Psychonomic Bulletin & Review

2004, 11 (6), 988-1010

An evaluation of
dual-process theories of reasoning
MAGDA OSMAN
University College London, London, England

Current theories propose that reasoning comprises two underlying systems (Evans & Over, 1996;
Sloman, 1996; Stanovich & West, 2000). The systems are identified as having functionally distinct roles,
differ according to the type of information encoded, vary according to the level of expressible knowl-
edge, and result in different responses. This article evaluates the arguments and the evidence from a
select number of key tasks that have been supportive of dual-reasoning theorists’ proposals. The review
contrasts the dualist approach with a single-system framework that conjectures that different types of
reasoning arise through the graded properties of the representations that are utilized while reasoning,
and the different functional roles that consciousness has in cognition. The article concludes by argu-
ing in favor of the alternative framework, which attempts to unify the different forms of reasoning iden-
tified by dual-process theorists under a single system.

Traditionally, logical thinking and intuition have been dual-process theory; Sloman’s (1996) two-systems the-
viewed as rival modes of thought. The former is deliber- ory; and Stanovich and West’s (2000) two-systems theory
ate, achieving accurate and justifiable representations of (dual-process theories, hereafter). These have been in-
the world, and the latter is ineffable, producing best-guess strumental in developing current research paradigms to
answers to problems without any discernable effort. This explore and examine the processes involved in all aspects
rivalry, summarized by Hammond (1996) as a distinction of reasoning behavior (i.e., deductive and inductive rea-
between intuitive and analytical thinking, has been the soning, judgment and decision making, and problem
basis for much psychological investigation in the study solving). Next, a contrasting single-system framework is
of reasoning. Recently, in their theoretical frameworks described. The origins of this framework are in the learn-
many theorists have appealed to a similar distinction ing literature, and so the introduction to this account is
proposing that reasoning comprises two underlying sys- situated therein, and then the empirical implications for
tems that serve functionally separate roles (Cosmides & reasoning are made. The third section focuses on three
Tooby, 2000; Epstein, 1973; Evans & Over, 1996; Goel, types of task that have provided supporting evidence for
1995; Kahneman & Frederick, 2002; Reyna & Brainerd, dual-reasoning processes. The final section summarizes
1990; Shafir & LeBoeuf, 2002; Sloman, 1996; Stanovich the main findings and considers their implications for
& West, 2000). dual-process theories and a single-system framework.
The aim of this article is to examine three key issues
that have arisen in the advent of dual-process accounts Development of Dual-Process Theories
of reasoning: What are the main distinctions between the of Reasoning
two reasoning systems? What evidence is used to sup- The psychological origins of the dualist distinction be-
port this view? Can a single-system framework accom- tween rational and irrational thinking can be traced back to
modate this evidence? To address these points, the re- James (1890/1950) and Freud (1900/1953). Both claimed
view begins by briefly outlining the development of the that reasoning takes the form of two different modes of
dualist perspective and introducing in detail three of the thought. James regarded reasoning as an experiential as-
main dual-reasoning theories: Evans and Over’s (1996) sociative type of thinking, as well as a separate analyti-
cal deliberate mode. Similarly, Freud’s dual theory of in-
formation processing distinguished between a primary
system that is associative and unconscious, and a sec-
The support of the Economic and Social Research Council (ESRC)
is gratefully acknowledged. The work was part of the program of the
ondary system that is conscious and capable of rational
ESRC Research Centre for Economic Learning and Social Evolution. I thought. Epstein (1973, 1994) integrated aspects of Freud’s
thank Axel Cleeremans, Darrell Collins, Celia Heyes, David Lagnado, psychodynamic account in his cognitive–experiential
Ben Newell, Maxwell Roberts, Matthew Saxton, and David Shanks for self-theory (CEST). The CEST was developed as a global
advice and guidance, and also Jonathan Evans, David Over, Brian Ross, theory of personality with two parallel systems. One is a
and Steven Sloman for helpful comments on an earlier draft of this ar-
ticle. Correspondence should be addressed to M. Osman, Department of nonexperiential information processor that is rational,
Psychology, University College London, Gower Street, London WC1E affect-free, abstract, and analytical, and the second is an
6BT, England (e-mail: [email protected]). experiential information processor that is emotionally

Copyright 2004 Psychonomic Society, Inc. 988

driven, encoding experiences in the form of concrete ex- select representations relevant to a particular problem
emplars. The theory claims that the two systems devel- space, in contrast to Tversky and Kahneman’s (1974)
oped to fulfill different functional roles and in turn have characterization of heuristics as strategies that provide
distinct evolutionary histories. The experiential system shortcuts to a solution. Analytic processes are conscious,
is the older of the two, operating in humans and nonhu- which means broadly that they are a type of deliberate,
mans, and although crude, it processes information ef- explicit thinking. Their role is to operate on representa-
fortlessly. The analytic system has a relatively short evo- tions deemed relevant by heuristic processes that are
lutionary history, is unique to humans, and was developed then used to generate inferences and form judgments
to operate in the medium of language. (e.g., Evans, 1995, 1996). They accomplish some forms
Early research on human judgment and decision making of logical analysis but are essentially context dependent
provided much evidence to suggest that there were com- because they are based on the individual’s experience.
peting modes of thought, one of which is intuitive and the Evans’s (2002; Evans & Over, 1996) current dual-
other extensional, and that they led to entirely different process account of reasoning incorporates his earlier pro-
types of responses (e.g., Hammond & Summers, 1972; posals and is strongly influenced by the kinds of distinc-
Tversky & Kahneman, 1983). Tversky and Kahneman tions between cognitive systems set out by implicit learn-
(1974) claimed that intuitive type reasoning is informal ing theorists (Berry & Dienes, 1993; A. S. Reber, 1993).
and unstructured, and relies on heuristics that include This theory divides reasoning into two systems. System 1,
such computations as similarity, representativeness (i.e., which is essentially pragmatic, is based on prior experi-
the assessment of the degree of correspondence between ences, beliefs, and background knowledge and achieves
a sample and a population), and attributions of causality. goals reliably and efficiently without necessarily accom-
Tversky and Kahneman shared Epstein’s (1973, 1994) panying awareness. It is characterized as implicit, asso-
view that the information the system acts upon is content ciative, fast, and highly robust, and it is spared by both
specific and affective, whereas extensional reasoning is aging (Gilinsky & Judd, 1994) and neurological damage
controlled, slow, and deliberate and acts as a regulator (e.g., Deglin & Kinsbourne, 1996). System 2 is explicit,
monitoring intuitive responses that it can choose to en- sequential, controllable, and makes high demands of
dorse or override. working memory. System 2 does not typically operate ac-
Early dual-process theories led to a diverse set of as- cording to normative logical conventions, but it is capable
sumptions about the nature and origins of the different of achieving solutions to logical problems as well as a
types of reasoning identified. While the research programs range of problem types (e.g., hypothesis testing, hypo-
developed from them center on different aspects of human thetical thinking, forecasting, and consequential decision
behavior, the basic claims made are the same. The basis making). Compared with System 1, System 2 is slow, but
for proposing separate reasoning systems stems from the in compensation it affords flexibility and controllability.
view that experiences are ordered very differently ac- Evans and Over (1996) make an important distinction
cording to how they are used to reason about the world. between the function of System 1, which is domain spe-
Experiences are arranged such that they form initial in- cific, and its mechanism, which is domain general. The
tuitive understandings of the world that, when organized, system is domain general in the sense that it is likened to
can be used to control and manipulate certain mental ac- a neural network where knowledge is expressed as activa-
tivities. These two alternative uses result in differences tions of particular units in the network, and not content-
in the way information is encoded, differences in behav- specific rules. The functioning is domain specific because
ior, and differences in expressible knowledge. the knowledge acquired under this system develops in
This article focuses on three of the most influential dual- highly localized domains. Evans and Over refer to Nisbett,
process theories currently pervading a wide range of work Krantz, Jepson, and Kunda’s (1983) study to illustrate
on thinking and reasoning. To begin, these theories will be this difference. Nisbett et al. presented problems based
described more thoroughly in terms of their characteriza- on the law of large numbers to participants without for-
tion of the two processes, the functions they serve, and the mal statistical knowledge. In one problem, participants
type of information they encode. In the remainder of this were asked to choose which of a range of descriptions
section, connections are drawn across the theories and then best explained why players showed better performance
discussed. Included in this section is the first of two ap- earlier rather than later in the sporting season. Those
praisals of the dual-process account, to provide some eval- with experience in the domain gave a statistical explana-
uation of the actual theoretical proposals made. At the end tion for changes in players’ form, but inexperienced par-
of this article comes the second evaluation, in which the ticipants chose a determinate account. The difference in
theories are reviewed in relation to empirical findings. descriptions showed that the application of the law was
based on domain-specific knowledge acquired through
Evans and Over’s Dual-Process Theory experience (i.e., based on exposure to sports matches).
Evans’s (1989, 1996) heuristic–analytic theory provided Moreover, the knowledge was likely to be implicit rather
the foundations for Evans and Over’s (1996) current than explicit since participants did not have any formal
dual-process theory. Evans (1984) proposed that heuris- understanding of the law itself and had acquired this
tic processes are preconscious, and their function is to knowledge incidentally.
990 OSMAN

In Evans’s (1984) heuristic–analytic account, he sug- The two systems are fundamentally distinguished by
gests that normative theories of rationality (e.g., logic, the underlying computational mechanisms, not by whether
decision theory) may not provide appropriate methods of they differentially involve implicit and explicit processing.
assessing human reasoning and often lead to the conclu- In fact, Sloman’s (1996) aligning of the two systems with
sion that we are irrational. However, if rationality is implicit and explicit processes is tentative. He suggests that
judged by the application of relevant knowledge to form although the content of awareness may be used to separate
inferences, humans are rational. This issue is expanded the two systems, it is neither a necessary or sufficient
in Evans and Over’s (1996) dual-process theory, leading way of doing so because, in his view, both System 1 and
to the proposal of two forms of rationality. One is per- System 2 can occur without conscious awareness. The
sonal (rationality1) and identifies reasoning behavior content of awareness reflects different conscious under-
that is generally reliable and efficient for achieving one’s standings of responses to a problem. Individuals using
goals. The other is impersonal (rationality2), a type of the associative system are conscious of the product of the
reasoning behavior that suggests acting when one has a computation—that is, the response, rather than the ac-
rationale for what one does, and is sanctioned by a nor- tual process used to achieve it. By contrast, the rule-
mative theory. based system generally occurs with full awareness of
both the process and the result.
Sloman’s Dual-System Theory The relationship between the systems is described as
The focus of Sloman’s (1996, 2002) dual-system the- interactive, and the degree to which they are each applied
ory is the computational distinction between two types to a problem will vary from individual to individual, de-
of reasoning systems. System 1 is associative and is at- pending on the reasoner’s skill, knowledge, and experi-
tuned to encoding and processing statistical regularities, ence. Moreover, neither system has an exclusive problem
frequencies, and correlations in the environment. Sys- domain to which they are applied. As Sloman (1996)
tem 2 is rule based. The representations in this system suggests: “It may not even be possible because both Sys-
are symbolic and unbounded, in that they are based on tems may contribute to a particular response” (p. 6). This
propositions that can be combined to form larger and forms an important basis for proposing a criterion to
more complex sets of propositions. identify dissociations between the systems. Sloman (1996,
Sloman (1996) uses Smolensky’s (1988) connectionist 2002) claims that one way of examining differences be-
framework to describe the computational differences be- tween the systems is by identifying tasks that satisfy
tween System 1 and System 2. Smolensky contrasted two what he terms Criterion S. This refers to situations where
types of inferential mechanisms within a connectionist individuals are led to respond in a manner consistent
framework: an intuitive processor and a conscious rule with System 1 but then come to realize, with or without
interpreter. Sloman claims that both System 1 (intuitive external prompting, an alternative response consistent
processor) and System 2 (conscious rule interpreter) are with System 2. System 1 remains influential despite its
implemented by the same hardware but use different subsequent replacement by System 2 thinking because
types of knowledge that are differently represented. The associative processes are highly salient. Sloman (1996)
representations within System 1 are characterized as as- describes a host of studies reporting findings that conform
sociative and are automatically generalizable, allowing to the conditions of Criterion S: for example, categoriza-
for a fast inferential process. Unlike System 2, the rep- tion tasks (e.g., Rips, 1989; Smith & Sloman, 1994), judg-
resentations of System 1 cannot be combined in novel ment and decision making (e.g., Fiedler, 1988; Tversky
ways, so that reasoning in System 1 is limited to what & Kahneman, 1983), conditional reasoning (e.g., Cheng
has already been represented. & Holyoak, 1985; Kirby, 1994), and syllogistic reason-
System 2 is a high-level processor abstracting infor- ing (e.g., Evans, Barston, & Pollard, 1983).
mation and expressing knowledge as production rules In response to the question of why reasoning should
(e.g., “if . . . , then . . .”), as described in J. R. Anderson’s be viewed as having two systems, Sloman (1996) sug-
(1993) ACT–R (atomic components of thought) model. gests, “One answer is that the Systems serve comple-
This knowledge is accessible, verbalizable, and shared, in mentary functions” (p. 18). Sloman claims that the sys-
that the rules that are formulated describe a task domain tems operate in concert but are developed to suit different
that conforms to predefined conventions or syntax that is types of knowledge and result in different outcomes that
relevant within a cultural community. Sloman (1996) are each useful in different ways. For example, disci-
uses chess to illustrate this. Players of chess have common plines such as mathematics and physics require a com-
knowledge of the rules, which have been sanctioned by a bination of intuitive thinking (System 1) and structured
community more extensive than the two players. The rules rule-based analysis (System 2).
refer to constituents of the game (i.e., the pieces and the
moves) that can be expressed in terms of a common ref- Stanovich and West’s Two-Systems Theory
erence. In this sense, the rule interpreter involves rea- Stanovich and West’s (2000, 2002) two-systems the-
soning about a set of moves at the same conceptual level ory of reasoning is primarily concerned with what causes
as the concepts within the problem domain—namely, the differences in the way individuals reason. Individuals
game itself. have been found to respond differently to reasoning tasks
 DUAL-PROCESS THEORIES OF REASONING 991

according to how they construe them (e.g., Osman & Of the three dual-process theories, that of Stanovich
Laming, 2001), their prior knowledge (e.g., Galotti, and West (2000) provides the most detailed explanation
Baron, & Sabini, 1986), the strategies they use (e.g., of how the two systems of reasoning developed, by out-
Klayman & Ha, 1987), cognitive and personality styles lining their evolutionary histories. System 1 serves evo-
(e.g., Stanovich & West, 1998b, 1998c), and their ability lutionary rationality; this type of rationality is designed
(e.g., Guilford, 1959; Stanovich & West, 1998b). to closely track increases in the reproduction probability
Stanovich and West used this source of data as a way of of genes and identifies natural regularities in the envi-
examining human rationality. Essentially, they distin- ronment. System 2 serves instrumental rationality: “Sys-
guish between three different accounts of individuals’ tem 2, while also clearly an evolutionary product, is also
departures from normative standards. One view is that primarily a control System focused on the interests of the
individuals’ poor performance on reasoning tasks re- whole person. It is the primary maximizer of an individ-
flects irrational tendencies (e.g., Nisbett & Ross, 1980; ual’s personal utility” (Stanovich & West, 2000, p. 656).
Tversky & Kahneman, 1974). However, others suggest This system is attuned to normative models and focuses
that individuals fail to perform well because of superfi- on making abstractions.
cial cognitive slips in attention, or memory lapses (e.g., On the issue of the evolutionary origins of reasoning,
Cohen, 1981). Finally, individuals’ performance might these distinctions position Stanovich and West (2000)
be consistent with a different normative model (e.g., apart from evolutionary psychologists (e.g., Fiddick,
Koehler, 1996), or the normative model may simply be Cosmides, & Tooby, 2000; Hertwig, Ortmann, & Gigeren-
inappropriately applied to assess responses to a particu- zer, 1997), who deny content-free logical processes as
lar task (e.g., Hilton, 1995; Schwarz, 1996). being part of the reasoning architecture. In Stanovich
To address these issues, Stanovich and West (1998a, and West’s (2000) view, many reasoning tasks trigger
1998b, 1998c) examined the magnitude of the correla- heuristics (e.g., conjunction problem, selection task) that
tion between performance on a range of tasks (deductive, make evolutionary sense (e.g., Hilton, 1995; Levinson,
inductive, statistical, decision making) and cognitive 1995; Stanovich & West, 2000), but individuals using
ability. They found that performance correlated with analytic processes adopt a normative construal of the
cognitive ability in every case, which suggested that task because they are motivated to maximize personal
computational limitations could be used as a predictor of utility (e.g., Baron, 1985; Stanovich, 1999). The goals
participants’ responses on reasoning and problem-solving programmed to suit the genes’ interests (i.e., the repro-
tasks. The examination of individual differences moti- duction probability of genes) and the goals by which the
vated Stanovich and West (2000) to present a unifying organism operates (i.e., to maximize the individual’s per-
framework proposing two functionally distinct reason- sonal utility) are differentially expressed in Systems 1
ing systems. and 2. In this sense, both systems are treated as optimally
System 1 is implemented automatically, is uncon- designed for different purposes and so produce different
scious, and is context dependent, relying on the types of outcomes.
heuristics described by Evans (1989). System 2 is a con-
trolled process that is purely analytical and is based on Commonalities Among the Theories
making abstractions that do not rely on context. The pri- In what ways do the claims of these theories overlap?
macy of System 1 leads to what Stanovich (1999) terms They share many common features both with each other
the fundamental computational bias, which is the ten- and earlier theoretical accounts (e.g., Epstein, 1973;
dency to automatically contextualize problems. The bias Tversky & Kahneman, 1983). They characterize the two
prevents individuals from reasoning about a task ac- systems in broadly the same way, suggesting that Sys-
cording to its logical properties; instead, they rely on tem 1 is associative, automatic, and fast, and that Sys-
cues from its context, which are interpreted in relation to tem 2 is rule-based, deliberate, flexible, and slow. The
real-life situations. systems are said to have an interactive relationship and
Stanovich and West propose that the large differences in have overlapping domains to which they apply, but they
cognitive ability result from the two systems cuing differ- also generate conflicting responses, which Sloman (1996)
ent responses. They elaborate on this by describing two and Stanovich (1999; Stanovich & West, 2000) regard as
types of intelligence. One embodies analytic intelligence support for the existence of separate reasoning systems.
and is measured by psychometric tests (e.g., Scholastic These theorists acknowledge the similarities in their
Aptitude Tests [SAT] ). The other is termed interactional proposals; however, there are three particular issues that
intelligence (Levinson, 1995) and is pragmatic in nature. Evans and Over (1996) and Stanovich and West (2000)
Individuals of higher cognitive ability are able to recruit stress in their theories.
analytical processes (System 2), avoiding the temptation Implicit and explicit processing. Evans and Over
to rely on simple general-purpose heuristics that would (1996) and Stanovich and West (2000) highlight the links
be unsuitable for a particular task. Conversely, individu- between their theories and implicit learning theorists’
als who perform less well are more likely to rely on (Berry & Dienes, 1993; A. S. Reber, 1993) conceptions
heuristics readily available in System 1. of implicit and explicit processing. Both theoretical do-
992 OSMAN

mains converge in suggesting that implicit processing tionality is broadly defined as the maximization of the
occurs incidentally and without concomitant awareness, goals of the individual organism. Evolutionary rational-
whereas explicit processing is deliberate and always ac- ity is defined in relation to Dawkins’s (1976) view of
companied by awareness. In addition, similar views are adaptive optimization at the level of the gene. Unlike
shared in the two processes’ differential relationship Evans and Over (1996), Stanovich and West (2000) pro-
with the central executive. Stanovich and West (2000) pose a one-to-one mapping between the two types of ra-
and Evans and Over suggest that System 1 makes virtu- tionality and Systems 1 and 2.
ally no demands on the central executive, and so allows
complex information processing to occur automatically Appraisal of Dual-Process Theories
(e.g., depth perception, face recognition, practiced motor How does a single-system perspective contrast the
behavior, intuition in problem solving and game playing, claims made by dual-process theorists? This section is
and language processing). concluded by a discussion on reasoning theorists’ (e.g.,
System 2 is bound to cognitive constraints, and so pro- Bucciarelli, 2000; Hammond, 1996; Margolis, 1987;
cessing is slow. The dependency of System 2 on the cen- Moshman, 2000; Newstead, 2000; Oberauer, 2000;
tral executive enables knowledge to be externalized and Roberts, 1997; Stevenson, 1997) challenges to the dual-
therefore reportable, which makes processing in this sys- ist framework of reasoning, who in turn propose a single-
tem explicit. According to Evans (2000), “Explicit pro- system account. The aim here is to examine these two
cesses also involve conscious control and can thus be positions, forming the basis for introducing the alterna-
modified by appropriate verbal instructions” (p. 6). Evans tive single-system framework in the next section.
and Over (1996) claim that the advantage of conscious One of the principal arguments made against dual-
reflective thought is that it provides the flexibility and process theories is that the proposed dichotomy does not
foresight that the implicit system cannot offer. Stanovich adequately accommodate the range of processes identi-
(1999; Stanovich & West, 2000) frames the advantages fied in studies of reasoning. Moshman (2000) argues
differently. He proposes that one of the main purposes of that the differences outlined between implicit heuristic
System 2 is to decouple information from the context au- reasoning and explicit analytical reasoning (e.g., Evans
tomatically supplied by System 1, which sometimes in- & Over, 1996; Stanovich & West, 2000) confound two
terferes with the depersonalized and decontextualized orthogonal distinctions—namely, those between implicit
representation that might be necessary for processing by and explicit processing and between heuristic and ana-
System 2. lytic processing. Evidence from studies of logical and
Two types of rationality and the evolutionary ori- mathematical reasoning (e.g., Braine & O’Brien, 1998;
gins of the two systems. Two further issues discussed in Dixon & Moore, 1997; Hawkins, Pea, Glick, & Scribner,
more detail by Evans and Over (1996) and Stanovich and 1984) shows that inferences consistent with normative
West (2000) are the systems’ relationship to conceptions principles are made automatically and that explicit rea-
of rationality, and their evolutionary origins. Evans and soning can also involve the deliberate discovery and appli-
Over’s two rationality types do not map directly onto the cation of heuristics (e.g., Kuhn, 2000; Moshman, 1999).
two reasoning systems but are described as loosely linked Moshman (2000) proposes that a multiple-systems frame-
to them. Their distinction is intended to account for what work should include four possible types of processing:
they identify as an “apparent paradox” in reasoning that implicit heuristic processing, implicit rule-based pro-
appears on the one hand to be irrational and fraught with cessing, explicit heuristic processing, and explicit rule-
biases and fallacies, and yet, on the other hand, is clearly based processing.
adapted to function well in everyday life. In their dis- Stevenson (1997) claims that criteria used to identify
cussion on the evolutionary basis of the two systems, the two reasoning systems do not accommodate exam-
Over and Evans (1997) claim that explicit thinking is “a ples of implicit and explicit processing that involve the
uniquely human facility added late in evolutionary terms acquisition and application of simple and complex rules
to a biologically tacit System, which is similar to that op- (e.g., Dienes & Fahey, 1998; Hayes & Broadbent, 1988).
erative in other animals” (p. 263). Implicit processes are Stevenson suggests that implicit processing involves the
said to originate from three sources: innate modules, im- encoding and application of fragmentary knowledge and
plicit learning, and automation of knowledge that was is situated at one end of a continuum of reasoning be-
first acquired explicitly. Over and Evans argue that Sys- havior (Burns & Vollmeyer, 2002; Geddes & Stevenson,
tem 2 is a prerequisite for rational2 thought, which is 1997). Explicit processing generates organized repre-
unique to humans, whereas rationality1 is instantiated by sentations of the current state of affairs that can be ma-
System 1, which is shaped both by past successes of an nipulated and combined with prior instances to form new
evolutionary nature and through the experiences ac- representations and is located at the other end of the con-
quired over the course of an individual’s lifetime. tinuum (Stevenson, 1997). Between these poles, repre-
Stanovich and West’s (2000) views on rationality are sentations of knowledge vary according to their accuracy
interlinked with their claims on the evolutionary basis of and relevance to the current problem space.
the two systems. They distinguish between two forms of Hammond’s (1996) cognitive continuum theory pro-
rationality: normative and evolutionary. Normative ra- poses that different forms of cognition (intuitive, analyt-
 DUAL-PROCESS THEORIES OF REASONING 993

ical, common sense) are situated in relation to one an- ing progression in the type of learning, from implicit, to
other along a continuum that places intuitive processing explicit, to automatic. For each form of learning, con-
at one end and analytical processing at the other. The sciousness has a different functional role (Cleeremans,
properties of reasoning (e.g., cognitive control, aware- Destrebecqz, & Boyer, 1998).
ness of cognitive ability, speed of cognitive activity) vary Cleeremans and Jiménez (2002) characterize con-
in degree, and the structural features of the tasks that in- sciousness as dynamic because it changes its states and
voke reasoning processes also vary along the continuum, dispositions (e.g., Cleeremans, 1997; Perruchet & Vinter,
according to the degree of cognitive activity they are pre- 2002). Rather than being an all-or-nothing state, it is
dicted to induce (Dunwoody, Haarbauer, Mahan, Marino, viewed as graded according to different types of subjective
& Tang, 2000). Hammond claims that the advantage of experience (e.g., Farah, 1994; Mathis & Mozer, 1996).
his approach, as an alternative to a dichotomous one, is Both of these characteristics lead to different functional
that it is parsimonious because it accommodates a broader roles in learned behavior. To illustrate, learning to cross
range of processing within a single-system framework. the road is first achieved by explicit instructions to look
The question of whether dual-cognitive systems can ad- left and right in order to anticipate where and when on-
equately capture the full range of processing types has coming traffic will appear; this requires active conscious
been raised in response to dual theories of learning and control of the behavior. Eventually, the steps involved in
memory (e.g., Block, 1995; Cleeremans, 1997; Farah, crossing the road become practiced and habitual, and the
1994; O’Brien & Opie, 1999). The proposal that there are learned behavior is carried out automatically, without the
separable learning systems has been challenged (e.g., need for conscious control. The subjective experience of
Meier & Perrig, 2000; Perruchet & Baveux, 1989; Shanks crossing the road changes as does the control that is ex-
& Channon, 2002; Shanks & Johnstone, 1999), as well as ecuted over the corresponding behavior.
the methodologies used to investigate the claimed di- Cleeremans and Jiménez (2002) use a connectionist
chotomy (see, e.g., Dulany, 1991; Holender, 1986; Shanks paradigm to describe the graded nature of representations.
& St. John, 1994; Velmans, 1991). In response to this, The cognitive system is viewed as a large set of intercon-
many propose (e.g., Cleeremans, 1997; Cleeremans & nected processing modules organized hierarchically, and
Jiménez, 2002; Farah, 1994; O’Brien & Opie, 1999) a as such, representations are dynamic because they are
single-system learning framework incorporating a con- transient patterns of activation over the units of each mod-
tinuum between implicit and explicit processes. This po- ule (Rumelhart & McClelland, 1986). Representations
sition is particularly relevant because some dual theorists allow the cognitive system to monitor the intermediate
(Evans & Over, 1996; Stanovich & West, 2000) empha- results of processing (J. R. Anderson, 1995; Karmiloff-
size the relationship between their dichotomy and that Smith, 1986), and for representations to enter awareness
proposed by implicit learning theorists (e.g., Berry & they need to be active enough to gain strength, stability
Dienes, 1993; A. S. Reber, 1989). It also provides the in time, and be sufficiently distinctive (e.g., Farah, 1994;
platform for which to discuss the different theoretical O’Brien & Opie, 1999; O’Reilly & Munakata, 2000).
claims that are made in the learning literature, and it es- These three properties define the “quality” of the repre-
tablishes the foundations by which the proposed alter- sentations on which the continuum is based.
ative framework can be discussed, first in terms of its ap- Strength is defined as the amount and the level of acti-
plications to learning and then to reasoning. The next vation of processing units. To illustrate, studies of com-
section focuses on Cleeremans and Jiménez’s (2002) plex implicit rule learning (e.g., Berry, 1991; Dienes &
framework, which will be used as a basis for contrasting Fahey, 1998) show that when learning is active and delib-
and evaluating the dualist reasoning approach. erate compared with nondirective, passive learning (e.g.,
observational learning), there is greater retention and con-
The Alternative Framework trol of the representations acquired, leading to transfer to
The following discussion first describes the proposals new task domains. Recall and transfer tests are often used
of Cleeremans and Jiménez’s (2002) dynamic graded as ways of measuring the extent to which representations
continuum (hereafter DGC) framework. The aim here is are strong enough to be retained and reliably called up by
not to advance a new theory of reasoning but to provide the central executive. Stability is the length of time a rep-
a framework that can be used to assess dual-process the- resentation remains active during processing. Subliminal
ories and the evidence used to support their claims. The perception studies (e.g., Cheesman & Merikle, 1984; Ku-
first part of this section outlines the DGC framework in nimoto, Miller, & Pashler, 2001) show that subjective ex-
relation to its origins, and then shows how it extends to perience varies (e.g., random guesses to accurate esti-
reasoning. The latter part of the section then discusses mates of detection of items presented) depending on the
the differences between the claims it makes, in contrast length of time that items are presented. As the duration of
to dual-process theories of reasoning. the items displayed decrease, so do participants’ subjec-
The DGC framework proposes that the quality of rep- tive experience and confidence in making accurate judg-
resentation (i.e., strength, distinctiveness, stability) lies ments about what they have been shown. Distinctiveness
along a continuum and the increase in the quality of the refers to the discriminability of representations. This can
representation along the continuum leads to a correspond- be measured according to the accuracy with which an in-
994 OSMAN

dividual can recognize fragments of knowledge acquired presented. In this example, the representation of the
through training. Studies of implicit memory (e.g., Crabb word sleep has been activated incidentally and without
& Dark, 1999; MacDonald & MacLeod, 1998) have shown any conscious awareness. Moreover, participants were
that accuracy in discriminating between old, learned items unable to control this representation by preventing its in-
and new items is dependent on how well attended the ma- clusion in the list of remembered items and for which
terial is at the time of encoding. there was overconfidence in its acquisition.
Farah’s (e.g., Farah, Monheit, & Wallace, 1991; Farah, During explicit learning, consciousness has an active
O’Reilly, & Vecera, 1993; Wallace & Farah, 1992) studies and deliberate role. The representations themselves are
help illustrate the proposed relationship between con- stable, strong, and distinctive enough to be translated into
sciousness and the graded quality of representations. Farah declarative statements. There is conscious control of the
et al. (1991) studied patients with extinction, a disorder of representations because one has metaknowledge of them
spatial attention, who were presented pairs of stimuli (e.g., and of their relevance at the time of processing. A simple
pictures of an apple and a comb), one of which was par- example of this is trying to memorize a telephone number.
tially occluded, and were later asked to judge whether they This process requires conscious control of the representa-
matched or differed. Farah et al. (1991) found that patients tions (i.e., the list of numbers) so that they can be accu-
were not consciously aware of the degraded portion of the rately stored in memory and reliably recalled in the future.
display and so were unable to correctly identify the stimu- In the case of automatic learning, representations have
lus but made accurate judgments as to whether the pairs of accumulated such strength that their influence on be-
items were the same or different. havior is no longer controlled. In contrast to implicit
Farah et al.’s (1991) findings are consistent with prior learning, during automatic learning the individual is able
studies (e.g., Karnath, 1988; Volpe, LeDoux, & Gaz- to show metaknowledge of representations and their in-
zaniga, 1979) that provide evidence for dissociations be- fluence on behavior. The representations have become
tween judgment and conscious identification. However, highly specialized because they are now optimally suited
Farah et al. (1991) repeated the task with forced-choice to a particular environment. For example, in the Stroop
questions (e.g., “did you see an apple or a comb?”) and task (Stroop, 1935) participants are instructed to name
found that correct identification of the stimuli was a re- the color of words presented to them, but there is inter-
liable index of the accuracy of judgments. Farah et al. ference from the automatic tendency to read the actual
(1991) concluded that identification and judgment tasks words, which results in slower reading times. According
made differential demands on the kind of representation to the DGC framework, automatic learning is defined by
held. Thus, the findings demonstrate a correlation be- an individual’s ability to consciously access behavior but
tween the graded quality of the perceptual representation not the ability to control it, and as the Stroop task shows,
(i.e., the partially occluded stimuli) and the contents of individuals are aware of the written words but have diffi-
conscious knowledge (i.e., identification of stimuli). culty restraining their automatic tendency to read them.
In order to draw the different proposals of the DGC
framework together, the following discussion charts the Applications to Reasoning
progress of representations along the continuum as they In this article, the DGC framework is used as an evalu-
gain strength, stability, and distinctiveness, and outlines ative tool to contrast dual processes with a single-system
with illustrations what form of learning occurs. account. Essentially, it describes how different forms of
Consciousness does not have a functional role in im- reasoning might occur and what the underlying relation-
plicit learning, and by definition there is no awareness or ship between these different forms is. Recasting Cleere-
control over the representations formed. For example, in mans and Jiménez’s (2002) DGC framework as an account
Roediger and McDermott’s (1995) memory illusion study of reasoning, the basic tenets remain the same. Reason-
participants were presented with lists of words (e.g., bed, ing involves generating and utilizing representations to
rest, awake, tired, dream, wake, snooze, blanket), all of infer the validity of the claims of a proposition, decision,
which were associates of a target (e.g., sleep) that was it- or judgment. As with learning, representations increase
self absent from the list. In a recall test, participants in quality along a continuum that leads to an equivalent
falsely recollected critical nonpresented words (e.g., progression in the type of reasoning, from implicit to ex-
sleep). Participants were also given recognition tests that plicit, to automatic, for which consciousness has a dif-
included three types of items: words previously shown, ferent functional role. The DGC framework adopts a
critical items (e.g., sleep), and noncritical items (e.g., number of proposals made by J. R. Anderson’s (1993)
spider) that were not presented. Each word was classi- ACT–R model, which describes some basic operations
fied according to whether it was new or old. There was of learning and reasoning, in order to specify more
a further classification for old items, according to whether clearly how the DGC framework applies to reasoning.
participants “remembered” or “just knew” that the item Anderson (J. R. Anderson & Betz, 2001; J. R. Anderson
had been presented before. Roediger and McDermott & Douglass, 2001; J. R. Anderson & Sheu, 1995) views
found that false recognition was high for critical words learning, categorization, problem solving, and causal in-
(e.g., sleep) and actual studied items, whereas noncriti- ference as operations composed of production rules (these
cal items were correctly identified as not having been are if–then or condition–action pairs), and by extension,
 DUAL-PROCESS THEORIES OF REASONING 995

reasoning. The “if ” specifies the circumstance under and West (2000) propose that System 1 also utilizes a
which the rule applies, and the “then” specifies to what, type of knowledge that is acquired consciously, which
although these are not necessarily formally stated. In this through practice becomes automatic.
sense, productions are functional units that associate the On the one hand, these definitions of implicit pro-
properties of a given domain (e.g., when there is P) with cessing refer to knowledge that is acquired and utilized
cognitive actions (e.g., then infer Q). without awareness, but on the other hand they include
Complex forms of reasoning are distinguished from knowledge that has been first learned explicitly and is then
simpler forms by the chains of production rules that are through practice applied to situations without awareness.
set to achieve particular goals; similarly, novice and ex- In the learning field, these are treated as different; Berry
pert reasoners are identified according to the complex- and Dienes (1993) claim that the former is what defines
ity of the chains of production rules they use. The ad- implicit learning, while the latter is automatic processing
vantage of using ACT–R to describe reasoning within or skill learning. The following discussion highlights the
the DGC framework is that it assumes that the same pro- differences between how the DGC framework treats im-
cedures operate in both learning and problem solving, plicit, explicit, and especially automatic processing and
for which there is empirical support (Bourne, 1970; Burns that of dual-process theories by defining with examples
& Vollmeyer, 2002; Dienes & Perner, 2003; Geddes & each of the three forms of reasoning outlined.
Stevenson, 1997; Klayman & Ha, 1987). In addition, The differences between implicit and automatic rea-
ACT–R does not attest to functional differences between soning are made more apparent in the DGC framework
forms of skill-based automatic processing, implicit pro- and clarified in the following definitions. Implicit rea-
cessing, and explicit processing. Finally, with more use, soning involves making a set of abstractions or infer-
production rules gather strength demonstrated in the ences without concomitant awareness of them. The ab-
speed and reliability of performance with practice, which stractions or inferences occur unintentionally, are not
maps onto the assumptions that the DGC framework susceptible to conscious control, and are therefore not
makes of the quality of representations. directly accessible to manipulation but are still capable of
It is important at this stage to acknowledge some of the influencing explicit processes. To illustrate, Siegler and
limitations of the DGC framework. As discussed already, Stern (1998) presented second-graders (8–9-year-olds)
it describes how implicit and explicit cognitive process- with a series of arithmetic problems over eight different
ing occurs. Dual-process theories also make claims con- learning sessions. Throughout the study, all participants
cerning these forms of processing, but they are particu- used a combination of strategies acquired incidentally
larly concerned with describing distinctions between and deliberately. Siegler and Stern used a number of
pragmatic and formal types of inference and the evolu- measures to examine the differences between these strate-
tionary origins of reasoning, about which the DGC can- gies, including accuracy of performance, solution times,
not make specific claims. Finally, the evidence from rea- and verbal reports. They found no differences between im-
soning studies discussed in the remainder of this section plicit and explicit strategies based on accuracy of perfor-
can only provide indirect support for the DGC frame- mance. Furthermore, there were no differences in working
work, because applying the framework to reasoning is an memory demands, which were measured by recording re-
innovation of this article. sponse latencies. The verbal descriptions of implicit strate-
gies were generally poor and less detailed, and partici-
Differences Between Dual Theories of Reasoning pants were unable to recognize when they had used them.
and the DGC Framework However, through repeated encounters with problems on
What are the differences between the DGC framework which the strategy was applicable, the representations
and dual-process theorists’ claims? The DGC framework formed became more coherent and distinct, which led to
proposes that differences in representation generate vari- greater awareness and more accurate understanding of
ation in forms of reasoning without assuming a multiple- the strategy. In turn, participants had more control over
system framework. The framework also makes an impor- them when they used this type of strategy.
tant distinction between implicit and automatic reasoning; The DGC defines explicit reasoning as having aware-
by contrast, dual-process theorists use the terms inter- ness of the abstractions or inferences that are made,
changeably and take them to refer to the same process. In which can be expressed as declarative knowledge. The
their definition of implicit processes, Evans and Over abstractions or inferences are available to conscious con-
(1996) suggest that they “operate in parallel, are compu- trol, and this allows them to be modified directly because
tationally extremely powerful, and present only their end they are accessible. There is conscious control of the rep-
products to consciousness” (p. 10). Similarly, Sloman resentations because one has metaknowledge of them
(1996) proposes that for automatic processing “a re- and of their relevance at the time of processing. They have
sponse is produced solely by the associative system [1], a high rate of activation and can be reliably recalled from
a person is conscious only of the result of the computa- memory because they are stable enough to become reg-
tion, not the process” (p. 6). Both definitions suggest istered in working memory.
that the knowledge utilized by this system is not con- Automatic reasoning is skill based and deliberately
scious. However, Over and Evans (1997) and Stanovich acquired through frequent and consistent activation of
996 OSMAN

relevant information that becomes highly familiarized. (2003) alternative explanation for these beneficiary ef-
This type of reasoning enables abstractions or inferences fects is in terms of the nested-set hypothesis. They as-
to be made without any control because the representa- sert that individuals can either construe class or category
tions are enduring, well defined, and stable through re- structures according to their instances, or by their inter-
peated use. The individuals possess metaknowledge of nal properties. Typically, individuals make probability
these representations’ influence and relevance to a task estimates on the basis of the properties of a category that
(J. R. Anderson, 1993), but not the opportunity to control can produce inaccurate judgments, whereas frequency
them. For instance, when interpreting the meaning of a formats disambiguate problems by inducing a represen-
sentence, skilled readers process the individual words tation of category instances and highlight how relevant
automatically (e.g., Jacoby, Levy, & Steinbach, 1992). instances correspond to each other. Essentially, clarify-
The processing is entirely relevant to the task and neces- ing set relations in probabilistic problems facilitates per-
sarily entails the recognition of their words and their formance, and frequency formats do this, but problems
meaning, but this is not invoked deliberately. When read- where set relations are transparent without using a fre-
ers’ recall and evaluation of the sentences was examined, quency frame are equally potent.
they expressed accurate metaknowledge of the words Dual theorists (Evans & Over, 1996; Stanovich & West,
they had processed automatically. 2000) propose that implicit /automatic processing of
Analogous to this, a skilled reasoner or logician will System 1 is heuristic based, whereas in System 2 explicit
automatically recognize which premises are relevant, processing is analytic. It is less clear how the findings
what the correct set of inferences are that follow, and discussed here would be accommodated within a dual-
what conclusion should be drawn. This chain of infer- theory account, since on the one hand they are consis-
ences will, if the task is highly familiar, be executed au- tent with System 1 processing, in that they are examples
tomatically, and with some effort, the individual will be of implicit reasoning, but on the other hand some of the
able to retrace the chain of inferences to explain how the findings are examples of System 2 because they demon-
conclusion was reached. For example, Galotti et al. (1986) strate analytic processing. One view of these studies is
compared poor, good, and expert syllogistic reasoning that they show that skill-based reasoning develops in a
and found that experts responded faster than both of the range of domains that require high-level deductive rea-
other groups and demonstrated accurate metaknowledge soning processes, as well as simple problem-solving
of their reasoning behavior. Implicit reasoning is likely skills, both of which the DGC views as automatic. The
to result from situations where reasoners are unfamiliar DGC framework does not assign a different functional
with the task environment. In this case, they will recog- role to these types of skill-based reasoning, since in both
nize few features of the task as relevant and will not cases the reasoning depends on the formation of higher
know why they are important. Furthermore, the chains quality representations and the degree of control that in-
of inferences that will lead to a conclusion are not re- dividuals have over them without reference to distinct
traceable because they are new and unfamiliar, and so no reasoning systems.
evidence of metaknowledge will be found.
The DGC framework and most dual-process theories What Are the Criteria for Demonstrating
agree that the knowledge utilized during automatic rea- Distinct Reasoning Systems?
soning is initially acquired explicitly and becomes au- This section begins by describing the type of evidence
tomatized through practice. Similarly, both types of ac- that has been used to support dual-process theories of
count propose that implicit reasoning only reveals the reasoning and discusses, in some detail, tasks that have
end products of its operations but not the processes that provided this sort of evidence. It also includes a brief
lead to them. Crucially, the DGC differentiates between analysis of the implications of the findings for dual-
implicit and automatic forms of skill-based reasoning process theories; however, the main evaluation is reserved
and provides a framework for understanding the under- for the final section of this article. The key evidence is
lying relationship between them on the basis of quality summarized, and the claims made by dual-process theo-
of representation. ries, as well as the proposals offered by the DGC frame-
Investigations of skill-based reasoning show that it in- work, are examined in relation to it.
volves both highly complex analytical processing (e.g., Support for dual-process theories comes from a wide
Galotti et al., 1986; Reingold, Charness, Pomplun, & range of studies (e.g., categorization, judgment and de-
Stampe, 2001; Reingold, Charness, Schultetus, & Stampe, cision making, problem solving, probabilistic, inductive,
2001) and simple algorithms that compute frequencies and deductive reasoning). Given such a range, this article
(e.g., Brase, Cosmides, & Tooby, 1998), although the ev- focuses on evidence from three particular tasks (Wason’s
idence for this is quite controversial. The case for a highly selection task, the conjunction problem, and syllogisms)
specialized frequency algorithm (Brase et al., 1998; Cos- taken from different domains of reasoning research. These
mides & Tooby, 1996) is based on evidence that cogni- tasks have been chosen because they have been widely
tive illusions diminish when problems are framed ac- researched and generate robust findings that the current
cording to a natural frequency format (Gigerenzer & dual-process theories have found to be highly supportive
Hoffrage, 1995). Sloman, Over, Slovak, and Stibel’s of their claims. In addition, these tasks are representa-
 DUAL-PROCESS THEORIES OF REASONING 997

tive of different forms of reasoning: Conditional reason- firming the statement (Wason, 1969), which was later
ing, decision making, and deductive reasoning further shown to be a bias for matching named cards with ex-
demonstrate the generality of the dual-process account. ample cards (Evans, 1972). The “matching” phenome-
The findings from the studies described here are classed non has since been widely researched (for a review, see
under four criteria (Criterion S, individual differences, Evans, 1998); however, as yet there is little agreement as
implicit vs. explicit processing, and neuroanatomical dif- to what underlying process generates this response.
ferences) that have been used to identify the two systems. Thematic or “contextualized” versions of reasoning
Criterion S. Sloman’s (1996) Criterion S, discussed tasks are examples where the task is embedded in a con-
earlier, outlines types of situations that demonstrate that text that makes references to real-world situations, and
the two reasoning systems operate simultaneously but these were devised to facilitate performance (e.g., Wason
lead to diverging outcomes. Sloman (1996) specifies that & Shapiro, 1971). Often, contextualized versions in-
Criterion S is also fulfilled when an individual’s initial clude deontic statements (e.g., “if you tidy your room,
response still exerts some influence although a second you can go out to play”), and these reliably generate a
answer provided either by them or the experimenter is high proportion of correct responses (e.g., Cheng &
known to be more acceptable. Holyoak, 1985; Griggs & Cox, 1982; Sperber, Cara, &
Individual differences. In addition to Sloman’s (1996) Girotto, 1996).
Criterion S, Stanovich (1999) predicts that “large differ- The general view (e.g., Manktelow & Over, 1995) is
ences in cognitive ability will be found only on problems that contextualized versions make references to situa-
that strongly engage both reasoning Systems and in which tions relevant to the reasoner. In such framings of the
the reasoning Systems cue opposite responses” (p. 143). task, reasoners are aware of exceptions and so are able to
An important aspect of such a task is that differences in select cards that undermine the statement. In this sense,
cognitive ability map onto the two types of responses. deontic and indicative versions are not comparable tasks
Those providing a normatively correct response also because they pose different questions: In indicative ver-
show higher cognitive ability compared with individuals sions, participants are asked to falsify the rule, whereas
whose answer is intuitively based. in deontic versions they are told to look for violations.
Implicit versus explicit processing. Dissociations
between implicit and explicit processing have been in- Criterion S
vestigated in studies that have used protocol measures, While there are no actual empirical tests of this crite-
transfer tasks, and tutoring. These measures indicate rion, Sloman proposes that evidence from studies of in-
that, unlike explicit processing, implicit processing is dicative selection tasks satisfies Criterion S. Through
difficult to access, and so participants are unable to ex- System 1, a computation of similarity between cards
press and control when and where it is applied. In addi- named in the rule and example cards generates the initial
tion, failure to improve performance following tutoring response of PQ selections. Corrective methods such as
supports the view that participants are unaware of the tutoring encourage System 2 rule-based processing, but
processes that contribute to their responses and so are the initial matching response is highly salient and still
unable to address them. Measuring the effects of tutor- influences participants’ responses to the task. Early tu-
ing in transfer tests also reveals that knowledge applied toring studies provide strong support for Sloman’s pro-
implicitly is utilized in domains specific to where it was posals. Wason (1968) first introduced tutoring proce-
first acquired. dures or “therapies” to invoke insight into the task. In
Neuroanatomical differences. Clinical and neuro- general, early studies using tutoring techniques failed to
imaging studies suggest that underlying neuroanatomi- improve performance (e.g., Johnson-Laird & Wason,
cal differences between the two forms of reasoning are 1970; Wason & Shapiro, 1971). When performance did
mediated by different hemispheric regions of the brain. significantly improve, following tutoring, it did not trans-
fer to other framings of the task (e.g., Wason, 1969; Wason
Wason’s Selection Task & Golding, 1974), thus, further suggesting that the knowl-
In Wason’s (1968) first published study of what now edge is context specific and therefore mediated by asso-
is referred to as the selection task, participants were ciative processing consistent with a dual-process account.
asked to decide which of the four cards (A [P], K [¬P], An alternative explanation for these results is that par-
4 [Q], and 7 [¬Q] ) should be turned over to prove whether ticipants were unresponsive to early tutoring methods
the conditional statement “if there is a vowel on one side because the techniques were inadequate, and they simply
(antecedent), then there is an even number on the other failed to identify and address problems related to cor-
(consequent)” is true or false. The correct solution re- rectly representing and understanding the task. In ab-
quires the selection of A (P) and 7 (¬Q), since they are stract versions of the task, the vast majority of partici-
the only combination that can potentially refute the state- pants (35%–60%) selected matched cards. However, a
ment. However, most participants tend to select the A (P) sizable proportion selected the P card (20%–35%), and
and 4 (Q) cards, which have been found to be a highly ro- a smaller proportion chose a range of other card combi-
bust choice. The first theory to account for this tendency nations (10%–20%). Rationalist accounts claim that the
claimed that individuals demonstrated a bias toward con- different card selections reflect differences in partici-
998 OSMAN

pants’ interpretations of the conditional statement but Newstead, Handley, Harley, Wright, and Farrelly’s
from these they are reasoning logically (Fillenbaum, (2004) study aimed to replicate Stanovich and West’s
1976; Osman & Laming, 2001; Stenning & van Lam- (1998b) findings. They found that the rational–experiential
balgen, 1999). inventory (REI) designed by Pacini and Epstein (1999),
Recent tutoring studies that tailor their techniques to which Newstead et al. (2004) used to measure intuition
the individual’s construal of the statement have found and rational thinking, failed to correlate with performance
improvements in performance that also transfer to com- on the selection task. Measures of intelligence (AH5 in-
plex propositional tasks (e.g., Margolis, 2000; Osman, telligence test) did correlate with responses to the task;
2002; Stenning & van Lambalgen, 2001). For example, however, this was not based on performance, but on the
Stenning and van Lambalgen (2001) identified a variety consistency of the cards selected across different versions,
of interpretations (e.g., “one side–other side” was taken irrespective of what that combination was. High cogni-
to mean “back and front”) of the conditional statement tive ability correlated with selections of the (P) card in
and the relevancy of the different cards, which were later abstract versions, a finding also reported by Stanovich
addressed in tutoring sessions. They found that correct and West (1998b). Conventionally, the selection of the P
performance transferred to other abstract versions of the card has been taken to indicate poor performance, whereas
task, as well as to a complex conditional problem. This a card choice that includes the combination P, Q, and ¬Q
task involved four example cards, and participants were is suggestive of partial insight (Wason & Johnson-Laird,
asked to choose from them in order to falsify two condi- 1970), which Stanovich and West (1998b) found corre-
tional statements simultaneously. Osman (2002) used lated with low cognitive ability. Newstead et al.’s (2004)
similar techniques to Stenning and van Lambalgen (2001) findings suggest that cognitive ability is predictive of
and achieved transfer to a deadline version of the ab- certain types of response pattern but that they do not
stract task (responses under limited time conditions) and clearly map onto the kinds of differences that Stanovich
to one that required participants to generate and test their and West (2000) take as support for different reasoning
own version of the selection task using statements and systems.
examples they devised themselves.
Criterion S proposes that the shifts from matched to cor- Implicit Versus Explicit Processing
rect card choices found in studies of tutoring result from a Protocols. In evaluating the evidence from protocol
switch from System 1 to System 2 reasoning. It also pro- measures, two related issues need to be considered. First,
vides an account of why tutoring fails. However, it is un- do the accounts that participants generate accurately cor-
clear whether Criterion S treats other erroneous card respond with behavioral evidence of their mental pro-
choices (e.g., P, ¬PQ) as products of an associative system, cesses? Evans (Evans & Wason, 1976; Wason & Evans,
and by the same token whether shifts from these to correct 1975) found that the verbal protocols that participants
responses are indicative of changes from associative to gave of their card selections were incompatible with
rule-based reasoning. Techniques used in tutoring studies their actual card choices, and were post hoc rationaliza-
vary; however, what is common to recent studies of tutor- tions rather than accurate descriptions of their reasoning
ing is that the large improvements in performance are the behavior. Moreover, there was no difference between
result of clarifying the task requirements and removing the participants’ confidence in the reasons given for valid
ambiguity in the conditional statement (e.g., Cheng, and for invalid card choices, suggesting that confidence
Holyoak, Nisbett, & Oliver, 1986; Klaczynski, Gelfand, & ratings of rationalizations were independent of the cards
Reese, 1989; Stenning & van Lambalgen, 2001). One ad- selected because participants do not have access to the
vantage of the rationalist account is that it accommodates processes that govern their own selections.
a wide range of card-selecting behavior by claiming that Later studies have reported associations between proto-
different choices are evidence of different construals of the cols and reasoning behavior. For example, Berry (1983)
statement. The underlying reasoning system does not used protocol techniques to investigate cross transfer of
change when individuals improve in their performance— correct responding in abstract and contextualized ver-
it is the interpretation of the task that changes. sions. She found that justifications for card selections
during concurrent protocols made participants aware of
Individual Differences the correct attributes of their own reasoning process, and
There is good evidence that cognitive ability and per- produced transfer from contextual to abstract and ab-
formance on the abstract selection task correlate (e.g., stract to contextual versions of the task. Despite unsuc-
Klaczynski, 2001; Stanovich & West, 1998b). This sup- cessful attempts to replicate Berry’s (1983) findings
ports Stanovich and West’s (2000) claims that the two concerning cross transfer (Chrostowski & Griggs, 1985;
reasoning systems often cue different responses that can Klaczynski et al., 1989), which has been thought to result
be discriminated according to high- and low-cognitive from differences in the type of contextual material used,
ability scores. Stanovich and West (2000) also claim that in Hoch and Tschirgi’s (1983) study a correspondence
deontic versions do not correlate with SATs, because was found between participants’ written protocols and
both System 1 and System 2 cue the correct response, so their card choices in both abstract and contextual ver-
participants of different cognitive abilities will perform sions. In a later replication, Hoch and Tschirgi (1985)
optimally. compared card choices before and after written protocols
 DUAL-PROCESS THEORIES OF REASONING 999

between students at different levels (high school, bache- of moving the mouse pointer across it, or including yes–no
lor’s and master’s degree) of education. Bachelor’s and decision buttons under each card, or a combination of
master’s students were more likely to improve in perfor- both) the effects are muted or in some cases reversed
mance after they explained their selections. Hoch and (Roberts, 1998). Roberts and Newton conclude that their
Tschirgi (1985) suggested that educated participants had evidence supports Evans’s (1989) heuristic–analytic ac-
more accurate representations of the problem structure to count but is also consistent with other single-process ac-
begin with and so, through protocols, were better at iden- counts such as mental models theory. Johnson-Laird and
tifying the errors in their thinking. Byrne’s (1991) mental models theory proposes that the
The second issue related to using protocol measures is initial construction of representations is often incom-
whether they should be treated as data in their own right. plete, but with time or the appropriate context they be-
There are mixed views on the credibility of data gathered come “fleshed out,” leading to correct responses. Thus,
from protocols: Nisbett and Wilson (1977) consider them it predicts that increasing the time to think about the task
to be an unreliable source of support for implicit processes, encourages participants to expand the number of repre-
but Ericsson and Simon (1980) suggest that protocols sentations, which alternatively describes the changes
are a valid way of demonstrating dissociations between from heuristic to analytic processes.
implicit and explicit processes. Certainly, early protocol
studies provide persuasive evidence for dual-reasoning Neuroanatomical Differences in the
systems and motivated the development of one of the Selection Task
earliest dual-process theories of reasoning (Wason & Golding’s (1981) study provided the earliest evidence
Evans, 1975). More recent studies (Green & Larking, for the role of the left hemisphere (LH) in content-
1995; Platt & Griggs, 1995) that have used protocol independent reasoning. Golding examined cards chosen
measures indicate that they are more usefully employed in an abstract version of the selection task by patients
when supplementing other measures of reasoning, partic- with right hemispheric (RH) and LH damage. Half the
ularly when developing ways to facilitate performance. patients with RH damage responded correctly to the ab-
Overall, the findings from work using this type of measure stract task, compared with only one patient with LH
suggest that participants do not spontaneously offer in- damage. Golding claimed that patients with RH damage,
formed knowledge of the processes by which they arrive including impairment to visual processing, were pre-
at their answers but with some cuing are able to articu- vented from making a perceptual match between the
late and capitalize on this knowledge. statement and the example cards and instead relied on
Inspection times. Dissociations between implicit and verbal reasoning ability.
explicit processes have also been studied using response Houde et al.’s (2001) positron emission tomography
time measures. Evans, Ball, and Brooks (1987) were the (PET) study compared two groups of normal partici-
first to develop a measure of attentional biases, which are pants, one receiving a short logic tutorial, and the other
thought to influence card choices (Evans, 1984). They a “logicoemotional” training session designed to elicit
recorded the order in which decisions for card selections feelings in conjunction with potential errors that could be
and rejections were made and found that participants first made. The groups were scanned while performing abstract
decided on cards they would select, followed by those versions of the selection task before and after training.
they would reject. This suggested that preconscious biases Houde et al. found that participants were more receptive
direct attention toward cards deemed relevant through to logicoemotional training compared with pure logical
linguistic cues from the task; if, instead, the task was training, as indicated by a higher proportion of correct
processed analytically, all four cards would be considered selections. In addition, they found that the right ventro-
equally. In a later study by Evans (1996), participants medial prefrontal cortex was only activated in the logico-
were presented all four cards onscreen and recorded the emotional group, which has been associated with decision
overall time taken for the inspection and selection (mea- making, emotion, and feeling (S. W. Anderson, Bechara,
sured by the time taken to move a mouse pointer to a Damasio, Tranel, & Damasio, 1999). In contrast to Gold-
card) of each individual card. Evans found that across ing’s (1981) claims, Houde et al. proposed that deduc-
abstract, negative, and thematic versions, participants tive reasoning depends on RH regions that are also de-
spent more time inspecting cards they selected than those voted to emotion. Knauff, Mulack, Kassubek, Salih, &
they rejected, a finding replicated by Ball, Lucas, Miles, Greenlee’s (2002) f MRI study of spatial brain regions in
and Gale (2003) and Roberts and Newton (2001). Evans deductive reasoning used relational and conditional
(1996) claimed that card selections are the result of heuris- problems. For both types of problem, activation of RH
tic processes that are invoked automatically, whereas an- regions was associated with the formation of spatial rep-
alytic processes that require time are invoked later and resentations, and occipito-parietal pathways were thought
are recruited to rationalize the decisions made. to support mental imagery, which is, again, inconsistent
Roberts and Newton (2001) proposed that with the ap- with Golding’s claims.
propriate methodology, inspection time effects are highly Neuropsychological studies using conditional reason-
replicable, but that with modifications (e.g., blacking out ing tasks provide conflicting results as to which hemi-
the cards until a card became highlighted only in the event spheric regions are involved in particular forms of reason-
1000 OSMAN

ing. This may in part be the result of the variability in the Hilton, 1991; Hilton, 1995; Margolis, 1987; Markus &
materials and the techniques used. However, the Houde Zajonc, 1985). Morier and Borgida (1984) examined the
et al. (2001) and Knauff et al. (2002) studies vary signif- extent to which they could reduce the fallacy by making
icantly in the actual task administered but agree on the the instructions more explicit. Participants were asked to
brain regions involved in abstract conditional reasoning. give probability estimates for each statement instead of
ranking them, and to give reasons for their estimates.
Conjunction Problem Often, respondents claimed that they used an averaging
Tverksy and Kahneman (1983) presented participants rule that weighted the traits of the character described.
with a description of “Linda” from which the following Morier and Borgida found that from this, they could at-
is an abbreviated version: “As a student, she was deeply tenuate the degree to which the fallacy was committed
concerned with issues of discrimination and social jus- by tailoring instructions to avoid application of the aver-
tice, and participated in anti-nuclear demonstrations” aging rule, which they found had a marked effect on per-
(p. 297). The participants were then asked to rank eight formance but did not completely eliminate the fallacy.
statements (e.g., “Linda is active in the feminist move- They concluded that participants’ misunderstandings of
ment,” “Linda is a bank teller,” “Linda is a bank teller the instructions influence the extent to which the fallacy
and is active in the feminist movement”) according to is committed.
how probable they were. The correct judgment relies on Agnoli and Krantz (1989) found that performance in-
participants applying the conjunction rule, which states creased following a short tutorial on Venn diagrams that
that the probability of the conjunction cannot exceed ei- was designed to encourage thinking about category ex-
ther of its constituents. Tverksy and Kahneman (1983) tensions rather than the category features of Linda’s
found that 80% of participants committed the conjunc- character. And Fong (Fong, Krantz, & Nisbett, 1986;
tion fallacy by rating “bank teller and feminist” as more Fong & Nisbett, 1991) reported the effectiveness of sta-
probable than either “bank teller” or “feminist” alone. tistical training on facilitation of correct performance on
Tverksy and Kahneman (1983) claimed that the fal- probabilistic tasks. Fong and Nisbett found that partici-
lacy was the result of an intuitive strategy, termed the pants possess rudimentary abstract knowledge of statis-
representativeness heuristic, which participants used to tical rules such as the law of large numbers and con-
assess “the degree of correspondence between a sample junction, and that because they possess some statistical
and a population, an instance and a category, an act and an intuitions, formal training procedures are able to build
actor or more generally an outcome and a model” (p. 295). directly on such knowledge. This is consistent with studies
This problem has consistently demonstrated reasoners’ er- that include protocol measures (e.g., Morier & Borgida,
rors in judgment and, like the selection task, has gener- 1984; Stolarz-Fantino, Fantino, & Kulik, 1996), which
ated a host of research, much of which has been directed show that as well as reporting the use of similarity as a
toward understanding why the fallacy is committed. strategy, participants develop strategies that utilize super-
ficial knowledge of probabilistic rules to solve the task.
Criterion S There is evidence that linguistic factors contribute to
Sloman (1996) proposed that despite knowledge of generating the fallacy, and that correcting misunderstand-
the conjunction rule, participants still tend to commit the ings can improve performance. However, other factors
fallacy, and this satisfies Criterion S because it is in- apart from ambiguity of the task framing generate the fal-
dicative of the two reasoning systems cuing contradic- lacy. Sides, Osherson, Bonini, and Viale (2002) attempted
tory responses. to reduce the source of misinterpretation by presenting un-
For example, Tversky and Kahneman (1974, 1983) ex- ambiguous conjunctive statements in a betting paradigm
amined whether prior knowledge of the rule would lead to in which participants bet on the conjunctive or one of its
improvements in performance. They presented the task to conjuncts. They found that the fallacy was committed as
three groups: naive (undergraduates from Stanford Uni- frequently in the betting paradigm as in the standard ver-
versity), informed (graduate students in psychology), and sion. This provides compelling evidence for Criterion S,
sophisticated (doctoral students in decision science pro- which suggests that the fallacy is generated by a highly re-
gram) with different experiences of probabilistic rules. In sistant bias resulting from associative processing.
all three groups, 85%–89% violated the conjunction rule.
As with the selection task, individuals are generally Individual Differences
insensitive to corrective measures. This suggests that Stanovich and West (1998a) proposed that the Linda
what generates the fallacy is difficult to alter, because of problem presents a host of features that support prag-
its associative basis, which may be why participants con- matic inferences (e.g., Adler, 1991; Dulany & Hilton,
tinue to lapse back into their initial beliefs after being in- 1991) invoked by System 1. Participants who solve the
formed of their erroneous thinking (e.g., Epstein, Dono- task correctly overcome the more obvious aspects of the
van, & Denes-Raj, 1999). task and apply the conjunction rule by using System 2
Studies that have attempted to induce insight into the processing. In their study, Stanovich and West (1998a)
task through training procedures propose that the fallacy compared performance and SAT scores in a standard
arises from misrepresenting the problem (e.g., Dulany & version of the Linda problem and two transparent ver-
 DUAL-PROCESS THEORIES OF REASONING 1001

sions (i.e., Reeves & Lockhart’s [1993] job problem and Neuroanatomical Differences in Judgment and
Fiedler’s [1988] student problem). Unlike in the selec- Decision Making
tion task, in which they found that performance on sim- Studies of schizophrenic patients’ reasoning (e.g.,
pler deontic versions was uncorrelated with ability, be- Kemp, Chua, McKenna, & David, 1997) that, among
cause both systems cue the same response, Stanovich many tasks, included the Linda problem, found that per-
and West (1998a) found that performance on the job formance was comparable with that of normals. This in-
problem correlated with ability. These findings can be dicates that neurological insult does not necessarily af-
taken to show that easier versions of the Linda problem fect judgments made in typical probabilistic tasks and
may actually engage System 2 processes, and this is why that the rudimentary processes involved are spared, which
performance correlates with cognitive ability. However, is consistent with some dual-reasoning theorists’ claims.
they are also consistent with Yates and Carlson’s (1986) However, other studies (Dudley, John, Young, & Over,
findings, showing that individuals use a variety of infor- 1997; Huq, Garety, & Hemsley, 1988) have found that
mal statistical rules that often lead to an incorrect re- neurological deficits do change the nature of the reason-
sponse but can also solve the task correctly, which does ing process involved in judgment-based tasks (e.g., judg-
not require a two-system explanation. ing the outcome of coin tosses). Dudley et al. (1997)
found that, while their general reasoning ability was not
Implicit Versus Explicit Processing impaired, schizophrenic patients required less informa-
Studies that use protocol measures extend much of the tion when making a statistical judgment. This was inter-
findings from tutoring studies suggesting that partici- preted as resulting from the patients’ pathology, which
pants invoke a variety of misrepresentations of the prob- made them mistrustful of the information presented, and
lem. Morier and Borgida (1984) categorized written pro- increased reliance on their own experience to form judg-
tocols according to two categories and found that most ments. As with the selection task, discrepancies in re-
participants explained their responses according to typ- ported findings might be due to differences in the type of
icality (“she sounds like a feminist, which makes it more material used and the population of patients tested.
likely than just bank teller”), and the remainder by way
of an averaging rule (“I averaged the two answers”). Syllogism
Yates and Carlson’s (1986) protocol study uncovered Syllogisms are complex arguments that include two
participants’ different understandings, from which a va- statements (premises) and a conclusion: For example, all
riety of strategies were developed to solve the problem. artists are beekeepers, all beekeepers are chemists, there-
They identified seven types of strategies, some of which fore all artists are chemists. Syllogisms vary in type and
they proposed were “formaliztic”: judgments made ac- complexity, and there is a substantial corpus of studies
cording to a combination of probabilistic rules, such as investigating the types of deductive reasoning that are in-
averaging. These findings suggest that protocols are an volved in solving them (for a review, see Johnson-Laird
accurate way of determining the methods used to solve & Byrne, 1991).
the problem, particularly because the types of misinter-
pretations identified in these studies have been corrected, Criterion S
leading to improved performance (e.g., Fisk & Pidgeon, One of the most robust effects in syllogistic reasoning
1996). This implies a correspondence between Systems 1 is belief bias. This is the tendency to base an assessment
and 2. Furthermore, Yates and Carlson’s findings, along of the validity of a syllogism on the believability of its
with tutoring studies (Benassi & Knoth, 1993; Betsch & content. When belief and logic agree, bias facilitates the
Fiedler, 1999; Crandall & Greenfield, 1986; Morier & logical response. For example, Evans et al. (1983) found
Borgida, 1984) suggest that improvements in perfor- that most participants correctly assessed as invalid the
mance can occur when participants supplement one type argument “No police dogs are vicious, some highly
of rule-based reasoning (e.g., applying an averaging trained dogs are vicious, therefore some police dogs are
rule) for another (e.g., the conjunction rule). This con- not highly trained.” When belief and logic disagree, par-
trasts the dual-process account, which would claim that ticipants often draw erroneous conclusions. Evans et al.
in order to induce correct responses the rule-based sys- (1983) found that the majority of participants incorrectly
tem is required to override the heuristic-based system. accepted as valid the argument “No addictive things are
Both protocol and tutoring studies suggest that partic- inexpensive, some cigarettes are inexpensive, therefore
ipants are aware of their reasoning processes, even in some addictive things are not cigarettes.” Sloman (1996)
cases where it is evident that they are using a type of proposed that the latter effect satisfies Criterion S be-
heuristic. Moreover, in tutoring studies (Benassi & Knoth, cause conflicting responses generated by the different
1993; Crandall & Greenfield, 1986), often where there is reasoning systems cue contradictory conclusions as to
no direct mapping from tutoring session to test items, the syllogism’s validity.
participants still are able to transfer newly acquired knowl- Accounts of bias differ in terms of where in the reason-
edge to similar and disparate task domains. Again, this ing process it is generated. This, in turn, has implications
indicates that participants can access and change what as to whether the bias is consistent with the conditions of
seems to be a highly resistant bias. Criterion S. Given the primacy of System 1, evidence of
1002 OSMAN

the bias occurring early in reasoning supports the pro- whereas participants of high cognitive ability evaluated
posal that it results from automatic associative processes syllogisms analytically, leading to correct performance.
(consistent with Criterion S), whereas evidence suggest- Other studies investigating correlations between intel-
ing the bias comes late in reasoning (suggestive of slower lectual ability and performance on syllogistic reasoning
System 2 processes) would trace the bias to analytic pro- have found that the number of models and alternative
cesses and therefore would be inconsistent with Crite- conclusions generated were more reliable predictors of
rion S. performance (Newstead, Thompson, & Handley, 2002;
Evans (Evans et al., 1983; Evans & Pollard, 1990) found Torrens, Thompson, & Cramer, 1999). Newstead et al.
that participants tended to check the conclusion of a syllo- (2002) used a range of measures (e.g., rationality, expe-
gism, and according to its believability, would accept it rientiality, creativity, generation of alternatives, motiva-
without further evaluation. Only when it mismatched their tion) and found that the number of alternative models
own beliefs would they scrutinize the conclusion. These falsifying a conclusion generated reliably predicted per-
findings were explained in terms of the selective scrutiny formance on syllogistic tasks, which is consistent with
two-stage theory (Evans, 1989; Evans & Pollard, 1990) findings reported by Galotti et al. (1986). Newstead et al.
and satisfy Criterion S. The theory proposes that the bias (2002) claimed that task variables such as instructions,
occurs early in the encoding stage and focuses reasoners’ content, and the complexity of the syllogism affected in-
attention on the conclusion rather than on the premises, dividuals’ willingness to produce alternatives. Their
thus preceding and preempting analytical reasoning. findings suggest that the number of alternatives gener-
Johnson-Laird and Byrne’s (1991) mental models the- ated was more reliable as an index of performance than
ory claims that solving syllogisms involves the genera- measures of cognitive ability.
tion of models (mental analogues of the premises) from
which conclusions are generated and evaluated accord- Implicit Versus Explicit Processing
ing to their consistency with the models. Johnson-Laird Evans et al.’s (1983) study investigated whether the
and Byrne proposed that the belief bias effect is pro- conflicts between heuristic and analytic processes found
duced at the final response stage of the reasoning pro- in Wason and Evans’s (1975) study of the selection task
cess, where believability influences the generation and were also present in syllogistic reasoning. They coded
evaluation of the conclusion (e.g., Markovits & Nantel, participants’ protocols according to the different possi-
1989; Oakhill, Garnham, & Johnson-Laird, 1990; Oakhill ble combinations of references that individuals made to
& Johnson-Laird, 1985). conclusions and premises of the syllogisms they were
Newstead, Pollard, Evans, and Allen (1992) investi- presented. Evans et al. (1983) concluded that those show-
gated the predictions of three theories: selective scrutiny, ing a belief bias often rationalized their responses by re-
mental models, and their own misinterpretation neces- ferring to irrelevant information, lending support to their
sity model, which predicted that the bias occurs late in dual-process account.
the response stage. Their results strongly supported the Other studies (e.g., Bucciarelli & Johnson-Laird,
mental models approach. They found that participants 1999; Ford, 1995) have found that participants’ proto-
did evaluate the arguments; however, poor understand- cols are consistent with behavioral evidence of their
ing of logical necessity, combined with logically inde- mental processes, suggesting that individuals have in-
terminate conclusions, led to reliance on beliefs and fail- sight into their reasoning processes. Galotti et al. (1986)
ure to generate alternative examples, which would have found that the protocols from three groups of reasoners
helped assess the validity of the conclusion. (poor, good, expert) provided direct evidence of differ-
Many of the findings from studies of belief bias have ences in their performance. Participants were able to re-
indicated that when solving contentful syllogisms, par- port the number of deduction rules they possessed and
ticipants begin with some rudimentary analytical pro- the type of misinterpretations of the premises they had,
cessing, and that the source of the bias is located at the both of which corresponded to their level of performance.
evaluation stage of reasoning. This contrasts with one of In addition, Ford found that verbal protocols, along with
the conditions set by Criterion S, which proposes that for the diagrams that participants constructed, indicated the
believable conclusions associative processes are domi- types of cognitive styles (i.e., visual or verbal representa-
nant and fast, which is why they precede analytically tions of premises) that they developed to solve syllogisms.
based evaluations of the problem.
Neuroanatomical Differences in
Individual Differences Syllogistic Reasoning
In their studies of individual differences, Stanovich Deglin and Kinsbourne (1996) examined neurologically
and West (Sá, West, & Stanovich, 1999; Stanovich, 1999; impaired patients’ (i.e., schizophrenics and manic depres-
Stanovich & West, 1998c) found that measures of intel- sives) performance on contentful syllogisms. Each pa-
ligence were a strong predictor of performance on con- tient was tested three times, once before ECT (electro-
tentful syllogisms. Participants making belief bias er- convulsive therapy), then once with ECT-induced RH and
rors, which Stanovich and West claim are generated by LH suppression. Prior to ECT, and with RH suppression,
System 1, scored lower on tests of general intelligence, 86% responded with logically correct answers. When the
 DUAL-PROCESS THEORIES OF REASONING 1003

LH was suppressed, 79% responded correctly. Deglin their relation to the main issues that distinguish dual-
and Kinsbourne concluded that the two hemispheres process theories and the DGC framework.
supported separate reasoning mechanisms. The LH was Criterion S. Dual-process theorists (Sloman, 1996;
claimed to mediate formal logical operations akin to Stanovich & West, 2000) claim that certain tasks (e.g.,
System 2 type processing, consistent with Golding’s abstract selection task, the standard Linda problem) gen-
(1981) findings, whereas the RH was dominant in context- erate conflicts that result in the two reasoning systems
dependent reasoning and was unable to decouple belief- cuing contradictory responses. It is said that the two sys-
based knowledge associated with System 1 processing. tems act simultaneously and that the faster of the two
Goel’s (Goel, Buchel, Frith, & Dolan, 2000; Goel & generates the first response, and the second system, which
Dolan, 2001; Goel, Gold, Kapur, & Houle, 1997) neuro- is slower to act, generates the correct response. Evidence
imaging studies lend further support to Deglin and Kins- from early tutoring studies of the selection task, which is
bourne’s (1996) claims. Normal participants were asked claimed to satisfy Criterion S, is alternatively explained
to assess the validity of contentful and abstract syllo- by tutoring methods failing to take into account the dif-
gisms under time pressure conditions. Goel et al. (2000) ferent types of erroneous beliefs that participants have
found that when participants solved contentful syllo- (e.g., Stenning & van Lambalgen, 1999).
gisms, brain regions in the LH were activated (e.g., ven- In the Linda problem, which also satisfies Criterion S,
tral networks involving temporal and frontal regions); conflicts are thought to arise between the correct under-
these have been found to be associated with syntactic standing and the initial highly compelling response, which
and semantic processing. In participants solving abstract still exerts its influence over participants’ later under-
syllogisms, Goel et al. (2000) found patterns of right standings of the task (e.g., Epstein et al., 1999; Tversky
parietal activity associated with spatial processing, sug- & Kahneman, 1983). Alternatively, the DGC framework
gesting that they had represented and manipulated spa- proposes that many individuals find some rule of inference
tial information. that takes the most salient premises first from which to
When comparing the activation of brain regions dur- form an answer, and with time arrive at the correct an-
ing probabilistic and deductive reasoning, Osherson’s swer. This is achieved by recombining the premises to
(Osherson et al., 1998; Parsons & Osherson, 2001) PET form a different or more complex set of inferences that
studies found that RH brain regions were exclusively ac- leads to the correct answer. The first answer is still highly
tive during deductive reasoning, and LH regions were influential because it is simple and is based on the most
uniquely activated when reasoning with probabilistic prominent features of the task, which are more likely
material. Parsons and Osherson propose that the large in- generalizable. For example, Sides et al.’s (2002) study,
consistencies between their findings and those of Goel which provides convincing evidence for dual-process ac-
are the result of differences in the stimuli and in the in- counts of the conjunction fallacy, shows that individuals
structions of the syllogistic tasks used. Participants in have difficulty overlooking their initial erroneous belief.
their study were presented with a series of contentful de- Even when sources of ambiguity from the instructions
ductive and probabilistic syllogisms and asked to evalu- were removed, the fallacy was found to generalize to
ate their validity. Reasoning on probabilistic syllogisms wider expressions of “and” than that of its original for-
activated brain regions (e.g., left inferior formal areas, mulation, to the extent that participants bet on the con-
parahippocampal and medial temporal areas) associated junctive even when there was a monetary incentive not to
with the recall and evaluation of a range of semantic do so. Here, the initial beliefs have become strongly rep-
knowledge, whereas deductive reasoning activated re- resented, which is why, when reconsidering the task, par-
gions (e.g., right basal ganglia, thalamus, and frontal ticipants cannot fail to revisit their initial inferences be-
cortex) associated with the transformation of mental rep- cause these are more distinctive than the more complex
resentations, working memory, executive strategy, and specialized chain of inferences required to achieve the
rule-based learning. correct solution (J. R. Anderson, 1993; Holland, Holy-
Studies, both clinical and imaging, suggest that each oak, Nisbett, & Thagard, 1989).
hemisphere performs functionally distinct types of rea- Criterion S assumes that System 1 generates the initial
soning, but the conflicting findings described here imply erroneous belief. But there are examples of more than
that using this evidence to adjudicate between single- one type of initially formed erroneous belief, and these
and dual-process theories is still premature. can arise from what can be viewed as analytical type rea-
soning (e.g., Bucciarelli & Johnson-Laird, 1999; Morier
Summarizing the Evidence & Borgida, 1984; Stenning & van Lambalgen, 1999;
This article began by discussing the distinctions that Yates & Carlson, 1986). The advantage of the DGC frame-
reasoning theorists have proposed. It also introduced a work is that it does not differentiate between types of er-
new framework that could alternatively account for differ- roneous understandings according to whether they are
ences in forms of reasoning and then presented evidence heuristic or analytic. The same underlying production
from tasks that have supported dual-process theorists’ rule can generate a variety of erroneous responses as a
claims. The aim of this final section is to summarize and result of the features of the task that an individual consid-
draw together key evidence from these tasks and discuss ers relevant. Moreover, because of this, the DGC frame-
1004 OSMAN

work accommodates individuals’ different initial erro- and skilled reasoners suggest that in any given task, some
neous construals that can be analytic (Stenning & van demonstrate knowledge that is highly familiarized and
Lambalgen, 1999; Yates & Carlson, 1986) as well as results in rapid responses. Galotti et al. (1986) reported
heuristic (Morier & Borgida, 1984), and in turn, accounts that experts performed faster than poor and good rea-
for the successes of tutoring studies that identify and cor- soners. Furthermore, studies of expertise in highly com-
rect these different construals, the benefits of which trans- plex task domains (e.g., forecasting probable odds in a
fer to similar and different task domains (e.g., Agnoli & betting environment) suggest that individuals utilize a
Krantz, 1989; Osman, 2002; Stenning & van Lambalgen, type of analytical processing that is uncorrelated with mea-
2001). sures of intelligence (e.g., Ceci & Liker, 1986; Ericsson,
Criterion S stresses that there are many examples of Krampe, & Tesch-Römer, 1993) and makes little demand
reasoning tasks that cause conflicts of interest between of executive functions.
initial and later inferences. Such a view highlights the Clearly, there is evidence to suggest that measures of
fact that individuals select out particular features of a performance on reasoning tasks correlate with tests of
task to test, and that in many cases particular features of cognitive ability (e.g., Guilford, 1959; Klaczynski et al.,
a task are found to be more salient than others. This pro- 1996; Stanovich & West, 1998b), but they can be inter-
vides important grounds for categorizing different rea- preted as relating to differences in degree rather than in
soning tasks. However, there is evidence suggesting that the kind of reasoning system used.
the conflicts identified may not necessarily arise from Implicit versus explicit processing. One of a number
distinct reasoning systems but can instead be interpreted of empirical routes to examining dissociations between
as a result of the varying strengths of the representations implicit and explicit forms of reasoning has been to use
utilized while reasoning. protocol measures (e.g., Evans et al., 1983; Wason &
Individual differences. Stanovich and West (2000) Evans, 1975). These are designed to show a lack of cor-
predict that in tasks where the two systems cue different respondence between verbal/written reports and behav-
responses, high scores of cognitive ability suggest the ioral data. There is much supporting evidence of disso-
use of analytic processes. By contrast, lower scores of ciations of this kind in studies of the selection task (e.g.,
cognitive ability suggest that individuals are reliant on Evans & Wason, 1976; Wason & Evans, 1975) and syl-
heuristic processes that often lead to incorrect responses. logistic reasoning (e.g., Evans et al., 1983). However,
These claims have been supported by a number of stud- others have shown that protocols are a good index of the
ies (e.g., Klaczynski, Gordon, & Fauth, 1996; Stanovich type of knowledge that individuals bring to bear on a task
& West, 1998b, 1998c). The DGC framework proposes (e.g., Galotti et al., 1986; Goodwin & Wason, 1972;
that what differentiates these groups is the amount of Stenning & van Lambalgen, 1999).
production rules and in turn, their complexity, and not One crucial difference between studies that report a
whether heuristic and analytic forms of reasoning are ap- correspondence and those that do not is that cues were
plied. Studies of deductive reasoning (e.g., Newstead often present (e.g., Ford, 1995; Hoch & Tschirgi, 1983),
et al., 2002; Torrens et al., 1999) suggest that alternative and these may prompt the individual in such a way as to
measures (e.g., the number of counterexamples to initial more accurately articulate his or her knowledge. In addi-
conclusions generated and evaluated) predict perfor- tion, online concurrent reports (e.g., Berry, 1983; Ford,
mance above and beyond measures of cognitive ability. 1995; Stenning & van Lambalgen, 1999) are made in the
Similarly, Galotti et al.’s (1986) findings suggest that the presence of many task cues, which may also explain why
frequency of alternative conclusions generated in syllo- they more accurately correspond to behavioral data than
gistic reasoning tasks reliably indexed individuals with do retrospective reports.
differing levels of reasoning expertise. Instead of appealing to functionally distinct systems,
Stanovich and West (2000) also predict that System 2, these findings can be interpreted as showing that partic-
identified by high scores of cognitive ability (e.g., SAT), ipants vary as to the insight they have into their own rea-
is slow and very cognitively demanding, whereas Sys- soning, and with enough cuing a correspondence can be
tem 1, identified by poor scores of cognitive ability, is found. Such an interpretation does not deny that knowl-
rapid and does not drain cognitive resources (e.g., Ep- edge unavailable to conscious awareness is therefore un-
stein, Pacini, Denes-Raj, & Heier, 1996; Stanovich & reportable, as described by Siegler and Stern (1998), but
West, 1998a, 1998b, 1998c). The DGC framework pro- suggests that some studies that have demonstrated dis-
poses that the different types of reasoning are dependent sociations may not have exhausted all efforts to uncover
on the representations from which participants reason, what knowledge is possessed.
and this is not reliably differentiated according to the de- Tutoring studies, which include transfer tests, provide
mands made on the central executive. For example, the some scope for examining knowledge thought to be in-
DGC proposes that skilled reasoners demonstrate a type accessible to the reasoner. Early studies suggested that
of automatic reasoning that is dependent on well-formed heuristic processing is insensitive to tutoring (e.g., Wason
representations of highly specialized rules; these have & Golding, 1974; Wason & Shapiro, 1971) and that erro-
acquired strength and become attuned to the circum- neous responses result from preconscious and inflexible
stances in which they apply. Comparisons between poor processes. Often in these examples, individuals returned to
 DUAL-PROCESS THEORIES OF REASONING 1005

a default incorrect response strategy (e.g., Wason, 1969). and experts should differ in their ability to accurately re-
Other demonstrations of differences between heuristic and spond because their representations of the task differ ac-
analytic processing (e.g., inspection paradigm) provide cording to whether they are implicit or automatic, whereas
evidence of differences in terms of speed of processing they should be equally poor if there is no distinction be-
(e.g., Ball et al., 2003; Evans et al., 1987). tween “automatic” and “implicit” representations.
However, recent tutoring studies were able to facili- Neuroanatomical differences. Some dual-reasoning
tate correct reasoning, which transferred to variants of theorists (Evans & Over, 1996; Stanovich & West, 2000)
the original task (e.g., Fong & Nisbett, 1991; Stenning & claim that different neural pathways subserve two func-
van Lambalgen, 1999). These findings suggest that pro- tionally distinct reasoning systems. There is, however,
cesses labeled implicit by dual theorists (e.g., Evans & some disagreement as to which hemispheric region is as-
Over, 1996; Stanovich & West, 2000) are accessible to sociated with particular forms of reasoning. For exam-
the reasoner and can be modified. Moreover, studies di- ple, clinical studies that examine patients with LH and
rectly contrasting implicit and explicit problem solving RH damage (Golding, 1981) or that induce suppression
suggest that they cannot be reliably identified according of these regions (Deglin & Kinsbourne, 1996) suggest
to the differential demands made on working memory that the LH is dominant in abstract deductive reasoning
(e.g., P. J. Reber & Kotovsky, 1997). and that the RH is associated with content-based rea-
Dual-process theorists (i.e., Evans & Over, 1996; soning (e.g., Goel et al., 2000) and probabilistic infer-
Stanovich & West, 2000) claim that implicit and auto- encing (e.g., Wharton & Grafman, 1998). Studies di-
matic forms of reasoning are examples of System 1, and rectly comparing probabilistic and deductive reasoning
so both are characterized as inaccessible. However, there have found that there is a logic-specific network in both
is evidence showing that automatic reasoning is analytic the LH and RH and that probabilistic reasoning activated
and accessible (e.g., Galotti et al., 1986), and that ana- mostly LH regions (Osherson et al., 1998; Parsons & Osh-
lytic reasoning is implicit and inaccessible to the rea- erson, 2001), which conflicts with other attempts to lo-
soner (e.g., Siegler & Stern, 1998). Both are examples of calize distinct reasoning mechanisms.
analytic processing and are inconsistent with the dualist Neural localization of different reasoning mechanisms
proposal that heuristic processes are implicit. The DGC is not tantamount to evidence for qualitatively distinct
suggests that what distinguishes implicit and automatic reasoning systems (Sherry & Schacter, 1987). Evidence
reasoning is that for the latter, individuals can demon- from both neuroimaging and clinical studies suggests
strate metaknowledge, which can be reliably reproduced that different framings (e.g., familiar vs. unfamiliar con-
because the representations utilized are strong, stable, text; concrete vs. abstract) of reasoning tasks such as syl-
and distinctive, whereas in the former, individuals lack logisms require functionally distinct forms of reasoning
metaknowledge of their behavior, which cannot be reli- and are handled by distinct regions of the brain. It is im-
ably reproduced. portant to ask whether the rules of operation differ or are
Clearly, further work is needed to disentangle differ- the same for each distinct region. For example, Goel et al.
ences between implicit and automatic forms of reason- (2000) found that there were no behavioral differences
ing. One way of examining these distinctions might in- between response latencies for content and noncontentful
volve comparing practiced with novice reasoners by syllogisms, and performance was also no different. How-
using a dual-task design. Typically, this design involves ever, there were distinct patterns of activation depending
presenting two tasks simultaneously, one that makes on the form that the syllogism took. Obviously, different
high demands on attention (e.g., counting back in 3 sec) representations are being formed when processing abstract
and a second problem-solving or reasoning task. One and contentful material, but it is not clear how to distin-
variant might involve repeating the task after a given in- guish the possibility that each region operates under a
terval (such as a day); if automatic and implicit knowl- different and entirely separate reasoning system, and
edge is the same, both groups will reproduce the same whether information is represented and stored differ-
responses, since implicit and automatic knowledge is ently but processed via a common reasoning mechanism.
highly inflexible. However, if there are differences be-
tween automatic and implicit knowledge, the novice Conclusions
group will not reliably reproduce the same response after The types of distinctions that dual-process theorists
an interval of time because they have developed unstable propose and the evidence that has been supportive of
and indistinct representations, whereas the expert group their claims have been evaluated in this review. Paralleling
will because they have formed strong and enduring rep- this, the review introduces the DGC framework, which
resentations. Furthermore, as an alternative to using ver- describes how different types of reasoning come about
bal protocols as tests of awareness, participants could be through the graded properties of the representations that
presented with a specific problem state they had already are generated while reasoning. These different approaches
experienced and be asked to predict the next move (this have been contrasted, and the review has sought to ex-
would measure the distinctiveness of their representa- amine whether the types of processes identified by dual-
tions) or recall a sequence of moves made (this would process theorists can be unified within a single-system
measure the strength of their representations). Novices framework that does not appeal to dissociable reasoning
1006 OSMAN

mechanisms. In so doing, it proposes that evidence used Bucciarelli, M. (2000). Reasoning strategies in syllogisms: Evidence
to support dual theories is consistent with single-system for performance errors along with computational limitations. Behav-
accounts. Moreover, the review suggests that certain ioral & Brain Sciences, 23, 669-670.
Bucciarelli, M., & Johnson-Laird, P. N. (1999). Strategies in syllo-
types of reasoning phenomena have been misclassified gistic reasoning. Cognitive Science, 23, 247-303.
by dual-process theorists as implicit, and should instead Burns, B. D., & Vollmeyer, R. (2002). Goal specificity effects on hy-
be classed as automatic reasoning. The grounds for mak- pothesis testing in problem solving. Quarterly Journal of Experi-
ing these claims arise from the DGC framework that as- mental Psychology, 55A, 241-261.
Ceci, S. J., & Liker, J. K. (1986). A day at the races: A study of IQ, ex-
signs different functional roles to consciousness and not pertise, and cognitive complexity. Journal of Experimental Psychol-
the reasoning system itself. ogy: General, 115, 255-266.
What the DGC is able to contribute to the reasoning Cheesman, J., & Merikle, P. M. (1984). Priming with and without
field is that it permits organization and understanding of awareness. Perception & Psychophysics, 36, 387-395.
reasoning phenomena within a more plausible single- Cheng, P. W., & Holyoak, K. J. (1985). Pragmatic schemas. Cognitive
Psychology, 17, 391-416.
system framework for thinking about mental representa- Cheng P. W., Holyoak, K. J., Nisbett, R. E., & Oliver, L. M. (1986).
tion. Crucially, it highlights the ways in which implicit Pragmatic versus syntactic approaches to training deductive reason-
and automatic processing differ but also how they relate ing. Cognitive Psychology, 18, 293-328.
to each other and explicit processing under a single rea- Chrostowski, J. J., & Griggs, R. A. (1985). The effects of problem
content, instructions, and verbalization procedure on Wason’s selec-
soning mechanism. tion task. Current Psychological Research & Reviews, 4, 99-107.
Cleeremans, A. (1997). Principles for implicit learning. In D. Berry
REFERENCES (Ed.), How implicit is implicit learning? New York: Oxford Univer-
sity Press.
Adler, J. E. (1991). An optimist’s pessimism: Conversation and con- Cleeremans, A., Destrebecqz, A., & Boyer, M. (1998). Implicit learn-
junctions. In E. Eells & T. Maruszewski (Eds.), Probability and ra- ing: News from the front. Trends in Cognitive Sciences, 2, 406-416.
tionality: Studies on L. Jonathan Cohen’s philosophy of science Cleeremans, A., & Jiménez, L. (2002). Implicit learning and con-
(pp. 251-282). Amsterdam: Rodopi. sciousness: A graded, dynamical perspective. In R. M. French &
Agnoli, F., & Krantz, D. H. (1989). Suppressing natural heuristics by A. Cleeremans (Eds.), Implicit learning and consciousness: An em-
formal instruction: The case of the conjunction fallacy. Cognitive pirical, philosophical and computational consensus in the making
Psychology, 21, 515-550. (pp. 1-40). Hove, U.K.: Psychology Press.
Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum. Cohen, L. J. (1981). Can human irrationality be experimentally demon-
Anderson, J. R. (1995). Learning and memory. New York: Wiley. strated? Behavioral & Brain Sciences, 4, 317-370.
Anderson, J. R., & Betz, J. (2001). A hybrid model of categorization. Cosmides, L., & Tooby, J. (1996). Are humans good intuitive statisti-
Psychonomic Bulletin & Review, 8, 629-647. cians after all? Rethinking some conclusions from the literature on
Anderson, J. R., & Douglass, S. (2001). Tower of Hanoi: Evidence for judgment under uncertainty. Cognition, 58, 1-73.
the cost of goal retrieval. Journal of Experimental Psychology: Learn- Cosmides, L., & Tooby, J. (2000). Consider the source: The evolution
ing, Memory, & Cognition, 27, 1331-1346. of adaptations for decoupling and metarepresentations. In D. Sperber
Anderson, J. R., & Sheu, C. F. (1995). Causal inferences as perceptual (Ed.), Metarepresentation: A multidisciplinary perspective (pp. 53-
judgments. Memory & Cognition, 23, 510-524. 117). Oxford: Oxford University Press.
Anderson, S. W., Bechara, A., Damasio, H., Tranel, D., & Damasio, Crabb, B. T., & Dark, V. J. (1999). Perceptual implicit memory re-
A. R. (1999). Impairment of social and moral behavior related to early quires attentional encoding. Memory & Cognition, 27, 267-275.
damage in human prefrontal cortex. Nature Neuroscience, 2, 1032-1037. Crandall, C. S., & Greenfield, B. (1986). Understanding the conjunc-
Ball, L. J., Lucas, E. J., Miles, J. N. V., & Gale, A. G. (2003). In- tion fallacy: A conjunction of effects? Social Cognition, 4, 408-419.
spection times and the selection task: What do eye-movements reveal Dawkins, R. (1976). The selfish gene. New York: Oxford University Press.
about relevance effects? Quarterly Journal of Experimental Psychol- Deglin, V. L., & Kinsbourne, M. (1996). Divergent thinking styles of
ogy, 56A, 1052-1077. the hemispheres: How syllogisms are solved during transitory hemi-
Baron, J. (1985). Rationality and intelligence. Cambridge: Cambridge sphere suppression. Brain & Cognition, 31, 285-307.
University Press. Dienes, Z., & Fahey, R. (1998). The role of implicit memory in con-
Benassi, V. A., & Knoth, R. L. (1993). The intractable conjunction fal- trolling a dynamic system. Quarterly Journal of Experimental Psy-
lacy: Statistical sophistication, instructional set, and training. Jour- chology, 51A, 593-614.
nal of Social Behavior & Personality, 8, 83-96. Dienes, Z., & Perner, J. (2003). Unifying consciousness with explicit
Berry, D. (1983). Metacognitive experience and transfer of logical rea- knowledge. In A. Cleeremans (Ed.), The unity of consciousness:
soning. Quarterly Journal of Experimental Psychology, 35A, 39-40. Binding, integration, and dissociation (pp. 214-232). Oxford: Oxford
Berry, D. (1991). The role of action in implicit learning. Quarterly University Press.
Journal of Experimental Psychology, 43, 881-906. Dixon, J. A., & Moore, C. F. (1997). Characterizing the intuitive rep-
Berry, D., & Dienes, Z. (1993). Implicit learning. Hove, U.K.: Erlbaum. resentation in problem solving: Evidence from evaluating mathe-
Betsch, T., & Fiedler, K. (1999). Understanding conjunction effects matical strategies. Memory & Cognition, 25, 395-412.
in probability judgments: The role of implicit mental models. Euro- Dudley, R. E. J., John, C. H., Young, A. W., & Over, D. E. (1997).
pean Journal of Social Psychology, 29, 75-93. Normal and abnormal reasoning in people with delusions. British
Block, N. (1995). On a confusion about a function of consciousness. Journal of Clinical Psychology, 36, 243-258.
Behavioral & Brain Sciences, 18, 227-287. Dulany, D. E. (1991). Conscious representation and thought systems.
Bourne, L. E. (1970). Knowing and using concepts. Psychological Re- In R. S. Wyer, Jr. & T. K. Srull (Eds.), The content, structure, and op-
view, 77, 546-556. eration of thought systems (pp. 97-120). Hillsdale, NJ: Erlbaum.
Braine, M. D. S., & O’Brien, D. P. (1998). The theory of mental propo- Dulany, D. E., & Hilton, D. J. (1991). Conversational implicature,
sitional logic: Description and illustration. In M. D. S. Braine & D. P. conscious representation, and the conjunction fallacy. Social Cogni-
O’Brien (Eds.), Mental logic (pp. 79-89). Mahwah, NJ: Erlbaum. tion, 9, 85-110.
Brase, G. L., Cosmides, L., & Tooby, J. (1998). Individuation, count- Dunwoody, P. T., Haarbauer, E., Mahan, R. P., Marino, C., & Tang,
ing, and statistical inference: The role of frequency and whole-object C. C. (2000). Cognitive adaptation and its consequences: A test of
representations in judgment under uncertainty. Journal of Experi- cognitive continuum theory. Journal of Behavioral Decision Making,
mental Psychology: General, 127, 3-21. 13, 35-54.
 DUAL-PROCESS THEORIES OF REASONING 1007

Epstein, S. (1973). The self-concept revisited: Or a theory of a theory. junction fallacy: Resolving signed summation and the low compo-
American Psychologist, 28, 404-416. nent model in a contingent approach. Acta Psychologica, 94, 1-20.
Epstein, S. (1994). Integration of the cognitive and the psychodynamic Fong, G. T., Krantz, D. H., & Nisbett, R. E. (1986). The effects of
unconscious. American Psychologist, 49, 709-724. statistical training on thinking about everyday problems. Cognitive
Epstein, S., Donovan, S., & Denes-Raj, V. (1999). The missing link Psychology, 18, 253-292.
in the paradox of the Linda conjunction problem: Beyond knowing Fong, G. T., & Nisbett, R. E. (1991). Immediate and delayed transfer
and thinking of the conjunction rule, the intrinsic appeal of heuristic of training effects in statistical reasoning. Journal of Experimental
processing. Personality & Social Psychology Bulletin, 25, 204-214. Psychology: General, 120, 34-45.
Epstein, S., Pacini, R., Denes-Raj, V., & Heier, H. (1996). Individual Ford, M. (1995). Two modes of mental representation and problem so-
differences in intuitive–experiential and analytical–rational thinking lution in syllogistic reasoning. Cognition, 54, 1-71.
styles. Journal of Personality & Social Psychology, 71, 390-405. Freud, S. (1953). The interpretation of dreams. In J. Strachey (Ed. &
Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role Trans.), The standard edition of the complete psychological works of
of deliberate practice in the acquisition of expert performance. Psy- Sigmund Freud (Vols. 4 and 5). London: Hogarth. (Original work
chological Review, 100, 363-406. published 1900)
Ericsson, K. A., & Simon, H. A. (1980). Verbal reports as data. Psy- Galotti, K. M., Baron, J., & Sabini, J. P. (1986). Individual differ-
chological Review, 87, 215-251. ences in syllogistic reasoning: Deduction rules or mental models?
Evans, J. S. B. T. (1972). Interpretation and “matching bias” in a rea- Journal of Experimental Psychology: General, 115, 16-25.
soning task. Quarterly Journal of Experimental Psychology, 24, 193- Geddes, B. W., & Stevenson, R. J. (1997). Explicit learning of a dy-
199. namic system with a non-salient pattern. Quarterly Journal of Ex-
Evans, J. S. B. T. (1984). Heuristic and analytic processes in reasoning. perimental Psychology, 50A, 742-765.
British Journal of Psychology, 75, 451-468. Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian
Evans, J. S. B. T. (1989). Biases in human reasoning: Causes and con- reasoning without instruction: Frequency formats. Psychological Re-
sequences. London: Erlbaum. view, 102, 684-704.
Evans, J. S. B. T. (1995). Relevance and reasoning. In S. E. Newstead Gilinsky, A. S., & Judd, B. B. (1994). Working memory and bias in
& J. S. B. T. Evans (Eds.), Perspectives on thinking and reasoning reasoning across the life span. Psychology & Aging, 9, 356-371.
(pp. 147-172). Hove, U.K.: Erlbaum. Goel, V. (1995). Sketches of thought. Cambridge, MA: MIT Press.
Evans, J. S. B. T. (1996). Deciding before you think: Relevance and Goel, V., Buchel, C., Frith, C., & Dolan, R. J. (2000). Dissociation
reasoning in the selection task. British Journal of Psychology, 87, of mechanisms underlying syllogistic reasoning. NeuroImage, 12,
223-240. 504-514.
Evans, J. S. B. T. (1998). Matching bias in conditional reasoning: Do Goel, V., & Dolan, R. J. (2001). Functional neuroanatomy of three-
we understand it after 25 years? Thinking & Reasoning, 4, 45-82. term relational reasoning. Neuropsychologia, 39, 901-909.
Evans, J. S. B. T. (2000). What could and could not be a strategy in rea- Goel, V., Gold, B., Kapur, S., & Houle, S. (1997). The seats of rea-
soning. In W. Schaeken, D. DeVooght, & A. D. D. Vandierendonck son: A localization study of deductive & inductive reasoning using
(Eds.), Deductive reasoning and strategies (pp. 1-22). Mahwah, NJ: PET (O15) blood flow technique. NeuroReport, 8, 1305-1310.
Erlbaum. Golding, E. (1981). The effect of unilateral brain lesion on reasoning.
Evans, J. S. B. T. (2002). Logical and human reasoning: An assessment Cortex, 17, 31-40.
of the deductive paradigm. Psychological Bulletin, 128, 978-996. Goodwin, R. Q., & Wason, P. C. (1972). Degrees of insight. British
Evans, J. S. B. T., Ball, L. J., & Brooks, P. G. (1987). Attention bias Journal of Psychology, 63, 205-212.
and decision order in a reasoning task. Quarterly Journal of Experi- Green, D. W., & Larking, R. (1995). The locus of facilitation in the
mental Psychology, 51A, 811-814. abstract selection task. Thinking & Reasoning, 1, 183-199.
Evans, J. S. B. T., Barston, J. L., & Pollard, P. (1983). On the con- Griggs, R. A., & Cox, J. R. (1982). The elusive thematic-materials effects
flict between logic and belief in syllogistic reasoning. Memory & in Wason’s selection task. British Journal of Psychology, 73, 407-420.
Cognition, 11, 295-306. Guilford, J. P. (1959). Three faces of intellect. American Psycholo-
Evans, J. S. B. T., & Over, D. E. (1996). Rationality and reasoning. gist, 14, 469-479.
Hove, U.K.: Psychology Press. Hammond, K. R. (1996). Human judgment and social policy. New
Evans, J. S. B. T., & Pollard, P. (1990). Belief bias and problem com- York: Oxford University Press.
plexity in deductive reasoning. In J.-P. Caverni, J.-M. Fabré, & Hammond, K. R., & Summers, D. A. (1972). Cognitive control. Psy-
M. Gonzalez (Eds.), Cognitive biases (pp. 131-154). Amsterdam: El- chological Review, 79, 58-67.
sevier, North-Holland. Hawkins, J., Pea, R. D., Glick, J., & Scribner, S. (1984). “Merds that
Evans, J. S. B. T., & Wason, P. C. (1976). Rationalisation in a reason- laugh don’t like mushrooms”: Evidence for deductive reasoning by
ing task. British Journal of Psychology, 63, 205-212. preschoolers. Developmental Psychology, 20, 584-594.
Farah, M. J. (1994). Visual perception and visual awareness after brain Hayes, N., & Broadbent, D. E. (1988). Two modes of learning for in-
damage: A tutorial overview. In C. Umiltà & M. Moscovitch (Eds.), teractive tasks. Cognition, 28, 249-276.
Attention and performance XV: Conscious and nonconscious infor- Hertwig, R., Ortmann, A., & Gigerenzer, G. (1997). Deductive
mation processing (pp. 37-76). Cambridge, MA: MIT Press. competence: A desert devoid of content and context. Cahiers de Psy-
Farah, M. J., Monheit, M. A., & Wallace, M. A. (1991). Uncon- chologie Cognitive, 16, 102-107.
scious perception of “extinguished” visual stimuli: Reassessing the Hilton, D. J. (1995). The social context of reasoning: Conversational
evidence. Neuropsychologia, 29, 949-958. inferences and rational judgment. Psychological Review, 118, 248-
Farah, M. J., O’Reilly, R. C., & Vecera, S. P. (1993). Dissociated 271.
overt and covert recognition as an emergent property of lesioned at- Hoch, S. J., & Tschirgi, J. E. (1983). Cue redundancy and extra logi-
tractor networks. Psychological Review, 100, 571-588. cal inferences in a deductive reasoning task. Memory & Cognition,
Fiddick, L., Cosmides, L., & Tooby, J. (2000). No interpretation with- 11, 200-209.
out representation: The role of domain-specific representations and Hoch, S. J., & Tschirgi, J. E. (1985). Logical knowledge and cue re-
inferences in the Wason selection task. Cognition, 77, 1-79. dundancy in deductive reasoning. Memory & Cognition, 13, 453-462.
Fiedler, K. (1988). The dependence of the conjunction fallacy on sub- Holender, D. (1986). Semantic activation without conscious activa-
tle linguistic factors. Psychological Research, 50, 123-129. tion in dichotic listening, parafoveal vision, and visual masking: A
Fillenbaum, S. (1976). Inducements: On the phrasing and logic of survey and appraisal. Behavioral & Brain Sciences, 9, 1-66.
conditional promises, threats, and warnings. Psychological Research, Holland, J. H., Holyoak, K. J., Nisbett, R. E., & Thagard, P. R.
38, 231-250. (1989). Induction: Processes of inference, learning and discovery.
Fisk, J. E., & Pidgeon, N. (1996). Component probabilities and the con- Cambridge, MA: MIT Press.
1008 OSMAN

Houde, O., Zago, L., Crivello, F., Moutier, S., Pineau, A., Ma- cial psychology. In G. Lindzey & E. Aronson (Eds.), Handbook of so-
zoyer, B., & Tzourio-Mazoyer, N. (2001). Access to deductive cial psychology (3rd ed.). New York: Random House.
logic depends on a right ventromedial prefrontal area devoted to emo- Mathis, W. D., & Mozer, M. C. (1996, July). Conscious and uncon-
tion and feeling: Evidence from a training paradigm. NeuroImage, scious perception: A computational theory. Paper presented at the
14, 1486-1492. 18th Annual Conference of the Cognitive Science Society, San Diego.
Huq, S. F., Garety, P. A., & Hemsley, D. R. (1988). Probabilistic Meier, B., & Perrig, W. J. (2000). Low reliability of perceptual prim-
judgements in deluded and non-deluded subjects. Quarterly Journal ing: Consequences for the interpretation of functional dissociations
of Experimental Psychology, 40A, 801-812. between explicit and implicit memory. Quarterly Journal of Experi-
Jacoby, L. L., Levy, B. A., & Steinbach, K. (1992). Episodic transfer mental Psychology, 53A, 211-233.
and automaticity: Integration of data-driven and conceptually driven Morier, D. M., & Borgida, E. (1984). The conjunction fallacy: A task
processing in rereading. Journal of Experimental Psychology: Learn- specific phenomenon? Personality & Social Psychology Bulletin, 10,
ing, Memory, & Cognition, 18, 15-24. 243-252.
James, W. (1950). The principles of psychology. New York: Dover. Moshman, D. (1999). Adolescent psychological development: Ratio-
(Original work published 1890) nality, morality, and identity. Mahwah, NJ: Erlbaum.
Johnson-Laird, P. N., & Byrne, R. M. J. (1991). Deduction. Hillsdale, Moshman, D. (2000). Diversity in reasoning and rationality: Meta-
NJ: Erlbaum. cognitive and developmental considerations. Behavioral & Brain Sci-
Johnson-Laird, P. N., & Wason, P. C. (1970). A theoretical analysis of ences, 23, 689-690.
insight into a reasoning task. Cognitive Psychology, 1, 134-148. Newstead, S. E. (2000). Are there two different types of thinking? Be-
Kahneman, D., & Frederick, S. (2002). Representativeness revisited: havioral & Brain Sciences, 23, 690-691.
Attribute substitution in intuitive judgment. In T. Gilovich, D. Griffin, Newstead, S. E., Handley, S. J., Harley, C., Wright, H., & Far-
& D. Kahneman (Eds.), Heuristics and biases: The psychology of in- relly, D. (2004). Individual differences in deductive reasoning.
tuitive judgment. Cambridge: Cambridge University Press. Quarterly Journal of Experimental Psychology, 57A, 33-60.
Karmiloff-Smith, A. (1986). From meta-processes to conscious ac- Newstead, S. E., Pollard, P., Evans, J. S. B. T., & Allen, J. L.
cess: Evidence from children’s metalinguistic and repair data. Cog- (1992). The source of belief bias effects in syllogistic reasoning. Cog-
nition, 23, 95-147. nition, 45, 257-284.
Karnath, H. O. (1988). Deficits in attention in acute and recovered vi- Newstead, S. E., Thompson, V. A., & Handley, S. J. (2002). Gener-
sual hemi-neglect. Neuropsychologia, 26, 27-43. ating alternatives: A key component in human reasoning? Memory &
Kemp, R., Chua, S., McKenna, P., & David, A. (1997). Reasoning and Cognition, 30, 129-137.
delusions. British Journal of Psychiatry, 170, 398-405. Nisbett, R. E., Krantz, D. H., Jepson, C., & Kunda, Z. (1983). The
Kirby, K. N. (1994). Probabilities and utilities of fictional outcomes in use of statistical heuristics in everyday inductive reasoning. Psycho-
Wason’s four-card selection task. Cognition, 5, 1-28. logical Review, 90, 339-363.
Klaczynski, P. A. (2001). Analytical and heuristic processing influ- Nisbett, R. E., & Ross, L. (1980). Human inference: Strategies and
ences on adolescent reasoning and decision making. Child Develop- shortcomings of social judgment. Englewood Cliffs, NJ: Prentice-Hall.
ment, 72, 844-861. Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can
Klaczynski, P. A., Gelfand, H., & Reese, H. W. (1989). Transfer of know: Verbal reports on mental processes. Psychological Review, 84,
conditional reasoning: Effects of explanations and initial problem 231-259.
types. Memory & Cognition, 17, 208-220. Oakhill, J., Garnham, A., & Johnson-Laird, P. N. (1990). Belief
Klaczynski, P. A., Gordon, D. H., & Fauth, J. (1996). Goal-oriented bias effects in syllogistic reasoning. In K. J. Gilhooly, M. T. G. Keane,
reasoning and individual differences in critical reasoning biases. R. H. Logie, & G. Erdos (Eds.), Lines of thinking: Reflections on the
Journal of Educational Psychology, 89, 470-485. psychology of thought (pp. 125-139). Chichester, U.K.: Wiley.
Klayman, J., & Ha, Y. W. (1987). Confirmation, disconfirmation and Oakhill, J., & Johnson-Laird, P. N. (1985). The effects of belief bias
information in hypothesis testing. Psychological Review, 94, 211-228. on the production of syllogistic conclusions. Quarterly Journal of Ex-
Knauff, M., Mulack, T., Kassubek, J., Salih, H. R., & Greenlee, perimental Psychology, 37A, 553-569.
M. W. (2002). Spatial imagery in deductive reasoning: A functional Oberauer, K. (2000). Do we need two systems for reasoning? Behav-
MRI study. Cognitive Brain Research, 13, 203-212. ioral & Brain Sciences, 23, 692-693.
Koehler, J. J. (1996). The base rate fallacy reconsidered: Descriptive, O’Brien, G., & Opie, J. (1999). A connectionist theory of phenomenal
normative and methodological challenges. Behavioral & Brain Sci- experience. Behavioral & Brain Sciences, 22, 127-196.
ences, 19, 1-53. O’Reilly, R. C., & Munakata, Y. (2000). Computational explorations
Kuhn, D. (2000). Metacognitive development. Current Directions in in cognitive neuroscience: Understanding the mind by simulating the
Psychological Science, 9, 178-181. brain. Cambridge, MA: MIT Press.
Kunimoto, C., Miller, J., & Pashler, H. (2001). Confidence and ac- Osherson, D., Perani, D., Cappa, S., Schnur, T., Grassi, F., & Fazio, F.
curacy of near-threshold discrimination responses. Consciousness & (1998). Distinct brain loci in deductive versus probabilistic reasoning.
Cognition, 10, 294-340. Neuropsychologia, 36, 369-376.
Levinson, S. C. (1995). Interactional biases in human thinking. In Osman, M. (2002, August). Is there evidence for unconscious reason-
E. Goody (Ed.), Social intelligence and interaction (pp. 221-260). ing processes? Proceedings of the 24th Annual Meeting of the Cog-
Cambridge: Cambridge University Press. nitive Science Society, Washington, DC.
MacDonald, P., & MacLeod, C. M. (1998). The influence of attention Osman, M., & Laming, D. (2001). Misinterpretation of conditional
at encoding on direct and indirect remembering. Acta Psychologica, statements in Wason’s selection task. Psychological Research, 65,
98, 298-310. 128-144.
Manktelow, K. I., & Over, D. E. (1995). Deontic reasoning. In S. E. Over, D., & Evans, J. S. B. T. (1997). Two cheers for deductive com-
Newstead & J. S. B. T. Evans (Eds.), Perspectives on thinking and petence. Current Psychology of Cognition, 16, 255-278.
reasoning (pp. 91-114). Hove, U.K.: Erlbaum. Pacini, R., & Epstein, S. (1999). The relation of rational and experi-
Margolis, H. (1987). Patterns, thinking, and cognition. Chicago: Uni- ential processing styles to personality, basic beliefs, and the ratio–
versity of Chicago Press. bias phenomenon. Journal of Personality & Social Psychology, 76,
Margolis, H. (2000). Wason’s selection task with a reduced array [On 972-987.
line] . Psycoloquy, 11 (5). Parsons, L. M., & Osherson, D. (2001). New evidence for distinct
Markovits, H., & Nantel, G. (1989). The belief-bias effect in the pro- right and left brain systems for deductive versus probabilistic rea-
duction and evaluation of logical conclusions. Memory & Cognition, soning. Cerebral Cortex, 11, 954-965.
17, 11-17. Perruchet, P., & Baveux, P. (1989). Correlational analyses of explicit
Markus, H., & Zajonc, R. B. (1985). The cognitive perspective in so- and implicit memory performance. Memory & Cognition, 17, 77-86.
 DUAL-PROCESS THEORIES OF REASONING 1009

Perruchet, P., & Vinter, A. (2002). The self-organising consciousness: Sloman, S. A. (2002). Two systems of reasoning. In T. Gilovich, D. Grif-
A framework for implicit learning. In R. M. French & A. Cleeremans fin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of
(Eds.), Implicit learning and consciousness: An empirical, philo- intuitive judgment (pp. 379-398). Cambridge: Cambridge University
sophical and computational consensus in the making (pp. 41-67). Press.
Hove, U.K.: Psychology Press. Sloman, S. A., Over, D., Slovak, L., & Stibel, J. M. (2003). Fre-
Platt, R. D., & Griggs, R.A. (1995). Facilitation and matching bias in quency illusions and other fallacies. Organizational Behavior &
the abstract selection task. Thinking & Reasoning, 1, 55-77. Human Decision Processes, 91, 296-309.
Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Smith, E. E., & Sloman, S. A. (1994). Similarity- versus rule-based
Experimental Psychology, 118, 219-235. categorization. Memory & Cognition, 22, 377-386.
Reber, A. S. (1993). Implicit learning and tacit knowledge: An essay Smolensky, P. (1988). On the proper treatment of connectionism. Be-
on the cognitive unconscious. New York: Oxford University Press. havioral & Brain Sciences, 11, 1-23.
Reber, P. J., & Kotovsky, K. (1997). Implicit learning in problem solv- Sperber, D., Cara, F., & Girotto,V. (1996). Relevance theory ex-
ing: The role of working memory capacity. Journal of Experimental plains the selection task. Cognition, 57, 31-95.
Psychology: General, 126, 178-203. Stanovich, K. E. (1999). Who is rational? Studies of individual differ-
Reeves, T., & Lockhart, R. S. (1993). Distributed versus singular ap- ences in reasoning. Mahwah, NJ: Erlbaum.
proaches to probability and errors in probabilistic reasoning. Journal Stanovich, K. E., & West, R. F. (1998a). Individual differences in
of Experimental Psychology: General, 122, 207-226. framing and conjunction effects. Thinking & Reasoning, 4, 289-317.
Reingold, E. M., Charness, N., Pomplun, M., & Stampe, D. M. Stanovich, K. E., & West, R. F. (1998b). Individual differences in ra-
(2001). Visual span in expert chess players: Evidence from eye move- tional thought. Journal of Experimental Psychology: General, 127,
ments. Psychological Science, 12, 48-55. 161-188.
Reingold, E. M., Charness, N., Schultetus, R. S., & Stampe, D. M. Stanovich, K. E., & West, R. F. (1998c). Who uses base rates and
(2001). Perceptual automaticity in expert chess players: Parallel en- P(D/~H)? An analysis of individual differences. Memory & Cogni-
coding of chess relations. Psychonomic Bulletin & Review, 8, 504- tion, 26, 161-179.
510. Stanovich, K. E., & West, R. F. (2000). Individual differences in rea-
Reyna, V. F., & Brainerd, C. J. (1990). Fuzzy processing in transitiv- soning: Implications for the rationality debate? Behavioral & Brain
ity development. Annals of Operations Research, 23, 37-63. Sciences, 23, 645-665.
Rips, L. J. (1989). Similarity, typicality, and categorization. In S. Vosni- Stanovich, K. E., & West, R. F. (2002). Individual differences in rea-
adou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 21- soning: Implications for the rationality debate? In T. Gilovich, D. Grif-
59). Cambridge: Cambridge University Press. fin, & D. Kahneman (Eds.), Heuristics and biases: The psychology
Roberts, M. J. (1997). On dichotomies and deductive reasoning re- of intuitive judgment (pp. 421-440). Cambridge: Cambridge Univer-
search. Current Psychology of Cognition, 16, 196-204. sity Press.
Roberts, M. J. (1998). Inspection times and the selection task: Are they Stenning, K., & van Lambalgen, M. (1999). Is psychology hard or
relevant? Quarterly Journal of Experimental Psychology, 51A, 781- impossible? Reflections on the conditional. In M. M. J. Gerbrandy,
810. M. Marx, M. de Rijke, & Y. Venema (Eds.), Liber amicorum for
Roberts, M. J., & Newton, E. J. (2001). Inspection times, the change Johan van Benthem’s 50th birthday (pp. 1-29). Amsterdam: Amster-
task, and the rapid-response selection task. Quarterly Journal of Ex- dam University Press.
perimental Psychology, 54A, 1031-1048. Stenning, K., & van Lambalgen, M. (2001). Semantics as a founda-
Roediger, H. L., III, & McDermott, K. B. (1995). Creating false mem- tion for psychology: A case study of Wason’s selection task. Journal
ories: Remembering words not presented in lists. Journal of Experi- of Logic, Language, & Information, 10, 273-317.
mental Psychology: Learning, Memory, & Cognition, 21, 803-814. Stevenson, R. J. (1997). Deductive reasoning and the distinction be-
Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed tween implicit and explicit processes. Current Psychology of Cogni-
processing: Explorations in the microstructure of cognition. Vol. 1: tion, 16, 222-229.
Foundations. Cambridge, MA: MIT Press. Stolarz-Fantino, S., Fantino, E., & Kulik, J. (1996). The conjunc-
Sá, W., West, R. F., & Stanovich, K. (1999). The domain specificity tion fallacy: Differential incidence as a function of descriptive frames
and generality of belief bias: Searching for a generalizable critical and educational context. Contemporary Educational Psychology, 21,
thinking skill. Journal of Educational Psychology, 91, 497-510. 208-218.
Schwarz, N. (1996). Cognition and communication: Judgmental bi- Stroop, J. (1935). Studies of inference in serial verbal reactions. Jour-
ases, research methods, and the logic of conversation. Mahwah, NJ: nal of Experimental Psychology, 18, 643-662.
Erlbaum. Torrens, D., Thompson, V. A., & Cramer, K. M. (1999). Individual
Shafir, E., & LeBoeuf, R. (2002). Rationality. Annual Review of Psy- differences and the belief bias effect: Mental models, logical neces-
chology, 53, 491-517. sity, and abstract reasoning. Thinking & Reasoning, 5, 1-28.
Shanks, D. R., & Channon, S. (2002). Effects of a secondary task on Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty:
“implicit” sequence learning: Learning or performance? Psycholog- Heuristics and biases. Science, 185, 1124-1131.
ical Research, 66, 99-109. Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive
Shanks, D. R., & Johnstone, T. (1999). Evaluating the relationship reasoning: The conjunction fallacy in probability judgment. Psycho-
between explicit and implicit knowledge in a sequential reaction time logical Review, 90, 293-315.
task. Journal of Experimental Psychology: Learning, Memory, & Velmans, M. (1991). Is human information processing conscious? Be-
Cognition, 25, 1435-1451. havioral & Brain Sciences, 14, 651-726.
Shanks, D. R., & St. John, M. F. (1994). Characteristics of dissocia- Volpe, D. T., LeDoux, J. E., & Gazzaniga, M. S. (1979). Information
ble human learning systems. Behavioral & Brain Sciences, 17, 367- processing of visual stimuli in an “extinguished” field. Nature, 282,
447. 722-724.
Sherry, D. F., & Schacter, D. L. (1987). The evolution of multiple Wallace, M. A., & Farah, M. J. (1992). Savings in relearning face–name
memory systems. Psychological Review, 94, 439-454. associations as evidence for “covert recognition” in prosopagnosia.
Sides, A., Osherson, D., Bonini, N., & Viale, R. (2002). On the real- Journal of Cognitive Neuroscience, 4, 150-154.
ity of the conjunction fallacy. Memory & Cognition, 30, 191-198. Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Ex-
Siegler, R. S., & Stern, E. (1998). Conscious and unconscious strat- perimental Psychology, 20, 273-281.
egy discoveries: A microgenetic analysis. Journal of Experimental Wason, P. C. (1969). Regression in reasoning. British Journal of Psy-
Psychology: General, 127, 377-397. chology, 60, 471-480.
Sloman, S. A. (1996). The empirical case for two systems of reason- Wason, P. C., & Evans, J. S. B. T. (1975). Dual processes in reasoning?
ing. Psychological Bulletin, 119, 3-22. Cognition, 3, 141-154.
1010 OSMAN

Wason, P. C., & Golding, E. (1974). The language of inconsistency. Wharton, C. M., & Grafman, J. (1998). Deductive reasoning and the
British Journal of Psychology, 65, 537-546. brain. Trends in Cognitive Sciences, 2, 54-59.
Wason, P. C., & Johnson-Laird, P. N. (1970). A conflict between se- Yates, F. J., & Carlson, B. W. (1986). Conjunction errors: Evidence
lecting and evaluating information in an inferential task. British Jour- for multiple judgment procedures, including “signed summation.”
nal of Psychology, 61, 509-515. Organizational Behavior & Human Decision Processes, 37, 230-253.
Wason, P. C., & Shapiro, D. (1971). Natural and contrived experience
in a reasoning problem. Quarterly Journal of Experimental Psychol- (Manuscript received August 8, 2003;
ogy, 23, 63-71. revision accepted for publication December 3, 2003.)