1. Logicism and normativism and their discontents

In everyday life, we are thoroughly accustomed to normative dictates wherever we turn. When we play chess, we conform to the rules of the game; when we drive, we try to heed traffic laws and know we would be sanctioned if we disobeyed them. In some countries, language is normatively regulated – L'Académie française is a prominent example. Voluntary or governmental bodies, such as the Advertising Standards Authority in the United Kingdom, impose normative constraints on advertisements. And occasionally, normative issues find their way into scientific theories, as well.

The research literature in higher mental processing – reasoning, judgement, and decision making – is rife with normative considerations. In the study of human reasoning, these considerations have traditionally taken the form of logicism – the idea that thinking (1) reflects some internalized form of extensional, classical logic and (2) should be measured against classical logic as a normative system (Evans Reference Evans2002) and ought in some clearly evaluative sense to conform with it (see Appendix for terminological clarifications). We will dub these two distinct meanings empirical versus prescriptive logicism, respectively. Inhelder and Piaget hold the dubious title of prototype logicists: In their monograph on the formal operations stage of cognitive development (Inhelder & Piaget Reference Inhelder and Piaget1958), they argued that normal adolescents and ultimately adults attain the ability to reason according to the rules of formal classical logic.

Half a century on, logicism in both its forms is not nearly as dominant in reasoning research as it used to be. Peter Wason's seminal work in the 1960s and '70s was motivated by an attack on empirical logicism in a period dominated by Piagetian theory. In support of this, he devised several ingenious reasoning problems, including the 2-4-6 task (Wason Reference Wason1960), the much researched selection task (e.g., Wason Reference Wason and Foss1966), and the THOG problem (Wason & Brooks Reference Wason and Brooks1979). However, Wason never seemed to doubt that human reasoning should conform to classical logic (i.e., prescriptive logicism), so that his interpretation of the many logical errors observed on his tasks was that people are illogical and therefore irrational (see Evans [2002] for a detailed account). Following the critique of Cohen (Reference Cohen1981), however, later researchers began to question whether logic was the right normative system against which to judge the rationality of people's reasoning, so that prescriptive logicism also came under attack. Some researchers have proposed that we should adopt alternative normative systems such as those based on information, probability, or decision theory (e.g., Oaksford & Chater Reference Oaksford and Chater1991; Reference Oaksford and Chater1998a; Reference Oaksford and Chater2007), while others proposed that at least some forms of rationality need not necessarily require a normative system at all (e.g., Evans Reference Evans, Manktelow and Over1993; Reference Evans2002; Evans & Over Reference Evans and Over1996; Gigerenzer & Selten Reference Gigerenzer and Selten2001). By this position, organisms are rational if they act in such a manner as to achieve personal goals, and such rationality need not involve any normative rule following.

Our concern here is not with logicism per se; in our view, logicism is but a special case of a more general attitude. Consider the empirical and prescriptive tenets of logicism. We could easily substitute for the word logic a name of another normative system, such as Bayesian probability:

• Empirical logicism: Thinking reflects logic.

• Prescriptive logicism: Rational thinking should be measured against logic as a normative system.

• Empirical Bayesianism: Thinking reflects Bayesian probability.

• Prescriptive Bayesianism: Rational thinking should be measured against Bayesian probability as a normative system.

Our own take on this is that both logicism and Bayesianism are special cases of the same paradigm. We call this paradigm normativism; analogous to what Stein (Reference Stein1996) calls “the standard picture,” it can be formulated in terms closely related to the ones we have already examined. These are:

• Empirical normativism: Thinking reflects S.

• Prescriptive normativism: Rational thinking should be measured against S as a normative system, and ought to conform to it.

Here S is a formal normative system such as logic (classical or otherwise), Bayesian probability, or decision theory. Note that a formal theory is not necessarily a normative theory unless taken as such by a specific normativist account. For example, extensional logic can be conceived as a useful computational tool rather than a normative standard for human reasoning. The notable exception is Subjective Expected Utility (SEU), which was developed with a normative goal in the first place (Savage Reference Savage1954, p. 19; von Neumann & Morgenstern Reference von Neumann and Morgenstern1947, pp. 8–9). Taken in this sense, widely diverse research programmes can be said to be normativist. For example, much of the judgement and decision making (JDM) literature is normativist in the prescriptive (albeit not in the empirical) sense, with SEU playing the role of the normative system. Even the most famous (and Nobel Prize–winning) descriptive theory of risky decision making – the prospect theory of Kahneman and Tversky (Reference Kahneman and Tversky1979) – was framed as a demonstration that the standard normative account provided by decision and probability theory failed to accurately describe human economic behaviour.

The prescriptive and empirical tenets of normativism can be considered as vectors defining a two-dimensional space, which makes normativism (and its subordinate paradigms, such as logicism and Bayesianism) a matter of varying degrees. Figure 1 maps out the normative space defined by empirical and prescriptive normativism, respectively, on which we have placed a number of leading authors for illustrative purposes. We realize that some readers would wish to debate the exact coordinates assigned, but that is not important for our purposes here. Note that the right-hand side – the high prescriptive side – of Figure 1 is much more crowded. This is hardly surprising. Historically, positions of prescriptive normativism tend to survive longer than positions of empirical normativism, because – for reasons we will explore later – they are much more difficult to eliminate. Two notable examples are, as already indicated, Wason's rejection of empirical logicism while continuing to uphold prescriptive logicism (Evans Reference Evans2002), and the heuristics and biases programme of Tversky and Kahneman (e.g., Kahneman & Tversky Reference Kahneman and Tversky1979). But a consequence of such a line of argument is that one must conclude people to be irrational. That is why those who feel that people should be rational have proposed alternative normative systems (e.g., Cohen Reference Cohen1981).

Note: Emp. Norm. = Empirical normativism: Thinking reflects normative system, S. Pres. Norm. = Prescriptive normativism: Thinking should be measured against S as a normative system and ought to conform to it. (For reasons of space, each research programme is identified in the figure by a single sample reference.)

Figure 1. Two-vector normativist space with sample references.

Note, too, that the upper-left quadrant of the normative space mapped out in Figure 1 is empty, highlighting that there is no coherent way of proposing high empirical normativism with low prescriptive normativism. In other words, the existence of a normative system is a necessary (albeit not sufficient) condition for the empirical facts of satisfying this system.

Empirical normativism can vary, from hardcore positions which consider thought processes to be isomorphic to the normative system (e.g., Inhelder and Piaget's formal operations), to positions which reject the normative system entirely. For example, Gigerenzer (e.g., Gigerenzer et al. Reference Gigerenzer and Todd1999) famously repudiates any form of normative system, arguing that heuristic rules of thumb outperform normative computations. In between are positions which might be termed “soft logicism,” and which postulate that some logical principles, such as non-contradiction, might underlie some thinking, but only to a limited extent (e.g., Over Reference Over and Roberts2007). Prescriptive normativism can vary according to factors such as the a priori status of the normative system, the position famously advocated by the philosophers Jonathan Cohen (Reference Cohen1981) and Daniel Dennett (Reference Dennett1987). Psychologists tend more to regard selecting the appropriate normative system as an empirical issue, a view shared by authors leading such diverse research programmes as Oaksford and Chater's rational analysis, which focuses on modal responses as a source for normative evaluations (Oaksford & Chater Reference Oaksford and Chater1998a; Reference Oaksford and Chater2007), and the earlier phase of Stanovich's individual differences programme (Stanovich Reference Stanovich1999), which focused on the normatively superior performance of cognitively able participants. We discuss the problems with this approach in section 4, and examine these research programmes in further detail in section 5.

Another factor that may vary is whether conformity to a normative system is considered both necessary and sufficient for rationality, or only necessary – the latter seems to be more common. Positions high on prescriptive normativism are also typically universalist, explicitly or implicitly taking the view that there is just one “right,” all-encompassing normative system, and all the others are “wrong.” However, this can still vary to some extent, with some authors (Oaksford & Chater Reference Oaksford and Chater1998a; Reference Oaksford and Chater2007) advocating one normative system across the board, while others are willing to accept a different normative solution for each specific task (Stanovich Reference Stanovich1999; Reference Stanovich2004; Stanovich & West Reference Stanovich and West2000b). A relativist position of the sort famously advocated by Stich (Reference Stich1990), and to some extent by Baron (Reference Baron2008) and by Stenning and van Lambalgen (Reference Stenning and van Lambalgen2008), would place these programmes lower on prescriptive normativism.

In what follows, our main concern is with the prescriptive tenet of normativism – the belief that people ought to conform to a normative standard – although we have a few things to say about empirical normativism as well. Our thesis is that prescriptive normativism is both problematic and unnecessary in scientific studies of human thinking. We start by outlining what we mean by normativism in reasoning and decision-making research, and how it differs from other forms of rationality. We then examine the possible relations between normative systems and psychological evidence, focusing in particular on the thorny problem of arbitration – that is, cases of conflict between competing normative systems. It will become clear that we have no quarrel with the use of formal theories per se, provided that they are used in a descriptive rather than normative manner.

We shall discuss several problems that normativist thinking has created. First, research programmes have been used to derive normative claims from empirical evidence, relying on the controversial inference from is to ought. We illustrate this with discussion of two leading research programmes in the study of human thinking. Next, we argue that normativism has systematically and harmfully biased the scientific study of thinking, affecting what is studied, how it is studied, and how findings are interpreted. We illustrate these further by discussion of the particular problems that normativist thinking has created in the study of dual processes in higher cognition. Finally, we argue that normativism is unnecessary: A descriptive approach aided by computational analysis can address all the relevant scientific questions, ridding the field of the research biases we discuss. We hence conclude that theories of reasoning, judgement, and decision making would be better off liberated from normative goals and augmented by a descriptivist agenda.

2. Normativism, rationality, and the three senses of “ought”

Normative rationality is not the only type. Here are some of the other concepts of rationality to be found in the literature (also see Nickerson Reference Nickerson2008, for a recent review):

• Instrumental rationality: Behaving in such a way as to achieve one's personal goals.

• Bounded rationality: Behaviour that is adaptive within the constraints of cognitive and biological capacity.

• Ecological rationality: Behaviour that is adapted to the environment in which the organism is operating.

• Evolutionary rationality: Behaviour that has been shaped by evolution and which serves the purpose of the genes.

What seems to set apart normative rationality from other types of rationality is the “oughtness” involved in normativism. Bounded rationality, for example, is not bounded because it ought to be so. Instead, there are just biological limits to how large brains can grow and how much information and how many computational algorithms they can store and execute. There is no “oughtness” to the Darwinian and Skinnerian algorithms that shape ecological rationality either. Adaptation to the environment is an “is,” not an “ought.” Darwinian principles are like Newton's laws of mechanics. Unsupported objects fall to the earth not because they ought to, but because that is what the laws of physics dictate. In the same way, there appears to be no scientific justification for “intelligent design” in evolution. Organisms develop adaptations in accordance with the laws of natural and sexual selection in much the same way as apples fall off trees in compliance with the law of gravity.

A possible argument here is that oughtness is part of what biological function is about; that the idea of function is basically a normative one.Footnote ¹ Often this argument is couched in adaptationist terms; for example, that the heart has a “proper function” (in the terminology suggested by Ruth Millikan; e.g., Millikan Reference Millikan1984; Reference Millikan1995; Reference Millikan, May, Friedman and Clark1996) to pump the blood, which is what it was selected for and therefore what it ought to do (although cf. Fodor Reference Fodor2008; and for response, Dennett Reference Dennett2008). One could even take this argument further and maintain that, by losing oughtness, we lose our ability to talk about function at all, biological, economic, or otherwise.Footnote ² However, our point is that functional “ought” is a different type of ought than the one involved in normativism. Ought, and its close relations must and should, can take at least three different meanings. Consider:

1. 1. Poverty should not exist.

2. 2. You must take the second exit from the roundabout.

3. 3. Ron should be able to catch the 4:25 to Birmingham.

Meanings (1) and (2) are deontic: they express evaluation and obligation; meaning (3), on the other hand, is epistemic, expressing belief or probability. In addition, there is a difference between the deontic function of (1), which is evaluative, and of (2), which is to direct a specific course of action. (Schurz [1997] makes a related distinction between what he terms “normative” and “valuative,” roughly equivalent to our directive and evaluative sense, respectively.) In everyday discourse, the directive and evaluative oughts are often combined, as in “I ought to donate to Oxfam.” However, as (2) demonstrates, these two deontic senses can be distinguished. Directive oughts are generally instrumentalFootnote ³ – for example, we need to take that second exit because it would bring us to our destination.

The “ought” of normativism is evaluative: it resembles (1). In contrast, the “ought” of selection-for and of functional analysis in general is directive, as in (2). There was no normative obligation for nature to select hearts for pumping. Natural selection can be said to contain a directive ought in the sense that function constrains (at least to some extent) evolution; what it does not have is the evaluative ought. With this caveat in place, we have no argument with the rational analysis approach of Anderson (Reference Anderson1990), whose main thesis is that rationality is best understood by formal task analysis. We do not need to take a position in the debate over the role of adaptations in evolution (see, e.g., Gould & Lewontin Reference Gould and Lewontin1979; and then Fodor Reference Fodor2008 and Dennett Reference Dennett2008, respectively) to be wary of normativism. Within bounds, behaviour is likely to be adaptive, so that analysis of the task and its environment may well be helpful in developing a formal account of human behaviour. Insofar as a research programme asks, as Oaksford and Chater's does in their adaptation of Anderson, which of several formal systems is most helpful to achieve one's goals, this falls under our definition of directive ought. It is only when formal systems are regarded as having a priori, unconditional value that the “ought” becomes an evaluative one. This is a very different argument than the one that leads from rational analysis to normative theory, and that, too, is part of Oaksford and Chater's research programme (see sect. 5).

With this distinction in mind, we can now rephrase some of the debate over instrumental rationality. The separation proposed by Evans and Over (Reference Evans and Over1996) between instrumental and normative rationality (i.e., achieving one's goals versus obeying a normative system, respectively) has been contested by various authors. Oaksford and Chater (Reference Oaksford and Chater1998a; Reference Oaksford and Chater2007) objected on the grounds that instrumental rationality needs to be justified, and that this justification should be normative, hence obliterating the boundaries between normative and instrumental rationality. In the terminology proposed here, Oaksford and Chater see the directive ought as inseparable from the evaluative ought, whereas we argue that these two senses are best kept apart.

So it appears to us that normativism is neither necessary nor helpful in discussions of function, adaptation, and ecological and instrumental rationality. Our task as scientists is to observe what people do and to construct and test theories of how they do it. That behaviour is typically well adapted and that people typically achieve their personal goals (with many exceptions, of course), can be described in some terms as “rational,” but without recourse to any normative theory of what people ought to be doing. It is an observation to be accounted for, rather than obligation to be fulfilled.

3. Normative systems and the problem of arbitration

For a normativist position to be coherent, in particular the prescriptive tenet, it has to have a selective notion of norm: There is a “right” or “appropriate” normative system for a paradigm (or even all across the board), and there are “wrong” ones. Nozick (Reference Nozick1993), for example, argues for “nomic universals” – scientific law–like statements, suggesting that norms cannot be particular. In some areas of cognition, deciding on the appropriate norm – and the closely related notion of error – does not seem to pose a practical problem. In most memory paradigms, for example, an error is when one falsely identifies a new stimulus as previously presented, or fails to identify an old stimulus – a practice that goes back to Ebbinghaus and the earliest days of experimental psychology. For psychologists, the problem becomes acute when one tries to adopt this sort of “signal detection” paradigm to reasoning and decision-making research, and this is where consensus on the normative system conspicuously fails. But without a clear-cut norm, normativism becomes far shakier.

Normativism thus faces a problem when more than one normative system seems to fit the bill; Evans (Reference Evans, Manktelow and Over1993) calls this the “normative system problem”; Stanovich (Reference Stanovich1999), “the inappropriate norm argument” (see also Cohen Reference Cohen1981; Reference Cohen1982; Gigerenzer Reference Gigerenzer, Stroebe and Hewstone1991; Lopes Reference Lopes1991). Deciding on an appropriate normative system for any set of experimental findings is, more often than not, far from obvious. Indeed, in contrast to memory, one is hard put to find an experimental paradigm in reasoning and decision making that has just one obvious norm to compare against and no competing alternative norms. In the following, we propose a typology of three normative situations, based on the nature and number of competing normative accounts of a particular experimental paradigm. Of the three types, two involve normative conflict and one involves no conflict. Table 1 summarizes them.

Table 1. The three types of normative conflict

With one established norm and no conflict, single-norm paradigms seem to offer the prototypical normativist situation. In other cognitive research domains, single-norm paradigms are indeed both basic and ubiquitous. Thus, in memory, in signal detection, and in most theory-of-mind paradigms, what is “right” and what is “wrong” is ordinarily beyond dispute. Either there is (for example) a visual signal or there isn't: the experimental environment is constructed so as to obviate the question. Not so, however, in reasoning and decision making, where single-norm paradigms are increasingly rare. One of the few remaining single-norm paradigms in reasoning seems to be conditional inference, or, more specifically, conditional elimination inferences. Such inferences are typically comprised of a “major” conditional premise of the form, if p, then q, and a categorical premise (e.g., p). The conclusion is categorical, eliminating the conditional form. There are also conditional introduction inferences, in which the conditional form is the conclusion of the inference rather than (one of) its premise(s). Table 2 presents several types of conditional inference.

Table 2. Types of conditional inference

The conditional elimination inferences constitute a single-norm paradigm: Regardless of one's theoretical position, MP (Modus Ponens) and MT (Modus Tollens) are generally considered valid types of inference, whereas DA (Denial of the Antecedent) and AC (Affirmation of the Consequent) are invalid. Although this validity can and has been contested under specific conditions (e.g., McGee Reference McGee1985), experimental paradigms are generally constructed to avoid these conditions. However, this is only half the story. When conditional inference is viewed as a whole, normative considerations are by no means uncontroversial. For example, the paradoxes of material implication (again, see Table 2) are the subject of some intensive dispute, considered valid in mental model theory (Johnson-Laird & Byrne Reference Johnson-Laird and Byrne2002; Schroyens Reference Schroyens, Oaksford and Chater2010) but deemed invalid in probabilistic approaches (e.g., Evans et al. Reference Evans, Over and Handley2005; Oaksford & Chater Reference Oaksford and Chater2007). Hence, when participants judge the paradoxes as invalid (Pfeifer & Kleiter Reference Pfeifer, Kleiter, Manktelow, Over and Elqayam2011), mental model theory judges their response as erroneous, whereas probabilistic approaches regard the same response as perfectly normative (and see Over et al. [Reference Over, Evans, Elqayam, Oaksford and Chater2010] for discussion of another type of conditional introduction inference with conflicting normative judgements).

While the conditional elimination inferences can be considered single-norm, conditional introduction inferences, then, are subject to dispute. We call these alternative-norm paradigms. Extensively covered by Stanovich (Reference Stanovich1999), alternative-norm paradigms are far more prevalent in the psychology of reasoning and JDM (judgement and decision making). In a typical debate of this type, a standard account of a particular observation competes with another, alternative account (or accounts), making an observed behaviour normatively rational according to the latter but not according to the former (and vice versa). Examples of alternative-norm paradigms are legion (Stanovich [Reference Stanovich1999] reviews some classic alternative-norm paradigms; Hahn & Warren [Reference Hahn and Warren2009] review some recent such developments in JDM). The longer a paradigm is studied, the more it tends to have alternative normative systems proposed.

The classic case is probably the Wason selection task (Wason Reference Wason and Foss1966), a hypothesis-testing task designed to test understanding of the logic of conditionals. In the abstract version of this famous task, participants are presented with four cards bearing values, such as A, G, 3, and 7, and are given a conditional rule of the general form, if p, then q, such as: “If there is an A on one side of the card, then there is a 3 on the other side.” Their task is to turn over all the cards – and only the cards – that need to be examined in order to decide whether the rule is true or false. The task is notoriously difficult in its standard, abstract form (for a recent review, see Evans & Over Reference Evans and Over2004). Only about 10% of participants (of higher IQ) typically find the standard normative solution: p and not-q (which in this example would be A and 7; A because a not-3 number on the other side would disprove the rule, and 7 because an A on the other side would do the same). However, Wason's normative departure point was logicist: the material conditional of the propositional calculus, according to which a conditional statement if p, then q is true whenever q is true or p is false. When measured against alternative normative systems, such as decision theory (Manktelow & Over Reference Manktelow and Over1991), Bayesian probability or information theory (Oaksford & Chater Reference Oaksford and Chater1994; Reference Oaksford and Chater1996), or default logics (Stenning & van Lambalgen Reference Stenning and van Lambalgen2008), the prevalent choices can be argued to be rational. For example, Oaksford and Chater (e.g., Reference Oaksford and Chater1994; Reference Oaksford and Chater2007) argue that participants select the optimal information in order to decide whether q depends on p or not, and are therefore normatively rational in terms of gaining information.

Clearly, alternative-norm paradigms pose a major challenge for normativism. If there is just one “correct” normative system, what are the mechanisms to arbitrate between the competing accounts? Another problem with the alternative-norm paradigm is that what makes one account “standard” and the other “alternative” is often hard to determine. Why, for example, should classical logic be considered “standard,” in the case of the selection task, and information theory considered “alternative”? Because classical logic was the first proposed or has been around the longest? Oaksford and Chater (Reference Oaksford and Chater2007) have recently argued that Bayesian probability is becoming the dominant paradigm in cognitive science. If this is true, then the current Kuhnian paradigm for the selection task is probabilistic, but the original normative system – and the one that has been around longest – is deductive. So which should we view as the standard and which the alternative?

The problem becomes even more striking when we consider multiple-norm paradigms, in which there are several normative systems available but none that appears to be standard. For example, consider the reasoning literature on metadeduction (e.g., Byrne & Handley Reference Byrne and Handley1997; Byrne et al. Reference Byrne, Handley and Johnson-Laird1995; Elqayam Reference Elqayam2006; Rips Reference Rips1989; Schroyens et al. Reference Schroyens, Schaeken and d'Ydewalle1999). In this paradigm, participants are presented with the Island of Knights and Knaves, whose inhabitants are either knaves (liars) or knights (truth-tellers). The task is to identify the speakers based on their statements. It is generally (albeit implicitly) assumed in the metadeduction literature that statements can be assigned truth-value based on partial information; for example, that one false conjunct is sufficient to make a conjunction false (so its speaker can be identified as a knave). But consider this sentence: “I am a knave, and snow is black,” described by most participants as indeterminate (Elqayam Reference Elqayam2006). Is such a response erroneous, then? The difficulty is that the statement “I am a knave” is paradoxical: it is a version of the Liar paradox (e.g., Martin Reference Martin1984). The issue now becomes evaluation of sentences with paradoxical constituents – which brings us to many-valued logics. As Elqayam (Reference Elqayam2003) argued, given the plethora of many-valued logics (for reviews, see Gottwald Reference Gottwald2001; Rescher Reference Rescher1969), there is little ground for preferring one type of system over the other.

The (increasing) scarcity of single-norm paradigms in reasoning and decision making poses a major problem for normativism, as the latter depends on an agreed norm for assessment. Psychologists can, of course, and do get involved in arguments about which norm is right – perhaps an odd activity for empirical scientists. In fact, the temptation to which they often succumb is to try to resolve the issue empirically. But this leads them into the questionable form of argumentation that involves is-ought inference, discussed below. First, we clarify the status of formal theories and their role in empirical science.

4. The computational, the competent, and the normative

From the foregoing discussion, the reader might have formed the impression that we reject formal systems entirely in favour of purely processing accounts. However, our objection is not to formal systems per se, but to their use as normative systems; it is the deontic, evaluative “ought” that we caution against. We have no problem with formal systems as competence or computational-level systems. Indeed, each of us separately has previously used formal systems as a major source of inspiration to construct a psychological theory, albeit on the computational rather than normative level. For example, Evans and Over (Reference Evans and Over2004) utilized the suppositional conditional (Edgington Reference Edgington1995; Reference Edgington and Zalta2008); Elqayam (Reference Elqayam2006) used Kripke's (Reference Kripke1975) theory of truth. We did so in much the same way that Chomskyan grammar provided and still provides inspiration to psycholinguistic and neurolinguistic research. (We will take this up again in more detail in section 7.)

This distinction between competence theory and normative theory is paramount to our argument. To illustrate it, we will start with linguistics, where a tradition going back to De Saussure (1916/Reference De Saussure1966) clearly separates descriptive from normative accounts in favour of the former. Here is a classic example. Consider double negation, as in “I don't know nothing.” Countless primary school teachers have lectured countless generations that double negation is not “Good English”. However, double negation is part of the grammar in some variants of English, such as African American Vernacular English (AAVE): A theory seeking to describe the linguistic competence of AAVE speakers would have to include it. Double negation, then, is part of a competence theory of AAVE, although it falls outside normative grammar. While descriptive competence theories aim to describe the rules of language actually used by speakers, normative approaches aim to regulate speech in particular ways, sometimes motivated by a social, educational, or political agenda that has little to do with the way human language works. For example, Mustafa Kemal Atatürk's reform of the Turkish language, “purging” it of centuries of Arabic influence (Lewis Reference Lewis1999), was grounded in nationalist normativism.

There are quite a few categorizations of levels of inquiry in the cognitive literature (for a review, see Stanovich Reference Stanovich1999), but Chomsky's and Marr's are probably the most influential ones in cognitive science, so we will limit ourselves to these two. We use the term competence here in the Chomskyan sense (e.g., Chomsky 1965), which is to say, a structural description of abstract knowledge that is quite value-free (although cf. Harris Reference Harris1980; Reference Harris1981). “Competence” is not intended to be contrasted with “incompetence,” but rather with performance, that is, the instantiation of linguistic competence in actual speech. The Chomskyan notion of competence is parallel to Marr's (Reference Marr1982) conception of the computational level of analysis – the level that describes what is being computed and why (e.g., the rules of arithmetic). Marr himself noted the analogue to Chomsky; what Marr's conception adds is the notion of function, which (as we have seen in sect. 2) has implications for our discussion. Additionally, Marr outlined an algorithmic level of analysis, which describes how the function is being computed (e.g., the calculator's chip). This is roughly analogous to the Chomskyan performance (although the latter is more heterogeneous; see Jackendoff Reference Jackendoff2002). This computational/algorithmic (or competence/performance) distinction is akin to the veteran product/process distinction, respectively: The structural description of the output (product) function is featured on the computational or competence level, while the actual processes involved in a specific task are on the algorithmic or performance level. [Marr also introduced a third level, the implementational (hardware/wetware) level, but this is not relevant to our discussion here.]

The essence of the difference between normative and computational/competence theories is in their respective research questions. As Marr noted, an algorithmic theory asks “how is…” questions; for example, how is a decision made in various frame contexts. A descriptive competence theory asks “what is…” questions; for example, what is the relation between the negative particle and the verb phrase in AAVE. A normative theory asks evaluative “ought” questions: “What ought to be the good use of negation in language?” A normative approach contains an element of evaluation, a sense of “goodness” and “badness,” of “right” and “wrong,” that is absent from a purely competence account. In short, normative theories are “ought”-type theories; computational theories are “is”-type theories. Note that competence theories and performance theories are both descriptive – what they share is the is.

In conclusion, our position is that the normative and the descriptive functions of competence theories are best kept strictly separate, as they are in mainstream linguistics. At the very least, it is not obvious that norm and competence are one and the same, and we suggest that the burden of proof is on anyone contesting the distinction. We therefore conceptualize competence-level explanations – alongside algorithmic-level explanations – as descriptive, “is”-type theories, rather than normative, “ought”-type theories. We will argue that failing to distinguish between is and ought inevitably invites a highly controversial type of inference. We now turn to examine this inference and its consequences.

5. Inferring ought from is

Differentiating between normative and competence accounts might not have mattered all that much were it not for the problem of arbitrating between competing normative accounts. As noted above, normativism has to be selective: where there are alternative systems, only one of them is “appropriate” (what Stanovich Reference Stanovich1999 calls “the inappropriate norm argument”). However, with alternative-norm and multiple-norm paradigms, arbitrating between competing normative systems is both crucial and far from easy. This is where the difference between normative and competence theories becomes critical. Competence theories are descriptive and can hence be supported by descriptive evidence. In contrast, can one support normative theory with descriptive evidence? Can one infer the ought from the is?

The short answer is “no”. Inferring an “ought”-type conclusion from “is”-type premises is highly controversial, and considered by many authors to be a logical fallacy. First identified by Hume (1739–1740/Reference Hume2000; although cf. MacIntyre Reference MacIntyre1959), is-ought inference is made whenever we attempt to derive a normative or evaluative conclusion from descriptive premises (although cf. Frankena Reference Frankena1939; Searle Reference Searle1964; Williams Reference Williams1985). For example:

• Human beings have natural fear of heights.

• Therefore, we should not fly in airplanes.

Since the premise has no normative value, inferring a normative conclusion is argued to be fallacious. Is-ought inference is closely related to what is called the “naturalistic fallacy” (Moore Reference Moore1903): deriving ethical norms from natural phenomena; for example, deriving ethics from evolution. The term is sometimes extended to any sort of evaluative norm derived from natural observation, and in that sense it overlaps to a great extent with is-ought inference. Our airplane example is problematic both in the is-ought sense and in the naturalistic sense. Note that one can argue that there is an implicit normative premise: the belief that we should act according to our natural emotions, including fear. With the implicit premise made explicit as a second premise, the normative term is included in the premises, and the argument no longer a fallacy. However, identifying – and justifying – the implicit “ought” premise can be rather tricky.

We should clarify at this stage that the is-ought question is a highly polemical one; whether it is always a fallacy is much contested in the philosophical literature (for reviews, see Hudson Reference Hudson1969; Schurz Reference Schurz1997). However, none of the proposed solutions suggests that is-ought inference is universally valid; solutions typically specify a set of conditions under which it is valid (e.g., for constitutive rules only; Searle Reference Searle1964). Cases that fall outside these conditions are indisputably invalid. Whether these conditions apply in the case of normativism is moot (Elqayam Reference Elqayam, Manktelow, Over and Elqayam2011), and we propose that the burden of proof is on normativism. We therefore submit that it is preferable to avoid such inference entirely. To do so, we must confine ourselves to competence, and not normative, theories. In what follows, we will look in detail into two examples of is-ought inference, both made by prominent normativist research programmes: Oaksford and Chater's (1998a; 2007) rational analysis programme, and Stanovich and West's (Stanovich Reference Stanovich1999; Stanovich & West Reference Stanovich and West2000b) individual differences research programme. We have chosen to focus on these two examples because they are high profile and well respected in the literature. Indeed, we ourselves admire both of these programmes in many respects. However, we also contend that each involves evaluative normativist thinking and a form of is-ought inference.

6. Normativist research biases

It may seem to some readers, as it did to a referee of an earlier draft of this paper, that we are objecting only to the style of research and writing about human thinking, and that our comments have few implications for the substance of such research programmes. This is far from the case. In fact, we wish to argue the opposite: that normativism has seriously biased and distorted the ways in which psychologists go about studying thinking, reasoning, and decision making (see Table 3). It makes a very substantial difference to how we practice our craft, and to the research questions we ask on both the processing and the computational levels. A descriptivist approach may free the psychology of reasoning and JDM from these research biases.

Table 3. Normativist research biases in psychology of reasoning and JDM

Let us start with the case of logic and the deduction paradigm (Evans Reference Evans2002). The standard practice in the psychology of reasoning, at least until the past decade or so, was as follows: You draw a sample of participants, specifically excluding any who have had formal training in logic. You present them with problems that are either abstract or designed in such a way that any influence of real-world beliefs can only be interpreted as a bias, since it is orthogonal to the logical structure. You then instruct participants to assume that all the information given is true, to base their reasoning only on the information given, and to draw only necessary conclusions. However, normally you do not provide them with any kind of instruction or training in logical principles, including that of necessary inference. You then assess performance against standard logical solutions to the problems, and also count as a cognitive bias any source of variance other than the logical structure. This describes the predominant practice in the psychology of reasoning over the past 50 years or so, and it explains the origins of such terms as belief bias (e.g., Evans et al. Reference Evans, Barston and Pollard1983; and see sect. 6.1 below) and matching bias (Evans Reference Evans1972); the balance has only started to shift in recent years.

We ask readers to reflect on whether the deduction paradigm could have developed this way without the logicist and normativist thinking that preceded it. The argument encouraged by Inhelder and Piaget and numerous philosophers is essentially this: (Classical) logic provides the laws of rational thought in all contexts. People are rational. Therefore, logic must be built into people's heads in some innate and a priori manner. We call this the prior rules bias. This is basically an empirical normativism approach, the idea that thinking reflects a normative system; and it has implications for computational-level analysis as well as processing accounts. From this, everything about the deduction paradigm follows, including the use of participants untrained in logic and contexts lacking helpful pragmatic cues. If people are rational, then they should still give the logical answers. Researchers then seem to be astonished when participants get the answers wrong, in contrast with the remarkable achievements of the human species in many specific fields of endeavour that require advanced reasoning.

Without logicism, the study of rationality in reasoning might have been entirely different. Why on earth, for example, should our notion of rationality exclude learning? Why not start with the observation that people can become expert reasoners in law, medicine, science, engineering, and so forth, noting that in every case they spend many years in specialized training to achieve this level of expertise? Why not focus on expert reasoning and how it is acquired? But no, we have spent the past half a century instead studying naïve participants with novel problems, resulting in a Kuhnian crisis as the field struggled to throw off the shackles of logicism (Evans Reference Evans2002; Reference Evans2010b; Oaksford & Chater Reference Oaksford and Chater1998a; Reference Oaksford and Chater2007). The new paradigm that is emerging utilizes a wide variation of methods, with a focus on uncertainty, belief, and pragmatic influences on reasoning. However, merely discarding logicism will not resolve the problem. As befits the topic of this target article, there is an as yet unresolved debate about whether the new paradigm requires an alternative normative theory, such as Bayesianism (Evans, in press b). The prior rules bias is still active – only the proposed rules have changed.

If we examine the study of judgement and decision making, we find that normativism has dictated research strategy in very similar ways. Here, too, researchers predominantly assess rationality by testing naïve participants on novel problems, carefully avoiding any instruction in the rules with which they need to reason. The prior rules bias is evident, for example, in the study of Bayesian reasoning. According to Bayes' theorem, posterior probability judgements should reflect a multiplicative function of prior probabilities (i.e., base rates) and diagnostic evidence. Since the pioneering work of Kahneman and Tversky (Reference Kahneman and Tversky1972) there has been much concern, and a very large number of research papers, about the finding that people neglect or underweight base rates in making posterior probability judgements (for review, see Barbey & Sloman Reference Barbey and Sloman2007). But think what is required to get the problem right. The participants must either know or somehow derive from first principles Bayes' theorem (or at least some sort of approximation), as this is never given by the experimenter. They must also perform the mental arithmetic required to multiply the relevant probabilities. Very few are able to do this, except when the information is presented in transparent “nested sets” clearly demonstrating the relations between superordinate and subordinate sets (Barbey & Sloman Reference Barbey and Sloman2007; Cosmides & Tooby Reference Cosmides and Tooby1996; Evans et al. Reference Evans, Handley, Perham, Over and Thompson2000; Gigerenzer & Hoffrage Reference Gigerenzer and Hoffrage1995). This facilitates a relatively simple mental representation of the problem, enabling its solution by general reasoning.

What kind of test of rationality do standard tests of Bayesian reasoning provide? Why should we expect people to reason with rules they do not possess, lacking the “mindware” for the task (Stanovich Reference Stanovich2010a)? Granted, many studies have shown base rate neglect in expert groups (Koehler Reference Koehler1996), with the obvious implication that such groups (doctors, lawyers, etc.) require training in this kind of reasoning. But this is where the enquiry should have begun. Some kinds of statistical reasoning can be learned by general experience, even if it remains domain specific (Nisbett et al. Reference Nisbett, Krantz, Jepson and Kunda1983), but other kinds cannot and require rule-based training. The question of whether people can be rational statistical reasoners must be assessed when appropriate training has been provided. Evolution may have provided “fast and frugal heuristics,” labour-saving rules of thumb that work well in some circumstances (Gigerenzer et al. 1999), but it certainly cannot provide us with the ability to be lawyers, engineers, and rocket scientists without training.

Normativism has affected not just the methodology of the psychology of reasoning but also the way in which findings are reported and interpreted on the processing level. We will call this the interpretation bias. In Jonathan Evans' first book on the topic (Evans Reference Evans1982), he argued that we should desist from the practice of reporting logical accuracy in reasoning tasks, and instead report what people actually did. This is particularly critical in the study of conditional inference. The standard paradigm focuses on the elimination inferences and it tests whether people will endorse each of the four inferences MP, DA, AC, and MT (see Table 2). The traditional practice, still quite common in the developmental literature (e.g., Barrouillet et al. Reference Barrouillet, Markovits and Quinn2001), is to score the number of logically correct inferences endorsed, which means adding yes answers to MP and MT (valid inferences) to no answers for DA and AC (invalid inferences). But this practice is highly interpretative and misleading. From a cognitive point of view, an inference is either drawn or it is not. The interpretation bias leads researchers to equate endorsing one kind of inference with refusing another, as though these were similar rather than opposite cognitive processes. Logicist thinking has even led leading advocates of the mental logic theory of reasoning to propose entirely different mechanisms to explain the drawing of valid and invalid conditional inferences (e.g., Braine & O'Brien Reference Braine, O'Brien, Braine and O'Brien1998), the former based on mental rules and the latter on pragmatic implicatures. But the experimental evidence supports no such distinction. For example, DA (invalid) and MT (valid) inferences are prone to exactly the same form of negative conclusion bias or double negation effect (Evans et al. Reference Evans, Newstead and Byrne1993). How could this be if different mechanisms are involved?

The field of judgement and decision making is, if anything, even more prone to interpretation bias than is the psychology of reasoning. Encouraged by the discipline of economics, from which the study of rational decision making derived, studies of JDM have focused again and again on conformity to or deviations from normative theory to the exclusion of psychological accounts of what people are actually doing. This may be why dual-process accounts of JDM were, until recently, mostly proposed by those working predominantly in the psychology of reasoning (Evans Reference Evans2007; Evans & Over Reference Evans and Over1996; Stanovich Reference Stanovich1999; Reference Stanovich2010a). Fortunately, following the explicit adoption of the theory by Kahneman and Frederick (Reference Kahneman, Frederick, Gilovich, Griffin and Kahneman2002), this is starting to change. However, JDM still lags behind the new paradigm psychology of reasoning in the use of process tracing methods such as protocol analysis, response times, eye-movement tracking, and neural imaging; but again, we are pleased to see this is now changing. But why has it taken so long for researchers to get focused on the cognitive processes underlying judgement and decision tasks? In a word: normativism.

Even at the purely computational level, normativist research biases may affect the very research puzzles that psychologists select to study. Recall our classification of conflict between formal systems in section 3: single-norm paradigms, where there is no conflict; alternative-norm paradigms, where an alternative system competes with the standard one; and multiple-norm paradigms, where there is multiplicity of formal systems, none of which can be said to have any precedence. Historically, the psychology of reasoning and decision making tended to be biased towards asking research questions drawing on single-norm paradigms, although they have a tendency to mutate into alternative-norm paradigms as researchers discover or invent alternative norms. The expectation that there will be a single, or at least standard, normative system is a natural consequence of empirical normativism: the belief that human thought follows a normative system. It is also crucial for prescriptive normativism, since a normative system has to be clearly identified for prescriptive normativism to make any sense. We call this the clear norms bias.

When an alternative norm is proposed, heated debate tends to follow, as normativism requires a clear standard. Moreover, the motivation for proposing alternative norms may be the observation that empirical normativism fails with the existing standard. Oaksford and Chater's (Reference Oaksford and Chater1996) account of selection task choices in terms of expected information gain, and the spate of critical notes that followed it, illustrate both aspects. In JDM, Hahn and Warren (Reference Hahn and Warren2009) have similarly taken on normative analysis of lay perception of randomness, long perceived as normatively incorrect in the JDM literature. Arguing that such perceptions are normatively correct when one takes into account the “finite attentional window” through which random strings are typically available to working memory, Hahn and Warren also partially exonerated the “gambler's fallacy” (the conviction that consecutive random draws should be balanced), again with the foreseeable flurry of critical notes.

As to multiple-norm paradigms, where no agreed normative standard exists, there is correspondingly little experimental work. Examples include embedded conditional statements, where “no theory has an intuitively adequate account” (Edgington Reference Edgington and Zalta2008), and conditional introduction, as opposed to the much studied elimination inferences (but see Over et al. [Reference Over, Evans, Elqayam, Oaksford and Chater2010] for recent discussion). These are by no means trivial paradigms: both have generated a great deal of discussion in philosophical logic (for review, see Edgington Reference Edgington and Zalta2008). Issues that the psychology of reasoning overlooked despite patent philosophical value tend to be multiple-norm paradigms. We find this suggestive to say the least.

7. Can we manage without a normative theory?

The previous three sections have reviewed the pitfalls of normativism. First, we have argued that in a quest to solve the thorny arbitration problem, theorists have fallen into the practice of dubious is-ought inference, which in the worst case can lead to circular reasoning: people ought to do whatever it is they actually do! Next, we have shown how normativist thinking has biased and constrained the relevant research programmes. Illustrating the problem with the case of research on dual processes, we have also identified a specific bias which we term the ought-is fallacy: the belief that System 2 is responsible for normative responding (and System 1 for errors and biases).

We now seem to be faced with a dilemma. On the one hand, the problems we have identified with normativism make it highly questionable as a meta-theoretical framework for the psychology of reasoning and JDM (judgement and decision making). On the other hand, the long and productive history of normative approaches in reasoning and JDM should give one pause before throwing them overboard. Formal systems such as logic and Bayesianism have provided major incentives and inspiration to countless research paradigms. Popper's logicist philosophy of science was the main motivation behind Wason's selection task and the 2-4-6 task; decision theory motivated Tversky and Kahneman's heuristics and biases programme. Can we make do in reasoning and JDM without normative theories altogether?

Evaluative normative considerations are just one way in which formal theories can be useful for psychological theorizing. There is a wide range of possible relations between formal systems and psychological theory, depicted in Figure 2. Formal theories can also constrain psychological theorizing; that is, psychological theory can be formed in a way that takes computational-level theories into account, and can provide a useful formal language.Footnote ⁴ Formal theories can also inspire psychological theory, which can be seen as a special case of weak constraining: a single idea or principle is taken from the formal theory, leaving a wide margin for psychological principles to be developed semi-independently. Psychological theorizing and data can also reflect back on formal theories: to arbitrate between formal accounts, either normatively or descriptively; and to judge the psychological validity of formal accounts.

Figure 2. Models of interaction between formal and psychological theories.

Normativism potentially utilizes almost all the relations shown in Figure 2 except validation: formal theory both inspires and constrains psychological theory. In contrast, with descriptivism, there is considerable variety. Although no theoretical approach seems to be explicitly committed to what we called “descriptivism,” some theoretical approaches can be characterized this way post hoc. One such approach has been adopted by Gigerenzer and his research group (e.g., Gigerenzer Reference Gigerenzer2007; Gigerenzer & Selten Reference Gigerenzer and Selten2001; Gigerenzer et al. 1999). Gigerenzer and Todd, in introducing their research programme on “fast and frugal” heuristics (Gigerenzer & Todd Reference Gigerenzer, Todd, Gigerenzer and Todd1999), appeal first to bounded rationality (Simon Reference Simon1982), then to ecological rationality, and finally to evolutionary rationality. What they specifically exclude, however, is normative rationality. The concept of normative rationality is replaced with the concept of ecological rationality, the rationality of adaptive behaviour. In noting the probabilistic revolution that has undermined logicism, Gigerenzer and Todd comment that their approach embraces its emphasis on uncertainty “without sharing its focus on probability theory, either as a description or an attainable norm of human behavior” (Gigerenzer & Todd Reference Gigerenzer, Todd, Gigerenzer and Todd1999, p. 6). Paradoxically perhaps, it is this very rejection of normativism that has led Stanovich (Reference Stanovich1999, pp. 57–58) to ascribe the naturalistic fallacy (see sect. 5) to Gigerenzer (Reference Gigerenzer, Stroebe and Hewstone1991), maintaining that he links the normative to the descriptive, even if only to reject it.

We have no argument with adaptive rationality per se; indeed, we keep to the position expressed by Over and Evans (Reference Over and Evans1997, pp. 255–56), which regarded any sort of instrumental rationality (adaptive rationality included) as primary. However, we agree with Stanovich (Reference Stanovich1999) that Gigerenzer's approach invites a strong is-ought inference: It seems that merely opting for a position of adaptive rationality does not inoculate a theory against it. Furthermore, given the heuristic value of formal theories, we are not convinced that a descriptivist theory should do without formal theories entirely; we argue that many of the relations depicted in Figure 2 can be maintained in a descriptivist approach.

The closest model we have is our preferred theoretical framework of hypothetical thinking theory (HTT), a dual-process framework of human thinking (Evans Reference Evans2006; Reference Evans2007). According to HTT, the main characteristic of analytic (or Type 2) processes is the ability to simulate hypothetical possibilities. HTT proposes that Type 2 processes use epistemic mental models. Unlike the semantic mental models of mental model theory (Johnson-Laird & Byrne Reference Johnson-Laird and Byrne1991; Reference Johnson-Laird and Byrne2002), which represent states of the world, epistemic mental models represent what we believe about the world. For this reason, they can include subjective probabilities, causal relations, and so on, which must be excluded from truth-verifiable semantic models. This emphasis on subjective belief makes HTT explicitly Bayesian (Evans Reference Evans2007, p. 33). Here is an interesting point of comparison with Oaksford and Chater's rational analysis programme. While both programmes are Bayesian (and decision-theoretic) in their treatment of deductive competence, HTT draws on Bayesian theory for its psychological features: subjectivity, belief, uncertainty. Indeed, Evans (Reference Evans2007) comments that Bayesian philosophy is a “more credible descriptive model” of scientific thinking than is the Popperian approach (p. 33, emphasis ours).

A similar case is the theory of the suppositional conditional, a special case of HTT (Evans & Over Reference Evans and Over2004; see also Evans et al. Reference Evans, Handley and Over2003; Reference Evans, Over and Handley2005). Strongly influenced by Edgington's philosophical work (e.g., Edgington Reference Edgington1995; Reference Edgington2003; Reference Edgington and Zalta2008), it draws on a famous philosophical tenet, the Ramsey test, which argues that, to judge a conditional sentence of the form if p, then q, we add p hypothetically to our stock of knowledge and evaluate q in this context (Ramsey 1931/Reference Ramsey and Mellor1990). In HTT, the philosophical Ramsey test is translated into a psychological principle of hypothetical thought and linked to Evans's (e.g., Reference Evans1989) earlier hypothesis that “if” triggers a focus on the conditional's antecedent (i.e., the p part of if p, then q). Evans and Over (Reference Evans and Over2004, p. vi) explicitly state that they do not try to answer normative questions. The philosophical foundation is there, but the concerns of the psychological theory are different: for example, it is strongly committed to a psychological dual-processing framework, while remaining uncommitted to any specific philosophical version of the Ramsey test (Evans & Over Reference Evans and Over2004, Ch. 2; for review of the philosophical literature, cf. Edgington Reference Edgington and Zalta2008). Evans et al. (Reference Evans, Handley and Over2003, p. 323) also refer to the Ramsey test as a source of inspiration.

More generally, HTT opts for an (implicitly) descriptivist framework by emphasizing the primacy of psychological processes. Evans (Reference Evans2007) specifically critiques the use of normative rationality in psychological inquiry. He discusses the issue in some detail, commenting that violations of normative theory are not an informative source for theorizing on cognitive processes (p. 108); and he creates a clear distinction between normative rationality and analytic thinking (pp. 159–61). Evans also addresses the issue of normative rationality explicitly, with these comments:

HTT, then, uses formal theories as a source of inspiration and even as weakly constraining psychological theorizing, without, however, accepting their normative role. While descriptivism can take different forms, we advocate the type that makes use of formal systems as fully as possible, with HTT as the closest existing realization of it.

In conclusion, we believe that descriptivism is a viable alternative to normativism. It can offer as much as normativism does in terms of the heuristic value of formal theories, without the problematic inferences and attendant research biases.

8. Final thoughts and conclusions

It is not our purpose to exclude normativism entirely from scientific endeavour. There is a need for research in education, planning, policy development, and so on, in all of which norms play a crucial role. The Meliorist position is a strong case in point, both the version advocated so powerfully by the individual differences research programme of Stanovich and West (Reference Stanovich and West2000b; also Stanovich Reference Stanovich1999; Reference Stanovich2004; Reference Stanovich2009b) and the version put forward by Baron (e.g., Reference Baron2008). Such authors wish to find ways to improve people's reasoning and decision making and therefore require some standard definition of what it means to be rational. This is an entirely different enterprise from the scientific investigation of the cognitive processes. Take the case of gambling. Gambling on games of chance that guarantee expected losses is commonplace in Western cultures, and as many as 1–2% of the population may become pathological gamblers (Raylu & Oei Reference Raylu and Oei2002). Such behaviour appears to be neither normatively nor instrumentally rational (although cf. Hahn & Warren [Reference Hahn and Warren2009] on the gambler's fallacy), if you define rationality as behaving in such a way as to achieve financial gain.

Gambling is interesting in part because of its apparent irrationality. However, this is not because we (scientists) regard gamblers as immoral people or sadly lacking in the theory of Bayesian decision making. What is striking is the apparent lack of instrumental rationality – people persist in behaviours that bring them more losses than gains. Among many interesting psychological findings in this field is that people hold false theories about chance and probability that reinforce their gambling habits (Wagenaar Reference Wagenaar1988), although it is unclear whether these are causal or confabulatory (Evans & Coventry Reference Evans, Coventry, Wiers and Stacy2006). In our view, normative theory has no role to play in the study of gambling behaviour, except perhaps to motivate interest in the topic. We need to understand what people are doing and why, rather than discussing what they are not doing. By contrast, normative theory has a clear role in the treatment of pathological gambling. Disabusing people of their false beliefs and teaching them the normative theory of probability has been shown to be an effective form of cognitive therapy for problem gamblers (Raylu & Oei Reference Raylu and Oei2002). The researchers who discovered this were clearly judgemental: they believed that pathological gambling was a bad thing, a severe problem for the individual, who should therefore be helped to stop doing it. We find this approach suitable – necessary, even – in applied research; but we see it as wholly inappropriate in basic theoretical research.

Meliorism does not have to rely on normative theories. As the gambling case shows, it is sufficient to focus on instrumental irrationality. While the earlier Meliorist research programme of Stanovich and West did employ normative theories, we have no argument with norms used to facilitate performance, so long as their evaluative aspect is acknowledged to precede research rather than to follow from it. We note also that Baron's Meliorist position is explicitly instrumental: “The best kind of thinking, which we shall call rational thinking, is whatever kind of thinking best helps people achieve their goals” (Baron Reference Baron2008, p. 61); “rational decisions can be defined relative to a person at a given time, with a given set of beliefs and goals” (p. 63).

For decades, the normativist agenda has reigned supreme in the psychology of human thinking – deductive reasoning and decision making – and it is still pretty much a dominant paradigm. However, its tenets and practices can no longer be taken for granted. The controversial inferences we have pointed out – is-ought and ought-is, respectively – are both a result of combining the descriptive with the prescriptive, as are several biases that affect the conduct and interpretation of research work. We do not deny that, as with gambling, normative thinking can attract interest in a phenomenon. In general, with a combination of evolutionary programming and instrumental learning, we would expect people to achieve most of their goals, most of the time. If we observe, for example, that many people buy high and sell low in the stock markets, then we are more likely to study their behaviour than if it were the other way around. Instrumentalism is the default behaviour and easily explained. The converse behaviour is more interesting, we agree. It may also be important to understand it from an applied perspective.

In conclusion, our argument is that psychology of human thinking would be better off with a descriptivist agenda. Normativism has played out its role in the history of the research on human thinking. The descriptivist approach views theoretical research on reasoning and decision-making research as descriptive rather than evaluative. The object is not to judge such behaviour, but to understand and predict it, using all relevant theoretical and methodological tools. Formal and computational models have an important role to play in this without being evaluative or being used to justify is-ought inference. We contend that evaluative considerations need only be invoked in educational and other applied research where the object is to improve human thinking and performance. A shift away from normativism and towards descriptivism has played a crucial role in the development of linguistics as a mature science. A similar change of direction may prove just as beneficial for the study of human reasoning and decision making.

Throwing the normative baby out with the prescriptivist bathwater Achourioti, Theodora, Fugard, Andrew and Stenning, Keith Institute for Logic, Language, and Computation/Philosophy Department, University of Amsterdam, 1012 GC Amsterdam, The Netherlands. t.achourioti@uva.nl; Center for Advanced Studies and Research in Information and Communication Technologies & Society, University of Salzburg, 5020 Salzburg, Austria. andy.fugard@sbg.ac.at; School of Informatics, The Forum, University of Edinburgh, Edinburgh EH1 1AB, Scotland, United Kingdom. k.stenning@ed.ac.uk Norms for reasoning about decisions Bonnefon, Jean-François CNRS and Université de Toulouse, CLLE, Maison de la Recherche, 31058 Toulouse Cedex 9, France. bonnefon@univ-tlse2.fr The unbearable lightness of “Thinking”: Moving beyond simple concepts of thinking, rationality, and hypothesis testing Brase, Gary L. and Shanteau, James Department of Psychology, Kansas State University, Manhattan, KS 66506. gbrase@ksu.edu shanteau@ksu.edu http://www.k-state.edu/psych/research/brase_gary.htm http://www.k-state.edu/psych/research/shanteau_james.htm Competence, reflective equilibrium, and dual-system theories Buckwalter, Wesley and Stich, Stephen Department of Philosophy, Graduate Center, City University of New York, New York, NY 10016-4309. jbuckwalter@gc.cuny.edu https://wfs.gc.cuny.edu/JBuckwalter/index.html; Department of Philosophy, Rutgers University, New Brunswick, NJ 08901-1107. sstich@ruccs.rutgers.edu http://www.rci.rutgers.edu/~stich/ A role for normativism Douven, Igor Faculty of Philosophy, University of Groningen, 9712 GL Groningen, The Netherlands. i.e.j.douven@rug.nl www.rug.nl/staff/i.e.j.douven/index The historical and philosophical origins of normativism Novaes, Catarina Dutilh Faculty of Philosophy, University of Groningen, 9712 GL Groningen, The Netherlands. C.DutilhNovaes@uva.nl Just the facts, and only the facts, about human rationality? Foss, Jeffrey Department of Philosophy, University of Victoria, Victoria, BC V8W 3P4, Canada. jefffoss@uvic.ca http://web.uvic.ca/philosophy/people/foss/index.php Overselling the case against normativism Fuller, Tim and Samuels, Richard Department of Philosophy, The Ohio State University, Columbus, OH 43210. fuller.193@osu.edu http://people.cohums.ohio-state.edu/fuller193/ samuels.58@osu.edu http://people.cohums.ohio-state.edu/samuels58/ Undisputed norms and normal errors in human thinking Girotto, Vittorio School of Arts and Design, University IUAV of Venice, 30123 Venice, Italy. vgirotto@iuav.it http://www.iuav.it/Ricerca1/Dipartimen/dADI/Docenti/girotto-vi/index.htm Normative theory in decision making and moral reasoning Gold, Natalie, Colman, Andrew M. and Pulford, Briony D. School of Philosophy, Psychology and Language Sciences, The University of Edinburgh, Edinburgh EH8 9AD, Scotland, United Kingdom. Natalie.Gold@ed.ac.uk http://homepages.ed.ac.uk/ngold; School of Psychology, University of Leicester, Leicester LE1 7RH, United Kingdom. amc@le.ac.uk http://www2.le.ac.uk/departments/psychology/ppl/amc bdp5@le.ac.uk http://www2.le.ac.uk/departments/psychology/ppl/staff/bdp5 Why rational norms are indispensable Hahn, Ulrike School of Psychology, Cardiff University, Cardiff CF10 3AT, United Kingdom. HahnU@cardiff.ac.uk http://psych.cf.ac.uk/contactsandpeople/academics/hahn.html Defending normativism Hrotic, Steven Center for the Study of Interdisciplinarity, University of North Texas, Denton, TX 76203-5017. stevenhrotic@yahoo.co.uk Cultural and individual differences in the generalization of theories regarding human thinking Kim, Kyungil and Park, Youngjun Department of Psychology, Ajou University, Suwon, 443-749, Korea. kyungilkim@ajou.ac.kr cyber13@ajou.ac.kr Norms and high-level cognition: Consequences, trends, and antidotes McNair, Simon and Feeney, Aidan School of Psychology, Queen's University Belfast, University Road, Belfast BT7 1NN, United Kingdom. a.feeney@qub.ac.uk smcnair01@qub.ac.uk http://www.psych.qub.ac.uk/Staff/Profiles/feeney/index.aspx Norms, goals, and the study of thinking Nickerson, Raymond S. Department of Psychology, Tufts University, Medford, MA 02155. r.nickerson@tufts.edu http://ase.tufts.edu/psychology/peopleNickerson.htm The “is-ought fallacy” fallacy Oaksford, Mike and Chater, Nick Department of Psychological Sciences, Birkbeck College, University of London, London, WC1E 7HX, United Kingdom. mike.oaksford@bbk.ac.uk http://www.bbk.ac.uk/psyc/staff/academic/moaksford; Behavioural Sciences Group, Warwick Business School, University of Warwick, Coventry CV4 7AL, United Kingdom. nick.chater@wbs.ac.uk http://www.wbs.ac.uk/faculty/members/Nick/Chater- Systematic rationality norms provide research roadmaps and clarity Pfeifer, Niki Ludwig-Maximilians-Universität München, Fakultät 10, Munich Center for Mathematical Philosophy, D-80539 München, Germany. Niki.Pfeifer@lrz.uni-muenchen.de www.pfeifer-research.de A case for limited prescriptive normativism Pothos, Emmanuel M. and Busemeyer, Jerome R. Department of Psychology, Swansea University, Swansea SA2 8PP, United Kingdom. e.m.pothos@swansea.ac.uk http://psy.swan.ac.uk/staff/pothos/; Department of Psychological and Brain Sciences, Indiana University, Bloomington, IN 47405. jbusemey@indiana.edu http://mypage.iu.edu/~jbusemey/home.html Epistemic normativity from the reasoner's viewpoint Proust, Joëlle Institut Jean-Nicod (EHESS-ENS), UMR CNRS 8129, Ecole Normale Supérieure, F-75005 Paris, France. jproust@ehess.fr http://joelleproust.hautetfort.com/ Naturalizing the normative and the bridges between “is” and “ought” Quintelier, Katinka J. P. and Fessler, Daniel M. T. Department of Philosophy and Moral Sciences, and Research Unit, “The Moral Brain”, Ghent University, B-9000 Ghent, Belgium. katinka.quintelier@ugent.be http://users.ugent.be/~kquintel; Department of Anthropology, and Center for Behavior, Evolution, & Culture, University of California, Los Angeles, Los Angeles, CA 90095-1553. dfessler@anthro.ucla.edu http://www.sscnet.ucla.edu/anthro/faculty/fessler Truth-conduciveness as the primary epistemic justification of normative systems of reasoning Schurz, Gerhard Department of Philosophy, University of Duesseldorf, D-40225 Duesseldorf, Germany. schurz@phil.uni-duesseldorf.de www.phil-fak.uni-duesseldorf.de/philo/personal/thphil/ Reason is normative, and should be studied accordingly Spurrett, David School of Philosophy and Ethics, University of KwaZulu-Natal, Durban 4041, South Africa. spurrett@ukzn.ac.za Normative models in psychology are here to stay Stanovich, Keith E. Department of Human Development and Applied Psychology, University of Toronto, Toronto, ON M5S 1V6, Canada. keith.stanovich@utoronto.ca http://web.mac.com/kstanovich/Site/Home.html Understanding reasoning: Let's describe what we really think about Sternberg, Robert J. Office of Academic Affairs and Department of Psychology, Oklahoma State University, Stillwater, OK 74078. Robert.sternberg@okstate.edu http://osu.okstate.edu/acadaffr/ Normative benchmarks are useful for studying individual differences in reasoning Stupple, Edward J. N. and Ball, Linden J. Centre for Psychological Research, University of Derby, Kedleston Road, Derby, DE22 1GB, United Kingdom. e.j.n.stupple@derby.ac.uk http://psychology.derby.ac.uk/staff/Ed_Stupple.html; Department of Psychology, Lancaster University, Lancaster LA1 4YF, United Kingdom. l.ball@lancaster.ac.uk http://www.psych.lancs.ac.uk/people/LindenBall.html Probability theory and perception of randomness: Bridging “ought” and “is” Sun, Yanlong and Wang, Hongbin School of Biomedical Informatics, University of Texas Health Science Center at Houston, Houston, TX 77030. Yanlong.Sun@uth.tmc.edu; Department of Psychology, Tsinghua University, Beijing, 100084 China. Hongbin.Wang@uth.tmc.edu Normativism versus mechanism Thompson, Valerie A. Department of Psychology, University of Saskatchewan, Saskatoon, SK, S7N 5A5, Canada. http://artsandscience.usask.ca/profile/VThompson Neurath's ship: The constitutive relation between normative and descriptive theories of rationality Waldmann, Michael R. Department of Psychology, University of Göttingen, 37073 Göttingen, Germany. michael.waldmann@bio.uni-goettingen.de http://www.psych.uni-goettingen.de/waldmann What is evaluative normativity, that we (maybe) should avoid it? Weinberg, Jonathan M. Department of Philosophy, University of Arizona, Tucson, AZ 85721-0027. jmweinberg@email.arizona.edu Towards a descriptivist psychology of reasoning and decision making Evans, Jonathan St. B. T. and Elqayam, Shira School of Psychology, Faculty of Science, University of Plymouth, Drake Circus, Plymouth PL4 8AA, United Kingdom. jevans@plymouth.ac.uk http://www.plymouth.ac.uk/staff/jevans; Division of Psychology, School of Applied Social Sciences, Faculty of Health and Life Sciences, De Montfort University, The Gateway, Leicester LE1 9BH, United Kingdom. selqayam@dmu.ac.uk http://www.psy.dmu.ac.uk/elqayam Is everyone Bayes? On the testable implications of Bayesian Fundamentalism – Erratum Speekenbrink, Maarten and Shanks, David R.