This target article falls into an odd genre in psychological metatheory – a genre where authors try to stop the field from doing something. In the present case, the authors want researchers to stop using normative theory in a certain way. Such papers are indeed sometimes highly cited and discussed, but in my experience they rarely succeed in actually shaping ongoing empirical research. The evolution of research programs continues apace, determined more by the yield in new empirical findings than by arguments about metatheory. This is likely to be even more true in the present case, where the authors are trying to constrain a particular practice in one of the most fruitful and influential research programs in all of psychology – the heuristics and biases approach. Using normative theory to study decision making was useful in the past, and it continues to be useful. The present target article is also embarrassingly timed, coming on the heels of the 2008/2009 financial crisis. Empirical and theoretical work in behavioral finance – much of it using normative approaches – now constantly appears in media articles attempting to explain aspects of the crisis (as well as famous scams such as the Madoff Ponzi).
In their target article, Elqayam & Evans (E&E) drive an odd wedge between instrumental rationality and Bayesianism (the latter defined as so-called normative rationality). The former is seen as appropriate, whereas the latter is seen as the type of approach that should be avoided. Contrary to this bifurcation, most theorists see the Bayesian strictures as being a crucial component of instrumental rationality, not a definition of rationality somehow separate from the instrumental view. Joyce (Reference Joyce, Mele and Rawling2004) gives what most decision scientists would see as the canonical view in The Oxford Handbook of Rationality:
Bayesianism provides a unified theory of epistemic and practical rationality based on the principle of mathematical expectation. In its epistemic guise it requires believers to obey the laws of probability. In its practical guise it asks agents to maximize subjective expected utility [SEU]. (Joyce Reference Joyce, Mele and Rawling2004, p. 132)
In short, most theorists incorporate Bayesianism, in Joyce's sense, into the notion of instrumental rationality, so the attempt to separate the two in the target article is a nonmainstream view. If it is only normative theory outside of approaches related to instrumental rationality (defined, as Joyce does, as including the Bayesian and associated SEU approaches) that is being attacked here, then 90% of the heuristics and biases tradition is spared, and the authors are left with a straw man that may apply to the four-card and a few other strange tasks. The whole attempt to drive a wedge between Bayesianism and instrumental rationality leads to another odd conclusion – that Bayesian norms do not help us as organisms achieve our goals. Precisely the reason that people should want to follow the axioms of utility theory (transitivity, etc.) as normative models is that failure to follow them means that a person is not maximizing utility. They should want to avoid becoming a money pump.
Another finding that is somewhat embarrassing for a view (like the one in the target article) that stresses the multiplicity of norms is that people most often retrospectively endorse the Bayesian and SEU norms that they violate (Shafir Reference Shafir, Manktelow and Over1993; Shafir & Tversky Reference Shafir, Tversky, Smith and Osherson1995). In introducing the collection of Amos Tversky's writings, Shafir (Reference Shafir2003) stresses this very point: “The research showed that people's judgments often violate basic normative principles. At the same time, it showed that they exhibit sensitivity to these principles' normative appeal” (p. x). For example, Koehler and James (Reference Koehler and James2009) found that non-normative “probability matchers rate an alternative strategy (maximizing) as superior when it is described to them” (p. 123).
E&E follow Cohen (Reference Cohen1981) in using the grammatical competence/performance distinction borrowed from linguistics as a structural analogue for understanding judgment and decision making. However, the analogy to linguistics here is inapt because it ignores the fact that the cultural evolution of reasoning tools has no analogy in language. Because normative models are tools of rationality and because these tools undergo cultural change and revision, there is no idealized human “rational competence” that has remained fixed throughout history. Thus, the analogy to the linguistic domain is forced and inappropriate, as argued many years ago by Jepson et al. (Reference Jepson, Krantz and Nisbett1983):
The analogy between language and inductive reasoning fails to recognize that there is far more to language than grammatical competence. Important cultural inventions, such as writing or new vocabulary, increase the effectiveness of language use… . We contend that effective inductive reasoning is also a skill; that it depends on cultural innovations, such as probability theory. (Jepson et al. Reference Jepson, Krantz and Nisbett1983, p. 495)
Psychologists are going to quite naturally study how beliefs are formed. That is inherently one of our topics. For example, the formation of stereotypes is a huge topic in social psychology. Of course, we could study beliefs without ever asking whether they were true. But what would be gained by refraining from asking if a particular stereotype studied in social psychology was generally true? When we do start to ask whether a belief is true – whether it maps to reality – then we have brought in the issue of epistemic rationality and its norms. It is the same with the study of decision making. We can link decisions with hedonic states and utility without ever asking whether a hedonic state or utility was maximized. But I cannot see the point of refraining from this question. Once we don't refrain from this step, norms of instrumental rationality (axioms of consistency and freedom from context effects) come in. In short, I do not think that psychology will heed any of these admonitions to refrain from normative language. I know I won't. So while this target article is provocative and full of deep issues, it is not going to change the type of psychology my colleagues and I practice. In our lab, subjects do make, what we unabashedly call – errors.
This target article falls into an odd genre in psychological metatheory – a genre where authors try to stop the field from doing something. In the present case, the authors want researchers to stop using normative theory in a certain way. Such papers are indeed sometimes highly cited and discussed, but in my experience they rarely succeed in actually shaping ongoing empirical research. The evolution of research programs continues apace, determined more by the yield in new empirical findings than by arguments about metatheory. This is likely to be even more true in the present case, where the authors are trying to constrain a particular practice in one of the most fruitful and influential research programs in all of psychology – the heuristics and biases approach. Using normative theory to study decision making was useful in the past, and it continues to be useful. The present target article is also embarrassingly timed, coming on the heels of the 2008/2009 financial crisis. Empirical and theoretical work in behavioral finance – much of it using normative approaches – now constantly appears in media articles attempting to explain aspects of the crisis (as well as famous scams such as the Madoff Ponzi).
In their target article, Elqayam & Evans (E&E) drive an odd wedge between instrumental rationality and Bayesianism (the latter defined as so-called normative rationality). The former is seen as appropriate, whereas the latter is seen as the type of approach that should be avoided. Contrary to this bifurcation, most theorists see the Bayesian strictures as being a crucial component of instrumental rationality, not a definition of rationality somehow separate from the instrumental view. Joyce (Reference Joyce, Mele and Rawling2004) gives what most decision scientists would see as the canonical view in The Oxford Handbook of Rationality:
Bayesianism provides a unified theory of epistemic and practical rationality based on the principle of mathematical expectation. In its epistemic guise it requires believers to obey the laws of probability. In its practical guise it asks agents to maximize subjective expected utility [SEU]. (Joyce Reference Joyce, Mele and Rawling2004, p. 132)
In short, most theorists incorporate Bayesianism, in Joyce's sense, into the notion of instrumental rationality, so the attempt to separate the two in the target article is a nonmainstream view. If it is only normative theory outside of approaches related to instrumental rationality (defined, as Joyce does, as including the Bayesian and associated SEU approaches) that is being attacked here, then 90% of the heuristics and biases tradition is spared, and the authors are left with a straw man that may apply to the four-card and a few other strange tasks. The whole attempt to drive a wedge between Bayesianism and instrumental rationality leads to another odd conclusion – that Bayesian norms do not help us as organisms achieve our goals. Precisely the reason that people should want to follow the axioms of utility theory (transitivity, etc.) as normative models is that failure to follow them means that a person is not maximizing utility. They should want to avoid becoming a money pump.
Another finding that is somewhat embarrassing for a view (like the one in the target article) that stresses the multiplicity of norms is that people most often retrospectively endorse the Bayesian and SEU norms that they violate (Shafir Reference Shafir, Manktelow and Over1993; Shafir & Tversky Reference Shafir, Tversky, Smith and Osherson1995). In introducing the collection of Amos Tversky's writings, Shafir (Reference Shafir2003) stresses this very point: “The research showed that people's judgments often violate basic normative principles. At the same time, it showed that they exhibit sensitivity to these principles' normative appeal” (p. x). For example, Koehler and James (Reference Koehler and James2009) found that non-normative “probability matchers rate an alternative strategy (maximizing) as superior when it is described to them” (p. 123).
E&E follow Cohen (Reference Cohen1981) in using the grammatical competence/performance distinction borrowed from linguistics as a structural analogue for understanding judgment and decision making. However, the analogy to linguistics here is inapt because it ignores the fact that the cultural evolution of reasoning tools has no analogy in language. Because normative models are tools of rationality and because these tools undergo cultural change and revision, there is no idealized human “rational competence” that has remained fixed throughout history. Thus, the analogy to the linguistic domain is forced and inappropriate, as argued many years ago by Jepson et al. (Reference Jepson, Krantz and Nisbett1983):
The analogy between language and inductive reasoning fails to recognize that there is far more to language than grammatical competence. Important cultural inventions, such as writing or new vocabulary, increase the effectiveness of language use… . We contend that effective inductive reasoning is also a skill; that it depends on cultural innovations, such as probability theory. (Jepson et al. Reference Jepson, Krantz and Nisbett1983, p. 495)
Psychologists are going to quite naturally study how beliefs are formed. That is inherently one of our topics. For example, the formation of stereotypes is a huge topic in social psychology. Of course, we could study beliefs without ever asking whether they were true. But what would be gained by refraining from asking if a particular stereotype studied in social psychology was generally true? When we do start to ask whether a belief is true – whether it maps to reality – then we have brought in the issue of epistemic rationality and its norms. It is the same with the study of decision making. We can link decisions with hedonic states and utility without ever asking whether a hedonic state or utility was maximized. But I cannot see the point of refraining from this question. Once we don't refrain from this step, norms of instrumental rationality (axioms of consistency and freedom from context effects) come in. In short, I do not think that psychology will heed any of these admonitions to refrain from normative language. I know I won't. So while this target article is provocative and full of deep issues, it is not going to change the type of psychology my colleagues and I practice. In our lab, subjects do make, what we unabashedly call – errors.