1 Less is more—more or less

Discovering that a higher degree of familiarity with a domain (e.g., the cities of one’s own country vs. those of another) can lead to poorer inferential accuracy is an important insight. It is no wonder the aptly titled “More or Less” BBC radio show wished to demonstrate the effect to its listeners. It is robust, exceedingly simple to demonstrate (as I and I am sure many of us do regularly in undergraduate lectures) and—at least initially—counterintuitive. The conditions under which such effects hold have been discussed at length in the pages of the special issue (e.g., Beaman, Smith, Frosch, & McCloy, Reference Beaman, Smith, Frosch and McCloy2010; Hoffrage, Reference Hoffrage2011; Katsikopoulus, Reference Katsikopoulos2010; Smithson, Reference Smithson2010). However, a counter-intuitive effect, though it might have led to the “discovery” of the recognition heuristic (RH, Hoffrage, Reference Hoffrage2011), is not sufficient evidence for the claim that people use recognition information in the way that the RH dictates.

Recognition and evaluation

G&G (Gigerenzer & Goldstein, Reference Gigerenzer and Goldstein2011) describe recognition and evaluation as two processes guiding the adaptive selection of the RH (i.e., recognition: “Do I recognize one object but not the other?” and evaluation: “If so is it reasonable to rely on the recognition heuristic in this situation?”). This formulation fits neatly with the interpretation that I and David Shanks proposed in 2004: “such an argument suggests that it is not pure recognition that determines an inference but recognition plus an appropriate reason for knowing why a particular object is recognized—or, at least, a correctly interpreted feeling of familiarity. It is not that an object is recognized and chosen without justification, but that the decision maker has a reasonable idea of why he or she recognizes the object and makes an inference on the basis of this secondary knowledge.” (Newell & Shanks, Reference Newell and Shanks2004, p. 933, emphasis added).

As we argued then, such a claim is far less newsworthy than the notion of a one-reason strategy for which further (i.e., secondary) knowledge is inconsequential (e.g., Goldstein & Gigerenzer, Reference Goldstein, Gigerenzer, Gigerenzer and Todd1999). The recognition and evaluation process sounds more like the description of a person searching (memory) for information and making a trade-off (i.e., evaluating) between relying (solely) on recognition or considering how further information about the decision environment and the object (situation—in G&G’s words) might affect a judgment (e.g., over-ride recognition, support it, etc.). Perhaps my confusion about these two seemingly different interpretations arises from a misunderstanding of the differences between cues, information and knowledge.

2 Cues, information and knowledge

G&G clarify that when in the past they said that further knowledge is inconsequential once a binary recognition judgment has been made (e.g., Goldstein & Gigerenzer, Reference Goldstein, Gigerenzer, Gigerenzer and Todd1999, p.56)—what they meant was not knowledge, or information but cue-values. While I can see that there is a distinction between direct knowledge of the criterion (e.g., knowing – for a fact - the population of a city), and a cue value that might be indicative of a criterion (e.g., knowing a city has an airport) – to say that the former is knowledge/information but that the latter is not is to engage in obfuscation. Little wonder so many researchers took G&G at their (original) word and concluded that there was little evidence for a non-compensatory RH of the kind originally implied (e.g., Bröder & Eichler, Reference Bröder and Eichler2006; Glöckner & Bröder, Reference Glöckner and Bröder2011; Hilbig & Pohl, Reference Hilbig and Pohl2009; Newell & Fernandez, Reference Newell and Fernandez2006; Newell & Shanks, Reference Newell and Shanks2004; Oeusoonthornwattana & Shanks, Reference Oeusoonthornwattana and Shanks2010; Richter & Spath, Reference Richter and Späth2006).

This clarification begs the question of when a cue becomes information and information becomes knowledge. If I search my memory to try to figure out if Milwaukee is bigger than Detroit there are lots of “things” I could use (beer companies, car companies, music labels, TV sitcoms)—are these cues, information or knowledge? G&G note in their open questions that choices between pairs of items in which only one object is recognised and some additional knowledge is known are better predicted by RH than when the object is merely recognised (Pohl, Reference Pohl2006—see also Newell & Fernandez Reference Newell and Fernandez2006 Experiment 1 for the same result). Unless this extra knowledge is somehow defined as not a cue-value then it seems very difficult to reconcile such results with the notion of sole reliance on recognition (though see Marewski, Gaissmaier, Schooler, Goldstein, & Gigerenzer, Reference Marewski, Gaissmaier, Schooler, Goldstein and Gigerenzer2010 for an attempt to do so.)

Perhaps more importantly—where do these cues and our knowledge of the direction of the cue-criterion correlation come from? G&G remind us that knowledge of such things (i.e., the recognition validity) is a prerequisite for the RH to operate, but how would a decision-maker know—how does she answer the “evaluation” question? How does she know when the environment is right—or, to put it in G&G’s terms the ecological rationality criterion is satisfied?

3 Ecological rationality and adaptive decisions

A cornerstone of the fast-and-frugal approach is the idea that heuristics are selected on the basis of their ecological rationality in particular environments. Despite the fundamental importance of this idea, ecological rationality is often defined in reference to itself. For example, G&G state “ecological rationality asks in which environment a given strategy (heuristic or otherwise) excels and in which it fails” (p. 101). Ecological rationality is therefore defined as a property of a heuristic that can only be determined from the perspective of the individual making the decision once a heuristic has been tried in a given environment. If one needs to try it out first then how can ecological rationality drive the initial selection? Without question people can learn to select strategies (or accumulate more or less evidence, if you prefer) in environments that favour one type of strategy over the other (e.g., Newell & Lee, Reference Newell, Lee, Taatgen, van Rijn, Nerbonne and Schomaker2009; Rieskamp & Otto, Reference Rieskamp and Otto2006) but the demonstration of this ability has nothing to do with ecological rationality per se. It simply shows that people can learn to adapt.

The concept of ecological rationality also sits uneasily with situations in which heuristics lead to poorer decisions. G&G discuss at some length the peanut butter tasting experiments of Hoyer and Brown (Reference Hoyer and Brown1990) but fail to point out that the behaviour observed therein is a clear example of recognition being maladaptive. Choosing the inferior-tasting peanut butter because it is in the better-known-brand jar would not seem like a good choice. And yet the emphasis of the RH is that simple reliance on recognition can get you a long way. Why does the decision-maker faced with inferior peanut butter not realise that use of the RH is ecologically irrational? Perhaps the ease with which the brand name is processed leads to a biased judgment (Tversky & Kahneman, Reference Tversky and Kahneman1973).

These are important and interesting questions, and some recent work points towards possible ways of answering them (e.g., Katsikopolous, Schooler, & Hertwig, Reference Katsikopoulos, Schooler and Hertwig2010). Moreover, by simply introducing these concepts important debates have been sparked (e.g., Dougherty, Franco-Watkins, & Thomas, Reference Dougherty, Franco-Watkins and Thomas2008); but to conclude that “ecological rationality guides heuristic selection” is to go beyond the available evidence (see also Hilbig, Reference Hilbig2010).

4 Individual differences and model fits

As noted by G&G analysis of mean levels of accordance with RH often belies important individual differences. For example, in the high validity condition of Newell and Fernandez (Reference Newell and Fernandez2006) Experiment 1, 12 participants chose the recognised object on less than 50% of occasions (with 3 never choosing it) and 11 chose it on more than 50% (with 5 always choosing it) giving a mean RH accordance rate of 55.5%. Can simple heuristics bounded by ecological rationality deal with such individual difference? G&G are optimistic that they can. An alternative approach is to test models (competitively against heuristics) that adopt a different set of assumptions about the distribution of behaviours in a given environment and see how they fare (e.g., Newell & Lee, in press; Lee & Cummins, Reference Lee and Cummins2004). I am also optimistic that new methods can shed light on how and why such individual differences might arise (e.g., Lee & Newell, Reference Lee and Newell2011; see also the excellent suggestions in Tomlinson, Marewski, & Dougherty, Reference Tomlinson, Marewski and Dougherty2011).