Bermúdez argues (see especially sect. 2.4) that the consensus view on classic framing effects a la Asian disease is that they are irrational, and that the irrationality view would be consensual among psychologists of reasoning, and in the cognitive sciences. I disagree: Rather, there is evidence that classic risky choice framing effects (i.e., risk-seeking with losses and risk-aversion with gains) can be construed as being rational. Notably, recent research has shown that risky choice framing effects can result from various cognitive processes, all being entirely intelligible and rational. Central is the idea that, rather than passively taking in information, people actively select and process information, taking also background knowledge into consideration. For instance, the Asian disease task includes a background of fighting diseases, rendering an option described as “200 people are saved” unlikely to be interpreted as “exactly 200 people are saved.” If, however, the bilateral “exactly” reading does not apply, but a lower-bound “at least”-reading does, extensionality of differently described option breaks down (see Mandel, Reference Mandel2014).

In addition, the options in most tasks modelled after the Asian disease task are incompletely described: It is only stated that “200 people are saved,” but nothing is said about the remaining people. Research shows that symmetric description (i.e., explicitly specifying all outcomes) makes the framing effect disappear (Kühberger, Reference Kühberger1995), indicating that there is more to the different descriptions than simply the framing. Finally, information equivalence of framed options implies that the choice of frame is arbitrary. This generally is not true, since people tend to choose descriptions, and frames, non-randomly. Specifically, they tend to describe objects by attributes that exceed, rather than fall short, of some reference point. In addition, frame selection may leak information about a speaker's attitude (Sher & McKenzie, Reference Sher and McKenzie2006).

All this goes to show that what initially was considered a most impressive demonstration of irrationality in many risky choice framing tasks follows from the naive, and often unsubstantiated, intuition that arithmetics (i.e., 200 of 600 saved = 400 of 600 dying) is all that counts in framing tasks. Pertinent research has shown this to be shortsighted: At semantic and pragmatic levels number expressions can have importantly different meanings. Thus, rather than showing irrational choices, the findings show that the (untested) assumption of equivalence stands on shaky grounds. If extensionality cannot be assured, the extensionality principle can hardly be violated.

Another curiosity in framing research should also be pointed out. In the Asian disease task, and all tasks modelled accordingly, one option (200 people saved; or 400 people die) is framed as “the sure option,” and the second option (600 people saved with p = 1/3, or no one saved with p = 2/3) is said to be “the risky option.” Bermúdez argues that “frames prime responses,” such that the gain frame primes risk-aversion (i.e., the choice of the sure option) and the loss frame primes risk-seeking (i.e., the choice of the risky option). Why is the option “200 people are saved” (or “400 people will die”) a sure option? Consider the respective situation: We have a population of 600 people contracting the disease. Two outcomes are possible: save or die; repeated 600 times. This is true for the risky option, but also for the sure option: any individual can be saved or not. Each option thus consists of 600 risky events. In other words, both options are risky, and they are identical in risk. They are only framed as sure or risky. The framing is done by using the word “probability” for the so-called risky option, while avoiding it (or using the notion “for sure”) for describing the so-called sure option. The impression of a sure option only follows from hiding the risk part. Taken together, in tasks following the Asian disease structure, a distinction of options in terms of risk does not make sense. If the task is modelled as a gamble, things are different. Imagine you have to choose among (A) winning €200 for sure, or (B) winning €600 with p = 1/3 or nothing with p = 2/3. Here, option A has only one possible outcome (€200), while option B has two (€600, or €0). Thus, A is sure, and B is risky.

The gambling situation is different from the disease situation in many respects. Most notably, the semantic and pragmatic aspects are weaker, or even nonexistent, with gambles. Gambles are critical for testing the irrationality argument, since with gambles extensionality can be best preserved. Using between-subjects designs, robust evidence exists for risk-aversion with gains and risk-seeking with losses also for gambles (e.g., Kühberger, Schulte-Mecklenbeck, & Perner, Reference Kühberger, Schulte-Mecklenbeck and Perner2002). However, little is known whether preferences also reflect in within-subjects designs. Note, however, that gambles are a very specific domain (unless you are a decision researcher). Rational cognition may be more adapted to general, rather than to specific situations. Finding irrationality in gambles may be too weak an argument for the verdict that human choices are irrational in general.