Reassurance that backfired

In 1986 a new disease was identified in cattle in the United Kingdom: bovine spongiform encephalopathy, or BSE, popularly known as mad cow disease. Over the next decade, as the epidemic grew, the UK government and its scientific advisers repeatedly told the public that beef was safe. In March 1996 the government announced that there was a probable link between BSE and a new variant of Creutzfeldt-Jakob disease in humans, a fatal brain disease. Public confidence collapsed, beef exports were banned by the European Union, and millions of cattle were slaughtered. The human death toll from variant CJD in the United Kingdom was in the low hundreds in the following decades, far fewer than some early projections but a grim outcome for the families.

The BSE Inquiry, chaired by Lord Phillips and reporting in 2000, found that the government's approach to communication had been shaped by a desire to avoid public alarm. Officials had given reassurance beyond what the evidence supported, and had not acknowledged uncertainty. When the link was announced, the public felt misled. The inquiry concluded that the risk should have been communicated candidly, with the uncertainty stated, and that the credibility of science advice had been damaged by the perception that it was serving political ends.

The episode is a standard case in the study of risk governance because it shows two things at once: that the science-policy interface is part of the governance system, and that how risk is perceived and communicated can matter as much as how it is measured. This lesson takes up both.

Why perception is part of the problem, not only an obstacle

Psychologists and risk analysts have shown that people judge risks by more than the expected number of deaths. In a series of studies in the 1970s and 1980s, Paul Slovic and colleagues, whose 1987 article in Science is widely cited, asked people to rate hazards on characteristics such as whether the risk is voluntary, how well it is known to science and to those exposed, how dreaded its consequences are, and whether it could cause catastrophic harm to many people at once. Risks that are involuntary, unfamiliar, hard to control and dreaded, such as nuclear power, were rated as far riskier than those that kill more people but are familiar and chosen, such as driving. The pattern is stable across many groups, including experts when they step outside their specialty.

The conclusion for governance is that perception is not simply a mistake to be corrected. It reflects values and concerns, such as fairness, consent and trust, that are legitimate inputs to deciding what risks are tolerable. That is why the IRGC framework includes a concern assessment alongside the scientific risk assessment. Ignoring concerns does not remove them; it leaves them to be expressed through opposition, delay or loss of trust.

Trust is the key variable. People who trust the institution responsible for a risk tend to accept it more readily, and trust is easier to lose than to gain. Honest communication of uncertainty, an admission of past errors, independence of advisers from the body that is being advised and openness of the evidence all build trust over time. Reassurance beyond the evidence, delay in releasing bad news and the appearance that advice is shaped by politics erode it, as BSE showed.

The science-policy interface

The science-policy interface is the set of arrangements by which evidence reaches decisions. Good arrangements have several features. Independence: advisers are chosen for expertise and are free of pressure to give the answer the government wants. Transparency: the advice, the evidence and the uncertainties are published, with the advisers' declarations of interest. Clear roles: scientists advise on what is known and with what confidence; decision makers decide, taking values, costs and public concerns into account. When scientists drift into making value judgements, or decision makers claim that "the science says" when the choice is about values, trust suffers on both sides. Two-way communication: decision makers need to ask questions scientists can answer, and scientists need to understand the decisions their work informs.

Several countries have formal structures for this, such as a chief scientific adviser and committees of independent experts, and some have formal rules for how uncertainty must be reported in advice. The precautionary principle, which appears in various forms in international environmental law and in European Union law, expresses one stance on uncertainty: that a lack of full certainty about serious or irreversible harm is not a reason to delay protective action. Applying it requires judgement about what is serious, what is proportionate and how to review a measure as evidence arrives.

Worked example (hypothetical): what a return period means over a planning horizon

A government tells a community that a river flood defence protects against a "1-in-100-year flood". Many residents hear this as "a flood like that happens once every hundred years, so it will not happen in my lifetime". The statistics say something different. A 1-in-100-year flood has a probability of 1 percent in any one year. The probability that it does not occur in a given year is 0.99. If years are independent, the probability that it does not occur in N years is 0.99^N, and the probability that it occurs at least once is 1 − 0.99^N.

• Over 10 years: 1 − 0.99^10 = 1 − 0.904 = 9.6 percent.
• Over 30 years, the length of a mortgage: 1 − 0.99^30 = 1 − 0.740 = 26.0 percent.
• Over 70 years, a lifetime: 1 − 0.99^70 = 1 − 0.495 = 50.5 percent.
• Over 100 years: 1 − 0.99^100 = 1 − 0.366 = 63.4 percent.

A resident who buys a house with a 30-year mortgage in the floodplain behind a defence designed for a 1-in-100 flood thus faces about a 1 in 4 chance of seeing that flood, or a larger one, during the mortgage. Now express the same facts as natural frequencies: "Imagine 100 households living there for 30 years. About 26 of them will see a flood that overtops a 1-in-100-year defence." People understand the second form far better than the first. The numbers are the same; the communication differs, and so will the response. A governance process that publishes only the return period has communicated the figure, not the risk.

Answers. Annual probability = 2 percent = 0.02. Chance of none in 25 years = 0.98^25 = 0.603; at least one = 1 − 0.603 = 39.7 percent. As a natural frequency: "Of 1,000 households living there for 25 years, about 397 will see at least one flood of that size or larger." Notice that the naive rule of thumb, 25 / 50 = 50 percent, overstates the chance. The naive rule is close only for short horizons, so the exact formula is the one to publish.

Evidence, values and the choice of what to measure

Evidence is never neutral in the sense that someone chooses what to count. A flood risk assessment that counts property damage will favour protection of valuable assets; one that counts people exposed will favour protection of dense low-income areas; one that counts lives lost may favour evacuation planning over embankments. The choice of metric is a value judgement, and governance should make it openly and with the people affected. The same holds for time horizons, discount rates and the treatment of rare, catastrophic outcomes. A good science-policy interface makes these choices visible, so that disagreement about values is not disguised as disagreement about facts.

Reading the chart as a resident would

The chart shows why the words "1-in-100-year" mislead. A resident who hears that phrase tends to think of a flood that happens once in a century and to expect not to see one. The numbers say otherwise. Over the length of a typical mortgage, 30 years, the chance of at least one flood of that size is about one in four. Over a lifetime of 70 years it is about half. A communicator who wants people to understand the risk can say, for example, "if you live here for 30 years, there is about a one in four chance that your home will be flooded at least once", and can add that this is a statement about the long run, not a prediction for any particular year. The same figure used in a planning decision tells a council that a development approved today is more likely than not to see such an event if its working life is 70 years or more.

What a strong answer looks like. A strong answer computes the probability correctly and states it in a way a resident can use. The message says something specific the reader can act on, includes a real uncertainty statement rather than boilerplate and commits to an update. The best answers avoid words such as "safe" and "protected" unless the claim is qualified with the design event.