C U R RE N T DI R EC TIO N S I N P SY CH O L O G I CA L SC I EN C E

How People Make Decisions

That Involve Risk
A Dual-Processes Approach
Valerie F. Reyna

Informatics and Decision-Making Laboratory, Departments of Surgery and Medicine, University of Arizona

ABSTRACT—Many health and safety problems, including war and psychological factors that shape the perception of many kinds of risks
terrorism, are by-products of how people reason about risk. and the decisions that involve those risks.
I describe a new approach to reasoning about risk that imple-
ments a modern dual-process model of memory called fuzzy- A SHORT HISTORY OF RESEARCH ON
trace theory. This approach posits encoding of both verbatim MEMORY AND REASONING
and gist representations, with reliance on the latter whenever
possible; dependence of reasoning on retrieval cues that access In the following section, I outline recent advances in research on the
stored values and principles; and vulnerability of reasoning to psychology of risk, drawing on modern concepts of memory re-
processing interference from overlapping classes of events, presentation, retrieval, and processing. New ideas about dual pro-
which causes denominator neglect in risk or probability judg- cesses in memory and reasoning make it possible to predict risk
ments. These simple principles explain classic and new findings, perceptions and decisions that involve risks, rather than merely ex-
for example, the finding that people overestimate small risks but plain them after the fact. Before proceeding, however, it is useful to
ignore very small risks. Fuzzy-trace theory differs from other discuss the background of research on memory and reasoning, which
dual-process approaches to reasoning in that it places intuition led up to these new developments regarding risk.
at the apex of development, considering fuzzy intuitive process- The conception of memory that has dominated psychology for
ing more advanced than precise computational processing (e.g., decades is the computer metaphor. That is, information is considered
trading off risks and rewards). The theory supplies a conception to be held in a temporary store, called working memory, that has
of rationality that distinguishes degrees of severity of errors in limited capacity. The idea that working memory has a limited capacity
reasoning. It also includes a mechanism for achieving con- goes back to George Miller’s research on the ‘‘magical’’ number seven
sistency in reasoning, a hallmark of rationality, by explaining (the number of chunks of information that he estimated could be
how a person can treat superficially different reasoning processed at one time). Miller’s work influenced many psychologists,
problems in the same way if the problems share an underlying notably Herbert Simon, who was a pioneer in research on judgment
gist. and decision making. Relying on the assumption that working memory
was limited, Simon introduced the concept of bounded rationality. In
KEYWORDS—risk perception; risky decision making; fuzzy-trace
short, Simon thought that human rationality is limited because of
theory; intuition; dual processes in reasoning
limitations in human information processing.
The assumption that human information processing is limited
continued to be the main motivating assumption behind subse-
As I write these words, coalition troops are at war in Iraq. The decision
quent approaches to judgment and decision making, including both
to go to war with Iraq, like many decisions, was based on a perception
heuristics-and-biases and fast-and-frugal approaches (see Gilovich,
of risk, that is, the perceived threat posed by Iraq under Saddam
Griffin, & Kahneman, 2002). According to these views, humans use
Hussein. Key government officials also believed that inaction with
heuristics or fast-and-frugal reasoning strategies as mental shortcuts
respect to Iraq posed a greater risk than taking action. Since Sep-
because of information processing limitations. The need to conserve
tember 11, 2001, foreign threats and risks of terrorism have loomed
limited mental resources was seen as the driving force in reasoning,
large in people’s perception, dwarfing perennial killers such as
judgment, and decision making.
highway accidents and heart disease. In this article, I discuss the
Although the computer metaphor guided research for decades, it
has recently come under attack. Memory researchers have begun to
Address correspondence to Valerie Reyna, Department of Psychol- criticize the concept of limited capacity in working memory (Nairne,
ogy, University of Texas at Arlington, Box 19528, Arlington, TX 2002). Moreover, evidence indicates that working memory capacity
76019-0528; e-mail: [email protected]. is unrelated to the accuracy of reasoning, judgment, and decision

60 Copyright r 2004 American Psychological Society Volume 13—Number 2

 Valerie F. Reyna

making. Researchers have studied many tasks, and each has shown differences is that fuzzy-trace theorists place intuition at the apex of
the same result: Reasoning is independent of memory (Reyna & development rather than at the nadir. This view of intuition is sup-
Brainerd, 1995). In addition, researchers have noted that dramatic ported by developmental studies of children’s learning and of adults’
errors in reasoning occur on tasks that impose few demands on acquisition of expertise, which have demonstrated a progression from
memory capacity (Shafir & LeBoeuf, 2002). detail-oriented and computational processes (e.g., trading off the
These findings of independence between reasoning and re- magnitudes of risk and reward) to fuzzy and intuitive processing
membering prompted the development of fuzzy-trace theory (Reyna & (people process less information more qualitatively as development
Brainerd, 1995). Fuzzy-trace theory accounts for this independence progresses; Jacobs & Klaczynski, 2002; Reyna & Ellis, 1994). For
through the assumption that people form two kinds of mental re- example, my colleagues and I found that in deciding whether to admit
presentations, verbatim and gist representations, but rely primarily a patient with chest pain to the hospital, expert cardiologists pro-
on gist. Gist representations are fuzzy (less precise than verbatim cessed fewer dimensions of information than less expert physicians
representations) traces of experience in memory, hence the name did and also processed those dimensions in a cruder all-or-none
fuzzy-trace theory. Because two kinds of representations are posited, fashion (patients were either at risk or not at risk of an imminent heart
fuzzy-trace theory is an example of a dual-process model of memory. attack; Reyna et al., 2003). Another difference between fuzzy-trace
Fuzzy-trace theory explains findings of reasoning-remembering theory and other recent dual-process approaches is the role they give
independence because responses to memory tests often require the emotion in decisions that involve risk. Some dual-process theorists
details found in verbatim representations, whereas responses to rea- elevate emotion above reason, arguing that decision makers ought to
soning tests often require only gist representations. Thus, reasoning rely on their gut feelings. As in other dual-process theories, emotion
accuracy is independent of memory accuracy because gist repre- is important in fuzzy-trace theory: Gist representations capture the
sentations are independent of verbatim representations. This ex- meaning of experience, including its emotional meaning. However,
planation was confirmed by experiments in which the reliance on emotion is not viewed as an unerring signal of what is adaptive (Reyna
verbatim versus gist representations was actively manipulated (e.g., et al., 2003).
through instructions or by varying the time delay prior to the memory Fuzzy-trace theory provides an alternative conception of rationality,
test), producing positive dependency, negative dependency, and in- compared with traditional theories of judgment and decision making.
dependence between memory and reasoning under theoretically pre- Any theory of rationality must provide a mechanism for achieving
dicted conditions (Reyna & Brainerd, 1995). Finally, assumptions of what is called the consistency criterion: Superficially different rea-
fuzzy-trace theory have been modeled mathematically to secure soning problems should be treated consistently. For example, if you
quantitative estimates of the contributions of verbatim and gist repre- choose surgery when it is described as having an 80% survival rate,
sentations, and associated judgment processes, in a variety of tasks you should still choose it if it is described as having a 20% mortality
(Reyna, Lloyd, & Brainerd, 2003). rate. In fuzzy-trace theory, the mechanism for achieving consistency is
Fuzzy-trace theory is a dual-process theory of reasoning as well as found in the sharing of gist. To the rational decision maker who is
memory. People can use either verbatim or gist representations to contemplating surgery, the gist of the risk of surgery is the same
solve reasoning problems (although they mainly use gist). Researchers whether it is described in terms of survival or mortality rates. Fuzzy
who study fuzzy-trace theory treat judgment-and-decision-making processing is also a source of irrational biases and inconsistencies,
tasks as examples of reasoning problems, but work from a conception however. My colleagues and I (Reyna et al., 2003) have developed a
of reasoning that is different from traditional approaches. Traditional taxonomy of these biases and inconsistencies in which degrees of
theories of reasoning are modeled on logic or computation; reasoning rationality are specified. That is, errors in reasoning, judgment, and
is said to occur in a series of ordered steps (e.g., premises are first decision making are categorized as more or less advanced according
understood and then integrated to draw conclusions), and precision is to the type of processing that produced them. Some kinds of errors are
considered a hallmark of good reasoning. In contrast, according to seen as more irrational—more serious failings—than others.
fuzzy-trace theory, reasoning processes unfold in parallel rather than In sum, ideas about working memory capacity that were tradition-
in series, often operating on the barest senses of ideas (the gist of ally invoked in judgment-and-decision-making research have been
a problem), and are fuzzy or qualitative rather than precise. Thus, challenged by findings that reasoning and remembering are inde-
a person presented with a reasoning problem encodes multiple pendent. Researchers such as Simon simply assumed that working
representations of the same problem facts, retrieves reasoning prin- memory capacity was an important factor, but research that has ac-
ciples from his or her stored knowledge (e.g., the principle that tually examined the relation between memory and reasoning does not
probability depends on the number of wins out of the total number of support this assumption. Fuzzy-trace theory emphasizes alternatives
plays), and applies the reasoning principles to the mental repre- to capacity explanations, independence of reasoning performance
sentations of the problem facts. In this view, human reasoning is a from memory constraints, and dual-process assumptions about mem-
messy process: Multiple perceptions of the problem are encoded, the ory and reasoning. Intuition has a special place in fuzzy-trace theory,
right reasoning principle might or might not be retrieved, and the and is considered an advanced form of reasoning because of devel-
execution of processing (applying principles to problem representa- opmental evidence about the typical sequence of errors as reasoners
tions) is unreliable. Processing is considered to be unreliable because gain expertise in reasoning. Intuitions in reasoning come about as a
of interference (getting bogged down in the execution of processing; result of parallel processing of multiple representations, uncertain
discussed in the next section), as opposed to logical incompetence or retrieval of reasoning principles, and an overarching preference
memory overload. for gist representations (as opposed to verbatim representations). What
Fuzzy-trace theory differs from other dual-process approaches do these assumptions predict about risk perception and decision
to reasoning in some important respects. One example of these making?

Volume 13—Number 2 61
 Making Decisions That Involve Risk

FUZZY PERCEPTIONS OF RISK mark quantitatively, but they have a better grasp of the correct gist
that the procedure involves taking a risk. Thus, understanding the gist
The period from the 1970s to the present produced a number of key that the surgery involves taking a risk is essential for informed con-
findings about how people perceive risk and make decisions that in- sent. More generally, the literature is replete with examples of how a
volve risk. For example, researchers found that people often over- person’s representation of gist (which reflects the person’s education,
estimate small risks (e.g., risks of complications from vaccinations) emotion, culture, and worldview), rather than verbatim information,
and that they choose options that involve taking a risk when potential governs the perception of risk.
outcomes are described as losses (e.g., a 2/3 chance that 600 people Table 1 summarizes additional examples of how fuzzy-trace theory’s
will die) but avoid taking a risk when the same outcomes are de- assumptions about gist predict results related to risk perception. For
scribed as gains (e.g., a 1/3 chance that 600 people will be saved; example, the table shows that very low risks described in relative
Gilovich et al., 2002). Table 1 provides examples of both over- terms, such as ‘‘half the risk,’’ are perceived differently than the same
estimation of small risks and shifts in risk taking associated with objective risks described in absolute terms, such as ‘‘a .0000030
framing the same options as gains versus losses. chance of injury for improved tires and a .0000060 risk for standard
These classic findings used to be interpreted from a psychophysical tires’’ (see Stone, Yates, & Parker, 1994).
viewpoint (e.g., perceptions of quantities such as probabilities were Two examples from Table 1 illustrate how people base their esti-
treated like perceptions of quantities of light or loudness), but are now mates of risks on the gist of a class of events, objects, or people. In
accounted for with dual-process explanations, such as those discussed studies involving hypothetical and real patients, physicians tended to
earlier. Dual-process explanations have also led to new, counter- equate patients’ probability of either coronary disease or heart attack
intuitive predictions (e.g., see Lloyd & Reyna, 2001; Reyna & Adam, with the overall probability of coronary disease, systematically ig-
2003; Reyna & Hamilton, 2001). For both old and new findings, three noring the probability that patients could be at risk of a heart attack
questions (about representation, retrieval, and processing) form the without having a significant probability of coronary disease (despite
basis of dual-process explanations in fuzzy-trace theory. Specifically, knowledge that such patients exist; Reyna et al., 2003). Physicians
researchers explain particular judgments or decisions that involve risk apparently estimated each risk by thinking of the gist of that class of
by asking: (a) How are classes of events that are involved in reasoning patients: The typical coronary disease patient is not at risk of an
about risk represented in memory? (b) What reasoning principles are imminent heart attack, but the typical patient at risk of a heart attack
cued in the particular context of reasoning? (c) Is reasoning subject to has coronary disease. Adding these two classes of risks together to
processing interference, especially from thinking about overlapping determine overall risk, physicians ignored the atypical patients who
classes of events (e.g., the intersection of the two classes of people are at risk of a heart attack but do not have significant coronary
with genetic mutations and people with breast cancer)? Answers to disease. Similarly, physicians and other health care professionals
these questions have been derived from research on judgments of overestimated risk reduction provided by condoms because the gist of
probability and relative frequency, which are closely related to sexually transmitted infections is that they are fluid-borne infections
judgments of risk. (Reyna & Adam, 2003). Condoms seem to provide more protection
than they actually do because they provide a barrier that reduces the
sharing of fluids. However, condoms do not provide as much protec-
Representations of Risk tion against skin-to-skin transmission; they do not provide as much
Consider the first question, about how classes of events involving risk reduction in risk as one might think based on a gist that diseases are
are represented in memory. Suppose that a patient must decide transmitted by sharing fluids. Again, the nongist class of risks—in-
whether to have surgery to prevent a 22% risk of stroke, and the fections transmitted in psychologically atypical ways, such as by skin-
surgeon informs the patient that the risk of dying from the surgery is to-skin contact—was systematically ignored.
2%. What will the patient remember about the risks that are relevant
to making this decision? According to fuzzy-trace theory, patients
encode both verbatim and gist representations into memory, but ver- Retrieval of Reasoning Principles
batim memories become rapidly inaccessible. Data about patients’ With respect to the second question, concerning retrieval of reasoning
memory for surgical risks confirmed this prediction: Many patients principles, it is important to note that people often know and endorse
misremembered the verbatim numbers indicating the risks of specific the correct reasoning principle even though their reasoning on a
treatment options (Reyna & Hamilton, 2001). As discussed, however, specific problem is not based on it. For example, for people given
decisions are usually based on gist representations of information, a bag containing 10 blue plastic tokens and 5 red plastic tokens, it is
rather than verbatim representations. Although the data showed that a simple matter to judge the risk of accepting a wager on blue versus
patients misremembered the verbatim numbers, they were able to red. However, if asked which is more probable, blue tokens or plastic
order the relevant classes of risks correctly: The risk of going without tokens, adults take a long time to answer, and children as old as 10 get
surgery was generally remembered as higher than the risk of surgery. the answer wrong. Children insist that there are more blue tokens than
Patients understood the gist of the relative risks, their relative plastic ones, and make the same mistake even with problems having
magnitudes, but did they understand the information essential for familiar content (e.g., given pictures of 7 cows and 3 horses, which
informed consent? According to fuzzy-trace theory, some errors are total 10 animals, they insist that there are more cows than animals).
worse than others. Patients who err by declaring that surgery with 2% Contrary to speculation, children do not get the token question wrong
risk has no risk have not extracted the proper gist for informed con- because of its unfamiliar content (tokens as opposed to farm animals)
sent, which requires recognition that the procedure involves some or because of linguistic ambiguity. Errors persist even when linguistic
risk. Patients who err by recalling the risk to be 5% are further off the ambiguity is eliminated (i.e., the plastic tokens are described as ‘‘red

62 Volume 13—Number 2
 Valerie F. Reyna

or blue’’ to avoid the possible misinterpretation that the phrase kens) are also more numerous or probable than the subclasses (blue
‘‘plastic tokens’’ refers to the smaller class of red tokens). Children fail plastic tokens) included in them, although they endorse that principle
to retrieve the principle that more inclusive classes (e.g., plastic to- in other contexts (Reyna, 1991).

TABLE 1
Explaining Biases in Risk Perception and Decision Making
Objective Subjective Gist Explanation
Type of bias Task risk risk representation of bias
Overestimation of Estimate risk of 8 deaths per 2.05  108 23 deaths per 2.05  108 Smallpox vaccine is Denominator
small risks smallpox vaccina- U.S. residents U.S. residents risky neglect
(Gilovich, Griffin, tion (1978)
& Kahneman,
2002)
Overestimation of Estimate risk of 17 people with heart Risk considered large ‘‘Despite the low Denominator
small risks (‘‘3rd smallpox vaccina- problems per about enough for IOM and two numbers’’ (IOM), neglect
Smallpox Vaccine tion (2003) 25,000 vaccinated states to recommend smallpox vaccine
Death Reported,’’ suspension is risky
2003)
Framing risks Choose between Gain frame: Gain frame: Gain frame: Qualitative
(Reyna & options to combat A: 200 people saved. The risky option is Saving some representation of
Brainerd, 1991) disease expected B: 1/3 probability frightening people is better options; retrieval
to kill 600 people 600 saved; 2/3 than saving none, of principles, such
probability no people so choose sure as saving some
saved. option people is better
Loss frame: Loss frame: Loss frame: than saving none
C: 400 people die. The risky option offers Nobody dying is
D: 1/3 probability no hope better than some
one dies; 2/3 dying, so choose
probability 600 risky option
people die.

Base-rate neglect Estimate risks of 20 accidents playing on 20 accidents are riskier Slides are dangerous Denominator
(Reyna & playground slides and 5 on than 5 accidents neglect
Brainerd, 1995) accidents swings; more children
play on slides

Biases in Estimate prob- Probability of being Physicians estimated: Probability typical Neglects nongist
integrating risks ability of CAD or CAD patient or being probability of CAD or CAD patient (not at class of patients
(Reyna, Lloyd, & risk of MI to MI patient 5 MI equals the risk of MI) plus at MI risk without
Brainerd, 2003) decide hospital CAD&MI 1 probability of CAD probability typical CAD
admission CAD&notMI 1 MI patient (also has
notCAD&MI CAD) 5 probability
of CAD (i.e.,
CAD&notMI 1
CAD&MI 5 CAD)
Relative vs. Choose between Absolute risk format: For very small risks Very small risks are Qualitative
absolute risk products and Example: Injury risk (tire blowouts, airplane edited to ‘‘essen- representation
(Stone, Yates, & estimate how is .0000030 for an accidents), people paid tially nil risk’’ (so of options;
Parker, 1994) much willing to improved product and more for risk reduction both risks are seen retrieval of
pay for safer .0000060 for the with the relative than as negligible in the principles, such as
products standard product absolute risk format; for absolute format more safe is better
Relative risk format: larger low risks, there and subjects are than less safe
Example: Injury risk was no difference not willing to pay
for improved product between formats more for one than
is half that for the other)
standard product

Volume 13—Number 2 63
 Making Decisions That Involve Risk

TABLE 1
Continued
Objective Subjective Gist Explanation
Type of bias Task risk risk representation of bias
Overestimation Estimate risk reduc- Risk reduction Adolescents, physicians, Gist of STIs for older Despite knowledge,
of reduction in tion for STIs provided by condoms and other health people (syphilis, professionals
risk (Reyna & provided by using limited because some care professionals over- gonorrhea) and neglect nongist
Adam, 2003) condoms, based STIs transmitted estimated the extent of younger people but highly prev-
on FDA labeling skin-to-skin risk reduction for STIs (HIV) is fluid- alent classes of
provided by condoms borne infection, infections
reduced with me- transmitted skin-
chanical barriers to-skin (HPV,
(e.g., condoms) HSV)

Note. IOM 5 Institute of Medicine; CAD 5 coronary artery disease, which leads to heart attacks; MI 5 myocardial infarction or heart attack; STIs 5 sexually
transmitted infections; FDA 5 Food and Drug Administration; HIV 5 human immunodeficiency virus; HPV 5 human papillomavirus; HSV 5 herpes simplex
virus.

The developmental psychologist Jean Piaget originally introduced Denominator neglect occurs regardless of whether risk is ex-
the class-inclusion task to study cognitive illusions in children (al- pressed with probabilities or frequencies, and instead results from
though, as predicted by fuzzy-trace theory, adults also find the class- interference caused by overlapping classes (Reyna, 1991; Wolfe,
inclusion question difficult because the classes are overlapping, a 1995). Denominator neglect accounts for overestimation of low-
situation discussed in the next section). In an analogue of this class- probability risks (it is the denominator that makes them low),
inclusion task, adults typically rank the chances of a stereotypical conversion errors in which the conditional probability of A given B
liberal activist being a ‘‘feminist bank teller’’ as greater than her being is confused with the probability of B given A (the numerators
a ‘‘bank teller’’ (again, even when possible ambiguity is eliminated— of conditional probabilities are identical; only the denominators
‘‘a bank teller whether or not she is a feminist’’), despite endorsing the differ1), and the neglect of base rates (the overall frequency of events,
inclusion principle that no one could be more likely to be a feminist the denominators, are neglected in favor of salient numerators; see
bank teller than a bank teller. (Naturally, the gist of a class could be Table 1).
more or less extensive than its technically more inclusive class be- The confusion of conditional probabilities is readily illustrated in
cause gist captures the meaning or intension of a class rather than its widespread misunderstanding of genetic risks, such as the probability
extension; Reyna et al., 2003.) The gist of the class of feminist bank of a woman developing breast cancer if she has BRCA1 or BRCA2
tellers fits the liberal activist better than the gist of the class of bank mutations versus the probability of a woman with breast cancer having
tellers (conservative types), contrary to the inclusion principle. In BRCA1 or BRCA2 mutations (Reyna, Lloyd, & Whalen, 2001). Pa-
another article (Reyna, 1991), I have described how retrieval cues in tients confuse the two probabilities and think the latter probability is
the questions posed to reasoners determine whether they retrieve the high, even after genetic counseling. Base-rate neglect occurs when
correct reasoning principle. people focus on the number of times a target event has happened
without thinking about the overall number of opportunities for it to
Interference Effects in Processing happen; 20 accidents on slides might be significant or not depending
The third question concerns how errors in reasoning about risk arise on the number of times children played on the slides. In fuzzy-trace
as a result of processing interference. According to fuzzy-trace theory, theory, none of these processing errors reflects a lack of logic or
overlapping classes cause processing interference. Reasoners focus on conceptual competence, nor do the errors reflect working memory
target members of a class and lose track of the larger universe of limitations. Rather, they are low-level bookkeeping errors that are
possibilities. That is, judgments of risk involve a target class of events made even late in development among advanced reasoners and are
(e.g., the winning tokens, the patients who survived) and a larger, more easily remedied by keeping classes of events distinct (see Reyna,
inclusive class of events that includes both targets and nontargets 1991, for effective interventions with children and Lloyd & Reyna,
(e.g., the losing tokens, the patients who did not survive). People 2001, for similar interventions with physicians).
compare target and nontarget events (e.g., Are there more winning Figure 1 illustrates the tenacity of denominator neglect even among
tokens or losing tokens?) and automatically extract the gist of which individuals with expertise relevant to the problem they are consider-
class of events is ‘‘bigger.’’ However, they pay less attention to the ing. The following identical problem was presented to 82 physicians,
more inclusive class, which is the denominator in the calculation of 34 other health care professionals, 93 health educators (who counsel
risk: Seventeen heart complications from smallpox vaccinations are high school students about risks), and 258 high school students:
more salient than the total number of vaccinations administered
(Table 1). (Note the similarity to the earlier example in which the class 1
The probability of A given B is calculated as the probability of both A and B
of feminist bank tellers is more salient in the context of judging a
occurring, divided by the probability of B; the probability of B given A is
liberal activist than is the total class of bank tellers, whether or not calculated as the probability of both A and B occurring, divided by the
they are feminists.) probability of A.

64 Volume 13—Number 2
 Valerie F. Reyna

Suppose the pre-test probability of disease is 10% and a diag-

Proportion Correct
nostic test has 80% sensitivity (80% of people with disease test
positive) and 80% specificity (80% of people without the disease
test negative). The test result is positive. What is the probability
0.35
of disease, 30% or 70%?

Ignoring the denominator (and thus confusing the conditional prob-

ability of a positive result given disease with the conditional prob- 0.3
ability of disease given a positive result) yields an answer of 80%,
which might be adjusted downward to 70% because of the low pretest
probability. However, the correct answer is closer to 30% (given that
90% of people do not have the disease, false positives outnumber true 0.25
positives). As Figure 1 shows, most people did not select the correct
answer. The disparity between actual and perceived risk was large for
people of all ages and levels of expertise. Those theorists who argue
that errors in these problems are actually smart ignore the pervasive 0.2
problem of false positives and negatives in medicine and other im-
portant real-life contexts (e.g., screening federal mail for biological
agents). It is important in the real world to discriminate good from bad
0.15
reasoning about risk.
Although processing interference remains a problem throughout life
for most reasoners, representation of information in memory and re-
trieval of reasoning principles undergo changes as knowledge and 0.1
experience are accumulated. As children grow to adulthood, they
change from quantitative reasoners, who trade off amounts of risk and
reward, to qualitative reasoners, who process categorical (e.g., some
money is better than no money, no risk is better than some risk) gist 0.05
(Reyna et al., 2003). For example, given a choice between a sure win
of 5 prizes and a 50% chance of winning 10 prizes or winning nothing,
adults prefer a sure win, whereas young children prefer the risky
0
option that offers the possibility of more prizes. Young children reason
MD HCP EDUC HIGH SCH
quantitatively about the numerical difference between 10 prizes and 5
prizes, and they weigh this difference in outcomes (which favors the Fig. 1. Illustration of denominator neglect that occurs despite expertise.
gamble) against the difference in the probability of winning between Participants were asked to indicate whether 30% or 70% was a better
the sure option and the gamble (Reyna & Ellis, 1994). Adults reason estimate of the risk of low-prevalence disease when a patient tested
positive using a less-than-perfect test (80% sensitivity and 80% specific-
qualitatively that winning some prizes is better than maybe winning ity). The respondents were physicians (MD), health care professionals
none: The quantity of prizes is not the determinative factor. (mainly nurses; HCP), health educators (EDUC), and high school stu-
Similarly, given a choice between hanging out at the mall or going dents (HIGH SCH). From Reyna and Adam (2003) and additional
to an unsupervised party, many adolescents prefer the risky option unpublished data.
that offers the possibility of more fun. Although they are more risk
averse than young children, adolescents analyze the decision by
weighing risks and benefits like young children do; the mall-versus- According to fuzzy-trace theory, the crude all-or-none categorization
party decision depends on the amount of fun and the degree of risk. that adults apply in this situation is more advanced than the subtle
Most parents would view these options in starker terms: No amount of shadings that adolescents perceive. Global categorical policies (e.g.,
fun can compensate for the risks, contrary to a strict cost-benefit avoid risk) exist on a higher plane of rationality, cutting across details
analysis. For adults (but not necessarily for children and adolescents), of amounts of risk and reward.
the number of bullets in the chamber is irrelevant to the decision of
whether to play Russian roulette. Adolescents deciding whether to go
to the party mentally represent the problem as involving whether they
would prefer to go to the mall and have a specific amount of fun or take CONCLUSIONS
a calculated, objectively small risk and have much more fun at the
party. Using the reasoning principle that more fun is better than less Perceptions of risk precipitate decisions to go to war, to vaccinate
fun, and applying it to the mental representation just described, which against smallpox, to reduce exposure to sexually transmitted diseases,
incorporates both risks and rewards, they decide that going to the to report suspicious illnesses to public heath authorities, or to admit to
party is preferred. In contrast, parents represent the decision as being the hospital a patient who has chest pain and is at risk of a heart
between (a) having fun at the mall and (b) taking a risk and either attack. Psychological science now offers predictive theories that ex-
having fun at the party or losing everything. They apply the reasoning plain dangerous biases in reasoning about risk and provide insights
principle that one should avoid risk and avoid catastrophic outcomes. into how these biases can be corrected. In particular, research on

Volume 13—Number 2 65
 Making Decisions That Involve Risk

memory provides a rich repertoire of empirically tested concepts Foundation, and the Academic Medicine and Managed Care Forum
that help explain the psychology of risk: the encoding of dual verbatim (Schering Plough/Integrated Therapeutics Group). This work was also
and gist representations of information and the reliance on the latter supported by a grant from the Agency for Healthcare Research and
whenever possible; the dependence of reasoning on retrieval cues Quality to the Arizona Center for Education and Research on Ther-
that elicit values and principles stored in long-term memory; apeutics (CERT). Special thanks are due to Allan Hamilton, whose
and the vulnerability of reasoning about risk to processing inter- support made it possible to sustain the work of the Informatics and
ference from overlapping classes of events, resulting in denominator Decision-Making Laboratory during the author’s recent term of gov-
neglect. ernment service.
These memory concepts are used in fuzzy-trace theory to explain
how dual processes operate in reasoning and decision making. The
basic idea is that intuitive gist-based reasoning increasingly supplants REFERENCES
analytical verbatim-based reasoning as children gain experience and
as novices become experts. As Figure 1 shows, this development does 3rd smallpox vaccine death reported. (2003, March 29). cbsnews.com. Retrieved
from http://www.cbsnews.com/stories/2003/03/29/health/main546669.
not mean that reasoning becomes immune to all sources of error. In
shtml
contrast to some other dual-process theories, this theory posits that
Gilovich, T., Griffin, D., & Kahneman, D. (2002). Heuristics and biases: The
intuition is advanced, errors are bad, and emotions are not necessarily psychology of intuitive judgment. New York: Cambridge University Press.
good. Within this theory, a detailed processing model distinguishes Jacobs, J.E., & Klaczynski, P.A. (2002). The development of judgment and
degrees of rationality based on observations of similar developmental decision making during childhood and adolescence. Current Directions in
progressions in types of errors across many tasks. The processing Psychological Science, 11, 145–149.
model also accounts for well-known findings about risk perception, Lloyd, F.J., & Reyna, V.F. (2001). A Web exercise in evidence-based medicine
such as overestimation of small risks because of denominator neglect, using cognitive theory. Journal of General Internal Medicine, 16,
as well as new findings, such as crude all-or-none processing of risks 94–99.
among expert cardiologists (compared with more fine-grained pro- Nairne, J.S. (2002). Remembering over the short-term: The case against the
standard model. Annual Review of Psychology, 53, 53–81.
cessing by physicians with less expertise).
Reyna, V.F. (1991). Class inclusion, the conjunction fallacy, and other cognitive
Gist-based reasoning requires neither exact nor inexact numbers
illusions. Developmental Review, 11, 317–336.
(although decision makers prefer to extract their own gist as opposed
Reyna, V.F., & Adam, M.B. (2003). Fuzzy-trace theory, risk communication,
to having it extracted for them). Such reasoning can proceed when ‘‘it and product labeling in sexually transmitted diseases. Risk Analysis, 23,
is impossible to quantify the risk,’’ to quote the director of the Centers 325–342.
for Disease Control and Prevention when she discussed the risk of Reyna, V.F., & Brainerd, C.J. (1991). Fuzzy-trace theory and framing effects in
terrorist attacks using smallpox. If gist-based reasoning is robust in choice: Gist extraction, truncation, and conversion. Journal of Behavioral
the face of this kind of ambiguity, as research suggests, and if it de- Decision Making, 4, 249–262.
velops with experience, it ought to be possible to marshal and hone it Reyna, V.F., & Brainerd, C.J. (1995). Fuzzy-trace theory: An interim synthesis.
to produce better decision making when outcomes are uncertain. In Learning and Individual Differences, 7, 1–75.
order to identify the most efficient nurturing conditions for better Reyna, V.F., & Ellis, S.C. (1994). Fuzzy-trace theory and framing effects in
children’s risky decision making. Psychological Science, 5, 275–279.
gist-based thinking, however, researchers need theories that hold
Reyna, V.F., & Hamilton, A.J. (2001). The importance of memory in informed
reasoners accountable for bad reasoning. A challenge in future re- consent for surgical risk. Medical Decision Making, 21, 152–155.
search will be to acknowledge that intuition is adaptive and even Reyna, V.F., Lloyd, F.J., & Brainerd, C.J. (2003). Memory, development, and
advanced, but nevertheless imperfect in systematic ways. rationality: An integrative theory of judgment and decision-making. In
S. Schneider & J. Shanteau (Eds.), Emerging perspectives on judgment and
decision research (pp. 201–245). New York: Cambridge University Press.
Recommended Reading Reyna, V.F., Lloyd, F.J., & Whalen, P. (2001). Genetic testing and medical
Gilovich, T., Griffin, D., & Kahneman, D. (2002). (See References) decision-making. Archives of Internal Medicine, 161, 2406–2408.
Stanovich, K.E., & West, R.F. (2000). Individual differences in reasoning: Shafir, E., & LeBoeuf, R.A. (2002). Rationality. Annual Review of Psychology,
Implications for the rationality debate? Behavioral and Brain Sciences, 53, 491–517.
23, 645–726. Stone, E.R., Yates, J.F., & Parker, A.M. (1994). Risk communication: Absolute
versus relative expressions of low-probability risks. Organizational Be-
havior and Human Decision Processes, 60, 387–408.
Wolfe, C. (1995). Information seeking on Bayesian conditional probability
Acknowledgments—Preparation of this article was supported by problems: A fuzzy-trace theory account. Journal of Behavioral Decision
grants from the National Institutes of Health, the National Science Making, 8, 85–108.

66 Volume 13—Number 2