In their target article, Elqayam & Evans (E&E) argue against prescriptive normativism and in favor of descriptivism. This commentary questions the assumption, implicit in E&E's article, that in the study of human reasoning and decision making there is a fundamental choice to be made between a normativist and a descriptivist approach. I argue, instead, that there is a role for both approaches. In particular, I argue that some research questions fit most naturally in a normativist framework, while others are better dealt with in a descriptivist framework. E&E's discussion demonstrates the validity of the latter framework for specific types of studies, but their discussion seems to overlook a number of worthwhile research questions that clearly call for a normativist approach. I would like to draw attention to two such questions.
The first question is suggested by a variety of related arguments that philosophers and mathematicians have proposed in defense of the postulates of Bayesian epistemology, that is, the probability axioms as static norms of rationality, and Bayes' rule as a dynamic norm of rationality (e.g., Joyce Reference Joyce1998; Rosenkrantz Reference Rosenkrantz1992). In sharp contrast with the more familiar Dutch book arguments, which purport to demonstrate that obeying the Bayesian norms has certain practical advantages, the newer arguments are explicitly meant to show that obedience to those norms is epistemically mandated. For instance, the newer arguments for Bayes' rule aim to show that by following this rule, in the long run one will be more accurate in one's assignment of probabilities than one could have been by following any other rule. (Accuracy here is measured by means of a so-called scoring rule.) This claim is then coupled with a (purportedly) a priori argument to the effect that long-run accuracy is what, qua rational epistemic agents, we aim at – which yields the desired epistemic defense of Bayes' rule. Even if we leave open the question of whether long-run accuracy has been rightly identified as our epistemic goal, the aforementioned arguments, if sound, entail that if an agent commits himself or herself to the goal of long-run accuracy, then he or she should update her subjective probabilities by dint of Bayes' rule. Does this hold true of real people? If not, what might explain the discrepancy between the norm and the observed behavior? These are research questions involving a normative “ought,” and nothing E&E say suggests that these questions are not worth investigating. Indeed, a discussion of the a priori arguments of the kind referred to here, and the implications they might have for psychological research, is altogether lacking from their article.
E&E rightly point out that when there are rival normative accounts of some aspect of human reasoning or decision making, experiments will not help us in deciding which of those (if any) is correct. That might suggest that empirical evidence has no normative value indeed. This suggestion appears dubitable, however, when we consider the situation in which we have simply no normative account of some aspect of our reasoning, while at the same time it does seem obvious that that aspect is subject to norms of rationality. In such a situation, empirical evidence might help us find those norms. That would go at least some way toward deriving an ought from an is.
To illustrate the point, consider the issue of updating on conditionals. In a standard Bayesian picture, we ought to update on incoming information by conditionalizing on it (i.e., by using Bayes' rule). Back in 1980, Brian Skyrms said that “we have no clear conception of what it might be to conditionalize on a conditional” (Skyrms Reference Skyrms1980, p. 169). Since then, only very little progress has been made on this matter. That is unfortunate, given that it is uncontentious that updating on conditionals is governed by some norms. Douven and Dietz (Reference Douven and Dietz2011) and Douven and Romeijn (in press) describe hypothetical cases in which people obtain conditional information. (They learn, for instance, that if it continues to rain, then a specific football match will be cancelled.) These works then point out that we have clear intuitions about these cases as to whether, given the context in which the conditional information is obtained, an update on that information should lead the person receiving the information to raise or, rather, to lower the probability he or she assigns to the conditional's antecedent, or whether that probability should remain unaltered. However, we are currently clueless as to what the underlying norms might be. Although the past decade has witnessed a surge of interest among psychologists for studying conditionals in a probabilistic setting, so far experimental work on updating on conditionals is still missing. The lacuna is noteworthy in itself, but for present purposes the important point is that getting some understanding of how people actually update on conditionals (or perhaps how cognitively superior people update on conditionals) might help to suggest, at least in outline form, the norms that philosophers assume to exist but to this day have been unable to pinpoint. Here, too, I cannot find anything in E&E's article that indicates that this approach is misguided.
E&E might seem to have provided a very general reason against normativism, to wit, that it has given rise to a number of research biases. I am not sure we must agree that all that they identify as research bias should be acknowledged as such, but some definitely are. Even so, pleading for normativism is not a call for thinking lightly of such problems. E&E certainly have not shown that if one embraces normativism, then one is bound to commit those errors. To the contrary, it is reasonable to think that once a bias has been identified as such, it should be possible to steer clear of it. Thus, these research biases give good reason for double- or even triple-checking the design of experiments conducted in a normativist framework, but they do not give good reason for abandoning the normativist approach wholesale.
You have
Access
In their target article, Elqayam & Evans (E&E) argue against prescriptive normativism and in favor of descriptivism. This commentary questions the assumption, implicit in E&E's article, that in the study of human reasoning and decision making there is a fundamental choice to be made between a normativist and a descriptivist approach. I argue, instead, that there is a role for both approaches. In particular, I argue that some research questions fit most naturally in a normativist framework, while others are better dealt with in a descriptivist framework. E&E's discussion demonstrates the validity of the latter framework for specific types of studies, but their discussion seems to overlook a number of worthwhile research questions that clearly call for a normativist approach. I would like to draw attention to two such questions.
The first question is suggested by a variety of related arguments that philosophers and mathematicians have proposed in defense of the postulates of Bayesian epistemology, that is, the probability axioms as static norms of rationality, and Bayes' rule as a dynamic norm of rationality (e.g., Joyce Reference Joyce1998; Rosenkrantz Reference Rosenkrantz1992). In sharp contrast with the more familiar Dutch book arguments, which purport to demonstrate that obeying the Bayesian norms has certain practical advantages, the newer arguments are explicitly meant to show that obedience to those norms is epistemically mandated. For instance, the newer arguments for Bayes' rule aim to show that by following this rule, in the long run one will be more accurate in one's assignment of probabilities than one could have been by following any other rule. (Accuracy here is measured by means of a so-called scoring rule.) This claim is then coupled with a (purportedly) a priori argument to the effect that long-run accuracy is what, qua rational epistemic agents, we aim at – which yields the desired epistemic defense of Bayes' rule. Even if we leave open the question of whether long-run accuracy has been rightly identified as our epistemic goal, the aforementioned arguments, if sound, entail that if an agent commits himself or herself to the goal of long-run accuracy, then he or she should update her subjective probabilities by dint of Bayes' rule. Does this hold true of real people? If not, what might explain the discrepancy between the norm and the observed behavior? These are research questions involving a normative “ought,” and nothing E&E say suggests that these questions are not worth investigating. Indeed, a discussion of the a priori arguments of the kind referred to here, and the implications they might have for psychological research, is altogether lacking from their article.
E&E rightly point out that when there are rival normative accounts of some aspect of human reasoning or decision making, experiments will not help us in deciding which of those (if any) is correct. That might suggest that empirical evidence has no normative value indeed. This suggestion appears dubitable, however, when we consider the situation in which we have simply no normative account of some aspect of our reasoning, while at the same time it does seem obvious that that aspect is subject to norms of rationality. In such a situation, empirical evidence might help us find those norms. That would go at least some way toward deriving an ought from an is.
To illustrate the point, consider the issue of updating on conditionals. In a standard Bayesian picture, we ought to update on incoming information by conditionalizing on it (i.e., by using Bayes' rule). Back in 1980, Brian Skyrms said that “we have no clear conception of what it might be to conditionalize on a conditional” (Skyrms Reference Skyrms1980, p. 169). Since then, only very little progress has been made on this matter. That is unfortunate, given that it is uncontentious that updating on conditionals is governed by some norms. Douven and Dietz (Reference Douven and Dietz2011) and Douven and Romeijn (in press) describe hypothetical cases in which people obtain conditional information. (They learn, for instance, that if it continues to rain, then a specific football match will be cancelled.) These works then point out that we have clear intuitions about these cases as to whether, given the context in which the conditional information is obtained, an update on that information should lead the person receiving the information to raise or, rather, to lower the probability he or she assigns to the conditional's antecedent, or whether that probability should remain unaltered. However, we are currently clueless as to what the underlying norms might be. Although the past decade has witnessed a surge of interest among psychologists for studying conditionals in a probabilistic setting, so far experimental work on updating on conditionals is still missing. The lacuna is noteworthy in itself, but for present purposes the important point is that getting some understanding of how people actually update on conditionals (or perhaps how cognitively superior people update on conditionals) might help to suggest, at least in outline form, the norms that philosophers assume to exist but to this day have been unable to pinpoint. Here, too, I cannot find anything in E&E's article that indicates that this approach is misguided.
E&E might seem to have provided a very general reason against normativism, to wit, that it has given rise to a number of research biases. I am not sure we must agree that all that they identify as research bias should be acknowledged as such, but some definitely are. Even so, pleading for normativism is not a call for thinking lightly of such problems. E&E certainly have not shown that if one embraces normativism, then one is bound to commit those errors. To the contrary, it is reasonable to think that once a bias has been identified as such, it should be possible to steer clear of it. Thus, these research biases give good reason for double- or even triple-checking the design of experiments conducted in a normativist framework, but they do not give good reason for abandoning the normativist approach wholesale.