Zero-risk bias

Zero-risk bias is a preference for eliminating one identifiable risk even when another option would reduce more total expected harm but leave some risk behind. The attraction is the clean endpoint: one danger moves from some to none.

That preference is not automatically mistaken. Removing a risk may be reasonable when its consequences are more severe, irreversible, unfairly distributed, legally prohibited, or difficult to monitor. The bias label fits only after the relevant costs and consequences have been compared on the same terms.

A fictional archive worksheet

Everything in this example is invented. The archive, error categories, rates, equal-consequence assumption, testing plans, and budget are teaching devices, not operational data or professional guidance.

A fictional archive digitization team can fund one testing plan. The worksheet stipulates that the two error types have equal consequences, both plans cost the same, and both cover the same 10,000 checks.

• Plan A reduces a rare label-placement error from 2 to 0 per 10,000 checks.
• Plan B reduces an indexing mismatch from 12 to 5 per 10,000 checks.

Plan A reaches zero and prevents 2 expected errors. Plan B leaves 5 expected errors but prevents 7. Under the worksheet's assumptions, Plan B prevents 5 more errors per 10,000 checks.

Choosing Plan A would demonstrate zero-risk bias only if its zero endpoint were the reason for overlooking that larger reduction. If label errors carried greater damage, or if the estimates had different uncertainty, the comparison would need to change.

What makes a zero-risk choice biased?

The word zero usually refers to one named category, not to total safety. Eliminating one failure mode does not erase unrelated hazards, and it may shift resources away from risks that could be reduced more efficiently.

A useful comparison starts with absolute reduction:

absolute risk reduction = starting risk - residual risk

When outcomes differ in severity, a simple decision model may instead compare expected harm:

expected harm = sum of probability × consequence across relevant outcomes

These calculations make tradeoffs visible. They do not settle every decision. Rights, equity, catastrophic loss, uncertainty, legal duties, reversibility, and who bears the remaining risk can all justify a choice that does not maximize the average expected reduction.

Zero-risk bias is therefore narrower than “choosing zero.” It is the extra preference for closing one risk category after the important differences between options have been held constant or made explicit.

What the classic cleanup study found

Jonathan Baron, Rajeev Gowda, and Howard Kunreuther used an abstract hazardous-waste questionnaire with 408 government workers and professionals. Participants considered two fictional cities with 8 and 4 expected cancer cases per year and ranked cleanup programs. (Baron, Gowda, and Kunreuther, 1993)

In the version designed to test zero risk:

• one program changed the case counts from 8 and 4 to 4 and 2, preventing 6 cases;
• a second changed them to 7 and 0, preventing 5 cases while reaching zero in the smaller city;
• a third changed them to 3 and 3, also preventing 6 cases.

The zero option prevented fewer cases than either alternative. Even so, 42% of respondents did not rank it last, and 11% ranked it first.

Those percentages belong to a mailed questionnaire built from simplified, hypothetical cases. They are not an estimate of how common the bias is in a general population, and the cancer counts were not epidemiological findings. The study establishes a choice pattern under a particular comparison.

A zero-risk premium is one way to measure the pattern

Some studies ask what people would pay rather than which program they would rank first.

W. Kip Viscusi, Wesley Magat, and Joel Huber studied risk-dollar tradeoffs involving household-product risks in a sample of more than 1,500 consumers. They found a premium for completely eliminating one risk, but no strong evidence of an additional premium for eliminating multiple risks. (Viscusi, Magat, and Huber, 1987) The result supports a zero-risk premium in those stated valuations, not one universal price people attach to certainty.

Kazuya Nakayachi asked 144 undergraduates to value protective actions that left 800, 400, or zero deaths in a stylized task. The willingness-to-pay increase from 400 to zero was larger than the increase from 800 to 400, although both steps removed the same number of deaths. The pattern appeared in both positive and negative frames. (Nakayachi, 1998) Because participants stated values for hypothetical actions, the study does not show what they would spend or implement in practice.

Related explanations are not synonyms

Several ideas can help explain why zero receives special weight, but they answer different questions.

The certainty effect describes a broader risky-choice pattern. Kahneman and Tversky observed that people can give disproportionate weight to outcomes received with certainty compared with outcomes that are merely probable. (Kahneman and Tversky, 1979) Zero-risk bias reverses the viewpoint: it concerns making a specified bad outcome impossible. The two ideas are related, but a sure gain in a lottery is not the same task as allocating resources across hazards.

Probability neglect refers to weak sensitivity to probability differences. Yuval Rottenstreich and Christopher Hsee found that affect-rich outcomes produced greater sensitivity near impossibility and certainty but less sensitivity to changes between intermediate probabilities than affect-poor outcomes. (Rottenstreich and Hsee, 2001) Emotionally charged hazards may therefore magnify the appeal of zero. That evidence does not show that every zero-risk choice results from ignoring probability.

Expected-value or expected-harm analysis is a comparison rule. It combines probabilities and consequences across options. Zero-risk bias is a descriptive departure from such a rule only when expected harm is the appropriate standard and the zero option performs worse under the agreed inputs.

Risk aversion concerns preferences under uncertainty and the shape of utility. A person may reasonably reject a higher average payoff to avoid a ruinous outcome. That is different from selecting a smaller risk reduction merely because one line on the worksheet reaches zero.

The effect changes with the task

The evidence does not support a fixed response that appears whenever zero is offered.

Ilana Ritov, Jonathan Baron, and John Hershey found that evaluations of multiple risk reductions changed with framing and reference points. The value assigned to elimination also fell when the source of the risk remained in place, and a status-quo effect appeared. (Ritov, Baron, and Hershey, 1993) Reaching zero on an outcome measure and removing the thing that generates the risk are psychologically distinct.

Four studies by Elisabeth Schneider and colleagues compared questionnaires with behavioral tasks, forced choices with resource allocation, and different decision domains. They concluded that the result was persistent across methods but highly sensitive to task abstractness, domain, and whether the zero option seemed appropriate. (Schneider et al., 2017)

A small professional survey shows why those limits matter. Among 54 U.S. formulary decision makers evaluating fictional oncology products, 90.7% chose the zero-risk option in one scenario, while 32.1% chose it in another. The authors treated the zero-risk findings as inconclusive and called for larger studies using real decisions. (Mezzio et al., 2018) One striking percentage cannot be separated from the scenario that produced it.

Policy allocation needs more than a slogan

Zero-risk bias can matter when a fixed budget is split across environmental, health, safety, or infrastructure risks. A policy that removes a small, visible hazard can be easier to explain than one that leaves residual risk everywhere while preventing more harm overall.

But “maximize expected cases prevented” is not a complete policy constitution. Decision makers may need to consider:

• the severity and reversibility of each outcome;
• uncertainty in probability and consequence estimates;
• whether the same people receive the benefit and bear the residual risk;
• legal obligations and rights that cannot simply be averaged away;
• the cost, timing, and durability of each intervention;
• risk transfer, substitution, and unintended effects;
• public reasons that can be audited rather than inferred after the choice.

The practical error is not caring about complete protection. It is letting the symbolic closure of one category replace comparison across the whole decision.