Women and taxis and dangerous judgments: content sensitive use of base‐rate information

Abstract

This study shows that the use or non-use of base-rate information in probability estimation depends not just on the form of the problem, but also on the content about which the base-rate information is given. When the information is stated about characteristics of types of humans it is used and recalled better than if the information is about arbitrary things. We speculate that the content specificity is a consequence of the human mind’s attentiveness to stereotypes. Making predictions for the uncertain future is part of our daily lives. We assess whether it is going to rain before deciding to take an umbrella, and we expect that traffic will stop at the read light before deciding to cross the road at an intersection. People also frequently make probability assessments as part of their job in organizations, ranging from judging the suitability of a candidate for a given job to making assessments for the future behavior of their fellow colleagues, competitors or customers. While these judgments are often made routinely and sometimes effortlessly there is no guarantee that these are also made optimally. According to Bayesian reasoning people can only make optimal probability assessments if they take into consideration both what they know about the specific entity they are considering (“individuating information”) and also rely on what they generally know about the broader population of which the entity belongs to (ie, “base-rate” or “prior” information). Doctors, for example, should assess the likelihood of a breast cancer for a patient with a positive mammogram, by not only considering the reliability of the mammogram (individuating information) but also considering the known likelihood that a women in the age group developing a breast cancer (base-rate information). Kahneman and Tversky (1973) and others (e.g., Lyon and Slovic, 1976; Nisbett and Borgida, 1975; Taylor and Thompson, 1982) found that people tend to deviate from the normative Bayesian model by ignoring baserate information in the presence of even flimsy individuating information. The claim that people generally ignore base-rate information is controversial largely because other studies have shown that at least under some circumstances people do take into account base-rate information (e.g., Carroll and Siegler, 1977; Manis, Dovalina, Avis and Cardoze, 1980; Bar-Hillel and Fischhoff, 1981). What is uncontroversial is that under some conditions people do not make proper use of base-rate information and at other times they do. As Fiske and Taylor (1991, p: 361) note “the appropriate question regarding the use of base-rate information is when is it used, rather than if it is used.” More ambitiously, we want more than just a catalog of when base-rate information is used. We should work toward a theory of human cognition that explains why it is used in some cases and not in others. We have noticed that when the object about which the base-rate informa-