It is uncontroversial that a full understanding of behaviour involves multiple aspects. Psychology seeks to identify lawful regularities: We seek to understand the “what” of behaviour such that we can predict it. We also seek to understand why these regularities obtain. This involves two distinct kinds of causal explanation: (1) an understanding of the mechanisms/processes that give rise to the behaviour, and (2) functional explanation, that is, an understanding of why this behaviour and not others. Finally, psychology considers how such understanding allows performance to be improved in practice.
For all these questions, normative standards, that is, characterizations of how something “ought to be,” seem indispensable. Trivially, performance cannot be improved without knowing what would count as “better.” Likewise, functional explanation will make reference to the fact that a behaviour maximizes some “desirable” criterion – where both “desirability” and “maximization” typically invoke normative considerations. Even the basic task of identifying behavioural regularities cannot afford blindness to normative considerations.
Rational standards provide essential interpretative tools. Human behaviour typically affords many different interpretations. In daily life we resolve this ambiguity with the help of “the principle of charity” (e.g., Govier Reference Govier1987). Given multiple possible interpretations of an utterance, we pick, as a default, the interpretation that renders it most sensible. Of course, this interpretation may prove wrong, and further evidence may lead us to abandon it; that there are default orderings over possible interpretations, however, simplifies massively the task of understanding.
Moreover this has consequences not only for the interpretation of a single utterance or act, but also for prediction. How we interpret what someone has said or done has direct implications for what we expect that person to do in other, similar situations. Furthermore, even without specific knowledge of an individual we can often make reasonably accurate predictions just on the basis of what would be “sensible” (though again, there is no guarantee that these predictions will be correct).
The principle of charity has both informal and more formal manifestations within psychological research. Informally, wherever we observe something surprising, we should ask ourselves whether there is an interpretation of participants' behaviour that renders it sensible and hence predictable. This may lead to re-evaluation of how our task is understood by the participants, to a different interpretation of what they are doing, and to quite different predictions and observations in response to changes in task parameters. Demonstrations of the (unintended) influence of the pragmatics of the experimental situation in giving rise to seeming “errors” and “biases” are a prime example of such re-evaluations (Hilton Reference Hilton1995).
Likewise, there are clear examples where a focus on the functional, computational level gave rise to more detailed predictions than research focused on process level explanations. For example, Oaksford and Chater's (Reference Oaksford and Chater1994) probabilistic approach to the conditional provided a level of specificity previously unknown in the field of logical reasoning by introducing quantitative predictions (Hahn Reference Hahn2009).
Ultimately, complementing this functional level of explanation with descriptively accurate process theory will provide even more accurate prediction by capturing not just those aspects where behaviour approximates the normative standard, but also its systematic deviations. But again, for identifying these underlying processes rational norms can provide powerful tools. This is illustrated by ideal observer analysis which has had tremendous success in the study of perception (e.g., Geisler Reference Geisler1987). Ideal observer models draw on the formal tools of probability and decision theory to specify a model of optimal performance, given the available input for a task. Actual human performance is compared to the performance of this ideal agent. In a process of iterative refinement, human performance and ideal observer are brought into ever closer correspondence by incorporating into the ideal observer details of the human system. This approach provides a tool for the elucidation of mechanism and process, embedded in an overall account that seeks to understand the system as “doing the best it can do” given the available hardware. In so doing, it inherently links behavioural prediction, mechanistic and functional explanation, and it might be viewed as a methodological formalization of the principle of charity. Similar programmes are found under the header of bounded rationality or bounded optimality within cognitive psychology. Here, it has been stressed how rational norms can aid the disambiguation between competing theories and assist in the identification of underlying cognitive universals above and beyond the demand characteristics of experimental tasks (Howes et al. Reference Howes, Lewis and Vera2009).
The fact that most of this work is outside the domain of classic reasoning, judgment, and decision-making research, calls into question Elqayam & Evans' (E&E's) claim that these latter areas would make more rapid progress by abandoning their focus on normative standards.
It also highlights the weakness in E&E's central claims about the “is-ought” fallacy. Spelling out what counts as “sensible” is itself a non-trivial task in that there are a number of normative frameworks – logic, probability theory, decision theory, ecological rationality, and so on. They can differ in the interpretive focus they provide on a given task, because they make salient different aspects of behaviour. However, lawful connections exist between these frameworks; they are not simply “competitors.” Even more importantly, choosing between them in the interpretation of behaviour is not an inference about their normativity. Classical logic, probability theory, and decision theory all have independent, normative justifications (e.g., in the instrumental rationality of “Dutch books”). Appealing to certain aspects of the data in order to claim, as do Oaksford and Chater, that, in standard logical reasoning tasks, participants behave as if considering these tasks to involve probabilistic inference is not a statement about whether probability theory itself is normative, but about whether it provides the standard that renders participants' behaviour “sensible” and “intelligible.”
There is no is-ought fallacy here. The normative theory is being used to provide interpretation and explanation of the behavioural data. In so doing, particular aspects of it are highlighted as relevant. As always, such interpretation may or may not turn out to be correct and the process may even be bi-directional; mismatch between what we consider to be “sensible” and what we observe may lead us to alter our understanding of “sensible” and what factors a behaviour should take into account. However, elucidating “sensible” will always involve considerations above and beyond the immediate data themselves – only then can it be of any use in interpretation, explanation, and prediction. The data themselves are prompts, not normative justification.
Normative theories are essential tools for understanding behaviour; it would be foolish to set them aside.
It is uncontroversial that a full understanding of behaviour involves multiple aspects. Psychology seeks to identify lawful regularities: We seek to understand the “what” of behaviour such that we can predict it. We also seek to understand why these regularities obtain. This involves two distinct kinds of causal explanation: (1) an understanding of the mechanisms/processes that give rise to the behaviour, and (2) functional explanation, that is, an understanding of why this behaviour and not others. Finally, psychology considers how such understanding allows performance to be improved in practice.
For all these questions, normative standards, that is, characterizations of how something “ought to be,” seem indispensable. Trivially, performance cannot be improved without knowing what would count as “better.” Likewise, functional explanation will make reference to the fact that a behaviour maximizes some “desirable” criterion – where both “desirability” and “maximization” typically invoke normative considerations. Even the basic task of identifying behavioural regularities cannot afford blindness to normative considerations.
Rational standards provide essential interpretative tools. Human behaviour typically affords many different interpretations. In daily life we resolve this ambiguity with the help of “the principle of charity” (e.g., Govier Reference Govier1987). Given multiple possible interpretations of an utterance, we pick, as a default, the interpretation that renders it most sensible. Of course, this interpretation may prove wrong, and further evidence may lead us to abandon it; that there are default orderings over possible interpretations, however, simplifies massively the task of understanding.
Moreover this has consequences not only for the interpretation of a single utterance or act, but also for prediction. How we interpret what someone has said or done has direct implications for what we expect that person to do in other, similar situations. Furthermore, even without specific knowledge of an individual we can often make reasonably accurate predictions just on the basis of what would be “sensible” (though again, there is no guarantee that these predictions will be correct).
The principle of charity has both informal and more formal manifestations within psychological research. Informally, wherever we observe something surprising, we should ask ourselves whether there is an interpretation of participants' behaviour that renders it sensible and hence predictable. This may lead to re-evaluation of how our task is understood by the participants, to a different interpretation of what they are doing, and to quite different predictions and observations in response to changes in task parameters. Demonstrations of the (unintended) influence of the pragmatics of the experimental situation in giving rise to seeming “errors” and “biases” are a prime example of such re-evaluations (Hilton Reference Hilton1995).
Likewise, there are clear examples where a focus on the functional, computational level gave rise to more detailed predictions than research focused on process level explanations. For example, Oaksford and Chater's (Reference Oaksford and Chater1994) probabilistic approach to the conditional provided a level of specificity previously unknown in the field of logical reasoning by introducing quantitative predictions (Hahn Reference Hahn2009).
Ultimately, complementing this functional level of explanation with descriptively accurate process theory will provide even more accurate prediction by capturing not just those aspects where behaviour approximates the normative standard, but also its systematic deviations. But again, for identifying these underlying processes rational norms can provide powerful tools. This is illustrated by ideal observer analysis which has had tremendous success in the study of perception (e.g., Geisler Reference Geisler1987). Ideal observer models draw on the formal tools of probability and decision theory to specify a model of optimal performance, given the available input for a task. Actual human performance is compared to the performance of this ideal agent. In a process of iterative refinement, human performance and ideal observer are brought into ever closer correspondence by incorporating into the ideal observer details of the human system. This approach provides a tool for the elucidation of mechanism and process, embedded in an overall account that seeks to understand the system as “doing the best it can do” given the available hardware. In so doing, it inherently links behavioural prediction, mechanistic and functional explanation, and it might be viewed as a methodological formalization of the principle of charity. Similar programmes are found under the header of bounded rationality or bounded optimality within cognitive psychology. Here, it has been stressed how rational norms can aid the disambiguation between competing theories and assist in the identification of underlying cognitive universals above and beyond the demand characteristics of experimental tasks (Howes et al. Reference Howes, Lewis and Vera2009).
The fact that most of this work is outside the domain of classic reasoning, judgment, and decision-making research, calls into question Elqayam & Evans' (E&E's) claim that these latter areas would make more rapid progress by abandoning their focus on normative standards.
It also highlights the weakness in E&E's central claims about the “is-ought” fallacy. Spelling out what counts as “sensible” is itself a non-trivial task in that there are a number of normative frameworks – logic, probability theory, decision theory, ecological rationality, and so on. They can differ in the interpretive focus they provide on a given task, because they make salient different aspects of behaviour. However, lawful connections exist between these frameworks; they are not simply “competitors.” Even more importantly, choosing between them in the interpretation of behaviour is not an inference about their normativity. Classical logic, probability theory, and decision theory all have independent, normative justifications (e.g., in the instrumental rationality of “Dutch books”). Appealing to certain aspects of the data in order to claim, as do Oaksford and Chater, that, in standard logical reasoning tasks, participants behave as if considering these tasks to involve probabilistic inference is not a statement about whether probability theory itself is normative, but about whether it provides the standard that renders participants' behaviour “sensible” and “intelligible.”
There is no is-ought fallacy here. The normative theory is being used to provide interpretation and explanation of the behavioural data. In so doing, particular aspects of it are highlighted as relevant. As always, such interpretation may or may not turn out to be correct and the process may even be bi-directional; mismatch between what we consider to be “sensible” and what we observe may lead us to alter our understanding of “sensible” and what factors a behaviour should take into account. However, elucidating “sensible” will always involve considerations above and beyond the immediate data themselves – only then can it be of any use in interpretation, explanation, and prediction. The data themselves are prompts, not normative justification.
Normative theories are essential tools for understanding behaviour; it would be foolish to set them aside.