At some level, human judgment is an unavoidable element of all human decisions. The question, then, naturally arises: How good is human judgment? Psychological studies of human judgment provide evidence for shortcomings and systematic biases in human decisionmaking. Furthermore, not only do peopleincluding expertssuffer various forms of myopia; they also often are oblivious of the fact. Indeed, statistical linear models summarizing the relationship between a set of predictor variables and a predicted outcome often (repeatedly) perform better than intuitive expert judgments (or subjective expert opinions). Burgeoning empirical evidence suggests that humans, including experts, can be inept at making judgments, particularly under conditions of high uncertainty.
Since the early 1970s, psychologists repeatedly have demonstrated human judgmental error and linked these errors to the operational nature of mental processes. The idea, spelled out in Kahneman et al. (1982), is that, because of limited mental processing capacity, humans rely on strategies of simplification, or mental heuristics, to reduce the complexity of judgment tasks. Although this strategy facilitates decisionmaking, these procedures are vulnerable to systematic error and bias.
In a classic series of publications, Tversky and Kahneman (1974, 1983) and Kahneman and Tversky (1979, 1996) claim that human judgment under uncertainty violates normative rules of probability theory. For example, Tversky and Kahneman (1983) invoke the "judgment by a representativeness" heuristic to explain evidence for the conjunction fallacy, whereby a conjunction of events is judged to be more likely than one of its constituents. This is a violation of a perfectly simple principle of probability logic: If A includes B, the probability of B cannot exceed A. Nevertheless, respondents consistently give a higher likelihood to the possibility of a subset or joint event than to the whole set, thereby violating the conjunction rule. Typically, respondents judge likelihood by representativeness (or stereotypes) and thus fail to integrate statistically relevant factors.
However, Gigerenzer (1994, 1996) argues that people are naturally adapted to reasoning with probabilities in the form of frequencies and that the conjunction fallacy "disappears" if reasoning is in the form of frequencies. Several studies report that violations of the conjunction rule are rare if respondents are asked to consider the relative frequency of events rather than the probability of a single event.
Kahneman and Tversky (1996) disagree and argue that the frequency format provides respondents with a powerful cue to the relation of inclusion between sets that are explicitly compared or evaluated in immediate succession. When the structure of the conjunction is made more apparent, respondents who appreciate the constraint supplied by the rule will be less likely to violate it.
Kahneman and Lovallo (1993) argue that people have a strong tendency to regard problems as unique although they would be viewed more advantageously as instances of a broader class. People pay particular attention to the distinguishing features of a particular case and reject analogies to other instances of the same general type as crudely superficial and unappealing. Consequently, they fall prey to fallacies of planning by anchoring their estimates on present values or extrapolations of current trends. Despite differing causal theories, both approaches find evidence for poor judgment under uncertainty or, alternatively, evidence that people are better off not attempting to assess probabilities for single events.
Nonetheless, public understanding of likelihood seems to be improved by adoption of frequentist formats. Several studies have shown that experts have great difficulty reasoning with subjective probabilities for unique or single events. However, respondents apparently are much more successful when the same problems are presented with frequencies rather than probabilities. Although experts have difficulties with the probability versionmost give wrong answersmost undergraduates readily provide the correct answer to similar problems constructed with frequencies.
Psychological research suggests that measures of risk that are communicated in terms of frequencies rather than probabilities will be more readily understood and rationally responded to, although IAMs need to translate these frequencies into probability distributions (e.g., Morgan and Dowlatabadi, 1996) to portray the wide range of outcomes that currently reflect estimates in the literature and by most IPCC authors.
Other reports in this collection