Framing effects are usually held to be irrational because they violate the principle Bermúdez calls extensionality: Preferences should be stable across different descriptions of the same outcomes. Bermúdez argues that extensionality is false: Preferences can be quasi-cyclical. An agent has quasi-cyclical preferences when she prefers A to B under one description while preferring B to A under another description, while knowing that the descriptions are different ways of framing the same outcome but believing that one way of framing the options is preferable to the other. Quasi-cyclical preferences are stable and therefore avoid many of the problems that might arise from cyclical preferences (such as failures of transitivity). More importantly, rational agents may reason their way to stable preferences by coming to see that one frame is normatively preferable to another. Some preferences adopted via framing are therefore rationally justifiable and we may rationally employ framing in decision-making, deliberation, and discussion.

Bermúdez's aim is to show that framing effects can be rational, especially in the “complicated” conditions that arise “outside the laboratory.” He does not attempt to vindicate the rationality of framing inside the laboratory, and in fact his strategy is unlikely to show that participants in classic experiments on framing choose rationally. Nevertheless, I suggest, there's a strong case for thinking that they do choose rationally. Frames provide agents with genuine evidence, and participants respond rationally to this evidence.

The extensionality principle is false when descriptions are coupled with recommendations. In most circumstances, agents with no prior knowledge of or preference between options who are told that A is better than B acquire a genuine reason to prefer A to B. Such an agent might rationally prefer A to B even though they also know they would have preferred B to A had their informant given them a different recommendation. If framing of options is testimonial, its rationality is vindicated.

There is evidence, both experimental (McKenzie, Liersch, & Finkelstein, Reference McKenzie, Liersch and Finkelstein2006) and from modelling (Carlin, Gervais, & Manso, Reference Carlin, Gervais and Manso2013) that options selected as the default are understood as authoritatively recommended. This is a special case of a broader phenomenon: To make an option salient – by making it the default, by placing it prominently, and so on – is in many contexts to recommend it or to implicate its relevance or importance. Framing is one more way of making options salient to individuals. In fact, there is evidence that framing is understood as conveying recommendations and that ordinary people use framing to this end. For example, someone might say that a researcher has had 40% of her journal submissions accepted (rather than conveying the same information by mentioning the proportion rejected) in order to recommend her to a recipient (Sher & McKenzie, Reference Sher and McKenzie2006). Further, the testimony will be understood as a recommendation, at least functionally – the recipient would form a more favourable impression of the person when her acceptance rate is mentioned, rather than her rejection rate (Fisher, Reference Fisher2020, Reference Fisher2022; McKenzie et al., Reference McKenzie, Liersch and Finkelstein2006).

A cognitive process might be assessed as rational against an ecological or a more direct standard. Ecological rationality is assessed by how well designed a process is to achieve agents' goals (Gigerenzer & Selten, Reference Gigerenzer and Selten2002; Todd & Gigerenzer, Reference Todd and Gigerenzer2012). A process is directly rational, as I define it here, if it transforms inputs in a way that reflects their actual informational content, such that it causes Bayesian belief update. Gigerenzer's influential defence of ecological rationality turns on the fact that the two kinds of rationality can dissociate: Human beings might systematically depart from Bayesian rationality yet do so in a way that allows us to achieve our goals, given the kinds of challenges we typically face.

Defences of human rationality that emphasise its ecological appropriateness face the problem that the environment in which we make decisions may depart significantly from the environment for which our decision-making processes are adapted, either because we deliberate and act in environments that depart significantly from the environment of evolutionary adaptiveness, or because hostile agents take advantage of our vulnerabilities to engineer cognitive traps for us (Evans & Stanovich, Reference Evans and Stanovich2013; Stanovich, Reference Stanovich2018). If our responses to framing are rational in the manner suggested, however, they are directly and not merely ecologically rational. Framing offers participants genuine evidence – technically, higher-order evidence; evidence about evidence – and it is rational for agents to respond to it accordingly (Fisher, Reference Fisher2020, Reference Fisher2022; Levy, Reference Levy2019, Reference Levy2021).

It is somewhat controversial under what conditions we ought to trust testimony; in particular, whether we need evidence of trustworthiness before we are justified in taking someone's word for something (Lackey & Sosa, Reference Lackey and Sosa2006). There is no doubt, however, that no matter how demanding the conditions for justified trust might be, they are satisfied in the classic experiments on framing effects. The stakes for participants are low, because nothing turns for them on their response. They have nothing else to base their decision on. And the testimony comes from experimenters, who participants have some reason to regard as more knowledgeable on the topic in question than they are (other things equal, the very act of offering testimony implicates knowledge). If framing functions as testimony, we ought to see participants being sensitive to the same features they are sensitive to when it comes to explicit testimony. For example, they should prefer frames associated with people who are competent or benevolent informants, or those associated with numerically more rather than fewer informants (Harris, Reference Harris2012; Sperber et al., Reference Sperber, Clément, Heintz, Mascaro, Mercier, Origgi and Wilson2010). In other work, I've suggested that inference to the best explanation supports understanding framing effects as functioning via the provision of testimony. Manipulation of properties like the apparent trustworthiness or competence of those seen to frame the information might provide a way to experimentally test the hypothesis.