P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD

05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

CHAPTER 1 2

A Model of Heuristic Judgment

Daniel Kahneman
Shane Frederick

The program of research now known as the of a stranger as menacing entails a prediction
heuristics and biases approach began with a of future harm.
study of the statistical intuitions of experts, The ancient idea that cognitive processes
who were found to be excessively confi- can be partitioned into two main families –
dent in the replicability of results from small traditionally called intuition and reason –
samples (Tversky & Kahneman, 1 971 ). The is now widely embraced under the general
persistence of such systematic errors in the label of dual-process theories (Chaiken &
intuitions of experts implied that their intu- Trope, 1 999; Evans and Over, 1 996; Ham-
itive judgments may be governed by funda- mond, 1 996; Sloman, 1 996, 2002; see Evans,
mentally different processes than the slower, Chap. 8). Dual-process models come in
more deliberate computations they had been many flavors, but all distinguish cognitive
trained to execute. operations that are quick and associative
From its earliest days, the heuristics and from others that are slow and governed by
biases program was guided by the idea that rules (Gilbert, 1 999).
intuitive judgments occupy a position – per- To represent intuitive and deliberate rea-
haps corresponding to evolutionary history – soning, we borrow the terms “system 1 ” and
between the automatic parallel operations “system 2” from Stanovich and West (2002).
of perception and the controlled serial op- Although suggesting two autonomous ho-
erations of reasoning. Intuitive judgments munculi, such a meaning is not intended.
were viewed as an extension of percep- We use the term “system” only as a label for
tion to judgment objects that are not cur- collections of cognitive processes that can
rently present, including mental represen- be distinguished by their speed, their con-
tations that are evoked by language. The trollability, and the contents on which they
mental representations on which intuitive operate. In the particular dual-process model
judgments operate are similar to percepts. we assume, system 1 quickly proposes intu-
Indeed, the distinction between perception itive answers to judgment problems as they
and judgment is often blurry: The perception arise, and system 2 monitors the quality of
2 67
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 68 the cambridge handbook of thinking and reasoning

these proposals, which it may endorse, cor- In the context of a dual-system view,
rect, or override. The judgments that are errors of intuitive judgment raise two
eventually expressed are called intuitive if questions: “What features of system 1 cre-
they retain the hypothesized initial proposal ated the error?” and “Why was the error not
with little modification. detected and corrected by system 2?” (cf.
The effect of concurrent cognitive tasks Kahneman & Tversky, 1 982). The first ques-
provides the most useful indication of tion is more basic, of course, but the second
whether a given mental process belongs to is also relevant and ought not be overlooked.
system 1 or system 2. Because the over- Consider, for example, the paragraph that
all capacity for mental effort is limited, ef- Tversky and Kahneman (1 974; p. 3 in
fortful processes tend to disrupt each other, Kahneman, Slovic, & Tversky, 1 982) used to
whereas effortless processes neither cause introduced the notions of heuristic and bias:
nor suffer much interference when com-
bined with other tasks (Kahneman, 1 973 ; The subjective assessment of probability re-
Pashler, 1 998). It is by this criterion that we sembles the subjective assessment of physi-
assign the monitoring function to system 2: cal quantities such as distance or size. These
People who are occupied by a demanding judgments are all based on data of lim-
ited validity, which are processed accord-
mental activity (e.g., attempting to hold in
ing to heuristic rules. For example, the ap-
mind several digits) are much more likely
parent distance of an object is determined
to respond to another task by blurting out in part by its clarity. The more sharply
whatever comes to mind (Gilbert, 1 989). By the object is seen, the closer it appears to
the same criterion, the acquisition of highly be. This rule has some validity, because in
skilled performances – whether perceptual any given scene the more distant objects
or motor – involves the transformation of an are seen less sharply than nearer objects.
activity from effortful (system 2) to effort- However, the reliance on this rule leads to
less (system 1 ). The proverbial chess master systematic errors in the estimation of dis-
who strolls past a game and quips, “White tance. Specifically, distances are often over-
mates in three” is performing intuitively estimated when visibility is poor because
the contours of objects are blurred. On the
(Simon & Chase, 1 973 ).
other hand, distances are often underesti-
Our views about the two systems are
mated when visibility is good because the
similar to the “correction model” proposed objects are seen sharply. Thus the reliance
by Gilbert (1 989, 1 991 ) and to other dual- on clarity as an indication leads to com-
process models (Epstein, 1 994; Hammond, mon biases. Such biases are also found in
1 996; Sloman, 1 996; see also Shweder, intuitive judgments of probability.
1 977). We assume system 1 and system 2
can be active concurrently, that automatic This statement was intended to extend
and controlled cognitive operations compete Brunswik’s (1 943 ) analysis of the percep-
for the control of overt responses, and that tion of distance to the domain of intuitive
deliberate judgments are likely to remain thinking and to provide a rationale for us-
anchored on initial impressions. We also ing biases to diagnose heuristics. However,
assume that the contribution of the two the analysis of the effect of haze is flawed:
systems in determining stated judgments It neglects the fact that an observer looking
depends on both task features and individ- at a distant mountain possesses two relevant
ual characteristics, including the time avail- cues, not one. The first cue is the blur of the
able for deliberation (Finucane et al., 2000), contours of the target mountain, which is
mood (Bless et al., 1 996; Isen, Nygren, & positively correlated with its distance, when
Ashby, 1 988), intelligence (Stanovich & all else is equal. This cue should be given
West, 2002), cognitive impulsiveness (Fred- positive weight in a judgment of distance,
erick, 2004), and exposure to statistical and it is. The second relevant cue, which
thinking (Agnoli, 1 991 ; Agnoli & Krantz, the observer can readily assess by looking
1 989; Nisbett et al., 1 983 ). around, is the ambient or general haziness.
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 69

In an optimal regression model for estimat- ships break up within a year?” may answer
ing distance, general haziness is a suppressor as if she had been asked “Do instances of
variable, which must be weighted negatively failed long-distance relationships come read-
because it contributes to blur but is uncor- ily to mind?” This would be an applica-
related with distance. Contrary to the argu- tion of the availability heuristic. A profes-
ment made in 1 974, using blur as a cue does sor who has heard a candidate’s job talk and
not inevitably lead to bias in the judgment now considers the question “How likely is it
of distance – the illusion could just as well that this candidate could be tenured in our
be described as a failure to assign adequate department?” may answer the much easier
negative weight to ambient haze. The effect question: “How impressive was the talk?”.
of haziness on impressions of distance is a This would be an example of one form of
failing of system 1 : The perceptual system is the representativeness heuristic.
not designed to correct for this variable. The The heuristics and biases research pro-
effect of haziness on judgments of distance gram has focused primarily on representa-
is a separate failure of system 2. Although tiveness and availability – two versatile at-
people are capable of consciously correcting tributes that are automatically computed
their impressions of distance for the effects and can serve as candidate answers to many
of ambient haze, they commonly fail to do different questions. It has also focused prin-
so. A similar analysis applies to some of the cipally on thinking under uncertainty. How-
judgmental biases we discuss later, in which ever, the restriction to particular heuristics
errors and biases only occur when both sys- and to a specific context is largely arbitrary.
tems fail. Kahneman and Frederick (2002) argued that
In the following section, we present this process of attribute substitution is a
an attribute-substitution model of heuris- general feature of heuristic judgment; that
tic judgment, which assumes that difficult whenever the aspect of the judgmental ob-
questions are often answered by substi- ject that one intends to judge (the target at-
tuting an answer to an easier one. This tribute) is less readily assessed than a related
elaborates and extends earlier treatments property that yields a plausible answer (the
of the topic (Kahneman & Tversky, 1 982; heuristic attribute), individuals may unwit-
Tversky & Kahneman, 1 974, 1 983 ). Fol- tingly substitute the simpler assessment. For
lowing sections introduce a research design an example, consider the well-known study
for studying attribute substitution, as well by Strack, Martin, and Schwarz (1 988) in
as discuss the controversy over the repre- which college students answered a survey
sentativeness heuristic in the context of a that included these two questions: “How
dual-system view that we endorse. The final happy are you with your life in general?” and
section situates representativeness within “How many dates did you have last month?”
a broad family of prototype heuristics, in The correlation between the two questions
which properties of a prototypical exemplar was negligible when they occurred in the
dominate global judgments concerning an order shown, but rose to .66 if the dating
entire set. question was asked first. We suggest that the
question about dating frequency automati-
cally evokes an evaluation of one’s romantic
Attribute Substitution satisfaction and that this evaluation lingers
to become the heuristic attribute when the
The early research on judgment heuris- global happiness question is subsequently
tics was guided by a simple and general encountered.
hypothesis: When confronted with a diffi- To further illustrate the process of at-
cult question, people may answer an eas- tribute substitution, consider a question in
ier one instead and are often unaware of a study by Frederick and Nelson (2004):
the substitution. A person who is asked “If a sphere were dropped into a open
“What proportion of long-distance relation- cube, such that it just fit (the diameter
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 70 the cambridge handbook of thinking and reasoning

of the sphere is the same as the interior these attributes must be drawn from the con-
width of the cube), what proportion of straints of ordinary language and are often
the volume of the cube would the sphere imprecise. For example, the conventional in-
occupy?” The target attribute in this judg- terpretation of overall happiness does not
ment (the volumetric relation between a specify how much weight ought to be given
cube and sphere) is simple enough to be un- to various life domains. However, it certainly
derstood but complicated enough to accom- does require that substantial weight be given
modate a wide range of estimates as plau- to every important domain of life and that
sible answers. Thus, if a relevant simpler no weight at all be given to the current
computation or perceptual impression ex- weather or to the recent consumption of a
ists, respondents will have no strong basis for cookie. Similar rules of common sense ap-
rejecting it as their “final answer.” Frederick ply to judgments of probability. For example,
and Nelson (2004) proposed that the areal the statement “John is more likely to survive
ratio of the respective cross-sections serves a week than a month” is clearly true, and,
that function; that is, that respondents an- thus, implies a rule that people would want
swer as if they were asked the simpler two- their probability judgments to follow. Ac-
dimensional analog of this problem (“If a cordingly, neglect of duration in assessments
circle were drawn inside a square, what pro- of survival probabilities would be properly
portion of the area of the square does the described as a weighting bias, even if there
circle occupy?”). As evidence, they noted were no way to establish a normative prob-
that the mean estimate of the “sphere inside ability for individual cases (Kahneman &
cube” problem (74%) is scarcely different Tversky, 1 996).
from the mean estimate of the “circle inside For some judgmental tasks, information
square” problem (77%) and greatly exceeds that could serve to supplement or correct the
the correct answer (5 2%) – a correct an- heuristic is not neglected or underweighted
swer that most people, not surprisingly, are but simply lacking. If asked to judge the rela-
surprised by. tive frequency of words beginning with K or
R (Tversky & Kahneman, 1 973 ) or to com-
pare the population of a familiar foreign city
Biases with one that is unfamiliar (Gigerenzer &
Whenever the heuristic attribute differs Goldstein, 1 996), respondents have little re-
from the target attribute, the substitution course but to base their judgments on ease
of one for the other inevitably introduces of retrieval or recognition. The necessary re-
systematic biases. In this treatment, we liance on these heuristic attributes renders
are mostly concerned with weighting bi- such judgments susceptible to biasing factors
ases, which arise when cues available to (e.g., the amount of media coverage). How-
the judge are given either too much or ever, unlike weighting biases, such biases of
too little weight. Criteria for determining insufficient information cannot be described
optimal weights can be drawn from sev- as errors of judgment because there is no way
eral sources. In the classic lens model, the to avoid them.
optimal weights associated with different
cues are the regression weights that opti-
Accessibility and Substitution
mize the prediction of an external criterion,
such as physical distance or the grade point The intent to judge a target attribute initi-
average that a college applicant will attain ates a search for a reasonable value. Some-
(Brunswik, 1 943 ; Hammond, 1 95 5 ). Our times this search ends quickly because the
analysis of weighting biases applies to such required value can be read from a stored
cases, but it also extends to attributes for memory (e.g., the answer to the question
which no objective criterion is available, “How tall are you?”) or a current experience
such as an individual’s overall happiness (e.g., the answer to the question “How much
or the probability that a particular patient do you like this cake?”). For other judg-
will survive surgery. Normative standards for ments, however, the target attribute does
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 71

not readily come to mind, but the search stitution in a wide variety of affect-laden
for it evokes other attributes that are con- judgments. Indeed, the evidence suggests
ceptually and associatively related. For ex- that a list of major general-purpose heuris-
ample, a question about overall happiness tics should include an affect heuristic (Slovic
may retrieve the answer to a related ques- et al., 2002). Slovic and colleagues (2002)
tion about satisfaction with a particular as- show that a basic affective reaction gov-
pect of life upon which one is currently erns a wide variety of more complex evalua-
reflecting. tions such as the cost–benefit ratio of various
We adopt the term accessibility to refer technologies, the safe level of chemicals, or
to the ease (or effort) with which particu- even the predicted economic performance
lar mental contents come to mind (see, e.g., of various industries. In the same vein, Kah-
Higgins, 1 996; Tulving & Pearlstone, 1 966). neman and Ritov (1 994) and Kahneman,
The question of why thoughts become ac- Ritov, and Schkade (1 999) proposed that an
cessible – why particular ideas come to mind automatic affective valuation is the principal
at particular times – has a long history in psy- determinant of willingness to pay for public
chology and encompasses notions of stimu- goods, and Kahneman, Schkade, and Sun-
lus salience, associative activation, selective stein (1 998) interpreted jurors’ assessments
attention, specific training, and priming. In of punitive awards as a mapping of outrage
the present usage, accessibility is determined onto a dollar scale of punishments.
jointly by the characteristics of the cogni- Attributes that are not naturally assessed
tive mechanisms that produce it and by the can become accessible if they have been re-
characteristics of the stimuli and events that cently evoked or primed (see, e.g., Bargh et
evoke it, and it may refer to different aspects al., 1 986; Higgins & Brendl, 1 995 ). The ef-
and elements of a situation, different ob- fect of temporary accessibility is illustrated
jects in a scene, or different attributes of an by the “romantic satisfaction heuristic” for
object. judging happiness. The mechanism of at-
Attribute substitution occurs when a rela- tribute substitution is the same, however,
tively inaccessible target attribute is assessed whether the heuristic attribute is chronically
by mapping a relatively accessible and re- or temporarily accessible.
lated heuristic attribute onto the target scale. There is sometimes more than one can-
Some attributes are permanent candidates didate for the role of heuristic attribute. For
for the heuristic role because they are rou- an example that we borrow from Anderson
tinely evaluated as part of perception and (1 991 ), consider the question “Are more
comprehension and therefore always acces- deaths caused by rattlesnakes or bees?” A re-
sible (Tversky & Kahneman, 1 983 ). These spondent who has recently read about some-
natural assessments include physical prop- one who died from a snakebite or bee sting
erties such as size and distance and more may use the relative availability of instances
abstract properties such as similarity (e.g., of the two categories as a heuristic. If no
Tversky & Kahneman, 1 983 ; see Goldstone instances come to mind, that person might
& Son, Chap. 2), cognitive fluency in per- consult his or her impressions of the “dan-
ception and memory (e.g., Jacoby & Dallas, gerousness” of the typical snake or bee, an
1 991 ; Schwarz & Vaughn, 2002; Tversky & application of representativeness. Indeed, it
Kahneman, 1 973 ), causal propensity (Hei- is possible that the question initiates both
der, 1 944; Kahneman & Varey, 1 990; Mi- a search for instances and an assessment of
chotte, 1 963 ), surprisingness (Kahneman & dangerousness, and that a contest of accessi-
Miller, 1 986), mood (Schwarz & Clore, bility determines the role of the two heuris-
1 983 ), and affective valence (e.g., Bargh, tics in the final response. As Anderson ob-
1 997; Cacioppo, Priester, & Berntson, 1 993 ; served, it is not always possible to determine
Kahneman, Ritov, & Schkade, 1 999; Slovic a priori which heuristic will govern the re-
et al., 2002; Zajonc, 1 980, 1 997). sponse to a particular problem.
Because affective valence is a natural as- The original list of heuristics (Tver-
sessment, it is a candidate for attribute sub- sky & Kahneman, 1 974) also included an
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 72 the cambridge handbook of thinking and reasoning

“anchoring heuristic.” An anchoring effect, get attribute has no upper bound. Kahne-
however, does not involve the substitution of man, Ritov, and Schkade (1 999) discussed
a heuristic attribute for a target attribute: It two situations in which an attitude (or af-
is due to the temporary salience of a particu- fective valuation) is mapped onto an un-
lar value of the target attribute. However, an- bounded scale of dollars: when respondents
choring and attribute substitution are both in surveys are required to indicate how much
instances of a broader family of accessibility money they would contribute for a cause,
effects (Kahneman, 2003 ). In attribute sub- and when jurors are required to specify an
stitution, a highly accessible attribute con- amount of punitive damages against a neg-
trols the evaluation of a less accessible one. ligent firm. The mapping of attitudes onto
In anchoring, a highly accessible value of dollars is a variant of direct scaling in psy-
the target attribute dominates its judgment. chophysics, where respondents assign num-
This conception is compatible with more bers to indicate the intensity of sensations
recent theoretical treatments of anchor- (Stevens, 1 975 ). The normal practice of di-
ing (see, e.g., Chapman & Johnson, 1 994, rect scaling is for the experimenter to pro-
2002; Mussweiler & Strack 1 999; Strack & vide a modulus – a specified number that
Mussweiler, 1 997). is to be associated with a standard stimu-
lus. For example, respondents may be asked
to assign the number 1 0 to the loudness of
Cross-Dimensional Mapping
a standard sound and judge the loudness
The process of attribute substitution in- of other sounds relative to that standard.
volves the mapping of the heuristic at- Stevens (1 975 ) observed that when the ex-
tribute of the judgment object onto the perimenter fails to provide a modulus, re-
scale of the target attribute. Our notion of spondents spontaneously adopt one. How-
cross-dimensional mapping extends Stevens’ ever, different respondents may pick moduli
(1 975 ) concept of cross-modality matching. that differ greatly (sometimes varying by a
Stevens postulated that intensive attributes factor of 1 00 or more); thus, the variability
(e.g., brightness, loudness, the severity of in judgments of particular stimuli is domi-
crimes) can be mapped onto a common scale nated by arbitrary individual differences in
of sensory strength, allowing direct matching the choice of modulus. A similar analysis
of intensity across modalities – permitting, applies to situations in which respondents
for example, respondents to match the loud- are required to use the dollar scale to ex-
ness of sounds to the severity of crimes. Our press affection for a species or outrage to-
conception allows other ways of compar- ward a defendant. Just as Stevens’ observers
ing values across dimensions, such as match- had no principled way to assign a number to
ing relative positions (e.g., percentiles) a moderately loud sound, survey participants
in the frequency distributions or ranges of and jurors have no principled way to scale
different attributes (Parducci, 1 965 ). An im- affection or outrage into dollars. The anal-
pression of a student’s position in the dis- ogy of scaling without a modulus has been
tribution of aptitude may be mapped di- used to explain the notorious variability of
rectly onto a corresponding position in the dollar responses in surveys of willingness to
distribution of academic achievement and pay and in jury awards (Kahneman, Ritov,
then translated into a letter grade. Note & Schkade, 1 999; Kahneman, Schkade, &
that cross-dimensional matching is inher- Sunstein, 1 998).
ently nonregressive: A judgment or predic-
tion is just as extreme as the impression
System 2 : The Supervision of
mapped onto it. Ganzach and Krantz (1 990)
Intuitive Judgments
applied the term “univariate matching” to a
closely related notion. Our model assumes that an intuitive judg-
Cross-dimensional mapping presents spe- ment is expressed overtly only if it is
cial problems when the scale of the tar- endorsed by system 2. The Stroop task
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 73

illustrates this two-system structure. Ob- beyond this simple strategy, but they need
servers who are instructed to report the color more than 5 seconds to do so. As this exam-
in which words are printed tend to stum- ple illustrates, judgment by heuristic often
ble when the word is the name of another yields simplistic assessments, which system 2
color (e.g., the word BLUE printed in green sometimes corrects by bringing additional
ink). The difficulty arises because the word is considerations to bear.
automatically read, and activates a response Attribute substitution can be prevented
(“blue” in this case) that competes with the by alerting respondents to the possibility
required response (“green”). Errors are rare that their judgment could be contaminated
in the Stroop test, indicating generally suc- by an irrelevant variable. For example, al-
cessful monitoring and control of the overt though sunny or rainy weather typically af-
response, but the conflict produces delays fects reports of well-being, Schwarz and
and hesitations. The successful suppression Clore (1 983 ) found that weather has no
of erroneous responses is effortful, and its effect if respondents are asked about the
efficacy is reduced by stress and distraction. weather just before answering the well-
Gilbert (1 989) described a correction being question. Apparently, this question re-
model in which initial impulses are often minds respondents that their current mood
wrong and normally overridden. He argued (a candidate heuristic attribute) is influ-
that people initially believe whatever they enced by a factor (current weather) that is
are told (e.g., “Whitefish love grapes”) and irrelevant to the requested target attribute
that it takes some time and mental effort to (overall well-being). Schwarz (1 996) also
“unbelieve” such dubious statements. Here found that asking people to describe their
again, cognitive load disrupts the control- satisfaction with some particular domain of
ling operations of system 2, increasing the life reduces the weight this domain receives
rate of errors and revealing aspects of intu- in a subsequent judgment of overall well be-
itive thinking that are normally suppressed. ing. As these examples illustrate, although
In an ingenious extension of this approach, priming typically increases the weight of that
Bodenhausen (1 990) exploited natural tem- variable on judgment (a system 1 effect), this
poral variability in alertness. He found that does not occur if the prime is a sufficiently
“morning people” were substantially more explicit reminder that brings the self-critical
susceptible to a judgment bias (the conjunc- operations of system 2 into play.
tion fallacy) in the evening and that “evening We suspect that system 2 endorsements of
people” were more likely to commit the fal- intuitive judgments are granted quite casu-
lacy in the morning. ally under normal circumstances. Consider
Because system 2 is relatively slow, its op- the puzzle “A bat and a ball cost $1 .1 0 in to-
erations can be disrupted by time pressure. tal. The bat costs $1 more than the ball. How
Finucane et al. (2000) reported a study in much does the ball cost?” Almost everyone
which respondents judged the risks and ben- we ask reports an initial tendency to answer
efits of various products and technologies “1 0 cents” because the sum $1 .1 0 separates
(e.g., nuclear power, chemical plants, cellu- naturally into $1 and 1 0 cents, and 1 0 cents
lar phones). When participants were forced is about the right magnitude. Many peo-
to respond within 5 seconds, the correlations ple yield to this immediate impulse. Even
between their judgments of risks and their among undergraduates at elite institutions,
judgments of benefits were strongly nega- about half get this problem wrong when it
tive. The negative correlations were much is included in a short IQ test (Frederick,
weaker (although still pronounced) when re- 2004). The critical feature of this problem
spondents were given more time to ponder is that anyone who reports 1 0 cents has ob-
a response. When time is short, the same viously not taken the trouble to check his
affective evaluation apparently serves as a or her answer. The surprisingly high rate
heuristic attribute for assessments of both of errors in this easy problem illustrates
benefits and risks. Respondents can move how lightly system 2 monitors the output of
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 74 the cambridge handbook of thinking and reasoning

system 1 : People are often content to trust one group of respondents provides judg-
a plausible judgment that quickly comes to ments of a target attribute for a set of ob-
mind. (The correct answer, by the way, is jects and another group evaluates the hy-
5 cents.) pothesized heuristic attribute for the same
The bat and ball problem elicits many er- objects. The substitution hypothesis im-
rors, although it is not really difficult and plies that the judgments of the two groups,
certainly not ambiguous. A moral of this when expressed in comparable units (e.g.,
example is that people often make quick percentiles), will be identical. This section
intuitive judgments to which they are not examines several applications of heuristic
deeply committed. A related moral is that elicitation.
we should be suspicious of analyses that ex-
plain apparent errors by attributing to re- Eliciting Representativeness
spondents a bizarre interpretation of the
question. Consider someone who answers a Figure 1 2.1 displays the results of two ex-
question about happiness by reporting her periments in which a measure of represen-
satisfaction with her romantic life. The re- tativeness was elicited. These results were
spondent is surely not committed to the ab- published long ago, but we repeat them here
surdly narrow interpretation of happiness because they still provide the most direct
that her response seemingly implies. More evidence for both attribute substitution and
likely, at the time of answering, she thinks the representativeness heuristic. For a more
that she is reporting happiness: A judgment recent application of a similar design, see
comes quickly to mind and is not obviously Bar-Hillel and Neter (1 993 ).
mistaken – end of story. Similarly, we pro- The object of judgment in the study from
pose that respondents who judge probabil- which Figure 1 2.1 (a) is drawn (Kahneman &
ity by representativeness do not seriously be- Tversky, 1 973 ; p. 1 27 in Kahneman, Slovic,
lieve that the questions “How likely is X to & Tversky, 1 982) was the following descrip-
be a Y?” and “How much does X resemble tion of a fictitious graduate student, which
the stereotype of Y?” are synonymous. Peo- was shown along with a list of nine fields of
ple who make a casual intuitive judgment graduate specialization:
normally know little about how their judg- Tom W. is of high intelligence, although
ment came about and know even less about lacking in true creativity. He has a need
its logical entailments. Attempts to recon- for order and clarity and for neat and tidy
struct the meaning of intuitive judgments by systems in which every detail finds its ap-
interviewing respondents (see, e.g., Hertwig propriate place. His writing is rather dull
& Gigerenzer, 1 999) are therefore unlikely and mechanical, occasionally enlivened by
to succeed because such probes require bet- somewhat corny puns and by flashes of
ter introspective access and more coherent imagination of the sci-fi type. He has a
beliefs than people normally muster. strong drive for competence. He seems to
have little feel and little sympathy for other
people and does not enjoy interacting with
others. Self-centered, he nonetheless has a
deep moral sense.
Identifying a Heuristic
Participants in a representativeness group
Hypotheses about judgment heuristics have ranked the nine fields of specialization by
most often been studied by examining the degree to which Tom W. “resembles a
weighting biases and deviations from nor- typical graduate student.” Participants in the
mative rules. However, the hypothesis that probability group ranked the nine fields ac-
one attribute is substituted for another in a cording to the likelihood of Tom W.’s spe-
judgment task – for example, representative- cializing in each. Figure 1 2.1 (a) plots the
ness for probability – can also be tested more mean judgments of the two groups. The
directly. In the heuristic elicitation design, correlation between representativeness and
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 75

a Tom W. b Linda
9 7
8
6
mean rank (likelihood)

mean rank (likelihood)

7
5
6

5 4
4
3
3
2
2

1 1
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7
mean rank (similarity) mean rank (similarity)
Figure 1 2 .1 . (a) Plot of average ranks for nine outcomes for Tom W. ranked by probability and by
similarity to stereotypes of graduate students in various fields. (b) Plot of average ranks for eight
outcomes for Linda ranked by probability and by representativeness.

probability is nearly perfect (.97). No The logic of probabilistic prediction in

stronger support for attribute-substitution this task suggests that the ranking of out-
could be imagined. However, interpreting comes by their probabilities should be in-
representativeness as the heuristic attribute termediate between their rankings by rep-
in these judgments does require two addi- resentativeness and by base rate frequencies.
tional plausible assumptions – that represen- Indeed, if the personality description is taken
tativeness is more accessible than probabil- to be a poor source of information, proba-
ity, and that there is no third attribute that bility judgments should stay quite close to
could explain both judgments. the base rates. The description of Tom W.
The Tom W. study was also intended to was designed to allow considerable scope
examine the effect of the base rates of out- for judgments of probability to diverge from
comes on categorical prediction. For that judgments of representativeness, as this logic
purpose, respondents in a third group esti- requires. Figure 1 2.1 (a) shows no such di-
mated the proportion of graduate students vergence. Thus, the results of the Tom W.
enrolled in each of the nine fields. By design, study simultaneously demonstrate the sub-
some outcomes were defined quite broadly, stitution of representativeness for probabil-
whereas others were defined more narrowly. ity and the neglect of known (but not explic-
As intended, estimates of base rates var- itly mentioned) base rates.
ied markedly across fields, ranging from 3 % Figure 1 2.1 (b) is drawn from an early
for Library Science to 20% for Humanities study of the Linda problem, the best-known
and Education. Also by design, the descrip- and most controversial example in the rep-
tion of Tom W. included characteristics (e.g., resentativeness literature (Tversky & Kahne-
introversion) that were intended to make man, 1 982) in which a woman named Linda
him fit the stereotypes of the smaller fields was described as follows:
(library science, computer science) better Linda is 3 1 years old, single, outspoken
than the larger fields (humanities and social and very bright. She majored in philoso-
sciences).1 As intended, the correlation be- phy. As a student she was deeply concerned
tween the average judgments of representa- with issues of discrimination and social jus-
tiveness and of base rates was strongly nega- tice and also participated in antinuclear
tive (−.65 ). demonstrations.
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 76 the cambridge handbook of thinking and reasoning

As in the Tom W. study, separate groups they were separated by a filler so respondents
of respondents were asked to rank a set of would not feel compelled to compare them.
eight outcomes by representativeness and In the anthropomorphic language used here,
probability. The results are shown in Fig- system 2 was given a chance to correct the
ure 1 2.1 (b). Again the correlation between judgment but was not prompted to do so.
these rankings was almost perfect (.99).1 In view of the confusing controversy that
Six of the eight outcomes that subjects followed, it is perhaps unfortunate that the
were asked to rank were fillers (e.g., ele- articles documenting base rate neglect and
mentary school teacher, psychiatric social conjunction errors did not stop with subtle
worker). The two critical outcomes were #6 tests. Each article also contained an experi-
(bank teller) and the so-called conjunction mental flourish – a demonstration in which
item #8 (bank teller and active in the fem- the error occurred in spite of a manipula-
inist movement). Most subjects ranked the tion that called participants’ attention to the
conjunction higher than its constituent, both critical variable. The engineer–lawyer prob-
in representativeness (85 %) and probabil- lem (Kahneman & Tversky, 1 973 ) included
ity (89%). The observed ranking of the two special instructions to ensure that respon-
items is quite reasonable for judgments of dents would notice the base rates of the
similarity, but not for probability: Linda may outcomes. The brief personality descriptions
resemble a feminist bank teller more than shown to respondents were reported to have
she resembles a bank teller, but she cannot been drawn from a set containing descrip-
be more likely to be a feminist bank teller tions of 3 0 lawyers and 70 engineers (or vice
than to be a bank teller. In this problem, re- versa), and respondents were asked “What
liance on representativeness yields probabil- is the probability that this description be-
ity judgments that violate a basic logical rule. longs to one of the 3 0 lawyers in the sample
As in the Tom W. study, the results make two of 1 00?” To the authors’ surprise, base rates
points: They support the hypothesis of at- were largely neglected in the responses, de-
tribute substitution and also illustrate a pre- spite their salience in the instructions. Sim-
dictable judgment error. ilarly, the authors were later shocked to dis-
cover that more than 80% of undergraduates
committed a conjunction error even when
asked point blank whether Linda was more
The Representativeness Controversy likely to be “a bank teller” or “a bank teller
who is active in the feminist movement”
The experiments summarized in Figure 1 2.1 (Tversky & Kahneman, 1 983 ). The novelty
provided direct evidence for the represen- of these additional direct or “transparent”
tativeness heuristic and two concomitant tests was the finding that respondents con-
biases: neglect of base rates and conjunc- tinued to show the biases associated with
tion errors. In the terminology introduced representativeness even in the presence of
by Tversky and Kahneman (1 983 ), the de- strong cues pointing to the normative re-
sign of these experiments was “subtle”: Ad- sponse. The errors that people make in trans-
equate information was available for partic- parent judgment problems are analogous to
ipants to avoid the error, but no effort was observers’ failure to allow for ambient haze
made to call their attention to that informa- in estimating distances: A correct response
tion. For example, participants in the Tom is within reach, but not chosen, and the fail-
W. experiment had general knowledge of the ure involves an unexpected weakness of the
relative base rates of the various fields of spe- corrective operations of system 2.
cialization, but these base rates were not ex- Discussions of the heuristics and biases
plicitly mentioned in the problem. Similarly, approach have focused almost exclusively
both critical items in the Linda experiment on the direct conjunction fallacy and on
were included in the list of outcomes, but the engineer–lawyer problems. These are
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 77

also the only studies that have been exten- A more lenient concept, reasoning ra-
sively replicated with varying parameters. tionality, only requires an ability to reason
The amount of critical attention is remark- correctly about the information currently
able because the studies were not, in fact, at hand without demanding perfect consis-
essential to the authors’ central claim. In tency among beliefs that are not simulta-
terms of the present treatment, the claim neously evoked. The best known violation
was that intuitive prediction is an operation of reasoning rationality is the famous “four
of system 1 , which is susceptible to both base card” problem (Wason, 1 960). The failure of
rate neglect and conjunction fallacies. There intelligent adults to reason their way through
was no intent to deny the possibility of sys- this problem is surprising because the prob-
tem 2 interventions that would modify or lem is “easy” in the sense of being easily
override intuitive predictions. Thus, the ar- understood once explained. What everyone
ticles in which these studies appeared would learns, when first told that intelligent peo-
have been substantially the same, although ple fail to solve the four-card problem, is
far less provocative, if respondents had over- that one’s expectations about human rea-
come base rate neglect and conjunction er- soning abilities had not been adequately cal-
rors in transparent tests. ibrated. There is, of course, no well-defined
To appreciate why the strong forms of metric of reasoning rationality, but whatever
base rate neglect and of the conjunction fal- metric one uses, the Wason problem calls
lacy sparked so much controversy, it is use- for a downward adjustment. The surprising
ful to distinguish two conceptions of human results of the Linda and engineer–lawyer
rationality (Kahneman, 2000b). Coherence problems led Tversky and Kahneman to a
rationality is the strict conception that re- similar realization: The reasoning of their
quires the agent’s entire system of beliefs subjects was less proficient than they had an-
and preferences to be internally consistent ticipated. Many readers of the work shared
and immune to effects of framing and con- this conclusion, but many others strongly
text. For example, an individual’s probabil- resisted it.
ity p (“Linda is a bank teller”) should be the The implicit challenge to reasoning ra-
sum of the probabilities p (“Linda is a bank tionality was met by numerous attempts to
teller and a feminist”), and p (“Linda is a bank dismiss the findings of the engineer–lawyer
teller and not a feminist”). A subtle test of and the Linda studies as artifacts of ambigu-
coherence rationality could be conducted by ous language, confusing instructions, conver-
asking individuals to assess these three prob- sational norms, or inappropriate normative
abilities on separate occasions under circum- standards. Doubts have been raised about
stances that minimize recall. Coherence can the proper interpretation of almost every
also be tested in a between-groups design. If word in the conjunction problem, including
random assignment is assumed, the sum of “bank teller,” “probability,” and even “and”
the average probabilities assigned to the two (see, e.g., Dulany & Hilton, 1 991 ; Hilton &
component events should equal the average Slugoski, 2001 ). These claims are not dis-
judged probability of “Linda is a bank teller.” cussed in detail here. We suspect that most
If this prediction fails, then at least some of them have some validity and that they
individuals are incoherent. Demonstrations identified mechanisms that may have made
of incoherence present a significant chal- the results in the engineer–lawyer and Linda
lenge to important models of decision the- studies exceptionally strong. However, we
ory and economics, which attribute to agents note a significant weakness shared by all
a very strict form of rationality (Tversky & these critical discussions: They provide no
Kahneman, 1 986). Failures of perfect coher- explanation of the essentially perfect con-
ence are less provocative to psychologists, sistency of the judgments observed in di-
who have a more realistic view of human rect tests of the conjunction rule and in
capabilities. three other types of experiments: subtle
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 78 the cambridge handbook of thinking and reasoning

comparisons, between-Ss comparisons, and are successfully overcome and some circum-
most important, judgments of representa- stances under which intuitions may not be
tiveness (see also Bar-Hillel & Neter, 1 993 ). evoked at all.
Interpretations of the conjunction fallacy
as an artifact implicitly dismiss the results statistical sophistication
of Figure 1 2.1 (b) as a coincidence (for an
exception, see Ayton, 1 998). The story of The performance of statistically sophisti-
the engineer-lawyer problem is similar. Here cated groups of respondents in different ver-
again, multiple demonstrations in which sions of the Linda problem illustrates the ef-
base rate information was used (see Koehler, fects of both expertise and research design
1 996, for a review) invited the inference that (Tversky & Kahneman, 1 983 ). Statistical ex-
there is no general problem of base rate ne- pertise provided no advantage in the eight-
glect. Again, the data of prediction by repre- item version in which the critical items were
sentativeness in Figure 1 2.1 (a) (and related separated by a filler and were presumably
results reported by Kahneman & Tversky, considered separately. In the two-item ver-
1 973 ) were ignored. sion, in contrast, respondents were effec-
The demonstrations that under some con- tively compelled to compare “bank teller”
ditions people avoid the conjunction fallacy with “bank teller and is active in the femi-
in direct tests, or use explicit base rate in- nist movement.” The incidence of conjunc-
formation, led some scholars to the blanket tion errors remained essentially unchanged
conclusion that judgment biases are artifi- among the statistically naive in this condi-
cial and fragile and that there is no need for tion but dropped dramatically for the statis-
judgment heuristics to explain them. This tically sophisticated. Most of the experts fol-
position was promoted most vigorously by lowed logic rather than intuition when they
Gigerenzer (1 991 ). Kahneman and Tversky recognized that one of the categories con-
(1 996) argued in response that the heuris- tained the other. In the absence of a prompt
tics and biases position does not preclude the to compare the items, however, the statis-
possibility of people’s performing flawlessly tically sophisticated made their predictions
in particular variants of the Linda and the in the same way as everyone else does – by
engineer–lawyer problems. Because laypeo- representativeness. As Stephen Jay Gould
ple readily acknowledge the validity of (1 991 , p. 469) noted, knowledge of the truth
the conjunction rule and the relevance of does not dislodge the feeling that Linda is a
base rate information, the fact that they feminist bank teller: “I know [the right an-
sometimes obey these principles is neither a swer], yet a little homunculus in my head
surprise nor an argument against the role of continues to jump up and down, shouting at
representativeness in routine intuitive pre- me – ‘but she can’t just be a bank teller; read
diction. However, the study of conditions the description.’”
under which errors are avoided can help us
understand the capabilities and limitations intelligence
of system 2. We develop this argument fur- Stanovich (1 999) and Stanovich and West
ther in the next section. (2002) observed a generally negative corre-
lation between conventional measures of in-
telligence and susceptibility to judgment bi-
Making Biases Disappear: A Task
ases. They used transparent versions of the
for System 2
problems, which include adequate cues to
Much has been learned over the years about the correct answer and therefore provide
variables and experimental procedures that a test of reasoning rationality. Not surpris-
reduce or eliminate the biases associated ingly, intelligent people are more likely to
with representativeness. We next discuss possess the relevant logical rules and also to
conditions under which errors of intuition recognize the applicability of these rules in
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 79

particular situations. In the terms of manipulations of attention

the present analysis, high-IQ respondents The weight of neglected variables can be in-
benefit from relatively efficient system 2 op- creased by drawing attention to them, and
erations that enable them to overcome er- experimenters have devised many ingenious
roneous intuitions when adequate informa- ways to do so. Schwarz et al. (1 991 ) found
tion is available. (However, when a problem that respondents pay more attention to base
is too difficult for everyone, the correlation rate information when they are instructed
may reverse because the more intelligent re- to think as statisticians rather than clini-
spondents are more likely to agree on a plau- cal psychologists. Krosnick, Li, and Lehman
sible error than to respond randomly, as dis- (1 990) exploited conversational conventions
cussed in Kahneman, 2000b.) about the sequencing of information and
confirmed that the impact of base rate in-
formation was enhanced by presenting that
frequency format information after the personality descrip-
Relative frequencies (e.g., 1 in 1 0) are more tion rather than before it. Attention to the
vividly represented and more easily under- base rate is also enhanced when partici-
stood than equivalent probabilities (.1 0) or pants observe the drawing of descriptions
percentages (1 0%). For example, the emo- from an urn (Gigerenzer, Hell, & Blank,
tional impact of statements of risk is en- 1 988) perhaps because watching the draw-
hanced by the frequency format: “1 person ing induces conscious expectations that re-
in 1 000 will die” is more frightening than a flect the known proportions of possible out-
probability of .001 (Slovic et al., 2002). The comes. The conjunction fallacy can also
frequency representation also makes it eas- be reduced or eliminated by manipulations
ier to visualize partitions of sets and detect that increase the accessibility of the rel-
that one set is contained in another. As a evant rule, including some linguistic vari-
consequence, the conjunction fallacy is gen- ations (Macchi, 1 995 ), and practice with
erally avoided in direct tests in which the logical problems (Agnoli, 1 991 ; Agnoli &
frequency format makes it easy to recog- Krantz, 1 989).
nize that feminist bank tellers are a subset of The interpretation of these attentional ef-
bank tellers (Gigerenzer & Hoffrage, 1 995 ; fects is straightforward. We assume most
Tversky & Kahneman, 1 983 ). For similar rea- participants in judgment studies know, at
sons, some base rate problems are more eas- least vaguely, that the base rate is rele-
ily solved when couched in frequencies than vant and that the conjunction rule is valid
in probabilities or percentages (Cosmides & (Kahneman & Tversky, 1 982). Whether they
Tooby, 1 996). However, there is little sup- apply this knowledge to override an intu-
port for the more general claims about the itive judgment depends on their cognitive
evolutionary adaptation of the mind to deal skills (education, intelligence) and on for-
with frequencies (Evans et al., 2000). Fur- mulations that make the applicability of a
thermore, the ranking of outcomes by pre- rule apparent (frequency format) or a rel-
dicted relative frequency is very similar to evant factor more salient (manipulations of
the ranking of the same outcomes by rep- attention). We assume intuitions are less sen-
resentativeness (Mellers, Hertwig, & Kahne- sitive to these factors and that the appear-
man, 2001 ). We conclude that the frequency ance or disappearance of biases mainly re-
format affects the corrective operations of flects variations in the efficacy of corrective
system 2, not the intuitive operations of sys- operations. This conclusion would be circu-
tem 1 . The language of frequencies improves lar, of course, if the corrective operations
respondents’ ability to impose the logic of were both inferred from the observation of
set inclusion on their considered judgments correct performance and used to explain that
but does not reduce the role of representa- performance. Fortunately, the circularity can
tiveness in their intuitions. be avoided because the role of system 2
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 80 the cambridge handbook of thinking and reasoning

can be verified – for example, by using ma- observed that people often average or add
nipulations of time pressure, cognitive load, where they should multiply.
or mood to interfere with its operations.
In summary, the factorial design is not
appropriate for testing hypotheses about bi-
within-subjects factorial designs
ases of neglect because it effectively guaran-
The relative virtues of between-subjects and tees that no manipulated factor is neglected.
within-subject designs in studies of judg- Figure 1 2.2 illustrates this claim by sev-
ment are a highly contentious issue. Facto- eral examples of an additive extension effect
rial designs have their dismissive critics (e.g., that we discuss further in the next section.
Poulton, 1 989) and their vigorous defenders The experiments summarized in the differ-
(e.g., Birnbaum, 1 999). We do not attempt ent panels share three important features:
to adjudicate this controversy here. Our nar- (1 ) In each case, the quantitative variable
rower point is that between-subjects designs plotted on the abscissa was completely ne-
are more appropriate for the study of heuris- glected in similar experiments conducted in
tics of judgment. The following arguments a between-subjects or subtle design; (2) in
favor this conclusion: each case, the quantitative variable com-
r Factorial designs are transparent. Partici- bines additively with other information; (3 )
in each case, a compelling normative ar-
pants are likely to identify the variables gument can be made for a quasimulti-
that are manipulated, especially if there plicative rule in which the lines shown in
are many trials and especially in a fully Figure 1 2.2 should fan out. For example, Fig-
factorial design in which the same stimu- ure 1 2.2(c) presents a study of categorical
lus attributes are repeated in varying com- prediction (Novemsky & Kronzon, 1 999) in
binations. The message that the design which respondent 5 judged the relative like-
conveys to the participants is that the ex- lihood that a person was a member of one
perimenter expects to find effects of ev- occupation rather than another (e.g., com-
ery factor that is manipulated (Bar-Hillel puter programmer vs. flight attendant) on
& Fischhoff, 1 981 ; Schwarz, 1 996). the basis of short personality sketches (e.g.,
r Studies that apply a factorial design
“shy, serious, organized, and sarcastic”) and
to judgment tasks commonly involve one of three specified base rates (1 0%, 5 0%,
schematic and impoverished stimuli. The or 90%). Representativeness and base rate
tasks are also highly repetitive. These were varied factorially within subjects. The
features encourage participants to adopt effect of base rate is clearly significant in this
simple mechanical rules that will allow design (see also Birnbaum & Mellers, 1 983 ).
them to respond quickly without forming Furthermore, the effects of representative-
an individuated impression of each stim- ness and base rate are strictly additive. As
ulus. For example, Ordóñez and Benson Anderson (1 996) argued, averaging (a spe-
(1 997) required respondents to judge the cial case of additive combination) is the most
attractiveness of gambles on a 1 00-point obvious way to combine the effects of two
scale. They found that under time pres- variables that are recognized as relevant (e.g.,
sure many respondents computed or esti- “She looks like a bank teller, but the base-rate
mated the expected values of the gambles is low.”). Additivity is not normatively ap-
and used the results as attractiveness rat- propriate in this case – any Bayes-like com-
ings (e.g., a rating of 1 5 for a 5 2% chance bination would produce curves that initially
to win $3 1 .5 0). fan out from the origin and converge again
r Factorial designs often yield judgments at high values. Similar considerations apply
that are linear combinations of the ma- to the other three panels of Figure 1 2.2 dis-
nipulated variables. This is a central cussed later. Between-subjects and factorial
conclusion of a massive research effort designs often yield different results in stud-
conducted by Anderson (1 996), who ies of intuitive judgment. Why should we
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 81

a Kahneman, Ritov, & Schkade Data b Schreiber & Kahneman Data

35 10
9
Mean contribution in $

30
8
25 7 71 dB

Aversiveness
Low
20 6
75 dB
Medium 5
15 78 dB
High 4
10 3 80 dB
5 2
1
0 0
0 20 40 60 80 0 5 10 15 20 25 30
Percentage Population Decline Duration in Seconds

c Novemsky & Kronzon Data d Ariely Data

90 80
80 75
Posterior Probability (%)

70 70
65
Aversiveness

60 Programmer Down
60
50 Surgeon Down&Up
55
40 Accountant
50 Up
`
30 Engineer 45 Up&Down
20 40
10 35
30
0
0 20 40 60 80 100 0 10 20 30 40 50

Base-rate (%) Duration in Seconds

Figure 1 2 .2 . (a) Willingness to pay to restore damage to species that differ in popularity as a function
of the damage they have suffered (from Kahneman, Ritov, & Schkade 2000); (b) global evaluations of
aversive sounds of different loudness as a function of duration for subjects selected for their high
sensitivity to duration (from Schreiber & Kahneman, 2000); (c) ratings of probability for predictions
that differ in representativeness as a function of base rate frequency (from Novemsky & Kronzon,
1 999); (d) global evaluations of episodes of painful pressure that differ in temporal profile as a
function of duration (Ariely, 1 998).

believe one design rather than the other? are made and is more likely to evoke the ca-
The main argument against the factorial de- sually intuitive mode of judgment that gov-
sign is its poor ecological validity. Encounter- erns much of mental life in routine situations
ing multiple judgment objects in rapid suc- (e.g., Langer, 1 978).
cession in a rigidly controlled structure is
unique to the laboratory, and the solutions
that they evoke are not likely to be typical.
Direct comparisons among concepts that Prototype Heuristics and the Neglect
differ in only one variable – such as bank of Extension
teller and feminist bank tellers – also provide
a powerful hint and a highly unusual oppor- In this section, we offer a common account
tunity to overcome intuitions. The between- of three superficially dissimilar judgmental
subjects design, in contrast, mimics the hap- tasks: (1 ) categorical prediction (e.g., “In a
hazard encounters in which most judgments set of 3 0 lawyers and 70 engineers, what is the
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 82 the cambridge handbook of thinking and reasoning

probability that someone described as ‘charm- ement of the set adds to the overall judg-
ing, talkative, clever, and cynical’ is one of the ment an amount that depends on the el-
lawyers?”); (2) summary evaluations of past ements already included. In simple cases,
events (e.g., “Overall, how aversive was it to conditional adding is just regular adding –
be exposed for 3 0 minutes to your neighbor’s the total weight of a collection of chairs is
car alarm?”); and (3 ) economic valuations the sum of their individual weights. In other
of public goods (e.g., “What is the most you cases, each element of the set contributes
would be willing to pay to prevent 2 00,000 mi- to the overall judgment, but the combina-
grating birds from drowning in uncovered oil tion rule is not simple addition and is most
ponds?”). We propose that a generalization typically subadditive. For example, the eco-
of the representativeness heuristic accounts nomic value of protecting X birds should be
for the remarkably similar biases that are ob- increasing in X, but the value of saving 2000
served in these diverse tasks. birds is for most people less than twice as
The original analysis of categorical pre- large as the value of saving 1 000 birds.
diction by representativeness (Kahneman & The logic of categorical prediction entails
Tversky 1 973 ; Tversky & Kahneman, 1 983 ) that the probability of membership in a cat-
invoked two assumptions in which the word egory should vary with its relative size, or
“representative” was used in different ways: base rate. In prediction by representative-
(1 ) A prototype (a representative exemplar) ness, however, the representation of out-
is used to represent categories (e.g., bank comes by prototypical exemplars effectively
tellers) in the prediction task, and (2) the discards base rates because the prototype of a
probability that the individual belongs to a category (e.g., lawyers) contains no informa-
category is judged by the degree to which the tion about the size of its membership. Next,
individual resembles (is representative of) the we show that phenomena analogous to the
category stereotype. Thus, categorical pre- neglect of base rate are observed in other
diction by representativeness involves two prototype heuristics: The monetary value at-
separate acts of substitution – the substitu- tached to a public good is often insensitive
tion of a representative exemplar for a cat- to its scope, and the global evaluation of a
egory and the substitution of the heuris- temporally extended experience is often in-
tic attribute of representativeness for the sensitive to its duration. These various in-
target attribute of probability. Perhaps be- stantiations of extension neglect (neglect of
cause they share a label, the two pro- base rates, scope, and duration) have been
cesses have not been distinguished in dis- discussed in separate literatures, but all can
cussions of the representativeness heuristic. be explained by the two-part process that
We separate them here by describing proto- defines prototype heuristics: (1 ) A category
type heuristics in which a prototype is sub- is represented by a prototypical exemplar,
stituted for its category, but in which repre- and (2) a (nonextensional) property of the
sentativeness is not necessarily the heuristic prototype is then used as a heuristic attribute
attribute. to evaluate an extensional target attribute of
The target attributes to which prototype the category. As might be expected from the
heuristics are applied are extensional. An ex- earlier discussion of base rate neglect, exten-
tensional attribute pertains to an aggregated sion neglect in all its forms is most likely to be
property of a set or category for which an observed in between-subjects experiments.
extension is specified – the probability that Within-subject factorial designs consistently
a set of 3 0 lawyers includes Jack, the over- yield the additive extension effect illustrated
all unpleasantness of a set of moments of in Figure 1 2.2.
hearing a neighbor’s car alarm, and the per-
sonal dollar value of saving a certain number
Scope Neglect in Willingness to Pay
of birds from drowning in oil ponds. Nor-
mative judgments of extensional attributes The contingent valuation method (CVM)
are governed by a general principle of con- was developed by resource economists (see
ditional adding, which dictates that each el- Mitchell & Carson, 1 989) as a tool for
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 83

assessing the value of public goods for pur- for the more complex target attribute of
poses of litigation or cost–benefit analysis. economic value. Other examples of radical
Participants in contingent valuation (CV) insensitivity to scope lend themselves to a
surveys are asked to indicate their willing- similar interpretation. Among others, Kah-
ness to pay (WTP) for specified public goods, neman (1 986) found that Toronto residents
and their responses are used to estimate the were willing to pay almost as much to clean
total amount that the community would pay up polluted lakes in a small region of On-
to obtain these goods. The economists who tario as to clean up all the polluted lakes in
design contingent valuation surveys inter- Ontario, and McFadden and Leonard (1 993 )
pret WTP as a valid measure of economic reported that residents in four western states
value and assume that statements of WTP were willing to pay only 28% more to protect
conform to the extensional logic of con- 5 7 wilderness areas than to protect a single
sumer theory. The relevant logic has been area (for more discussion of scope insensitiv-
described by a critic of CVM (Diamond, ity, see Frederick & Fischhoff, 1 998).
1 996), who illustrates the conditional adding The similarity between WTP statements
rule by the following example: In the ab- and categorical predictions is not limited
sence of income effects, WTP for saving X to such demonstrations of almost complete
birds should equal WTP for saving (X − k) extension neglect. The two responses also
birds, plus WTP to save k birds, where the yield similar results when extension and
last value is contingent on the costless prior prototype information are varied factori-
provision of safety for (X − k) birds. ally within subjects. Figure 1 2.2(a) shows
Strict adherence to Bayes’ rule may be the results of a study of WTP for pro-
an excessively demanding standard for intu- grams that prevented different levels of
itive predictions; similarly, it would be too damage to species of varying popularity
much to ask for WTP responses that strictly (Ritov & Kahneman, unpublished observa-
conform to the “add-up rule.” In both cases, tions, cited in Kahneman, Ritov, & Schkade,
however, it seems reasonable to expect some 1 999). As in the case of base rate [Figure
sensitivity to extension – to the base rate 1 2.2(c)], extensional information (levels of
of outcomes in categorical prediction and to damage) combines additively with nonex-
the scope of the good in WTP. In fact, several tensional information. This rule of combina-
studies have documented nearly complete tion is unreasonable; in any plausible theory
neglect of scope in CV surveys. The best- of value, the lines would fan out.
known demonstration of scope neglect is an Finally, the role of the emotion evoked
experiment by Desvouges et al. (1 993 ), who by a prototypical instance was also exam-
used the scenario of migratory birds that ined directly in the same experiment, us-
drown in oil ponds. The number of birds said ing the heuristic elicitation paradigm intro-
to die each year was varied across groups. duced earlier: Some respondents were asked
The WTP responses were completely insen- to imagine that they saw a television pro-
sitive to this variable; the mean WTPs for gram documenting the effect of adverse eco-
saving 2000, 20,000, or 200,000 birds were logical circumstances on individual mem-
$80, $78, and $88, respectively. bers of different species. The respondents
A straightforward interpretation of this indicated, for each species, how much con-
result involves the two acts of substitution cern they expected to feel while watching
that characterize prototype heuristics. The such a documentary. The correlation be-
deaths of numerous birds are first repre- tween this measure of affect and willingness
sented by a prototypical instance – perhaps to pay, computed across species, was .97.
an image of a bird soaked in oil and drown-
ing. The prototype automatically evokes
Duration Neglect in the Evaluation
an affective response, and the intensity of
of Experiences
that emotion is then mapped onto the dol-
lar scale – substituting the readily accessi- We next discuss experimental studies of the
ble heuristic attribute of affective intensity global evaluation of experiences that extend
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 84 the cambridge handbook of thinking and reasoning

over some time, such as a pleasant or a a simple average of the peak affect recorded
horrific film clip (Fredrickson & Kahne- during a film and the end affect reported as
man, 1 993 ), a prolonged unpleasant noise the film was about to end. This has been
(Schreiber & Kahneman, 2000), pressure called the peak/end rule. However, the cor-
from a vise (Ariely, 1 998), or a painful med- relation between retrospective evaluations
ical procedure (Redelmeier & Kahneman, and the duration of the films was negligible –
1 996). Participants in these studies provided a finding that Fredrickson and Kahneman la-
a continuous or intermittent report of hedo- beled duration neglect. The resemblance of
nic or affective state, using a designated scale duration neglect to the neglect of scope and
of momentary affect (Figure 1 2.3 ). When base rate is striking and unlikely to be ac-
the episode had ended, they indicated a cidental. In this analysis, all three are mani-
global evaluation of “the total pain or dis- festations of extension neglect caused by the
comfort” associated with the entire episode. use of a prototype heuristic.
We first examine the normative rules that The peak/end rule and duration neglect
apply to this task. The global evaluation of have both been confirmed on multiple oc-
a temporally extended outcome is an exten- casions. Figure 1 2.3 presents raw data from
sional attribute, which is governed by a dis- a study reported by Redelmeier and Kahne-
tinctive logic. The most obvious rule is tem- man (1 996), in which patients undergoing
poral monotonicity: There is a compelling colonoscopy reported their current level of
intuition that adding an extra period of pain pain every 60 seconds throughout the proce-
to an episode of discomfort can only make dure. Here again, an average of peak and end
it worse overall. Thus, there are two ways pain quite accurately predicted subsequent
of making a bad episode worse – making global evaluations and choices. The duration
the discomfort more intense or prolonging of the procedure varied considerably among
it. It must therefore be possible to trade off patients (from 4 to 69 minutes), but these
intensity against duration. Formal analyses differences were not reflected in subsequent
have identified conditions under which the global evaluations in accord with duration
total utility of an episode is equal to the neglect. The implications of these psycho-
temporal integral of a suitably transformed logical rules of evaluation are paradoxical. In
measure of the instantaneous utility associ- Figure 1 2.3 , for example, it appears evident
ated with each moment (Kahneman, 2000a; that patient B had a worse colonoscopy than
Kahneman, Wakker, & Sarin, 1 997). patient A (on the assumption they used the
Next, we turn to the psychology. scale similarly). However, it is also appar-
Fredrickson and Kahneman (1 993 ) proposed ent that the peak/end average was worse
a “snapshot model” for the retrospective for patient A, whose procedure ended at
evaluation of episodes, which again involves a moment of relatively intense pain. The
two acts of substitution: First, the episode is peak/end rule prediction for these two pro-
represented by a prototypical moment; next, files is that patient A would evaluate the
the affective value attached to the represen- procedure more negatively than patient B
tative moment is substituted for the exten- and would be more likely to prefer to un-
sional target attribute of global evaluation. dergo a barium enema rather than a repeat
The snapshot model was tested in an exper- colonoscopy. The prediction was correct for
iment in which participants provided con- these two individuals and confirmed by the
tinuous ratings of their affect while watch- data of a large group of patients.
ing plotless films that varied in duration and The effects of substantial variations of du-
affective value (e.g., fish swimming in coral ration remained small (although statistically
reefs, pigs being beaten to death with clubs), robust) even in studies conducted in a fac-
and later reported global evaluations of their torial design. Figure 1 2.2(d) is drawn from a
experiences. The central finding was that the study of responses to ischemic pain (Ariely,
retrospective evaluations of these observers 1 998), in which duration varied by a factor of
were predicted with substantial accuracy by 4. The peak/end average accounted for 98%
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 85

Patient A unimportant . . . people may be aware of

duration and consider it important in the
8 abstract [but] what comes most readily to
7 mind in evaluating episodes are the salient
moments of those episodes and the affect
6
Pain Intensity

associated with those moments. Duration

5 neglect might be overcome, we suppose, by
4 drawing attention more explicitly to the
3 attribute of time. (p. 5 4)
2
This comment applies equally well to
1 other instances of extension neglect: The ne-
0 glect of base rate in categorical prediction,
0 10 20
the neglect of scope in willingness to pay, the
Time (minutes)
neglect of sample size in evaluations of ev-
idence (Griffin & Tversky, 1 992; Tversky &
Patient B Kahneman, 1 971 ), and the neglect of prob-
ability of success in evaluating a program of
8
species preservation (DeKay & McClelland,
7
1 995 ). More generally, inattention plays a
Pain Intensity

6
similar role in any situation in which the in-
5 tuitive judgments generated by system 1 vio-
4 late rules that would be accepted as valid by
3 the more deliberate reasoning that we asso-
2 ciate with system 2. As we noted earlier, the
1 responsibility for these judgmental mishaps
0 is properly shared by the two systems: Sys-
0 10 20
tem 1 produces the initial error, and system
Time (minutes)
2 fails to correct it, although it could.
Figure 1 2 .3. Pain intensity reported by two
colonoscopy patients.
Violations of Dominance
The conjunction fallacy observed in the
of the systematic variance of global evalua-
Linda problem is an example of a domi-
tions in that study and for 88% of the vari-
nance violation in judgment: Linda must be
ance in a similar factorial study of responses
at least as likely to be a bank teller as to
to loud unpleasant sounds [Schreiber & Kah-
be a feminist bank teller, but people be-
neman, 2000, Figure 1 2.2(b)]. Contrary to
lieve the opposite. Insensitivity to extension
the normative standard for an extensional at-
(in this case, base rate) effectively guaran-
tribute, the effects of duration and of other
tees the existence of such dominance viola-
determinants of evaluation were additive
tions. For another illustration, consider the
[Figures 1 2.2(b) and 1 2.2(d)].
question: “How many murders were there
The participants in these studies were
last year in [Detroit/Michigan]?” Although
well aware of the relative duration of their
there cannot be more murders in Detroit
experiences and did not consciously de-
than in Michigan, because Michigan con-
cide to ignore duration in their evalua-
tains Detroit, the word “Detroit” evokes a
tions. As Fredrickson and Kahneman (1 993 )
more violent image than the word “Michi-
noted, duration neglect is an attentional
gan” (except of course for people who im-
phenomenon:
mediately think of Detroit when Michigan
. . . duration neglect does not imply is mentioned). If people use an impres-
that duration information is lost, nor sion of violence as a heuristic and neglect
that people believe that duration is geographic extension, their estimates of
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 86 the cambridge handbook of thinking and reasoning

murders in the city may exceed their esti- ishing pain. In a replication, Schreiber and
mates for the state. In a large sample of Uni- Kahneman (2000, experiment 2) exposed
versity of Arizona students, this hypothesis participants to pairs of unpleasant noises in
was confirmed – the median estimate of the immediate succession. The participants lis-
number of murders was 200 for Detroit and tened to both sounds and chose one to be re-
1 00 for Michigan. peated at the end of the session. The “short”
Violations of dominance akin to the con- noise lasted 8 seconds at 77 db. The “long”
junction fallacy have been observed in sev- noise consisted of the short noise plus an
eral other experiments involving both indi- extra period (of up to 24 seconds) at 66 db
rect (between-subjects) and direct tests. In a (less aversive, but still unpleasant and cer-
clinical experiment reported by Redelmeier, tainly worse than silence). Here again, the
Katz, and Kahneman (2001 ), half of a large longer noise was preferred most of the time,
group of patients (N = 682 ) undergoing a and this unlikely preference persisted over a
colonoscopy were randomly assigned to a series of five choices.
condition that made the actual experience The violations of dominance in these di-
strictly worse. Unbeknownst to the patient, rect tests are particularly surprising because
the physician deliberately delayed the re- the situation is completely transparent. The
moval of the colonoscope for approximately participants in the experiments could eas-
1 minute beyond the normal time. The in- ily retrieve the durations of the two experi-
strument was not moved during the extra pe- ences between which they had to choose,
riod. For many patients, the mild discomfort but the results suggest that they simply
of the added period was an improvement ignored duration. A simple explanation is
relative to the pain than they had just ex- that the results reflect “choosing by liking”
perienced. For these patients, of course, pro- (see Frederick, 2002). The participants in
longing the procedure reduced the peak/end the experiments simply followed the nor-
average of discomfort. As expected, retro- mal strategy of choice: “When choosing be-
spective evaluations were less negative in tween two familiar options, consult your ret-
the experimental group, and a 5 -year follow- rospective evaluations and choose the one
up showed that participants in that group that you like most (or dislike least).” Lik-
were also somewhat more likely to comply ing and disliking are products of system 1 ,
with recommendations to undergo a repeat which do not conform to the rules of ex-
colonoscopy (Redelmeier, Katz, & Kahne- tensional logic. System 2 could have inter-
man, 2001 ). vened, but in these experiments it generally
In an experiment that is directly analo- did not. Kahneman et al. (1 993 ) described a
gous to the demonstrations of the conjunc- participant in their study, who chose to re-
tion fallacy, Kahneman et al. (1 993 ) exposed peat the long cold-pressor experience. Soon
participants to two cold-pressor experiences, after the choice was recorded, the partic-
one with each hand: a “short” episode (im- ipant was asked which of the two expe-
mersion of one hand in 1 4 ◦ C water for riences was longer. As he correctly identi-
60 seconds), and a “long” episode (the short fied the long trial, the participant was heard
episode, plus an additional 3 0 seconds during to mutter “the choice I made doesn’t seem
which the water was gradually warmed to to make much sense.” Choosing by liking
1 5 ◦ C). The participants indicated the inten- is a form of mindlessness (Langer, 1 978),
sity of their pain throughout the experience. which illustrates the casual governance of
When they were later asked which of the system 2.
two experiences they preferred to repeat, Like the conjunction fallacy in direct
a substantial majority chose the long trial. tests, which we discussed earlier, violations
These choices violate dominance, because of temporal monotonicity in choices should
after 60 seconds in cold water anyone will be viewed as an expendable flourish. Be-
prefer the immediate experience of a warm cause the two aversive experiences occurred
towel to 3 0 extra seconds of slowly dimin- within a few minutes of each other and
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 87

respondents could accurately recall the dura- others (see also Kahneman, Ritov, &
tion of the two events, system 2 had enough Schkade, 1 999). We attribute extension ne-
information to override choosing by liking. glect and violations of dominance to a lazy
Its failure to do so is analogous to the fail- system 2, and to a prototype heuristic that
ures observed in direct tests of the Linda combines two processes of system 1 : the rep-
problem. In both cases, the violations of resentation of categories by prototypes and
dominance tell us nothing new about sys- the substitution of a nonextensional heuris-
tem 1 ; they only illustrate an unexpected tic attribute for an extensional target at-
weakness of system 2. Just as the theory of tribute. We also propose that people have
intuitive categorical prediction would have some appreciation of the role of extension
remained intact if the conjunction fallacy in the various judgment tasks. Consequently,
had not “worked” in a direct test, the model they will incorporate extension in their judg-
of evaluation by moments would have sur- ments when their attention is drawn to this
vived even if violations of dominance had factor – most reliably in factorial experi-
been eliminated in highly transparent situa- ments, and sometimes (although not always)
tions. The same methodological issues arise in direct tests. The challenge for compet-
in both contexts. Between-subjects experi- ing interpretations is to provide a unified ac-
ments or subtle tests are most appropriate count of the diverse phenomena that have
for studying the basic intuitive evaluations of been considered in this section.
system 1 , and also most likely to reveal com-
plete extension neglect. Factorial designs in
which extension is manipulated practically
guarantee an effect of this variable, and al- Conclusions and Future Directions
most guarantee that it will be additive, as
in Figures 1 2.2(b) and 1 2.2(d) (Ariely, 1 998; The original goal of the heuristics and biases
Ariely, Kahneman, & Loewenstein, 2000; program was to understand intuitive judg-
Schreiber & Kahneman, 2000). Finally, al- ment under uncertainty. Heuristics were de-
though direct choices sometimes yield sys- scribed as a collection of disparate cognitive
tematic violations of dominance, these vio- procedures, related only by their common
lations can be avoided by manipulations that function in a particular judgmental domain –
prompt system 2 to take control. choice under uncertainty. It now appears,
In our view, the similarity of the re- however, that judgment heuristics are ap-
sults obtained in diverse contexts is a com- plied in a wide variety of domains and share
pelling argument for a unified interpreta- a common process of attribute substitution,
tion, and a significant challenge to critiques in which difficult judgments are made by
that pertain only to selected subsets of this substituting conceptually or semantically re-
body of evidence. A number of commenta- lated assessments that are simpler and more
tors have offered competing interpretations readily accessible.
of base rate neglect (Cosmides & Tooby, The current treatment explicitly ad-
1 996; Koehler, 1 996), insensitivity to scope dresses the conditions under which intu-
in WTP (Kopp, 1 992), and duration ne- itive judgments are modified or overridden.
glect (Ariely & Loewenstein, 2000). How- Although attribute substitution provides an
ever, these interpretations are generally spe- initial input into many judgments, it need
cific to a particular task and would not carry not be the sole basis for them. Initial impres-
over to analogous findings in other domains. sions are often supplemented, moderated, or
Similarly, the various attempts to explain the overridden by other considerations, includ-
conjunction fallacy as an artifact do not ex- ing the recognition of relevant logical rules
plain analogous violations of dominance in and the deliberate execution of learned al-
the cold-pressor experiment. The account gorithms. The role of these supplemental or
we have offered is, in contrast, equally ap- alternative inputs depends on characteristics
plicable to all three contexts and possibly of the judge and the judgment task.
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 88 the cambridge handbook of thinking and reasoning

Our use of the dual-process terminology ments that are made in a split second. Thus,
does not entail a belief that every mental one need not be committed, a priori, to as-
operation (including each postulated heuris- signing a process to a particular system; the
tic) can be definitively assigned to one sys- data will dictate the best characterization.
tem or the other. The placement of di- The two-system model is a framework
viding lines between “systems” is arbitrary that combines a set of empirical generaliza-
because the bases by which we characterize tions about cognitive operations with a set
mental operations (difficulty of acquisition, of tests for diagnosing the types of cognitive
accessibility to introspection, and disrupt- operations that underlie judgments in spe-
ability) are all continua. However, this does cific situations. The generalizations and the
not make distinctions less meaningful; there specific predictions are testable and can be
is broad agreement that mental operations recognized as true or false. The framework
range from rapid, automatic, perception-like itself will be judged by its usefulness as a
impressions to deliberate computations that heuristic for research.
apply explicit rules or external aids.
Many have questioned the usefulness
of the notion of heuristics and biases by Acknowledgments
pointing to inconsistencies in the degree to
which illusions are manifested across differ-
This chapter is a modified version of a
ent studies. However, there is no mystery
chapter by Kahneman and Frederick (2002).
here to explain. Experimental studies of “the
Preparation of this chapter was supported by
same” cognitive illusions can yield different
grant SES-021 3 481 from the National Sci-
results for two reasons: (1 ) because of vari-
ence Foundation.
ation in factors that determine the accessi-
bility of the intuitive illusion, and (2) be-
cause they vary in factors that determine the
accessibility of the corrective thoughts that Note
are associated with system 2. Both types of
variation can often be anticipated because 1 . The entries plotted in Figure 1 2.1 are averages
of the vast amount of psychological knowl- of multiple judgments, and the correlations are
edge that has accumulated about the differ- computed over a set of judgment objects. It
ent sets of factors that determine the ease should be noted that correlations between av-
erages are generally much higher than corre-
with which thoughts come to mind – from
sponding correlations within the data of indi-
principles of grouping in perception to prin- vidual respondents (Nickerson, 1 995 ). Indeed,
ciples that govern transfer of training in rule group results may even be unrepresentative if
learning (Kahneman, 2003 ). Experimental they are dominated by a few individuals who
surprises will occur, of course, and should produce more variance than others and have
lead to refinements in the understanding of an atypical pattern of responses. Fortunately,
the rules of accessibility. this particular hypothesis is not applicable to
The argument that system 1 will be ex- the experiments of Figure 1 2.1 , in which all re-
pressed unless it is overridden by system 2 sponses were ranks.
sounds circular, but it is not, because empir-
ical criteria can be used to test whether a par-
ticular characterization of the two systems is References
accurate. For example, a feature of the situ-
ation will be associated with system 2 if it is Agnoli, F. (1 991 ). Development of judgmental
shown to influence judgments only when at- heuristics and logical reasoning: Training coun-
tention is explicitly directed to it (through, teracts the representativeness heuristic. Cogni-
say, a within-subjects design). In contrast, a tive Development, 6, 1 95 –21 7.
variable will be associated with system 1 if it Agnoli, F., & Krantz, D. H. (1 989). Suppressing
can be shown to influence even those judg- natural heuristics by formal instruction: The
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 89

case of the conjunction fallacy. Cognitive Psy- Birnbaum, M. H. (1 999). How to show that 9 >
chology, 2 1 , 5 1 5 –5 5 0. 221 : Collect judgments in a between-subjects
Anderson, N. H. (1 991 ). Contributions to informa- design. Psychological Methods, 4(3 ), 243 –249.
tion integration theory. Vol. I: cognition. Hillsdale, Birnbaum, M. H., & Mellers, B. A. (1 983 ).
NJ: Erlbaum. Bayesian inference: Combining base rates with
Anderson, N. H. (1 996). A functional theory of opinions of sources who vary in credibility.
cognition. Mahwah, NJ: Erlbaum. Journal of Personality and Social Psychology, 45 ,
Ariely, D. (1 998). Combining experiences over 792–804.
time: The effects of duration, intensity changes Bless, H., Clore, G. L., Schwarz, N., Golisano,
and on-line measurements on retrospective V., Rabe, C., & Wolk, M. (1 996). Mood and
pain evaluations. Journal of Behavioral Decision the use of scripts: Does a happy mood really
Making, 1 1 , 1 9–45 . lead to mindlessness? Journal of Personality and
Ariely, D. (2001 ). Seeing sets: Representation Social Psychology, 71 (4), 665 –679.
by statistical properties. Psychological Science, Bodenhausen, G. V. (1 990). Stereotypes as judg-
1 2 (2), 1 5 7–1 62. mental heuristics: Evidence of circadian vari-
Ariely, D., Kahneman, D., & Loewenstein, G. ations in discrimination. Psychological Science,
(2000). Joint comment on “When does dura- 1 (5 ), 3 1 9–3 22.
tion matter in judgment and decision making?” Brunswik, E. (1 943 ). Organismic achievement
(Ariely & Loewenstein, 2000). Journal of and environmental probability. Psychological
Experimental Psychology: General, 1 2 9, 5 24– Review, 5 0, 25 5 –272.
5 29. Cacioppo, J. T., Priester, J. R., & Berntson,
Ariely, D., & Loewenstein, G. (2000). When does G. G. (1 993 ). Rudimentary determinants of
duration matter in judgment and decision mak- attitudes: II. Arm flexion and extension have
ing. Journal of Experimental Psychology: Gen- differential effects on attitudes. Journal of Per-
eral, 1 2 9, 5 08–5 23 . sonality and Social Psychology, 65 , 5 –1 7.
Ayton, P. (1 998). How bad is human judgment? Chaiken, S., & Trope, Y. (1 999). Dual-process the-
In G. Wright & P. Goodwin (Eds.), Forecasting ories in social psychology. New York: Guilford
with judgment. West Sussex, UK: Wiley. Press.
Bargh, J. A. (1 997). The automaticity of everyday Chapman, G., & Johnson, E. (2002). Incorpo-
life. In R. S. Wyer, Jr. (Ed.), Advances in social rating the irrelevant: Anchors in judgments of
cognition (Vol. 1 0). Mahwah, NJ: Erlbaum. belief and value. In T. Gilovich, D. Griffin, &
Bargh, J. A., Bond, R. N., Lombardi, W. J., & Tota, D. Kahneman (eds.), Heuristics and biases: The
M. E. (1 986). The additive nature of chronic psychology of intuitive judgment (pp. 1 20–1 3 8).
and temporary sources of construct accessibil- New York: Cambridge University Press.
ity. Journal of Personality and Social Psychology, Chase, W. G., & Simon, H. A. (1 973 ). Perception
5 0(5 ), 869–878. in chess. Cognitive Psychology, 4, 5 5 –81 .
Bar-Hillel, M., & Fischhoff, B. (1 981 ). When do Cohen, J. (1 969). Statistical power analysis for the
base-rates affect predictions? Journal of Per- behavioral sciences. San Diego: Academic Press.
sonality and Social Psychology, 41 (4), 671 – Cohen, J. (1 992). A power primer. Psychological
680. Bulletin, 1 1 2 (1 ), 1 5 5 –1 5 9.
Bar-Hillel, M., & Neter, E. (1 993 ). How alike is Cosmides L., & Tooby J. (1 996). Are humans
it versus how likely is it: A disjunction fallacy good intuitive statisticians after all? Rethinking
in probability judgments. Journal of Personality some conclusions from the literature on judg-
and Social Psychology, 41 (4), 671 –680. ment under uncertainty. Cognition, 5 8(1 ), 1 –
Begg, I., Anas, A., & Farinacci, S. (1 992). Disso- 73 .
ciation of processes in belief: Source recollec- DeKay, M. L., & McClelland, G. H. (1 995 ). Prob-
tion, statement familiarity, and the illusion of ability and utility components of endangered
truth. Journal of Experimental Psychology: Gen- species preservation programs. Journal of Ex-
eral, 1 2 1 (4), 446–45 8. perimental Psychology: Applied, 2 , 60–83 .
Begg, I., & Armour, V. (1 991 ). Repetition and Desvouges, W. H., Johnson, F., Dunford, R.,
the ring of truth: Biasing comments. Cana- Hudson, S., Wilson, K., & Boyle, K. (1 993 ).
dian Journal of Behavioural Science, 2 3 , 1 95 – Measuring resource damages with contingent
21 3 . valuation: Tests of validity and reliability.
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 90 the cambridge handbook of thinking and reasoning

In Contingent valuation: A critical assessment. European review of social psychology (Vol. 2,

Amsterdam: North Holland. pp. 83 –1 1 5 ). Chichester, UK: Wiley.
Diamond, P. (1 996). Testing the internal consis- Gigerenzer, G., Czerlinski, J., & Martignon,
tency of contingent valuation surveys. Journal L. (2002). How good are fast and frugal
of Environmental Economics and Management, heuristics? In T. Gilovich, D. Griffin, & D.
2 8, 1 5 5 –1 73 . Kahneman (Eds.), Heuristics and biases: The
Dulany, D. E. & Hilton, D. J. (1 991 ). Conver- psychology of intuitive judgment (pp. 5 5 9–5 81 ).
sational implicature, conscious representation, New York: Cambridge University Press.
and the conjunction fallacy. Social Cognition, 9, Gigerenzer, G., & Goldstein, D. G. (1 996).
85 –1 1 0. Reasoning the fast and frugal way: Models
Epstein, S. (1 994). Integration of the cognitive of bounded rationality. Psychological Review,
and the psychodynamic unconscious. American 1 03 (4), 65 0–669.
Psychologist, 49(8), 709–724. Gigerenzer, G., Hell, W., & Blank, H. (1 988). Pre-
Evans, J., Handley, S. J., Perham, N., Over, D. E., sentation and content – The use of base rates
& Thompson, V. A. (2000). Frequency versus as a continuous variable. Journal of Experimen-
probability formats in statistical word prob- tal Psychology – Human Perception and Perfor-
lems. Cognition, 77, 1 97–21 3 . mance, 1 4(3 ), 5 1 3 –5 25 .
Evans, J. St. B. T., & Over, D. E. (1 996). Rationality Gigerenzer, G., & Hoffrage, U. (1 995 ). How
and reasoning. Hove, UK: Psychology Press. to improve Bayesian reasoning without
instruction: Frequency formats. Psychological
Finucane, M. L., Alhakami, A., Slovic, P., &
Review, 1 02 , 684–704.
Johnson, S. M. (2000). The affect heuristic in
judgments of risks and benefits. Journal of Be- Gigerenzer, G., Todd, P. M., & the ABC Group.
havioral Decision Making, 1 3 , 1 –1 7. (1 999). Simple heuristics that make us smart.
New York: Oxford University Press.
Frederick, S. W. (2002). Automated choice
heuristics. In T. Gilovich, D. Griffin, & D. Gilbert, D. (1 989). Thinking lightly about others:
Kahneman (eds.), Heuristics and biases: The Automatic components of the social inference
psychology of intuitive judgment (pp. 5 48–5 5 8). process. In J. Uleman & J. A. Bargh (Eds.),
New York: Cambridge University Press. Unintended thought (pp. 1 89–21 1 ). New York:
Guilford Press.
Frederick, S. W. (2004). (Im)patient, (un)-
adventurous, (in)sensitive, and (im)moral: Cog- Gilbert, D. (1 991 ). How mental systems believe.
nitive (in)ability as a predictor of time preference, American Psychologist, 46(2), 1 07–1 1 9.
risk preference, magnitude sensitivity, and cheat- Gilbert, D. (1 999). What the mind’s not. In S.
ing. Working paper, Massachusetts Institute of Chaiken & Y. Trope (Eds.), Dual-process theo-
Technology, Cambridge, MA. ries in social psychology (pp. 3 –1 1 ). New York:
Frederick, S. W., & Fischhoff, B. (1 998). Scope Guilford Press.
(in)sensitivity in elicited valuations. Risk, Deci- Gould, S. J. (1 991 ). Bully for brontosaurus. Reflec-
sion, and Policy, 3 , 1 09–1 23 . tions in natural history. New York: Norton.
Frederick, S. W., & Nelson, L. (2004). Attribute Griffin, D., & Tversky, A. (1 992). The weighing of
substitution in physical judgments, Working pa- evidence and the determinants of confidence.
per, Massachusetts Institute of Technology, Cognitive Psychology, 2 4(3 ), 41 1 –43 5 .
Cambridge, MA. Hammond, K. R. (1 95 5 ). Probabilistic function-
Fredrickson, B. L., & Kahneman, D. (1 993 ). Du- ing and the clinical method. Psychological Re-
ration neglect in retrospective evaluations of view, 62 , 25 5 –262.
affective episodes. Journal of Personality and So- Hammond, K. R. (1 996). Human judgment and
cial Psychology, 65 (1 ), 45 –5 5 . social policy. New York: Oxford University
Ganzach, Y., & Krantz, D. H. (1 990). The psy- Press.
chology of moderate prediction: I. Experience Heider, F. (1 944). Social perception and phenom-
with multiple determination. Organizational enal causality. Psychological Review, 5 1 , 3 5 8–
Behavior and Human Decision Processes, 47, 3 74.
1 77–204. Hertwig, R., & Gigerenzer, G. (1 999). The ‘con-
Gigerenzer, G. (1 991 ). How to make cognitive junction fallacy’ revisited: How intelligent in-
illusions disappear: Beyond “heuristics and bi- ferences look like reasoning errors. Journal of
ases.” In W. Stroebe & M. Hewstone (Eds.), Behavioral Decision Making, 1 2 (4), 275 –3 05 .
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 91

Higgins, E. T., & Brendl, C. M. (1 995 ). Accessi- study in the headline method. Journal of Risk
bility and applicability: Some “activation rules” and Uncertainty, 9, 5 –3 8.
influencing judgment. Journal of Experimental Kahneman, D., Ritov, I., & Schkade, D. (1 999).
Social Psychology, 3 1 , 21 8–243 . Economic preferences or attitude expressions?
Hilton, D. J., & Slugoski, B. R. (2001 ). Conversa- An analysis of dollar responses to public issues.
tional processes in reasoning and explanation. Journal of Risk and Uncertainty, 1 9, 203 –23 5 .
In A. Tesser & N. Schwarz (Eds.), Blackwell Kahneman, D., Schkade, D. A., & Sunstein, C. R.
handbook of social psychology. Vol. 1 : Intrain- (1 998). Shared outrage and erratic awards: The
dividual processes (pp. 1 81 –206). Oxford, UK: psychology of punitive damages. Journal of Risk
Blackwell. and Uncertainty, 1 6, 49–86.
Isen, A. M., Nygren, T. E., & Ashby, F. G. (1 988). Kahneman, D., Slovic, P., & Tversky, A. E. (1 982).
Influence of positive affect on the subjective Judgment under uncertainty: Heuristics and bi-
utility of gains and losses – It is just not worth ases. New York: Cambridge University Press.
the risk. Journal of Personality and Social Psy- Kahneman, D., & Tversky, A. (1 972). Subjective
chology, 5 5 (5 ), 71 0–71 7. probability: A judgment of representativeness.
Jacoby, L. L., & Dallas, M. (1 981 ). On the re- Cognitive Psychology, 3 , 43 0–45 4.
lationship between autobiographical memory Kahneman, D., & Tversky, A. (1 973 ). On the psy-
and perceptual learning. Journal of Experimen- chology of prediction. Psychological Review, 80,
tal Psychology: General, 3 , 3 06–3 40. 23 7–25 1 .
Kahneman, D. (1 986). Valuing environmen- Kahneman, D., & Tversky, A. (1 982). On the
tal goods: An assessment of the contin- study of statistical intuitions. Cognition, 1 1 ,
gent valuation method. In R. Cummings, D. 1 23 –1 41 .
Brookshire, & W. Schulze (Eds.), Valuing envi- Kahneman, D., & Tversky, A. (1 996). On the re-
ronmental goods: An assessment of the contingent ality of cognitive illusions: A reply to Gigeren-
valuation method. Totowa, NJ: Rowman and zer’s critique. Psychological Review, 1 03 , 5 82–
Allanheld. 5 91 .
Kahneman, D. (2000a). Experienced utility and Kahneman, D., & Varey, C. A. (1 990). Propensi-
objective happiness: A moment-based ap- ties and counterfactuals: The loser that almost
proach. In D. Kahneman & A. Tversky (Eds.), won. Journal of Personality and Social Psychol-
Choices, values, and frames. New York: Cam- ogy, 5 9(6), 1 1 01 –1 1 1 0.
bridge University Press and the Russell Sage
Kahneman, D., Wakker, P. P., & Sarin, R. (1 997).
Foundation.
Back to Bentham? Explorations of experienced
Kahneman, D. (2000b). A psychological point of utility. Quarterly Journal of Economics, 1 1 2 ,
view: Violations of rational rules as a diagnostic 3 75 –405 .
of mental processes (commentary on Stanovich Keysar, B. (1 989). On the functional equivalence
and West). Behavioral and Brain Sciences, 2 3 , of literal and metaphorical interpretations in
681 –683 . discourse. Journal of Memory and Language, 2 8,
Kahneman, D. (2003 ). A perspective on judg- 3 75 –3 85 .
ment and choice: Mapping bounded rational- Koehler, J. (1 996). The base-rate fallacy reconsi-
ity. American Psychologist, 5 8, 697–720. dered: Descriptive, normative, and method-
Kahneman, D., & Frederick, (2002). In T. ological challenges. Behavioral and Brain Sci-
Gilovich, D. Griffin, & D. Kahneman (Eds.), ences, 1 9, 1 –5 3 .
Heuristics of intuitive judgment. New York: Kopp, R. (1 992). Why existence value should be
Cambridge University Press. used in cost–benefit analysis. Journal of Policy
Kahneman, D., Fredrickson, D. L., Schreiber, Analysis and Management, 1 1 , 1 23 –1 3 0.
C. A., & Redelmeier, D. A. (1 993 ). When more Krosnick, J. A., Li, F., & Lehman, D. R. (1 990).
pain is preferred to less: Adding a better end. Conversational conventions, order of informa-
Psychological Science, 4, 401 –405 . tion acquisition, and the effect of base rates and
Kahneman, D., & Miller, D. T. (1 986). Norm individuating information on social judgment.
theory: Comparing reality with its alternatives. Journal of Personality and Social Psychology, 5 9,
Psychological Review, 93 , 1 3 6–1 5 3 . 1 1 40–1 1 5 2.
Kahneman, D., & Ritov, I. (1 994). Determinants Langer, E. J. (1 978). Rethinking the role of
of stated willingness to pay for public goods: A thought in social interaction. In Harvey, Ickes,
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

2 92 the cambridge handbook of thinking and reasoning

& Kidd (Eds.), New directions in attribution re- Redelmeier, D., Katz, J., & Kahneman, D. (2001 ).
search (Vol. 2). Potomac, MD: Erlbaum. Memories of colonoscopy: A randomized trial.
Macchi, L. (1 995 ). Pragmatic aspects of the base- Redelmeier, D., & Kahneman, D. (1 996). Pa-
rate fallacy. Quarterly Journal of Experimental tients’ memories of painful medical treatments:
Psychology, 48A, 1 88–207. Real-time and retrospective evaluations of
Mandler, G., Hamson, C., & Dorfman, J. (1 990). two minimally invasive procedures. Pain, 66,
Tests of dual process theory – Word priming 3 –8.
and recognition. Quarterly Journal of Experi- Schreiber, C., & Kahneman, D. (2000). Deter-
mental Psychology, 42 (4), 71 3 –73 9. minants of the remembered utility of aver-
Margolis, H. (1 987). Patterns, thinking & cognition. sive sounds. Journal of Experimental Psychology:
Chicago: University of Chicago Press. General, 1 2 9, 27–42.
McFadden, D., & Leonard, G. K. (1 993 ). Issues Schwarz, N. (1 996). Cognition and communi-
in the contingent valuation of environmental cation: Judgmental biases, research methods, and
goods: Methodologies for data collection and the logic of conversation. Mahwah, NJ: Erlbaum.
analysis. In Hausman (Ed.), Contingent valua- Schwarz, N., Bless, H., Strack, F., Klumpp, G.,
tion. A critical assessment. Amsterdam: North- Rittenauer-Schatka, H., & Simons, A. (1 991 ).
Holland. Ease of retrieval as information: Another look
Mellers, B., Hertwig, R., & Kahneman, D. (2001 ). at the availability heuristic. Journal of Personal-
Do frequency representations eliminate con- ity and Social Psychology, 61 , 1 95 –202.
junction effects? An exercise in adversarial col- Schwarz, N., & Clore, G. L. (1 983 ). Mood, mis-
laboration. Psychological Science, 1 2 , 269–275 . attribution, and judgments of well-being: In-
Michotte, A. (1 963 ). The perception of causality. formative and directive functions of affective
New York: Basic Books. states. Journal of Personality and Social Psychol-
Mitchell, R., & Carson, R. (1 989). Using surveys ogy, 45 (3 ), 5 1 3 –5 23 .
to value public goods: The contingent valuation Schwarz, N., Strack, F., Hilton, D., & Naderer, G.
method. Washington, DC: Resources for the (1 991 ). Base rates, representativeness, and the
Future. logic of conversation: The contextual relevance
Nickerson, C. (1 995 ). Does willingness-to-pay of “irrelevant” information. Social Cognition, 9,
reflect the purchase of moral satisfaction? A re- 67–84.
consideration of Kahneman and Knetsch. Jour- Schwarz, N., & Vaughn, L. A. (2002). The avail-
nal of Environmental Economics and Manage- ability heuristic revisited: Ease of recall and
ment, 2 8, 1 26–1 3 3 . content of recall as distinct sources of infor-
Nisbett, R. E., Krantz, D. H., Jepson, C., & Kunda, mation. In T. Gilovich, D. Griffin, & D. Kahne-
Z. (1 983 ). The use of statistical heuristics in man (Eds.), Heuristics & biases: The psychology
everyday inductive reasoning. Psychological Re- of intuitive judgment (pp. 1 03 –1 1 9). New York:
view, 90(4), 3 3 9–3 63 . Cambridge University Press.
Novemsky, N., & Kronzon, S. (1 999). How are Shweder, R. (1 977). Likeness and likelihood in
base-rates used, when they are used: A compar- everyday thought – Magical thinking in judg-
ison of Bayesian and additive models of base- ments about personality. Current Anthropology,
rate use. Journal of Behavioral Decision Making, 1 8(4), 63 7–65 8.
1 2 , 5 5 –69. Simon, H. A., & Chase, W. G. (1 973 ). Skill in
Ordóñez, L., & Benson, L., III. (1 997). Decisions chess. American Scientist, 61, 3 94–403 .
under time pressure: How time constraint af- Sloman, S. A. (1 996). The empirical case for
fects risky decision making. Organizational Be- two systems of reasoning. Psychological Bulletin,
havior and Human Decision Processes, 71 (2), 1 1 9, 3 –22.
1 21 –1 40. Sloman, S. A. (2002). Two systems of reason-
Parducci, A. (1 965 ). Category judgment: A range- ing. In T. Gilovich, D. Griffin, & D. Kahne-
frequency model. Psychological Review, 72 , man (Eds.), Heuristics & biases: The psychology
407–41 8. of intuitive judgment (pp. 3 79–3 96). New York:
Pashler, H. E. (1 998). The psychology of attention. Cambridge University Press.
Cambridge, MA: MIT Press. Slovic, P., Finucane, M., Peters, E., & MacGre-
Poulton, (1 989). Bias in quantifying judgments. gor, D. G. (2002). The affect heuristic. In T.
London: Erlbaum. Gilovich, D. Griffin, & D. Kahneman (Eds.),
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

a model of heuristic judgment 2 93

Heuristics & biases: The psychology of intu- probability. Cognitive Psychology, 5 (2), 207–
itive judgment (pp. 3 97–420). New York: Cam- 23 2.
bridge University Press. Tversky, A., & Kahneman, D. (1 974). Judgment
Stanovich, K. E. (1 999). Who is rational?: Studies under uncertainty: Heuristics and biases. Sci-
of individual differences in reasoning. Mahwah, ence, 1 85 (41 5 7), 1 1 24–1 1 3 1 .
NJ: Erlbaum. Tversky, A., & Kahneman, D. (1 982). Judgments
Stanovich, K. E., & West, R. (2002). Individual of and by representativeness. In D. Kahneman,
differences in reasoning: Implications for the P. Slovic, & A. Tversky (Eds.), Judgment under
rationality debate? In T. Gilovich, D. Griffin, uncertainty: Heuristics and biases (pp. 84–98).
& D. Kahneman (Eds.), Heuristics & biases: New York: Cambridge University Press.
The psychology of intuitive judgment (pp. 421 – Tversky, A., & Kahneman, D. (1 983 ). Extensional
440). New York: Cambridge University vs. intuitive reasoning: The conjunction fallacy
Press. in probability judgment. Psychological Review,
Stevens, S. S. (1 95 7). On the psychophysical law. 90, 293 –3 1 5 .
Psychological Review, 64, 1 5 3 –1 81 . Tversky, A., & Kahneman, D. (1 986). Rational
Stevens, S. S. (1 975 ). Psychophysics: Introduction choice and the framing of decisions. Journal of
to its perceptual, neural, and social prospects. Business, 5 9, S25 1 –S278.
New York: Wiley. Wason, P. C. (1 960). On the failure to elimi-
Strack, F., Martin, L. L., & Schwarz, N. (1 988). nate hypotheses in a conceptual task. Quarterly
Priming and communication: The social deter- Journal of Experimental Psychology, 1 2 , 1 29–
minants of information use in judgments of life- 1 40.
satisfaction. European Journal of Social Psychol- Zajonc, R. B. (1 980). Feeling and thinking: Pref-
ogy, 1 8, 429–442. erences need no inferences. American Psychol-
Tversky, A., & Kahneman, D. (1 971 ). Belief in the ogist, 3 5 (2), 1 5 1 –1 75 .
law of small numbers. Psychological Bulletin, 76, Zajonc, R. B. (1 997). Emotions. In D. T. Gilbert,
1 05 –1 1 0. S. T. Fiske, & G. Lindzey (Eds.), Handbook of
Tversky, A., & Kahneman, D. (1 973 ). Availa- social psychology (4th ed., pp. 5 91 –63 2). New
bility: A heuristic for judging frequency and York: Oxford University Press.
P1 : IKB-IRK/KAB P2: IKB-IYP/KAB QC: IYP/JZO T1 : KOD
05 21 8241 76c1 2.xml CB798B/Holyoak 0 5 21 8241 7 6 February 2, 2005 0:41

294