Confirmation Bias in Safety: 4 Decisions That Hide Contrary Evidence

Confirmation bias rarely appears as an obviously careless choice. It appears when a supervisor sees a familiar control, recognizes a familiar task, and accepts evidence that supports the belief that the work is under control. This article shows how to challenge that confidence before it becomes routine exposure.

Why does confirmation bias survive competent safety teams?

Confirmation bias in safety is the tendency to notice, interpret, and remember evidence that supports an existing risk judgment while discounting evidence that threatens it. A competent team can still fall into this pattern because experience makes familiar work feel easier to classify, especially when production pressure rewards quick closure.

The danger is not that people ignore every warning. They often notice the warning and then explain it away. A loose guard becomes a temporary defect, an unusual alarm becomes a sensor problem, and a worker's hesitation becomes a personal preference. The explanation protects the original decision, which is why the bias can look like practical judgment.

James Reason's model of active and latent failures helps separate the person from the system that shaped the judgment. The supervisor may be the last visible decision-maker, but the interpretation was influenced by staffing, schedule, prior success, local norms, and the way the organization responds to bad news.

Across 25+ years leading EHS at multinationals, Andreza Araujo has seen that the strongest field conversations do not ask only whether the rule was followed. They ask what evidence would prove the control is weakening, who can see that evidence, and what decision follows when it appears.

Decision 1: Treating a familiar task as a familiar risk

The first confirmation-bias decision is assuming that a task remains low risk because the team has completed it many times without a serious event. Familiarity changes attention, but it does not remove exposure, and a small change in material, sequence, weather, staffing, or equipment condition can alter the failure path.

That assumption feels reasonable because the work has a history of success. Yet past completion is not proof that the control was strong. It may only show that several weak layers happened to align. A task can remain uneventful while its margin for recovery is shrinking.

Andreza Araujo's work on risk perception, including the ideas developed in *80 Ways to Expand Risk Perception*, treats perception as something that must be widened deliberately. The relevant question is not whether the task looks ordinary. The question is which new condition could make the old control inadequate.

Before the shift starts, ask the operator to identify one condition that would make the usual method unreliable. Then verify that condition in the work area. If the answer never changes across days, the conversation is probably measuring recall rather than perception.

Decision 2: Choosing the evidence that makes the control look healthy

The second confirmation-bias decision is selecting convenient evidence while ignoring evidence that is harder to collect. A signed pre-job brief, completed observation card, or closed action may be real evidence, but none of them automatically proves that the critical control worked under the actual exposure.

Managers prefer tidy evidence because it travels well through dashboards and review meetings. Field evidence is less comfortable. It includes a worker bypassing a step to protect a handoff, a supervisor accepting a degraded barrier, or an operator explaining that the procedure cannot be completed within the planned cycle time.

250+ companies have been served worldwide through Andreza Araujo's work, and that breadth reinforces one practical lesson: the evidence closest to the task is often the evidence most likely to contradict the report. The contradiction is not a nuisance to remove. It is the diagnostic signal that tells leaders where the written control and the lived control have separated.

Use a two-column verification habit. Record what the system says should be visible, then record what the worker must actually do when the task changes. If the second column is empty, the audit is confirming documentation rather than testing protection.

Decision 3: Explaining away contrary signals

The third confirmation-bias decision is assigning a reassuring explanation to a signal that deserves investigation. Teams often call repeated alarms nuisance events, treat near misses as isolated mistakes, or describe a worker who stops the task as overly cautious.

A contrary signal matters because it weakens the story that the control is stable. The right response is not to dramatize every deviation. It is to classify the signal by consequence, recurrence, and proximity to a serious injury or fatality, then decide whether the current control can still be trusted.

As Andreza Araujo argues in *Safety Culture: From Theory to Practice*, culture becomes visible in the treatment of small decisions before they become large events. A team that rewards only smooth production will learn to make inconvenient evidence disappear, even when its leaders say that speaking up is welcome.

Supervisors can interrupt the pattern with one question: “What would we say if this happened three more times next week?” That question moves the conversation away from whether the current event is acceptable and toward whether the trend is telling the truth about the control.

Decision 4: Confusing agreement with a tested judgment

The fourth confirmation-bias decision is treating fast agreement as proof that the risk judgment is sound. Agreement may reflect shared knowledge, but it may also reflect hierarchy, fatigue, production pressure, or a team norm that makes dissent expensive.

Silence is especially misleading during toolbox talks and pre-job briefings. People may understand the hazard and still avoid raising a concern because they expect delay, blame, or a vague instruction to “be careful.” The absence of objections therefore cannot serve as a leading indicator by itself.

During her PepsiCo South America tenure, where the accident ratio fell 50% in six months under a 180-day plan, Andreza Araujo learned that safer behavior required changing what supervisors tolerated, not merely repeating the message. A question that invites disagreement is stronger than a closing request for “any concerns?” because it gives the team permission to test the plan.

Ask one person to argue that the control will fail, even when the group believes the task is ready. Then ask what field observation would settle the disagreement. The goal is not artificial debate. It is to make the judgment testable before exposure begins.

What should a supervisor compare before accepting a reassuring signal?

A supervisor should compare the declared control with the observable behavior, the expected condition with the current condition, and the reported confidence with the evidence that could disprove it.

When a team uses this comparison consistently, confirmation bias becomes easier to see because the judgment has a defined test. That is the difference between asking people to be more alert and redesigning the conversation so inconvenient evidence has somewhere to go.

Every shift in which a reassuring explanation replaces a tested judgment increases the chance that a weak control will become normal work, especially when the next change arrives without a fresh challenge.

How can leaders make contrary evidence useful?

Leaders make contrary evidence useful by defining in advance what must be checked, who can challenge the decision, and what action follows when the evidence does not fit the plan.

That process does not require suspicion toward operators. It requires disciplined curiosity from the people who authorize work. A supervisor who asks for disconfirming evidence is not undermining confidence. The supervisor is protecting the team from confidence that has not been tested.

In more than 250 cultural transformation projects, Andreza Araujo observes that behavior changes when the local consequence changes. If a worker raises an inconvenient signal and receives a faster technical response, the team learns that accurate information is valuable. If the signal only produces blame or delay, silence becomes the rational choice.

Confirmation bias is therefore a safety-control problem, not merely a personal thinking error. Strong teams do not eliminate judgment. They build routines that challenge judgment before familiar evidence becomes permission to proceed. Safety is about coming home, and that requires leaders to test the story that the work is safe, not just repeat it.