Thinking and Deciding

506 RISK
Government agencies often incorporate this attitude into their policies. When the
U.S. government regulates toxic substances, it tries first to establish the frequency
of effects (poisoning, death, and other health effects) resulting from each chemical.
The frequency is equivalent to the probability for the average person. Each estimate
has some uncertainty around it, and this is represented as a 95% confidence interval.
For example, the best guess might be “ten cases per year” but the interval might be
“a 95% chance that the number of cases will fall between five and twenty.” Government
agencies typically use the upper bound of the 95% confidence interval as their
estimate for policy purposes (Zeckhauser and Viscusi, 1990). In some cases, this
policy leads to expensive cleanups that would not be done if the decision were based
on the expected benefit (Viscusi, Hamilton, and Dockins, 1997).
Such responses to ambiguity can result in too many resources being spent on reducing
ambiguous risks, when the same resources could do more good elsewhere.
Suppose two risks have ten expected fatalities per year. For risk A (like auto accidents),
we have had extensive experience and we know that this figure is accurate.
For risk B, the best guess is eight, but the 95% confidence interval ranges as high
as sixteen. If we think it is equally costly to reduce each risk by 50%, whatever its
level, and if we must allocate funds between these two risks, we will allocate all the
funds to B. If we go according to the best guess, we will allocate all the funds to A.
If we allocate funds according to the upper bound, then we will — on the average,
across many decisions of this type — do a little less well than we could do. On the
average, we will cut the total fatality rate by four instead of five.
This argument depends on one crucial assumption. This is that the “best guess”
is unbiased for unknown risks. If the best guess for unknown risks is systematically
too low, then the strategy of using a safety factor is normatively correct. (Of course,
it could be that the bias goes the other way.) To my knowledge, this assumption has
not been tested.
Catastrophic versus individual
Slovic et al. (1984) found that people are more frightened of risks with potentially
catastrophic outcomes, such as nuclear power-plant meltdowns, than of risks that are
expected to cause greater harm at a predictable rate, such as air pollution from the
burning of fossil fuels. This was also found in the psychometric research described
earlier. Public reactions seem to follow the same principle. Every day in the United
States, a dozen children are killed by guns, one at a time. The public and its representatives
were galvanized into action by a couple of well-publicized incidents in which
several school children were shot at once. This was, as several observers noted, a
“statistical blip” in the overall problem. But the fact that the deaths happened all
at once made people more concerned. Can our concern with catastrophic risks be
justified?
One possible justification is that the utility of a life lost increases with the number
of lives already lost. A disaster that kills 10,000 people could be 100,000 times worse
than an accident that kills ten. But, to a first approximation, the people who matter