Confirmation Bias in Safety: 6 Blind Spots Leaders Miss

The ILO's 2023 global estimates report nearly 3 million deaths each year from work-related accidents and diseases. Confirmation bias does not explain all of that harm, but it explains a pattern leaders recognize too late: once a team believes the job is safe, evidence that supports the belief gets louder and evidence that disturbs it gets softened.

On the Headline Podcast, Andreza Araujo and Dr. Megan Tranter often return to the same leadership discipline: real safety depends on the quality of the conversation before the event, not the elegance of the explanation afterward. Confirmation bias is dangerous because it makes the conversation feel rational while it is already narrowing.

This article is a diagnostic for senior EHS leaders, plant managers, and supervisors who need to see where bias enters safety decisions before a weak signal becomes a serious incident.

Why confirmation bias is a leadership problem

Confirmation bias is the tendency to notice, prefer, and remember information that supports an existing belief while discounting information that challenges it. Daniel Kahneman's work on judgment and cognitive bias helps explain why this is not a character flaw. It is a predictable feature of human decision making under pressure.

In safety, the bias often appears after a leader has already decided that a site is improving, a supervisor is competent, a contractor is under control, or a procedure is working. From that point forward, the dashboard, audit, field visit, and investigation can become evidence collection for a conclusion that was chosen too early.

The market usually treats confirmation bias as an investigator problem, but that is too narrow. The same bias shapes capital decisions, production meetings, safety walks, pre-task briefings, contractor reviews, and executive dashboards. If leaders only correct it after incidents, the organization has already paid the highest price for a preventable learning delay.

1. Clean metrics receive more trust than field evidence

The first blind spot appears when leaders trust clean indicators more than uncomfortable field evidence. A low TRIR, stable LTIFR, or quiet near-miss channel can feel like proof that the system is working, especially when the board wants a short safety story.

The contrary evidence is usually less tidy. A mechanic says the isolation sequence is rushed. A contractor reports that the permit is signed before the field is ready. A supervisor admits that the rescue plan exists on paper but has not been practiced. Those signals rarely arrive as a dashboard trend, which is why leaders who depend only on clean numbers can miss the condition that matters most.

Andreza's Portuguese title Muito Alem do Zero, often explained in English as Far Beyond Zero, challenges this exact comfort. Zero harm as a slogan can make leaders defend the number instead of testing whether the number reflects real control.

Use safety KPI weighting as a reference point: every comforting metric should sit beside a bad-news indicator, because confirmation bias grows fastest where the dashboard only rewards agreement.

2. Supervisors hear agreement and call it alignment

The second blind spot is conversational. A crew meeting can sound aligned because nobody objects, although silence may mean fatigue, resignation, fear of delay, or the belief that the decision has already been made.

Confirmation bias enters when the supervisor interprets silence as consent because consent is the answer the plan needs. The job can then move forward with an untested assumption, and the leader later says that no one raised a concern.

The correction is not to demand dramatic dissent. It is to ask questions that make contrary evidence normal. What part of the plan depends on perfect attention? Which control would fail first if we lost time? What did we ignore because the job is routine? These questions make disagreement usable before it becomes personal.

The same discipline appears in safety coaching after shortcuts, where the leader tests the context behind the behavior instead of treating the first answer as the whole truth.

3. Investigations protect the first theory

The third blind spot appears after an incident or high-potential near miss. Once the team forms an early theory, every interview, document, and photograph can be pulled toward that story.

If the first theory is operator error, the investigation notices missed steps. If the first theory is contractor weakness, it notices contract gaps. If the first theory is supervision, it notices handover problems. Some of those facts may be real, but confirmation bias turns a partial truth into the whole explanation.

James Reason's work on latent failures gives safety teams a stronger anchor because it forces the question upstream. What conditions made the active failure more likely? Which planning, design, staffing, maintenance, or management decision shaped the event before the worker arrived at the task?

This is why operator blame in RCA remains one of the most important cross-checks for confirmation bias. If the cause story became simple too quickly, the investigation probably stopped where the evidence became comfortable.

4. Familiar contractors get lighter questions

The fourth blind spot is relationship-based. A contractor with a long history, friendly supervisors, and a clean recent record often receives less challenge than a new contractor, even when the work package contains the same high-energy exposure.

Familiarity can be useful because trust supports coordination. It becomes dangerous when leaders confuse a familiar company with a verified control. The question is not whether the contractor usually performs well. The question is whether this task, on this day, with this crew, under this pressure, has controls that are visible and tested.

Confirmation bias also works in reverse. A contractor with a poor reputation may receive harsh scrutiny even when the current crew has corrected the previous weakness. Both patterns weaken judgment because the leader is reading the label before reading the work.

A practical defense is to standardize the critical-control questions for all high-risk work. Familiarity may change the tone of the conversation, but it should not change the evidence required before exposure begins.

5. Production pressure becomes invisible when the plan succeeds

The fifth blind spot appears when the team completes difficult work under time pressure and leaders remember the success more vividly than the compromises that made it possible.

A shutdown finishes on schedule. A late order ships. A maintenance task restarts the line. Because the outcome is positive, the organization may treat the method as proof of resilience rather than asking which controls were stretched, skipped, improvised, or dependent on one experienced person.

This is how production pressure becomes culture. The first recovery is treated as exceptional. The second becomes expected. The third becomes the new planning assumption, and confirmation bias filters out the warnings because the operation can point to previous success.

The related article on production pressure decisions shows the same mechanism from the behavior side. Leaders must review successful recoveries with the same curiosity they bring to failures, because success can hide the next exposure.

6. Leaders ask for evidence after the decision is already made

The sixth blind spot is the most senior one. Leaders sometimes request evidence only after the budget, deadline, staffing model, or public commitment has already been set.

At that point, the organization is no longer asking whether the plan is safe enough. It is asking the safety team to find a way to support a plan that has political momentum. The EHS leader may still produce a risk assessment, but the assessment is operating inside a narrowed decision space.

On Headline Podcast, the phrase visible felt leadership matters because workers need to feel that leaders will change a decision when reality changes. A leader who asks for dissent after approval has already taught the organization that challenge is welcome only if it does not disturb the plan.

The fix is to move safety evidence earlier. Before deadlines are promised, contracts awarded, or staffing assumptions frozen, leaders should ask which controls make the plan feasible and which assumptions would make the plan unsafe.

Comparison: biased review versus disciplined review

2 competing explanations should be the minimum standard before a serious safety decision closes, because the second explanation forces the team to examine evidence that the first story would ignore.

How leaders can break confirmation bias before it hardens

Leaders reduce confirmation bias by designing friction into decisions. The goal is not endless debate. The goal is to make the system pause long enough for inconvenient evidence to surface while the decision can still change.

Start with three controls. First, require a dissent question in every high-risk pre-task briefing. Second, ask investigators to write two plausible cause paths before finalizing the report. Third, pair every green metric with one indicator that would reveal underreporting, weak controls, or hidden recovery work.

Co-host Andreza Araujo has explored the cultural side of this discipline in Safety Culture: From Theory to Practice, where culture is treated as repeated behavior under real pressure. Confirmation bias is one of the reasons declared culture and operated culture separate.

Each month that leaders accept only evidence that confirms the plan makes the next weak signal harder to hear, because people learn which facts the organization prefers.

Conclusion

Confirmation bias in safety does not usually arrive as ignorance. It arrives as confidence, familiarity, success, clean metrics, and a story that feels complete before the evidence has been tested.

The leadership task is to protect the organization from premature certainty. When leaders ask for contrary evidence early, reward useful dissent, and review success with the same rigor as failure, they make real safety more likely before the dashboard has a chance to prove them wrong.