Advanced Review

Decision making under risk

and uncertainty
Joseph G. Johnson1∗ and Jerome R. Busemeyer2

Decision making is studied from a number of different theoretical approaches.

Normative theories focus on how to make the best decisions by deriving algebraic
representations of preference from idealized behavioral axioms. Descriptive
theories adopt this algebraic representation, but incorporate known limitations
of human behavior. Computational approaches start from a different set of
assumptions altogether, focusing instead on the underlying cognitive and
emotional processes that result in the selection of one option over the other. This
review comprehensively but concisely describes and contrasts three approaches in
terms of their theoretical assumptions and their ability to account for behavioral and
neurophysiological evidence from experimental research. Although each approach
contributes substantially to our understanding of human decision making, we
argue that the computational approach is more fruitful and parsimonious for
describing and predicting choices in both laboratory and applied settings and for
understanding the neurophysiological substrates of decision making.  2010 John
Wiley & Sons, Ltd. WIREs Cogn Sci

D ecision making is a faculty that is evident in

nearly everything we do. From the commonplace
to the consequential, our lives are guided by the
are interested in the explanation for a specific course
of action; some prefer to know what decision should
be made, while others strive to understand why. In
decisions we make. Therefore, it is important to the current review, we identify three major streams of
understand how we make decisions, so that we may development in decision theory that can be classified
be aware of how various factors may have exerted an according to the focal behaviors and functional nature
influence on past decisions, and so that we may be able of the corresponding decisions.
to improve upon future decisions. Indeed, one could First, a great deal of foundational decision
easily argue that our decision-making ability and the research was focused on the notion of making ‘opti-
agency it provides us is what separates us from lower mal’ decisions. Given a particular situation, how
order animals. should one go about selecting the best among compet-
Because decision making is so central to our ing alternatives? This normative research stream has
lives, it is not surprising that it receives research the goal of reducing a decision situation essentially
attention from a wide range of disciplines: cognitive to a mathematical optimization problem and finding
psychology, economics, political science, marketing, the correct solution to this problem. It is responsible
social psychology, engineering, philosophy, and more. for treating decision outcomes as random variables,
Although this breadth in contributing disciplines casting decision problems in expectation terms, and
is beneficial in bringing multiple perspectives to deriving solutions that maximize the expected utility
bear, it is also (at least partly) responsible for among probability distributions of outcomes pro-
somewhat divergent or inconsistent research goals. duced by different actions.
Some researchers are interested in how to make the Second, this treatment gave rise to a counterpoise
‘best’ decision under specific conditions, while others among researchers who wanted to impart a more
∗ Correspondence to: [email protected] psychological and constrained view of decision
1 Department of Psychology, Miami University, Oxford, OH 45056, making. Because humans often make ‘suboptimal’
USA decisions, how can we describe and predict the choices
2 Psychological and Brain Sciences, Indiana University, Bloomington, that one will make in a particular situation? Research
IN 47405, USA with this descriptive focus attempts to describe how
DOI: 10.1002/wcs.76 humans actually make decisions, rather than trying to

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

find ideal decisions for any given situation. People are selection of a single course of action X described
not likely to be able to apply the analytic machinery by the value of the possible outcomes {x1 , x2 ,
developed within the normative approach, and the . . ., xn } that could result from selecting the action
descriptive perspective can be characterized by the and the associated probability that each outcome
addition of psychological factors that embellish this would occur if the action were selected {p1 , p2 ,
basic machinery. That is, this approach retains a form . . ., pn }. This representation reduces the choice task
of the utility maximization goal but is focused on to one of selecting from among competing simple
what psychological adjustments need to be made to random variables (see Ref 1, for a critique of this
account for observed human decisions. ‘gambling metaphor’, or Ref 2, for an alternative
Third, beyond providing psychological meaning ‘naturalistic’ research paradigm). The simplest rule,
and justification to descriptive modifications of the mathematically, is then to select the option X that has
normative theories, many recent researchers examine the highest expected value EV(X):
the decision processes themselves, rather than just
the final choice. What mental or neural operations

n

are taking place that lead to the selection of one EV(X) = pi xi (1)
option over another in a given situation? This i=1

computational approach seeks to understand what

the underlying (cognitive and emotional) processes For example, take a decision with two options:
are that produce the observable actions predicted by (A) a certain outcome valued at $1 million, and
the descriptive theories. Rather than beginning with (B) an uncertain option with an 89% chance of
utility maximization goals derived from the normative $1 million, a 10% chance of 5 million, and a 1%
approach, and then modifying them as needed in the chance of receiving nothing. The expected value
descriptive approach, the computational approach is calculation in Eq. (1) suggests that one should take
built directly from cognitive and emotional processing the second option, because EV(B) = $1.39 million >
assumptions. It attempts to formally define the $1 million = EV(A).
dynamic processes—whether neural, componential, The EV rule seems reasonable for gambles played
or holistic—that over time determine a final decision. repeatedly many times. But for gambles with high
This advanced review will provide an update stakes that are only played once, it is easy to see
on the status and contemporary research for each of that this objective may not be so appealing. Bernoulli3
these major research streams. It will provide sufficient observed that most people did not make choices in
background in the development of each approach line with the expected value rule when the values
but focus on how they deal with current issues and (x) were determined with large objective amounts
challenges in the realm of decision making. It will (e.g., $1 million). He proposed that people did not
focus on individual (rather than collective or group) view (monetary) outcomes objectively, but rather they
decision making. It will also focus on situations did so subjectively. That is, $1000 does not have
dealing mainly with risk in the form of known possible the same subjective value to both a miller and a
outcomes with well-specified probabilities. These millionaire—the former places more subjective worth
situations are distinct from situations of uncertainty on the same objective dollar amount. In fact, given
involving ambiguity in the probability distribution the hypothetical choice between A and B above, the
over outcomes, or situations of certainty where choice majority of experimental participants select A even
outcomes are clearly defined. Although these different though it has a lower expected value. Presumably,
domains share some similarities, they are treated this is due to the fact that the subjective experience of
distinctly in the extant literature; space constraints receiving $5 million instead of $1 million is not five
prevent detailed treatment of each. times as pleasurable as receiving $1 million instead
of $0. Rather, as wealth increases, the additional
value placed on subsequent increments decreases (an
NORMATIVE APPROACHES TO additional $1 million means more if you are broke
DETERMINE ‘OPTIMAL’ CHOICE than if you already have $4 million).

We make many different types of decisions everyday.

What should I do this weekend? Should I pay off
DIMINISHING MARGINAL UTILITY
my credit card or wait? Should I take the job offer
or not? Do I take this job or keep looking? Most The increments in subjective value corresponding
theories of decision making assume any of these to increases in objective value decrease as the
decisions can be abstracted and represented as the initial objective value increases; this is termed as

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

diminishing marginal utility. This explains why a raise evidence about human choice behavior challenged
of $10,000 per year would be quite meaningful to SEU on other grounds.
the average reader, but probably not to Bill Gates. Allais8 presented people with the choice between
The concept is similar to the Weber–Fechner Law in A and B introduced earlier, as well as a choice between
psychophysics, where changes in stimulus intensity two other options: (C) an 11% chance of receiving $1
have different psychological sensations depending on million, otherwise nothing and (D) a 10% chance of
the initial magnitude. Mathematically, this is typically receiving $5 million, otherwise nothing. Here, the
represented with a simple power utility function, options C and D are created simply by changing
U(x) = xα . This form also allows for describing an a ‘common consequence’ of an 89% chance at $1
individual’s risk attitudes with a single parameter: million in A and B, respectively, to an 89% chance at
when 0 < α < 1, the utility function is concave and $0. If one chooses A over B in the first choice, then
risk-averse behavior is predicted, whereas a convex this implies a utility function that predicts one should
function predicting risk-seeking behavior emerges if still take option C over D, because SEU(A) > SEU(B)
α > 1. implies SEU(C) > SEV(D). However, although most
Mathematically, this involves a function that people select A instead of B, they select D instead
transforms objective value into subjective utility, U(x). of C. This choice pattern is inconsistent with SEU,
A simple modification then suggests one should select regardless of the form of U(x). Specifically, it violates
an option with the highest expected utility EU(X): one of the axioms (independence) set forth by von
Neumann and Morgenstern5 and Savage.4


n This empirical inconsistency prompted re-
EU(X) = pi U(xi ) (2) searchers to explore further modifications to the SEU
i=1 framework. At this point, although the basic algorithm
was retained (maximization of a mathematical
Bernoulli’s concept is intuitively plausible and could expectation), theories began to depart substantially
explain actual choice behavior. Furthermore, it is easy from these previous ‘rational’ ideals in order to explain
to impute psychological meaning on the utility func- the decisions of we ‘irrational’ humans.
tion, such as risk attitudes (see section Diminishing Another important advance in utility theory was
Marginal Utility). However, this approach was crit- the extension of the theory to outcomes described by
icized by some theorists who adhered strongly to multiple conflicting attributes.9 For example, when
the normative approach, because there was no ratio- choosing a medical insurance plan, one needs to
nal foundation for why people should use the EU consider not only the cost of the plan but also
for choices only played once. In 1944, a seminal the breadth of the coverage, the quality of the
book by von Neumann and Morgenstern overcame care provided by the coverage, and other attributes
this limitation by providing an axiomatic foundation of the plan. Thus this decision involves evaluating
for expected utility theory. The original EU theory consequences with respect to several conflicting
was restricted to gambles with objectively known objectives. Should one spend more money to achieve
probabilities, but Savage4 is credited with further greater coverage or should one save money but take
extending the axiomatic foundation of von Neumann a risk with lower coverage? The most commonly
and Morgenstern5 beyond subjective utility to addi- used multiattribute utility model combines the values
tionally include subjective probability for uncertain of the conflicting attributes according to a weighted
events with no objectively known probabilities, a additive rule (much like the utility theory for gambles),
notion raised earlier by Ramsey6 and de Finetti,7 where the weights reflect the tradeoffs among the
as well as by von Neumann himself. Mathematically, attributes. The weighted additive rule is considered to
the subjective expected utility (SEU) of an option then be a compensatory rule which allows deficits on one
becomes attribute to be compensated by advantages on other
attributes.


n
SEU(X) = πi U(xi ) (3)
i=1
DESCRIPTIVE APPROACHES
Here, the events are assigned subjective probabilities,
TO EXPLAIN OBSERVED CHOICE
πi . Savage’s4 axiomatization is still considered as a Descriptive theories in decision making, as their name
rational theory, as the subjective probabilities were suggests, are more concerned with describing the
still constrained by the laws of probability. However, choices people actually make rather than providing
it did not take long before additional empirical a ‘rational’ basis for making choices, as EV, EU, and

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

SEU aimed to do. This shift is due, in large part, to the (a) UG(x)
fact that psychologists began to relax the idealistic
models heretofore introduced by mathematicians,
statisticians, and economists. The most popular
descriptive theory of choice is termed as prospect
theory, introduced by Kahneman and Tversky.10
S x
Prospect Theory
Prospect theory introduced four important aspects
from cognitive psychology to impart a more human-
centered view of decision making.10 First, it suggested
a predecisional ‘editing’ stage where the decision UL(x)
problem is prepared, such as by eliminating clearly
inferior choice options and simplifying and mentally
ordering outcomes. Second, it introduced the notion (b)
of reference dependence, where outcomes are not p(p)
evaluated absolutely but relative to some benchmark,
such as one’s current wealth or ‘status quo’.11
Third, it suggested that outcomes could be evaluated
differentially based on whether they were seen as gains
or losses relative to the status quo—that is, there were
separate utility functions for gains UG (x) and losses
UL (x). Fourth, specifically, it proposed the concept of
loss aversion, that the marginal utility of a constant
change is greater for losses (a $100 loss is more
aversive than a $100 gain is pleasant).
Formally, these assumptions can be incorpo-
rated into Eq. (3) with the appropriate specifica-
p
tion (Figure 1): UG (x) = f (x − S) for x − S > 0 and
UL (x) = −λf (S − x) for x − S < 0 where S is the sta- FIGURE 1 | Cumulative prospect theory’s value and weighting
tus quo and f is concave for gains, convex for losses, functions.
and steeper for losses (λ is a parameter to model
the degree of loss aversion). Kahneman and Tver-
sky also introduced the term decision weight for the Kahneman and Tversky10 introduced the probability
multiplier attached to each outcome. Although they weighting function, which could be interpreted in
assumed decision weights were based on the objective terms of concepts such as discriminability and
probabilities, π (p), they explicitly distinguished this attractiveness,13 or affective notions such as elation
notion from a purely probabilistic evaluation.12 They and disappointment.14 Computational models of
put forth a strictly convex form for π (p) that implied decision weighting describe how these weights may
overweighting of small probabilities and underweight- result from probability judgments based on memory
ing of large probabilities; a revised form suggests retrieval15 or as the result of differential attention and
concavity for small probabilities (Figure 1).25 . These ‘dwelling’ on specific outcomes or events.16
restrictions on Eq. (3), spurred by reflecting on human Craig Fox’s extension of support theory to
thinking rather than any rational calculus, produced decision making17–19 describes how individuals in
a theory that was much more accurate at describing circumstances of uncertainty might estimate proba-
actual choices—but only to a degree. bilities, which in turn can then be used to derive
decision weights. Support theory distinguishes among
different descriptions of events as the carriers of belief
SUBJECTIVE PROBABILITY (rather than the objective events themselves) and is
based on support for a focal or salient description
AND UNCERTAINTY
relative to other possible descriptions. This is an
Prospect theory introduced the important notion of important theory for extending decisions under risk
decision weights, but this in turn raises the question (known event probabilities) to situations of uncer-
of how these weights are psychologically determined. tainty (unknown event probabilities).

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

Rank-Dependent Theories weight was simply equal to the subjective probability,

Even prospect theory was unable to account for all w(p) = π (p).
the observed trends in human choice behavior.10,20,21
For example, it predicted that people would choose
Configural Weight Theories
options even if they were clearly dominated by other
Michael Birnbaum and his colleagues advocate a
options in the choice set (see Ref 22, for a review)
descriptive theory that is similar in some respects to
and was severely limited to choice options with only
RDU theories, but incorporates important differences
two outcomes. It also implied that the weight given
and makes competing predictions regarding some
to an outcome was independent of the value of the
important empirical phenomena.26,27 For example,
outcome. That is, the weight given to a probability
RDU theories predict that individuals should never
of 0.10 would be the same regardless whether this
choose an option that is stochastically dominated
probability corresponded to the worst or the best
by another. If we define XQ as the random value
outcome of a gamble, which is not true for human
produced by gamble Q and XR as the random value
choices.23
produced by gamble R, then option Q stochastically
In the late 1980s, several researchers indepen-
dominates option R if and only if Pr(XQ > x) ≥
dently arrived at an idea that served as the next
Pr(XR > x) for all x, and the inequality is strict for at
major modification to the expected utility frame-
least one x (i.e., the cumulative distribution function
work. In particular, it allowed the decision weight
of Q is always above that of R). In fact, for certain
of an outcome to be dependent on the rank order
gamble types, people seem to robustly select an option
of the outcome (e.g., whether it was the worst or that is stochastically dominated by another option in
the best), and hence these theories are called rank- the choice set.27
dependent utility (RDU) theories (see also rank-and- Birnbaum’s configural weight utility (CWU)
sign-dependent utility in Ref 24). This was achieved theories are able to explain these violations of
by changing the basis of the weighting function stochastic dominance, as well as the other behaviors
from the probability of winning x to the proba- covered previously. CWU theories retain the algebraic
bility of winning x or more (decumulative prob- representation and expectation maximization rule of
ability). Tversky and Kahneman25 extended their all previous utility theories. The key difference is in
original theory into cumulative prospect theory, the specification of the weighting function, which
perhaps the most popular RDU theory. Formally, is exemplified by the transfer of attention exchange
RDU theories propose the following general func- CWU model (TAX; see Ref 28). In the TAX model,
tion for the decision weight assigned to a positive similar to the RDU models, lower values of a gamble
outcome xj : ‘steal’ or ‘tax’ decision weight from higher values.
 n   n  But in addition, the TAX model assigns a separate


weight to each outcome listed in a gamble even
w(xi ) = π pj − π pj (4) if the same value is listed more than once. For
i i+1
example, the single weight assigned to $100 for
the gamble ‘0.5 chance to win $100 or else win
A notational change produces the following overall nothing’ is not the same as the total weight assigned
utility of an option X: to both of the $100 outcomes in the gamble ‘0.25
chance to win $100, 0.25 chance to win $100, and


n
0.5 chance to win nothing’. RDU theories assign
RDU(X) = w(xi )U(xi ) (5)
weight based on the cumulative probability of an
i=1
outcome, and so these two gambles would be treated
as the same. The violations of stochastic dominance
RDU theories, via Eqs (4) and (5), introduce
indicate that people do not treat these gambles as the
a subtle but important distinction between the
same.
subjective probability and the decision weight. The
subjective probability refers to the distortion of
decumulative probability, π (·), and is thus essentially
HISTORICAL PERSPECTIVES
a psychophysical measure. Decision weight, w(·), is
a further transformation that describes the relative
ON CONFIGURAL WEIGHTING
weight given to an outcome when integrating across The configural weight models formally applied to
other outcomes to determine the holistic value of decision-making behavior are found as early as Birn-
an option. In earlier SEU models, the decision baum and Sutton.29 However, it is notable that the

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

idea of configural weighting can be traced back even those found in RDU theories. However, Lopes allows
earlier to explanations of social judgment biases in for evaluation of outcomes in both a low-to-high,
Birnbaum and Stegner,30 , based on work by Birnbaum cumulative fashion and the decumulative, high-to-low
et al.31 Thus, configural weighting theories actually fashion posited by RDU theories. Her reasoning is that
predate both the decision-theoretic work on rank- individuals may exhibit security-minded behavior that
dependent weighting functions and even prospect focuses on the probability of obtaining an outcome
theory. with a value of x or less, and/or a potential-minded
analysis in line with RDU that focuses on the probabil-
ity of obtaining an outcome with a value of x or more.
Regret Theory Mathematically, Lopes allows for a parameter that
Prospect theory, RDU theories, and CWU theories moderates the degree to which the security-minded
strove to incorporate human tendencies into the eval- versus potential-minded analyses contribute to the
uation of outcomes and their associated probabilities, assessment of outcome utility.
or weights. Other theories sought to redefine the basic Second, SP/A theory includes the notion of an
currency of a choice option, such as by introduc- aspiration level or goal achievement component. That
ing utility derived not just from the actual outcome is, in addition to the value assigned to options based on
values, but also by comparisons to outcomes of fore- the assessment of their outcome value (as described
gone options.32,33 Loomes and Sugden33 introduced in the preceding paragraph), options are evaluated
their regret theory in response to prospect theory and favorably if they allow a decision maker to achieve
showed how it could explain the same empirical results some preset goal. Mathematically, this aspiration
put forth by Kahneman and Tversky10 as evidence for criterion evaluation for an option is based on the
the latter. Essentially, regret theory assumes that util- probability that the option provides an outcome at
ity U(x) is composed of two distinct components, or above the aspiration level. If one has a goal of
an evaluation of the outcome that is obtained and a winning $80 in the coin flip choice from the previous
difference between that outcome and those forgone. section, then A has a 50% chance of meeting this
For example, assume one is choosing between aspiration level (corresponding to the ‘heads’ value
two gambles, A and B, determined by a coin flip. of $100 > $80) and B has no chance of meeting the
If one chooses A, then $100 is won if the coin aspiration level (neither outcome is >$80).
lands on heads, and nothing is won if it lands on SP/A theory assumes that a decision maker inte-
tails; B offers $70 for heads and $30 for tails. In grates the two components into a holistic utility value
evaluating option A, regret theory proposes that the for each option and again maximizes the expec-
utility assigned to the outcome ‘heads’ will be a (linear) tation of this utility. Each single component may
combination of the utility of $100 and the additional produce competing predictions that produce internal
utility or ‘rejoice’ associated with the fact that, had conflict. For example, with an aspiration level of $80,
B been chosen, then only $70 would have been won. A is preferred using the goal criterion; however, a
Conversely, evaluation of the outcome ‘heads’ for security-minded decision maker who focuses on the
option B involves the utility of $70 as well as the low outcomes may prefer option B on this criterion
disutility or ‘regret’ associated with the fact that one (due to its advantageous low outcome, relative to A).
could have obtained $30 more had A been chosen. The Mathematically, model parameters can specify the rel-
basic psychological mechanisms involved in regret ative degree to which each component contributes to
theories are similar to those studied extensively in choice behavior.
other psychological domains, such as work in social
psychology on counterfactual thinking.34 Mellers
et al.35 extended these ideas and developed a more COMPUTATIONAL APPROACHES TO
detailed model of the emotional basis for these regrets. MODEL LATENT CHOICE PROCESSES
Descriptive theories of choice embellish the basic
Security-Potential/Aspiration Theory framework of maximizing an expectation with obser-
Lola Lopes introduced additional psychological con- vations from human psychology. Prospect theory,
structs such as hope, fear, and goal achievement to RDU theories, and CWU theories make claims about
develop a descriptive theory of decision making called the specific nature of utility and probability assess-
security-potential/aspiration (SP/A) theory.36,37 This ment that depart from rational norms and laws of
theory assumes that a decision maker simultaneously probability. Other theories include additional consid-
considers two distinct criteria in making decisions. erations beyond an expected utility assessment, such as
First, one considers a utility component similar to the potential satisfaction or disappointment resulting

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

from comparing outcomes to those forfeit (regret the- choice set and not considered any further. This
ory), or the desire for a choice option to fulfill some sequential aspect selection and elimination process
goal (SP/A theory). In contrast to all of these descrip- continues until only a single choice option ‘survives’.
tive approaches which focus on choice as determined Tversky40 illustrated the ability of this model to
by the maximization of some utility function, compu- account for violations of rational choice axioms
tational approaches focus on the underlying cognitive, (e.g., violations of independence from irrelevant
motivational, and emotional processes from which alternatives; see Ref 41 for discussion). Although
choices dynamically emerge. In this section, we will Tversky40 also showed how EBA could be represented
review several popular research streams that adhere as a (random) utility model, it has a decidedly
to this philosophy.38 different flavor through its presentation in terms of
simple rules and is not ‘rational’ in the sense that
it is noncompensatory. An option can be eliminated
PROCESSING ASSUMPTIONS AND from consideration simply on the basis of a single
MODEL REPRESENTATION attribute even though it may be holistically the
‘best’ because of its many advantages on other
Computational models do not begin with the algebraic
attributes.
utility maximization assumptions of the normative
and (most) descriptive approaches. However, it
Thorngate’s Heuristics for Gamble Forms
could be that choices in line with the normative
Thorngate42 proposed 10 distinct decision heuristics,
goals of utility maximization evolve from the
based largely on the work of Coombs et al.43 that were
underlying processes. If so, it could be that utility
formulated for application to choices among gambles
maximization is indeed representative of human
like those presented earlier. For example, his minimax
choice behavior, even if the algebraic representation
heuristic selects the alternative with the highest min-
is merely paramorphic—thus, computational and
imum outcome value, or max[x1 ], and the maximax
descriptive approaches are not mutually exclusive.
heuristic chooses according to max[xn ]. His least-
likely heuristic chooses the alternative with the lowest
Heuristic and Rule-Based Approaches probability of its worst outcome, min[p1 ], whereas the
Perhaps the most intuitive computational approaches most likely heuristic chooses according to max[p1 ].
specify simple procedures for making choices, often Other suggested heuristics include elimination heuris-
called heuristics. Heuristics are typically expressed as tics like EBA, and an equal-weighting heuristic that
verbalizable rules or flowcharts for applying discrete selects based on the highest average outcome value
steps to make a decision (see Ref 39, for a review and (ignoring probabilities; see also Ref 44). Thorngate’s
organizing framework). analyses showed that these simple heuristics often
selected options that were normatively optimal (in
Elimination by Aspects terms of expected utility) or very close to it.
One of the earliest popular heuristics was Tversky’s40
elimination by aspects (EBA) model. Consider for The Adaptive Decision Maker Hypothesis
example the problem of buying a new digital Payne et al.45,46 proposed that decision strategies,
camera. This is a multiattribute decision involving the including utility maximization algorithms as well as
consideration of attributes such as price, resolution simple heuristics, could be formalized in terms of
of the camera, size of the camera, etc. Rather than what they called elementary information processing
maximizing a weighted average of attribute values, (EIP) units, such as ‘retrieve’, ‘add’, ‘multiply’,
as suggested by multiattribute utility theories, the and ‘compare’. Implementing maximax among two
EBA model proposes that individuals sequentially alternatives A and B would involve four EIPs:
consider different aspects, such as whether the retrieve a1 , retrieve b1 , compare a1 , b1 , choose
price is within budget, whether the resolution is max [a1 , b1 ]. This specification welcomes precise
satisfactory, and whether size is sufficiently small for implementation in computer simulations and allows
a new digital camera. The probability of considering for the derivation of measures such as decision time
an aspect is proportional to its importance, so and information acquisition. It is worthy to note that,
that if price is the most important attribute to a although Payne et al.45,46 did not introduce novel
consumer, it is most likely to be considered first.a heuristics per se, their method of formalizing and
When considering an aspect, any choice option that studying heuristics has been enormously influential
does not meet a minimum criteria (e.g., a price on subsequent computational modeling. Furthermore,
over one’s budget of $300) is eliminated from the they introduced an adaptive view of strategy selection,

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

based on an efficient frontier involving a tradeoff makes specific quantitative predictions about infor-
between desired levels of effort (in terms of EIPs) mation acquisition and response times, in addition to
and accuracy (relative to a normative algorithm). This choices.
represents an important advance in understanding First, DFT assumes that attention shifts to dif-
which heuristic from among many may be applied in ferent dimensions of the choice task over time. For
any given situation. gambles, these shifts occur (independently) across
the outcomes of each option, with the probability of
Gigerenzer’s Adaptive Toolbox attending to an outcome proportional to its objective
Gigerenzer and Todd47 and the ABC Research probability (see Ref 16, for details). For preferen-
Group also advance the notion of a collection of tial choice, these transitions are typically assumed
decision heuristics. Many of their heuristics are to be defined across attributes, with the simplifying
very similar to the earlier mentioned heuristics in assumption that attention to a specific attribute (e.g.,
terms of the process description. For example, the the price of all camera models) at any moment is pro-
priority heuristic48 involves sequential application of portional to the attribute’s importance51 (see Ref 55,
thresholded versions of the maximin, most-likely, for alternative assumptions).
and maximax heuristics. First, choose the option Second, the current focus of attention generates
that maximizes the minimum possible outcome; but a relative evaluation for each choice option. When
only if the minimum outcome value is sufficiently price is under consideration, then those options with
larger than the other options’ minimum outcomes. the highest prices will receive low evaluations. Specif-
Otherwise, consider the probability p1 of each option’s ically, an option’s evaluation is based on the affective
lowest outcome, and so on. Although the actual reaction to the option’s value on the focal attribute,
heuristics are very similar to those already mentioned, relative to the average reaction of all the other com-
this research stream is notable for three additional peting options’ values.
characteristics. First, it is applied to prediction, Third, these momentary evaluations are accu-
inference, and categorization, as well as decision mulated over time to describe the current preference
making. Second, like Payne et al.,46 it stresses the for each option at each point during deliberation
adaptive nature of the development and application (Figure 2). To the extent that attention focuses on fea-
of simple heuristics (hence the term ‘adaptive toolbox’) tures that are favorable for a particular option, that
in terms of ecological fit between the heuristics and the option will have a greater value of preference over
environment. Third, they decompose the majority of time. This accumulation process can be subject to spe-
their heuristics into three distinct components: a rule cific effects such as gradual decay to produce recency
for guiding information search, a rule for determining effects, as well as inhibitory influences from compet-
when to stop search, and a decision rule applied ing options (i.e., as one option becomes preferred, it
to the information collected. This strikes a balance inhibits or reduces the preference for other similar
between the low-level EIP-based description and the options). An option is chosen, ending deliberation,
presentation of holistic rules. when it reaches a preset threshold level of preference
used as a criterion for being ‘good enough’ to merit
selection.
Decision Field Theory DFT has been successful in accounting for var-
The most influential type of decision model in cog- ious puzzling phenomena in pairwise choice between
nitive science is the sequential sampling/accumulation gambles under risk and uncertainty,50 as well as robust
model. This type of model has been applied to neu- paradoxes arising in multialternative and multiat-
roscience, sensation, perception, memory, and catego- tribute choice problems51 and pricing.56 It provides
rization domains.49 The first application of sequential a measure of preference strength (rather than just
sampling models to decision making under risk and direction) and has recently been extended to predict
uncertainty was decision field theory (DFT; Refs decision confidence as well.57 It also uniquely accounts
50–53, for reviews; see also Ref 54, for a neural for effects of decision time such as speed–accuracy
network representation of DFT). Most broadly, DFT tradeoffs50 and changes in preference under time
is a mathematical model based on cognitive princi- pressure.58 DFT has also been extended to model rule
ples of selective attention and relative evaluation, that learning and rule-based decision making, including
models deliberation as a dynamic system accumulat- strategy switching.59 It has been successfully applied
ing evidence in favor of each choice option. The first to engineering problems such as human-in-the-loop
option to reach a criterion level of evidence is selected. control systems60 and agent-based models of emer-
In contrast to descriptive utility theories, DFT thus gency evacuation decisions.61

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

P(t)

θ2
Options O1 O2 O3

θ1
A

Attributes A1 A2 A3

B
Attention X

FIGURE 3 | Generic neural network representation of a decision

t0 t1 t2 t3 t problem. Each of three choice options is described by three attributes.
An attention node determines which attribute(s) is/are processed at
FIGURE 2 | Decision field theory (DFT) representation of preference each time step and thus embodies decision weight. Links between
accumulation for two options. Preference P(t) accumulates for each attributes and options represent the value of each option on the
option, shown as separate trajectories, over time t. At time t1 , option B corresponding attribute. All solid connections are assumed to be
is preferred with a higher value of P(t) ; at time t2 , preference is equal bidirectional for parallel constraint satisfaction (PCS) models, and
between the two options, after which option A is consistently preferred. feedforward for leaky competing accumulator (LCA) and decision field
Choice is determined when an option’s trajectory reaches the threshold theory (DFT). Inhibitory bidirectional connections among options
level of preference, P(t) = θ. A decision maker with a threshold of θ 2 (dashed lines) model competition among options in LCA, PCS, and DFT
would thus select option A at time t3 ; a more impulsive individual models. LCA and DFT assume an additional layer between options and
modeled by θ 1 would select option B at time t1 . attributes to compute differences.

Connectionist Approaches Most recently, Andreas Glöckner et al.68,69 have

Several contemporary computational models of successfully applied one of these parallel constraint
decision making besides DFT have been cast in neu- satisfaction (PCS) models to various decision-making
ral network architectures. While these models may tasks. PCS models involve a search for coherence or
be less transparent and in some sense more com- consistency among a set of choice options, such as in
plex, they are popular in many cognitive domains and trying to resolve conflicting preferences across options
have the advantage of neurally plausible mechanisms (one option may have a lower price, but another has
(Figure 3). better resolution, and so on). Upon presentation of a
choice problem, PCS mechanisms are activated to find
Leaky Competing Accumulator Model the best interpretation of the problem in a perception-
Usher and McClelland 62 proposed a connectionist like process. Rather than explicitly describing any
model that employs a recursive network to describe necessarily conscious strategy or heuristic that is
how preference builds for various choice options over applied to a choice problem, these models rely on
time. This model is very closely related to DFT a more holistic or Gestalt conceptualization where a
(see Ref 63, for a comparison of the two). It also preferred option ‘emerges’ instantly or over the course
involves sequential comparison of attributes where of deliberation.
the options ‘compete’ for preference based on their Choice options and outcome values are rep-
relative excellence, and these momentary comparisons resented as network nodes, and links represent the
are ‘accumulated’ over time into a holistic preference possession of certain aspects as well as their decision
value for each option, with some degree of decay (or weight (based on link strength; see (Figure 3). The
‘leaking’). In contrast to DFT, it includes the notion of basic intuition is that there are some sets of node acti-
loss aversion from prospect theory as a fundamental vations (option preference strengths) that will produce
assumption. a stable network (consistent representation), based on
the constraints in the form of attribute values and
Coherence-Seeking Network Models weights. Given the decision problem representation,
Thagard and Milgram64 applied Thagard’s network initial advantages of one option are automatically
model called ECHO to decision making, and other highlighted by increasing the activation of supporting
researchers have since extended this formulation.65–67 and decreasing the activation of contrary information.

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

An activation updating algorithm continues to adjust and predict their choices.24 This allows for a strict and
the node activations until a stable state is achieved, at concise way of expressing a decision policy, and lends
which point the node (choice option) with the highest itself to easy derivation of closed-form predictions.
activation is predicted to be chosen. Importantly, these The problem arises when individuals or people in gen-
models involve bidirectional links between attributes eral fail to adhere to these principles. Computational
and options and are thus able to explain ‘restruc- approaches benefit from increased attention to men-
turing’ of the choice problem,70,71 such as changes tal and emotional processes and thus psychological
in attribute importance or decision weight across the plausibility. They can also account for many of the
choice task (i.e., coherence shifts) that are not possible violations of these principles and often of collections
in static, descriptive approaches. of violations. However, these models are less transpar-
ent and thus often more difficult to treat analytically,
often requiring simulation or direct application to a
Memory-Based Approaches specific context to derive predictions. Finally, both
Several researchers have acknowledged the crucial within and across classes of models, it is important
role that memory plays in decision making.72,73 to understand the tradeoff between flexibility and
In fact, simple recognition memory can be used robustness of models. For example, does the increased
to make inferential decisions when the likelihood predictive power of prospect theory over earlier SEU
of recognition based on salience is correlated with theories justify the ‘cost’ of increasing the number of
the decision criterion.74 Dougherty et al.15 use a free parameters? Furthermore, is the increase in fit to
memory-based judgment model to account for several the data theoretically meaningful above and beyond
robust phenomena in judgment and estimation tasks, that afforded by this increased flexibility (see Refs
such as base-rate neglect, hindsight bias, and the 78–80, for excellent discussions of these issues)?
conjunction fallacy.75 Elke Weber, Eric Johnson, and This review has provided a comprehensive but
colleagues also propose that memory processes can concise account of the development of theories in
be used to model decision tasks.76,77 This approach, decision making under risk and uncertainty. Although
most recently dubbed ‘Query theory’, assumes that
decision research has come a long way, there are still
preferences that drive choice and other decisions are
many open questions that are not fully addressed,
based on a collection of serially posed queries to
even by the more sophisticated theories covered
memory concerning relevant characteristics of the
in this review (see Ref 81). How independent are
task. For example, if deciding whether to buy a
evaluations of attributes and/or alternatives? Is the
certain digital camera, an individual might attempt
evaluation of probability (or weight) separable from
to recall experiences with similar models or generate
the evaluation of outcome value? Furthermore, this
the pros and cons of buying the camera. Query
review is not exhaustive of the theories and approaches
theory is able to explain some empirical trends
in decision research. For example, an entire class of
in human decision behavior by embellishing this
random utility theories82 is beyond the scope of this
simple notion with what is known about human
review, as are some of the more recent computational
memory, such as serial position effects, priming, and
models. A ‘dual systems’ approach recognizing the
interference. Although the theory’s assumptions have
role of automatic or intuitive processes, in addition
been empirically supported, at this point it has not
to more directed and deliberative processes, is also
been formally introduced as a mathematical model
becoming quite popular83–85 (see Ref 86 and the
or at a specific algorithmic level, as the preceding
related commentaries for various perspectives; and
computational models have.
Ref 87, for the historical precedent in cognitive
science). Finally, the field has relatively recently
focused a great deal on understanding the role of
SUMMARY AND FUTURE DIRECTIONS affect or emotion in decision making35,88–90 (see Ref
We do not propose that any of the approaches 91, for an earlier treatment).
described above is privileged in any objective sense. An important recent development in decision
Rather, each approach may be seen as possessing research is the advent of neuroscientific methods
inherent strengths and weaknesses, or as differen- to better understand decision making under risk
tially applicable across domains or academic pursuits. and uncertainty (see Ref 92, for a concise review
The class of utility models born from the normative and organizing framework; for more extensive
approach has the appeal of an axiomatic foundation, summative treatment, see Refs 93,94). In fact, this has
meaning that adherence to specific principles ensures spawned an entire ‘subfield’ called neuroeconomics
that a utility representation can be created to describe or decision neuroscience that attempts to verify

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

the underlying neural substrates associated with the surpasses a threshold.94,101,102 Thus, in contrast to
various components of decision theories and their pur- the normative and descriptive utility maximization
ported mechanisms.95 Work in this vein has indeed theories, there is considerable neuroscientific evidence
found evidence for brain regions responsible for rep- for neuronal populations that may be responsible for
resenting the components of utility theories such as the computational process that produces observable
probability and reward value96 (see Ref 97, for a dis- decision behavior.
cussion), as well as evidence for distinctions between In closing, we would like to convey the excite-
gains and losses consistent with prospect theory.98 ment and opportunity that face the field of decision
Although this work supports the necessary condition making. Current advances are beginning to produce
of an adequate neural representation underlying utility fruitful practical results. For example, prospect theory
theories, it is not sufficient evidence for the maximiza-
has impacted economic theory, and computational
tion goal process. That is, there is evidence for the
models (heuristic rule-based models and dynamic
ingredients of utility theories, but not necessarily for
accumulation models) are being incorporated into
the mechanism that uses this information to produce
engineering and agent-based models of mixed human
choice (action selection).
There is now substantial neurophysiological evi- and machine systems. The number and nature of
dence supporting the mechanisms hypothesized by tools at our disposal continue to grow (including
computational accumulation models such as DFT experimental techniques for process tracing; see Refs
and leaky competing accumulator (LCA) (see Refs 45,46,103), as does the number and nature of fields
99,100, for reviews). Specifically, recent research indi- involved in studying decisions. As they do, we hope to
cates that neuronal activation accumulates over time better understand how people make decisions, predict
during decisions under risk and uncertainty, and an what decisions might be made in given situations, and
action is performed when the accumulated evidence reflect and improve upon those already made.

NOTES
a Note
that this model is closely related to the earlier lexicographic models of Coombs104 and Fishburn.105 These
models, however, specified a deterministic order for attribute selection.

REFERENCES
1. Goldstein WM, Weber EU. Content and discontent: 7. de Finetti B. Foresight: its logical laws, its subjective
Indications and implications of domain specificity in sources. In: Kyburg H, Smokler H, eds. Studies in Sub-
preferential decision making. In: Busemeyer JR, Hastie jective Probability: New York: John Wiley & Sons;
R, Medin DL, eds. Decision Making from a Cognitive 1937/1964.
Perspective. The Psychology of Learning and Moti-
vation, vol. 32. New York: Academic Press; 1995, 8. Allais M. Le comportement de l’homme rationnel
83–136. devant le risque: critique des postulats et axiomes de
l’école AmØricaine [The behavior of the rational man
2. Zsambok CE, Klein GA, eds. Naturalistic Decision facing risk: Critique of the postulates and axioms of the
Making. Mahwah, NJ: Lawrence Erlbaum Associates; American school]. Econometrica 1953, 21:503–546.
1997.
3. Bernoulli D. Specimen theorie novae de mensura sortis 9. Keeney RL, Raiffa H. Decisions with multiple objec-
[Exposition of a new theory of the measurement of tives: Preferences and value tradeoffs. New York: John
risk]. Econometrica 1738/1954, 22:23–36. Wiley & Sons; 1976.
4. Savage LJ. The Foundations of Statistics. New York: 10. Kahneman D, Tversky A. Prospect theory: an anal-
John Wiley & Sons; 1954. ysis of decision under risk. Econometrica 1979,
5. von Neumann J, Morgenstern O. Theory of Games and 47:263–292.
Economic Behavior. Princeton, NJ: Princeton Univer-
sity Press; 1944. 11. Markowitz HM. Portfolio Selection. New York: John
Wiley & Sons; 1959.
6. Ramsey, FP. Truth and probability. The Foundations
of Mathematics and other Logical Essays: Routledge & 12. Edwards W. Subjective probabilities inferred from deci-
Kegan Paul; 1931. sions. Psychol Rev 1962, 69:109–135.

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

13. Gonzalez R, Wu G. On the shape of the probability 33. Loomes G, Sugden R. Regret theory: an alternative the-
weighting function. Cognit Psychol 1999, 38:129–166. ory of rational choice under uncertainty. Econ J Nepal
14. Brandstätter E, Kühberger A, Schneider F. A cognitive- 1982, 92:805–824.
emotional account of the shape of the probability 34. Epstude K, Roese NJ. The functional theory of
weighting function. J Behav Decision Making 2002, counterfactual thinking. Pers Soc Psychol Rev 2008,
15:79–100. 12:168–192.
15. Dougherty MRP, Gettys CF, Ogden EE. MINERVA- 35. Mellers BA, Schwartz A, Ho K, Ritov I. Decision affect
DM: a memory process model for judgments of likeli- theory: emotional reactions to the outcomes of risky
hood. Psychol Rev 1999, 106:108–209. options. Psychol Sci 1997, 8:423–429.
16. Johnson JG, Busemeyer JR. Process weighting model. 36. Lopes LL. Between hope and fear: the psychology of
Unpublished manuscript [under review], 2009. risk. Adv Exp Soc Psychol 1987, 20:255–295.
17. Fox CR, Tversky A. A belief-based account of decision 37. Lopes LL. Algebra and process in the modeling of risky
under uncertainty. Manage Sci 1998, 44:879–895. choice. In: Busemeyer J, Hastie R, Medin DL, eds. Deci-
sion Making from a Cognitive Perspective. San Diego,
18. Fox CR. Strength of evidence, judged probability,
CA: Academic Press; 1995, 177–220.
and choice under uncertainty. Cognit Psychol 1999,
38:167–189. 38. Busemeyer JR, Johnson JG. Computational models of
decision making. In: Kohler DJ, Harvey N, eds. Black-
19. Tversky A, Fox CR. Weighing risk and uncertainty.
well Handbook of Judgment and Decision Making.
Psychol Rev 1995, 102:269–283.
Oxford: Blackwell Science; 2004, 133–154.
20. Quiggin J. A theory of anticipated utility. J Econ Behav
39. Shah AK, Oppenheimer DM. Heuristics made easy:
Org 1982, 3:323–343.
an effort-reduction framework. Psychol Bull 2008,
21. Yaari ME. The dual theory of choice under risk. Econo- 134:207–222.
metrica 1987, 55:95–115.
40. Tversky A. Elimination by aspects: a theory of choice.
22. Starmer C. Developments in non-expected utility the- Psychol Rev 1972, 79:281–299.
ory: the hunt for a descriptive theory of choice under
41. Rieskamp J, Busemeyer JR, Mellers BA. Extending the
risk. J Econ Lit 2000, 38:332–382.
bounds of rationality: evidence and theories of prefer-
23. Birnbaum MH, McIntosh WR. Violations of branch ential choice. J Econ Lit 2006, 44:631–661.
independence in choices between gambles. Organ
42. Thorngate W. Efficient decision heuristics. Behav Sci
Behav Hum Dec 1996, 67:91–110.
1980, 25:219–225.
24. Luce, RD. Utility of Gains and Losses: Mahwah, NJ:
43. Coombs C, Dawes R, Tversky A. Mathematical Psy-
Lawrence Erlbaum Associates; 2000.
chology. Englewood Cliffs, NJ: Prentice Hall; 1970.
25. Tversky A, Kahneman D. Advances in prospect the-
44. Dawes RH, Corrigan B. Linear models in decision
ory: cumulative representations of uncertainty. J Risk
making. Psychol Bull 1974, 81:95–106.
Uncertain 1992, 5:297–323.
45. Payne JW, Bettman JR, Johnson EJ. Behavioral decision
26. Birnbaum MH. New paradoxes of risky decision mak-
research: a constructive processing perspective. Annu
ing. Psychol Rev 2008, 115:463–501.
Rev Psychol 1992, 43:87–131.
27. Birnbaum MH, Navarette JB. Testing descriptive utility
46. Payne JW, Bettman JR, Johnson EJ. The Adaptive Deci-
theories: violations of stochastic dominance and cumu-
sion Maker. New York: Cambridge University Press;
lative independence. J Risk Uncertain 1998, 17:49–78. 1993.
28. Birnbaum MH, Chavez A. Tests of theories of decision 47. Gigerenzer G, Todd PM, the ABC Research Group.
making: violations of branch independence and distri- Simple Heuristics that make us Smart. New York:
bution independence. Organ Behav Hum Decis Process Oxford University Press; 1999.
1997, 71:161–194.
48. Brandstätter E, Gigerenzer G, Hertwig R. Risky
29. Birnbaum MH, Sutton SE. Scale convergence and utility choice with heuristics: Reply to Birnbaum (2008),
measurement. Organ Behav Hum Decis Process 1992, Johnson, Schulte-Mecklenbeck, and Willemsen (2008),
52:183–215. and Rieger and Wang (2008). Psychol Rev 2008,
30. Birnbaum MH, Stegner SE. Source credibility in social 115:281–290.
judgment: bias, expertise, and the judge’s point of view. 49. Ratcliff R, Smith PL. A comparison of sequential sam-
J Pers Soc Psychol 1979, 37:48–74. pling models for two-choice reaction time. Psychol Rev
31. Birnbaum MH, Parducci A, Gifford RK. Contextual 2004, 111:333–367.
effects in information integration. J Exp Psychol 1971, 50. Busemeyer JR, Townsend JT. Decision field theory:
88:158–170. a dynamic-cognitive approach to decision making
32. Bell DE. Regret in decision making under uncertainty. in an uncertain environment. Psychol Rev 1993,
Oper Res 1982, 30:961–981. 100:432–459.

 2010 Jo h n Wiley & So n s, L td.

 WIREs Cognitive Science Decision making

51. Roe RM, Busemeyer JR, Townsend JT. Multi- Hillsdale, NJ: Lawrence Erlbaum Associates; 2002,
alternative decision field theory: a dynamic connec- 393–398.
tionist model of decision-making. Psychol Rev 2001, 66. Holyoak KJ, Simon D. Bidirectional reasoning in deci-
108:370–392. sion making by constraint satisfaction. J Exp Psychol
52. Busemeyer JR, Diederich A. Survey of decision field Gen 1999, 128:3–31.
theory. Math Soc Sci 2002, 43:345–370. 67. Simon D, Krawczyk DC, Holyoak KJ. Construction
53. Busemeyer JR, Johnson JG. Micro-process models of of preferences by constraint satisfaction. Psychol Sci
decision making. In: Sun R, ed. Cambridge Hand- 2004, 15:331–336.
book of Computational Psychology. Cambridge, MA:
68. Glöckner A, Betsch T. Modeling option and strat-
Cambridge University Press; 2008, 302–321.
egy choices with connectionist networks: towards an
54. Busemeye JR, Jessup RK, Johnson JG. Townsend integrative model of automatic and deliberate decision
JTBuilding bridges between neural models and complex making. Judgm Decis Mak 2008, 3:215–228.
human decision making behavior. Neural Networks
69. Glöckner A, Betsch T, Schindler N. Coherence shifts in
2006, 19:1047–1058.
probabilistic inference tasks. J Behav Decision Making
55. Diederich A. Dynamic stochastic models for decision (in press).
making under time constraints. J Math Psychol 1997,
70. Montgomery H. From cognition to action: the search
41:260–274.
for dominance in decision making. In: Montgomery H,
56. Johnson JG, Busemeyer JR. A dynamic, stochastic, com- Svenson O, eds. Process and Structure in Human Deci-
putational model of preference reversal phenomena. sion Making. New York: John Wiley & Sons; 1989,
Psychol Rev 2005, 112:841–861. 23–49.
57. Pleskac TJ, Busemeyer JR. Two-Stage Dynamic Signal 71. Svenson O. Differentiation and consolidation theory of
Detection: A theory of confidence, choice, and response human decision making: a frame of reference for the
time. Psychological Review, 2009 In press. study of pre- and post-decision processes. Acta Psychol
58. Diederich A. MDFT account of decision making under (Amst) 1992, 80:143–168.
time pressure. Psychon Bull Rev 2003, 10:157–166. 72. Hastie R, Park B. The relationship between memory and
59. Johnson JG, Busemeyer JR. Rule-based Decision Field judgment depends on whether the judgment is memory-
Theory: a dynamic computational model of transi- based or on-line. Psychol Rev 1986, 93:258–268.
tions among decision-making strategies. In: Betsch T, 73. Weber EU, Goldstein WM, Barlas S. And let us not
Haberstroh S, eds. The Routines of Decision Making. forget memory: the role of memory processes and
Mahwah, NJ: Lawrence Erlbaum Associates; 2005, techniques in judgment and choice. In: Busemeyer JR,
3–20. Hastie R, Medin DL, eds. Decision Making from the
60. Gao G, Lee JD. Extending the decision field theory to Perspective of Cognitive Psychology. Psychology of
model operator’s reliance on automation in supervisory Learning and Motivation Series. New York: Academic
control systems. IEEE Trans Syst Man Cybern 2006, Press; 1995.
36:943–959.
74. Goldstein DG, Gigerenzer G. Models of ecological
61. Lee S, Son Y, Jin J. Integrated human decision making rationality: the recognition heuristic. Psychol Rev 2002,
and planning model for evacuation scenarios under BDI 109:75–90.
framework. ACM Trans Model Comput Simulation
75. Tversky A, Kahneman D. Extension versus intuititve
(in press).
reasoning: the conjunction fallacy in probability judg-
62. Usher M, McClelland JL. Loss aversion and inhibition ment. Psychol Rev 1983, 90:293–315.
in dynamic models of multi-alternative choice. Psychol
76. Weber EU, Johnson EJ. Constructing preferences from
Rev 2004, 111:757–769.
memories. In: Lichtenstein S, Slovic P, eds. The Con-
63. Busemeyer JR, Townsend JT, Diederich A, Barkan R. struction of New York, NY: Cambridge University
Contrast effects or loss aversion? Comment on M. Press. 2006, 397–410.
Usher, J. L. McClelland’s. (2004). ‘Loss aversion and
77. Johnson EJ, Haubl G, Keinan A. Aspects of endow-
inhibition in dynamical models of multi-alternative
ment: a query theory of value construction. J Exp
choice.’ Psychol Rev 2005, 112:253–255.
Psychol Learn Mem Cogn 2007, 33:461–474.
64. Thagard P, Millgram E. Inference to the best plan:
a coherence theory of decision. In: Ram A, Leake 78. Myung IJ. The importance of complexity in model
DB, eds. Goal-driven Learning. Cambridge, MA: MIT selection. J Math Psychol 2000, 44:190–204.
Press; 1995, 439–454. 79. Pitt MA, Myung IJ. When a good fit can be bad. Trends
65. Guo FY, Holyoak KJ. Understanding similarity in Cogn Sci 2002, 6:421–425.
choice behavior: A connectionist model. In: Gray W, 80. Roberts S, Pashler H. How persuasive is a good fit?
Schunn C, eds. Proceedings of the Twenty-Fourth A comment on theory testing. Psychol Rev 2000,
Annual Conference of the Cognitive Science Society. 107:358–367.

 2010 Jo h n Wiley & So n s, L td.

 Advanced Review wires.wiley.com/cogsci

81. Hastie R. Problems for judgment and decision making. 93. Camerer C, Lowenstein G, Prelec D. Neuroeconomics:
Annu Rev Psychol 2001, 52:653–683. How neuroscience can inform economics. J Econ Lit
82. McFadden D. Econometric Models of Probabilistic 2005, 43:9–64.
Choice. In: Manski C and McFadden D, eds. Structural 94. Gold JI, Shadlen MN. Representation of a perceptual
Analysis of Discrete Data with Econometric Applica- decision in developing oculomotor commands. Nature
tions. Cambridge: MIT Press; 1981. 2000, 404:390–394.
83. Sloman SA. The empirical case for two systems of 95. Sanfey AG, Loewenstein G, McClure SM, Cohen
reasoning. Psychol Bull 1996, 119:3–22. JD. Neuroeconomics: cross-currents in research on
84. Sloman SA. Two systems of reasoning. In: Gilovich T, decision-making. Trends Cogn Sci 2006, 10:108–116.
Griffin D, Kahneman D, eds. Heuristics and Biases: The
96. Platt ML, Glimcher PW. Neural correlates of deci-
Psychology of Intuitive Judgment. Cambridge: Cam-
sion variables in parietal cortex. Nature 1999,
bridge University Press; 2002.
400:233–238.
85. Kahneman D, Frederick S. Representativeness revis-
97. Sanfey AG. Neural computations of decision utility.
ited: Attribute substitution in intuitive judgment. In:
Trends Cogn Sci 2004, 8:519–521.
Gilovich T, Griffin D, Kahneman D, eds. Heuristics
and Biases: The Psychology of Intuitive Judgment. New 98. Tom S, Fox CR, Trepel C, Poldrack RA. The neural
York: Cambridge University Press; 2002, 49–81. basis of loss aversion in decision making under risk.
86. Stanovich KE, West RF. Individual differences in rea- Science 2007, 315:515–518.
soning: implications for the rationality debate? Behav 99. Schall JD. Neural basis of deciding, choosing and act-
Brain Sci 2000, 23:645–665. ing. Nat Rev Neurosci 2001, 2:33–42.
87. Shiffrin RM, Schneider W. Controlled and automatic 100. Smith PL, Ratcliff R. Psychology and neurobiology of
human information processing: II. Perceptual learning, simple decisions. Trends Neurosci 2004, 27:161–168.
automatic attending, and a general theory. Psychol Rev
1977, 84:127–190. 101. Ratcliff R, Cherian A, Segraves M. A comparison of
macaque behavior and superior colliculus neuronal
88. Bechara A, Damasio H, Damasio AR. Emotion, deci- activity to predictions from models of simple two-
sion making, and the orbitofrontal cortex. Cereb Cor- choice decisions. J Neurophysiol 2003, 90:1392–1407.
tex 2000, 10:295–307.
102. Schall JD. Neural correlates of decision processes: eural
89. Damasio AR. Descartes’ Error: Emotion, Reason, and
and mental chronometry. Curr Opin Neurobiol 2003,
the Human Brain. New York: Grosset/Putnam; 1994.
13:182–186.
90. Lowenstein G, Lerner J. The role of emotion in decision
making. In: Davidson RJ, Goldsmith HH, Scherer KR, 103. Svenson O. Decision making and the search for funda-
eds. Handbook of Affective Science. Oxford: Oxford mental psychological regularities: What can be learned
University Press; 2003. from a process perspective? Organ Behav Hum Decis
Process 1996, 65:252–267.
91. Zajonc RB. Feeling and thinking: preferences need no
inferences. Am Psychol 1980, 35:151–175. 104. Coombs CH. A Theory of Data. New York, NY: Wiley;
1964.
92. Rangel A, Camerer C, Montague PR. A framework
for studying the neurobiology of value-based decision- 105. Fishburn PC. Utility theory. Management Science 1968,
making. Nat Rev Neurosci 2008, 9:545–556. 14:335–378.

FURTHER READING
Brandstätter E., Gigerenzer G, Hertwig R. The priority heuristic: making choices without trade-offs. Psychol
Rev 2006, 113:409–432.
Busemeyer JR, Jessup RK, Johnson JG, Townsend JT. Building bridges between neural models and complex
decision making behavior. Neural Netw 2006, 19:1047–1058.
Gold JI, Shadlen MN. The neural basis of decision making. Annu Rev Neurosci 2007, 30:535–574.
Koehler D, Harvey N, eds. Blackwell Handbook of Judgment and Decision Making. Oxford, UK: Blackwell
Science; 2004.
Lichtenstein S, Slovic P, eds. The Construction of Preference. New York: Cambridge University Press; 2006.

 2010 Jo h n Wiley & So n s, L td.