did it stop triggering investigation? What language changed (from "violation" to "out of spec" to "acceptable" to "normal")? Plot frequency and severity over time. Ask: "What would our past selves think of current practices?" Identify the moment when extraordinary became ordinary.

Classic Application¶

NASA's Challenger disaster again—not just for hindsight bias but for normalization. O-ring erosion was initially classified as an anomaly requiring investigation. After repeated occurrences without failure, it was reclassified as "acceptable risk," then as "in-family" (normal expected behavior). The normalization process took years and was well-documented in meeting minutes, showing how the organization talked itself into accepting increasing risk.

Surprising Application¶

Personal fitness and health. You notice occasional knee pain during running. First time: "I should rest and investigate this." After it happens several times without acute injury: "I guess this is just normal for me now." You've normalized a deviation from healthy function. Or in work environments: first week of 60-hour weeks is exhausting, after a year it's "just how we work here." Tracking when "temporary" becomes "permanent" reveals normalization in action.

Failure Modes¶

Can become a conspiracy theory where any deviation is treated as dangerous, preventing legitimate learning and adaptation. Not all normalization is bad—some initial standards are overly cautious and rational updating occurs. Requires distinguishing normalization (forgetting why a rule exists) from calibration (learning actual risk levels). Hard to apply in real-time since normalization is invisible while happening—requires historical comparison.

Go Deeper¶

Vaughan, The Challenger Launch Decision, Chapter 5; Snook, Friendly Fire, Chapter 6; Dekker, Drift into Failure, Chapters 4-5.