• Choose a category
• All
• Classic-Fiction

About a month after the earthquake, some seismologists affiliated with the working group published an article in the prestigious journal Science that included the following statements: The earthquake “fulfilled a long-term forecast,” “two decades of research . . . allowed an accurate long-term forecast of both the occurrence and consequences” of the earthquake, and the earthquake occurred “where it was anticipated.” Obviously there were significant misunderstandings about the meaning of probabilities and their validation. First, the earthquake was less likely to occur (probability 0.3) than not (which therefore had a probability of 0.7) in the next thirty years on the Loma Prieta segment, so how did this earthquake fulfill a long-term forecast? Second, if an earthquake occurred in the San Francisco area, it was more likely to occur on one of the other four segments, so why did this earthquake occur where it was anticipated? The quality of probabilistic forecasts is a complex concept and accuracy cannot be determined from a single event.

Interpreting probabilities even when there is a significant amount of data is often perplexing. A couple of years ago at a military academy, in a questionnaire prior to a seminar, I asked the following: “Suppose a rare medical condition is present in 1 of 1,000 adults, but until recently there was no way to check whether you have the condition. However, a recently developed test to check for the condition is 99 percent accurate. You decide to take the test and the result is positive, indicating that the condition is present. What is your judgment of the chance, in percent, that you have the condition?” Almost half of the more than 300 respondents thought the chance they had the disease was 99 percent, approximately a quarter had an answer between 3 percent and 15 percent, and the remaining quarter were spread thinly over the entire range from 0 to 95 percent. The correct answer is “9 percent,” which is almost shocking to many people.

The logic can be explained in simple terms as follows. Suppose there are 1,000 random adults who are to be tested. On average, we would expect only 1 of these adults to have the disease and the test for that individual would almost surely be positive. This gives us 1 positive test. Also, we would expect 999 adults not to have the disease. With an error rate of 1 percent, we would expect essentially 10 of these individuals to have a positive test result. Thus, for the 1,000 individuals, we would expect an average of 11 positive tests, only 1 of which was for an individual with the rare medical condition. Thus, only 1 of 11—or 9 percent—of those with the positive test result really have the rare medical condition. Research in the decision sciences has frequently shown that individuals’ intuitions in probabilistic situations are far from accurate.

THE AMPLIFICATION OF RISK

The social amplification of risk is another concept relevant to numerous public policy decisions. Eight years prior to the introduction of the social amplification of risk framework in 1988, I conceptualized an amplification model, without using the word amplification. The model separated the direct personal impacts of fatalities due to a specific cause and the induced indirect societal impact of those fatalities. Then, I constructed a value model and did assessments to preliminarily examine the importance of such amplification. Specifically, this model was developed for government agencies that have responsibility for public safety.

In conceptualizing the model, I stated that “there should be two major governmental concerns regarding impact of fatalities on the public. The first reflects the direct personal impacts of pain, suffering, and economic hardship. These fall most heavily on a very small percent of the total public, those individuals who are the fatalities and their friends and relatives. The second concern involves indirect societal impacts, which include general political, social, and economic turmoil which may occur as