weber_mindful judgment and decisionmaking(1).pdf

7/29/2019 Weber_Mindful judgment and decisionmaking(1).pdf

1/36

Mindful Judgmentand Decision Making

Elke U. Weber and Eric J. Johnson

Center for the Decision Sciences (CDS), Columbia University, New York, New York 1email: [email protected]

Annu. Rev. Psychol. 2009. 60:5385

First published online as a Review in Advance onSeptember 17, 2008

The Annual Review of Psychology is online atpsych.annualreviews.org

This articles doi:10.1146/annurev.psych.60.110707.163633

Copyright c 2009 by Annual Reviews.All rights reserved

0066-4308/09/0110-0053$20.00

Key Words

choice, preference, inference, cognition, emotion, attention, memor

learning, process models

Abstract

A full range of psychological processes has been put into play to explajudgment and choice phenomena. Complementing work on attentioinformation integration, and learning, decision research over the p

10 years has also examined the effects of goals, mental representatioand memory processes. In addition to deliberative processes, automa

processes have gotten closer attention, and the emotions revolutihas put affective processes on a footing equal to cognitive ones. Ps

chological process models provide natural predictions about individu

differences andlifespan changes andintegrate across judgment anddesion making( JDM) phenomena. MindfulJDMresearch leverages o

knowledge about psychological processes into causal explanations

important judgment and choice regularities, emphasizing the adaptiuse of an abundance of processing alternatives. Such explanations suplement and support existing mathematical descriptions of phenome

such as loss aversion or hyperbolic discounting. Unlike such descritions, they also provide entry points for interventions designed to he

people overcome judgments or choices considered undesirable.

53


2/36

JDM: judgment anddecision making

Contents

INTRODUCTION . . . . . . . . . . . . . . . . . . 54ATTENTION . . . . . . . . . . . . . . . . . . . . . . . 56

Exogenous Influences . . . . . . . . . . . . . . 57Endogenous Influences. . . . . . . . . . . . . 58

ENCODING AND EVALUATION . . 59

Evaluation is Relative . . . . . . . . . . . . . . 59Choice from External Search . . . . . . . 60

Inferences from External Search . . . . 62Goal and Framing Effects . . . . . . . . . . 62

MEMORY PROCESSES . . . . . . . . . . . . . 62Memory Storage and Retrieval . . . . . 62

Memory and Inference . . . . . . . . . . . . . 63MULTIPLE INFORMATION

PROCESSES . . . . . . . . . . . . . . . . . . . . . . 65The Emotions Revolution . . . . . . . . . . 65

Affective Processes . . . . . . . . . . . . . . . . . 65

Dual-Process Explanations . . . . . . . . . 67Dual-Representation Models . . . . . . . 69

LEARNING . . . . . . . . . . . . . . . . . . . . . . . . . 70

Predictive Accuracy . . . . . . . . . . . . . . . . 71

CHARACTERISTICS OF THEDECISION MAKER . . . . . . . . . . . . . . 72

Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

Personality . . . . . . . . . . . . . . . . . . . . . . . . 73Cognitive Traits/Styles . . . . . . . . . . . . . 73

INCREASING POLICY

RELEVANCE . . . . . . . . . . . . . . . . . . . . . 74Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74Wealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

Implications: The BehavioralAdvantage . . . . . . . . . . . . . . . . . . . . . . 75

CONCLUSIONS . . . . . . . . . . . . . . . . . . . . 75

INTRODUCTION

Since its origins in the 1950s, judgment and de-

cision making ( JDM) research has been dom-inated by mathematical functional relationship

models that were its point of departure in theform of normative models. This focus on eco-

nomics and statistics may have led JDM re-search to underutilize the insights and meth-

ods of psychology. Aided by the recent arrival

of neuroscience methodologies to complem

behavioral research, the field has started talize, however, that the brain that decides

to invest pension money and what car to

is the same brain that also learns to recogand categorize sounds and faces, resolves

ceptual conflicts, acquires motor skills suthose usedin playing tennis, and remember

fails to remember) episodic and semantic inmation. In this review, we make a strong

for the utility of this realization.JDM reviews are often structured by

categories, with section headings such as erences, beliefs, and decisions under

and uncertainty (Payne et al. 1992), and choice, intertemporal choice, and s

decisions (Loewenstein et al. 2007). In

trast, our review employs headings that mbe found in a cognitive psychology textb

It capitalizes on the 50 years of researccognitive and motivational processes that

followed Simons (1957) depiction of hudecision makers as finite-capacity informa

processors and decision satisficers. Attent(in particular, perceptual) and learning

cesses have a longer history of considerawith phenomena such as diminishing s

tivity of outcomes or reference point ening for perception and the illusion of vali

for learning. Affective, memory, and pre

tion processes have only more recently emeas explanations of judgment and ch

phenomena.We retain some task category distinc

to organize specific content where approprThus, we distinguish between preference

inference. Preferences involve value judgmand are therefore subjective, such as deci

how much to charge for an item on eBay. Iences are about beliefs, such as the judgedli

hood that a political candidate will win theelection, and typically have objectively v

able answers. Although this distinction refl

tradition, it may not reflect psychological ity. Preferences and inferences seem to dra

the same cognitive processes.Our ability to organize our review by

chological processes is a sign of the grow

54 Weber Johnson


3/36

maturity of the field. JDM research no longer

simply generates a growing list of phenomenathat show deviations from the predictions of

normative models. Instead, it has been devel-

oping and testing hypotheses about the psycho-logical processes that give rise to judgments and

choices and about the mental representationsused by these processes. Although the number

of JDM articles in major social psychology jour-nals remained constant over the past 10 years,

the number of JDM articles in major cognitivepsychology journals increased by 50% over that

period, reflecting the increased interest in inte-grating judgment and choice phenomena with

the frameworks of hot and cold cognition.New tools have undoubtedly contributed to

this trend.This includes functionalimaging and

other neural and physiological recordings, pro-cess tracing tools (see sidebar Process Models

and Process Tracing), and, increasingly, mod-eling tools such as mediation (Shrout & Bolger

2002) and multilevel analysis (Gelman & Hill2007). A focus on psychological mechanisms

has guided the decomposition of JDM taskbehavior into contributing cognitive processes

and their variation across groups (Busemeyer& Diederich 2002, Stout et al. 2004, Wallsten

et al. 2005, Yechiam et al. 2005). An increasedfocus on individual differences has been a no-

ticeable feature of behavioral decision research

over the past decade. Increased use of Web-based experimentation (Birnbaum & Bahra

2007) allows access to respondents with muchbroader and representative variation on demo-

graphic and cognitive variables, with new in-sightsabout individual, group, andlife-spandif-

ferences on JDM tasks, topics that are discussedin the second section of our review. More af-

fordable genotyping has led to examinationsof the heritability of economic traits like trust

(Cesarini et al. 2007).JDM research attracts public and media at-

tention because it addresses real-world phe-

nomena, from myopic dietary decisions to ex-cessive stock market trading. Policy makers

have increasingly utilized JDM theory and re-sults when designing or changing institutions

(Shafir 2008), the topic of our last major sec-

PROCESS MODELS AND PROCESS TRACING

Early models in decision research attempted to explain changein judgments or decisions (the output) as a result of change

in information considered (the inputs) using tools such as re

gression and analysis of variance. This approach is problemati

because it considers only a subset of observable behavior and because different models can predict one set of outputs from a giveset of inputs. Process models help because they consider mor

variables and add multiple constraints. By virtue of hypothesizing a series of psychological processes that precede a judgmen

or choice, they make predictions about intermediate states of thdecision maker, between the start and end of the decision (Wha

external information is sought out? What facts are recalled frommemory?). Process models alsomake predictions about the tem

poral order of these states (What will a decision maker thinabout first, second, etc.?). Process data are the data used to tes

hypotheses about these intervening processes and intermediat

states. They include functional imaging and other measures olocalized brain activation, response times, verbal protocols, eye

movement tracking, and other information-acquisition tools (sewww.mouselabweb.org).

Preferences: ineconomics, inferrefrom choices andassumed to reflectutilities. In psychothought to be

constructed in ordmake a choice

Inferences: decismakers judgmentsabout the world uslogic and oftenimperfect anduncertain informa

tion. The recognition that preferences are typ-ically constructed rather than stored and re-

trieved (Lichtenstein & Slovic 2006) may be

psychologys most successful export to behav-ioral economics and the policy community and

illustrates the utility of psychological processexplanations. We now know how, and increas-

ingly why, characteristics of choice options andtask guide attention, and how internal memory

or external information search and option com-parison affect choice in path-dependent ways.

This not only explains apparent inconsistenciesin choice, but also provides insights and recipes

for decision aiding and interventions, includ-ing the design of decision environments that

nudge people to construct their preferences in

ways they will not regret after the fact (Thaler& Sunstein 2008).

Psychological process explanations cast lighton areas obscured in the shadows of statistical

decision-process approaches. For example,years of work with Egon Brunswiks lens model,

which provided valuable insights into the

www.annualreviews.org Mindful Judgment and Decision Making 55


4/36

performance of human decision makers, may

have hidden the important distinction betweenautomatic and deliberative (controlled) pro-

cesses and their properties (Schneider & Chein

2003). Process explanations also serve an in-tegrative function by explaining multiple phe-

nomena, providing an organizing principle fora field criticized for being long on effects and

short on unifying explanations. Judgments andchoices typically engage multiple psychological

processes, from attention-guided encoding andevaluation, to retrieval of task-relevant infor-

mation from memory or external sources, pre-diction, response, and postdecision evaluation

of consequences and resulting updating. Dif-ferent tasks involve these processes to differ-

ent degrees. For example, attention accounts

for a larger proportion of response variance indecisions from description, where the decision

maker is explicitly provided with all relevantinformation in numeric or graphic form. In

contrast, memory and learning will be moreimportant in decisions from experience, where

information about outcomes and their likeli-hood is acquired by trial and error sampling

of choice options over time (see Hertwig et al.2004). Similarly, affective processes are more

important in dynamic decisions under uncer-tainty, whereas analytic evaluations play a larger

role in static risky decisions (Figner et al. 2008).

The last comprehensive Annual Reviewarticle on JDM was published more than

10 years ago (Mellers et al. 1998). Two reviewssince then have addressed specialtopics, namely

rationality (Shafir & LeBoeuf 2002) and un-solved problems in decision research (Hastie

2001). Given this time span between JDMarticles, our review had to be extremely se-

lective. Our mandate, to review research oncognitive processes in judgment and choice,

necessitated the omission of papers that de-scribe JDM phenomena without emphasizing

psychological process interpretations. We also

had to limit the scope of psychological pro-cesses covered. With a few exceptions, we omit-

ted very basic perceptual processes (e.g., cate-gorization) and processes that go beyond the

individual (e.g.,groupjudgments and decisions;

interdependent, competitive, and strategi

cisions; advice giving; social judgments; inmation aggregations; and prediction mark

We were unable to go beyond judgmentchoice processes, not covering problem

ing, reasoning, or positive psychology. The

geoning field of neuroeconomics recentlceived its own review (Loewenstein et al. 2

When multiple papers could have been for a given point, we restricted ourselv

the most important, innovative, or compresive examples, and omitted citations for cl

phenomena.

ATTENTION

Decision makersface a wealth of potentially

evant information in the external environmand memory. Given the processing limita

ofHomo sapiens, selectivity is a central comnent of goal-directed behavior. Selective a

tion operates at very basic levels of percepidentification (Lachter et al. 2004). It also o

ates at higher cognitive levels, including thtial perception of the situation and assess

of the task at hand (framing, goal elicitatevidence accumulation (which can be ext

or internal, and usually is a combination o

two), and judgment or choice (determiningoffs or decision rules).

A focus on attention as a finite resorequiring selectivity, goes back to the be

nings of scientific psychology. William Jain 1890 considered attention a necessary

dition for subsequent memory, distinguibetween voluntary and nonvoluntary a

tion, and suggested the use of eye mments to track attentional focus. More rece

Daniel Kahneman (1973) summarizedwhaknown about attention during the postbe

iorist period when attention was used as abel for some of the internal mechanisms

determine the significance of stimuli (p

Kahneman emphasized capacity limitationthe selective aspect of attention and di

guished betweentwo determinants, momen(voluntary) task intentions and more endu

dispositions such as the (involuntary) orien

56 Weber Johnson


5/36

response to novel stimuli. Herbert Simon

(1978) identified conscious attention as a scarceresource for decision makers in the year of his

Nobel prize; Kahnemans Nobel lecture (2003)

reiterates that this scarce resource needs to beallocated wisely and points to automatic (ori-

enting) processes and fast emotional reactionsas means to that end.

Exogenous Influences

Orienting responses. Some features of theenvironment attract attention because respond-

ing to them has survival value. Changes in the

environment, and especially the appearance ofnovel stimuli, introduce the possibility of op-

portunity and/or threat. Constant exposure toa stimulus leads to habituation, i.e., reduced re-

sponding, as things not previously responded toare likely to be neither dangerous nor promis-

ing.Ontheotherhand,achangeintheenviron-ment results in dishabituation and an orienting

response (Posner & Rothbart 2007).As a result of the orienting response to

changes in the environment, things that varyautomatically attract and maintain attention. A

siren that wails will attract attention longer thana siren that operates at a constant frequency.

This has implications for a wide range of is-

sues, from research design to human factorsand institutional design, with salient continu-

ous changes in the level of key decision vari-ables as a recipe for keeping peoples attention

on the task, a manipulation perfected by videogames. Arguments by Birnbaum (1983) about

the consequences of within- versus between-subject manipulations of base rates have re-

cently been revived in the context of quan-tity (in)sensitivity in protected value tradeoffs.

Bartels & Medin (2007) reconcile conflictingresultsby showing that between-subject designs

leadto quantity insensitivity(e.g., thesame will-

ingnesstopaytorestorethepHlevelofonelakeor of ten lakes) (Baron & Ritov 2004), whereas

within-subject designs, which attract attentionto variation in quantity, show sensitivity to the

variable (Connolly & Reb 2003).

Task characteristics. In the same spirit of in-

tegrating across apparently contradictory re-search results, a range of JDM tasks and con-

text characteristics have beenexaminedfor theireffect of guiding attention and thus decision

weight to different outcome dimensions. Vi-

olations of procedure invariance are one ofthe most vexing cases of deviation from nor-

mative models of preference. Selling pricestypically exceed buying prices by a factor of

two, even when strategic misrepresentation iseliminated, and discounting of future bene-

fits is much steeper when people are askedto delay rather than accelerate consumption

(Kahneman & Tversky 2000). Below, we re-view information-recruitment mechanisms that

explain how the direction of an economic trans-actions (e.g., acquiring or giving up ownership;

switchingfrom immediate to delayed consump-

tion or vice versa) can affect valuation. Relatingsuch valuation asymmetries to attentional pro-

cesses, Carmon & Ariely (2000) show that de-cision makers focus their attention on the fore-

gone, i.e., the status quo and its characteristicsattract more attention and thus importance and

decision weight than do other choice options.

Judgment versus choice. It has long been knownthat judgment versus choice tasks can direct at-

tention to different characteristics of choice op-tions, from preference reversal studies of risky

decisions in the 1970s to the theory of task-

contingent weighting of multiattribute choice(see Lichtenstein & Slovic 2006). Editing oper-

ationscancelout commonoutcomes for choicesbut cannot do so for judgments, with result-

ing differences in attentional allocation and in-formation use that translate into differences in

preference. Consumer purchases are typicallythe result of choice from among multiple alter-

natives, where alignable features receivegreaterattention, whereas postpurchase consumer sat-

isfaction is the result of judging the product inisolation, where features that are easily evalu-

ated in an absolute sense receive greater atten-

tion (Hsee & Zhang 2004). Many task-detail-induced inconsistencies in judgment and choice

can be explained by differences in attentional



6/36

PT: prospect theory

Beta-delta model:explains greaterdiscounting of futureoutcomes whenimmediate rewards areavailable than when allrewards are in thefuture by anexponential deltaprocess that alwaysoperates and anadditional exponentialbeta process that onlyoperates whenimmediate rewards arepresent

focus, although the inconsistencies are not ex-

clusively due to attentional mechanisms. Moststable JDM phenomena such as preference

reversals are probably stable because they are

multiply determined.

Description of choice options. The way in whichinformation about choice options is communi-

cated to decision makers influences preferenceconstruction through selective attention, even

though variants may be informationally equiv-alent. One of these ways is the order in which

options are presented. Candidate name orderon ballots, for example, has been shown to in-

fluencepreference andvoting sufficiently to de-termine election results (Krosnick et al. 2004).

Options encountered first capture atten-

tion, leading to reference-dependent subse-quent evaluations and comparisons (Kahneman

2003). In decisions from description, some out-come dimension values (namely certainty on

the probability dimension and immediacy onthe delay dimension) are given special status,

i.e., extra attention and decision weight of amore categorical than continuous nature, as

captured by prospect theorys (PT) decisionweight function and Laibsons (1997) beta-delta

model of time discounting. Weber & Chapman(2005) show that certainty and immediacy

are connected, in that adding delay undoes

the special preference given to certainty, andadding uncertainty removes the special prefer-

ence given to immediacy.

Process of knowledge provision. In decisions fromdescription, attention is shared between out-

come and probability information, which areboth explicitly provided. In decisions from ex-

perience, the series of sequentially experiencedoutcomes focuses attention on this dimension,

with more recent outcomes looming larger(Weber et al. 2004). The emergent evidence

that rare events get underweighted in decisions

from experience but overweighted in decisionsfrom description, as captured by PT, can be ex-

plained by differences in attentional focus dur-ing information acquisition (Erev et al. 2008),

because attention directed by both external and

internal factors has been shown to translate

decision weight (Weber & Kirsner 1997).

Endogenous Influences

In addition to external influences, the inte

state of the decision maker guides attenDecision makers generally have more con

over their internal states,thus allowing formvoluntary allocations of attention.

Goals. JDMresearchovertheperiodofou

view has started to interpret behavior in tof goals and plans rather than (or in add

to) utilities (Krantz & Kunreuther 2007). vival and economic well-being dictate that

terial goals play an important role in peo

plans and decisions. Material goals are ressible for the effectiveness of financial incen

in shaping behavior. However, people hamany other goals, some of which relate to

material dimensions of the choices made being defensible (Lerner & Tetlock 19

whereas others relate to the nature of thecision process [e.g., wanting a procedurally

process (Tyler 2005) or a process that feels(Higgins 2005)]. With multiple and often

flicting goals in play, selective attention toferent subsets of goals has been shown t

fluence how a decision is made and wh

selected (Krantz & Kunreuther 2007). A rof factors has been shown to situationally

vate goals or chronically elevate their accbility, including cultural values of the dec

maker (Weber et al. 2005a), the content doof the decision, e.g., risky choices about co

grades versus stock investments (RettingHastie 2001),and task characteristics such a

quired accountability (Tetlock 2002). Activgoals determinewhether the decision rules

are deontological (Whatis right?) versussequentialist (What has the best outcom

versus affective (What feels right?) (Bart

Medin 2007). Ariely et al. (2000) point toimportance of goals in the context of cho

between different streams of experience time. Similar to the discussion above a

quantity (in)sensitivity in the contex

58 Weber Johnson


7/36

protected value tradeoffs, people are more or

less duration sensitive when evaluating expe-riences over time as a function of how their

attention is focused by how they report their

experiences and why.

Affect as spotlight. Emotions experienced bythe decision maker, in addition to the many

cognitive factors mentioned above, focus atten-tion on features of the environment that mat-

ter for emotion-appropriate action tendencies.Mood-congruent perception focuses attention

on either upside opportunity or downside risk(Chou et al. 2007). Feelings of fear or worry

focus attention on the source of the apparentthreat and ready flight responses (Loewenstein

et al. 2001). Feelings of anger focus attention

on information about motives and responsibil-ity and make decision makers eager to act and

punish. Sadness elicits a desire to change onesstate, resulting in reduced selling and inflated

buying prices, whereas disgust triggers a desireto purge or acquire less,with the opposite effect

on willingness to pay (Lerner et al. 2004).

ENCODING AND EVALUATION

One clear finding from behavioral decision re-

search is that information is acquired by deci-sionmakersin ways not addressed by normative

models. Goal-relevant and context-sensitiveencoding of information is one of the ways in

which people execute their task with minimaleffort and, perhaps, maximal satisfaction. One

important distinction to make is between in-formation obtained from a search of external

sources (external search; e.g., when choosinga cereal by studying product information in a

supermarket aisle) versus information retrievedfrom memory (internal search; e.g., when re-

trieving options about which route to take on a

drive home). Most decisions involve both kindsof search. The cereal choice probably involves

recalling how much the previously purchasedbrand was enjoyed, and the choice of a route

home uses external retrieval cues and infor-mation about traffic congestion. The distinc-

tion matters, however, because the properties

Variability or riskthe risk in risky choptions is introducby not knowing whoutcome will occueconomics and fina

the variance of posoutcomes is used ameasure of risk

of external search (reviewed in this section) are

demonstrably different from the properties ofretrieval from memory (reviewed in the next

section on Memory Storage and Retrieval).

Evaluation is Relative

Outcomes. The humorist Thurber was once

asked how he liked his new wife. His re-sponse Compared to what? reflects one of

prospect theorys (Kahneman & Tversky 1979)major insights, namely that evaluation is rela-

tive. This insight continues to gather support,albeit in more complex ways than formalized

by PT. Since neurons encode changes in stim-ulation (rather than absolute levels), absolute

judgments on any dimension are much more

difficult than relative judgments. The list of ref-erence points used in relative evaluation con-

tinues to grow and includes other observed orcounterfactual outcomes from the same or dif-

ferent choice alternatives, as well as expecta-tions. For example, the range of options of-

fered as potential certainty equivalents has beenshown to affect peoples valuation of gambles

(Stewart et al. 2003). One important area forfuture research is to understand better the se-

lection among reference points and how multi-ple reference points might be used.

Most discussions of relative evaluation have

focused on the evaluation of a single outcomeby comparing it to a reference point, typically

by computing their difference in value. How-ever, differences themselves may be in need

of relative evaluation. If asked how good his$5000 salary increase was, Thurber probably

would have also asked, compared to what?Gonzalez-Vallejos (2002) proportional differ-

ence model is a stochastic model of choice thatanswers this question. Differences in attribute

values of two choice options are normalized bydividing them by the best (for positive) or worst

(for negative) possible outcome. These propor-

tional differences are then integrated across at-tributes by a stochastic decision process, al-

lowing the model to account for a broaderrange of choice patterns than other models

(Gonzalez-Vallejo et al. 2003). Normalization



8/36

Expected value: theaverage outcome onegets from some riskychoice; e.g., $50 is theexpected value of acoin toss for $100 or

$0 [$50 = 0.5($100)+0.5($0)]

of outcome differences in ratio form also ap-

pears to hold for implicit evaluations of variabil-ity or risk. The coefficient of variation, defined

as the standard deviation of possible choice

outcomes divided by their expected value (i.e.,risk per unit of return), predicts peoples risky

choices and risky foraging of animals far betterthan does the typical nonnormalized measures

of variability or risk (standard deviation or vari-ance) employed in finance (Weber et al. 2004).

The discriminability of differences is a cen-tral concern for relative evaluations. It lies at the

root of Ernst Webers 1834 basic law about thepsychophysical coding of just-noticeable differ-

ences, which captures the observation that de-tectable increases in visual or auditory signal

intensity are proportional to the starting value,

i.e., need to be larger for larger starting values.Furlong & Opfer (2008) provide provocative

evidenceabouttheeffectofoutcomemagnitudeon the discriminability of differences. In their

studies of humans and orangutans in the pris-oners dilemma game, changing the currency in

which the usual payoffs for defection or coop-eration are issued (for humans, dollar outcomes

multipliedby 100to produce outcomes in cents;for orangutans, grapes issued intact or cut into

tiny pieces) increases the rate of cooperation,presumably because thedifference in payoffs for

defection over cooperation is less discriminable

with the larger numeraires.

Probabilities. Traditionally, explicitly pro-vided probability judgments of events were

thought to reflecteither a frequentist evaluationor an expression of a degree of belief. How-

ever, more recent formulations have positedtransformations of explicitly provided outcome

probabilities in choice into decision weightsthat are a function of the amount of atten-

tion paid to the different potential states of theworld, which is affected by more than the states

likelihood of occurrence. Events may attract

greater attention for perceptual and motiva-tional reasons (Weber & Kirsner 1997). Thus,

small-probability events may be overweightedby PT relative to their stated likelihood of

occurrence because decision makers attention

is regressive. In other words, decision ma

pay more equal attention to all possible comes than is warranted by their (typically

equal) probabilities, and decision makers liat extreme outcomes to assess best- and wo

case scenarios. Rank-dependent models of

choice have provided such a reinterpretatiothe way in which explicitly stated probabi

are evaluated in choice. They also provide aternative way to think about risk-averse or

seeking behaviors. In cumulative PT (Tve& Kahneman 1992),the subjectiveweight g

to a given outcome no longer is simply a noear transformation of its objective probabil

occurring, but also reflects the relative ranthe outcome in the distribution of possible

comes. Cumulative PT is only one way in wthe evaluation of outcome probabilities can

pend on the position of the outcome in the

figuration of outcomes (Lopes & Oden 1More complex ways, such as those in B

baums transfer of attention model (Birnb2005),havebeenshowntoaccountforabro

range of choice phenomena. These attenteffects become even more important w

choiceoptionscontainmorethantwooutcor when the gambles are mixed (Luce 2

Payne 2005).

Choice from External SearchHeuristics for risky choice. Brandst

et al.s (2006) priority heuristic (PH) triaccount for many phenomena in risky ch

in simpler ways than do models that invtradeoffs, such as PT. The model is note

thy for making not just choice predictionsalso predictions about response times andi

mation acquisition. The PH has been criticfor its use of discrete measures of error (R

& Wang 2008) and for making choice pr

tions that are not observed (Birnbaum 20Johnson et al. (2008) found that although s

implications of the PH were supportedcritical test, namely that decision maker

not integrate probabilities and payoffs, not borne out by process measures. De

the mixed empirical support surrounding

60 Weber Johnson


9/36

heuristic, the research exchange triggered by it

demonstrates that process predictions and theirtests can improve choice models.

Sampling and evaluation in external search.

If we believe that decision makers often attend

selectively to a subset of possible information,it is important to understand the properties of

such samples, the processes used to producethem, and the consequences these samples have

on decisions.A class of what might be called middle-level

sampling models ambitiously attempts to de-scribe a large set of empirical regularities or

stylized facts. Each model has its own set of as-sumptions about cognitive processes and rep-

resentations and thus makes predictions not

just for observed choices, but also for processmeasures such as response times (Ratcliff et al.

2006). Although these models share a concernwith the accumulation of evidence via sampling,

they emphasize different aspects of the decisionprocess.

Prototypical of a class of models that couldbe characterized as stimulus sampling mod-

els are recent extensions of Busemeyer &Townsends (1993) decision field theory (DFT)

to multiattribute choice (Roe et al. 2001) andto models of value judgments as well as choice

( Johnson & Busemeyer 2005). The key idea in

DFT is that attributes of choice alternatives arerepeatedly randomly sampled andthat evidence

accumulates over samples. This process of in-formation retrieval, whether from the external

environment or from memory, is assumed to beindependent of the evaluation of the object, i.e.,

is not path dependent. When applied to choice,DFT posits a race between options, with each

additional acquisition of evidence increasing ordecreasing the valuation for an option, ending

when the first option exceeds a preset thresh-old. In addition to having a closed-form mathe-

matical formulation, DFTcan alsobe expressed

as a multilayer connectionist network and hasbeen applied to explain context effects such as

the similarity, attraction, and compromise ef-fects (Roe et al. 2001). By adding a set of po-

tential responses (in a comparison layer) to its

Decision field the(DFT): amathematical andprocess modelsuggesting thatdecisions are made

aggregating samplrandomly drawn frthe informationavailable about a salternatives

neural network version, DFT can generate pre-

dictions for several preference reversals (Buse-meyer&Diederich2002).DFT(anditsdecom-

position) has also provided a useful framework

to analyze group differences on the Becharagambling task, as described below. Computa-

tional considerations have led to a modifica-tion of DFT that incorporates loss aversion

into the accumulation of evidence (Usher &McClelland 2004), thus extending stimulus

sampling models to explain the endowment ef-fect and other JDM phenomena attributed to

loss aversion.Decision by sampling (Stewart et al. 2006),

another mid-level model, is an interesting at-tempt to explain several stylized facts with two

simple mechanisms: (a) value is constructed by

simple ordinal comparisons between an objectat hand and consecutive repeated samples of

objects drawn from memory, and (b) the sam-ples reflect the external ecological frequency

of objects. Using archival data, these two as-sumptions are able to reproduce the PT value

and probability weighting function and a time-discounting function that looks hyperbolic.

Decision by distortion. Stimulus sampling

models typically assume samples that are un-biased reflections of the environment and are

path-independent. In contrast, two streams of

research suggest that choice involves a biased,and path-dependent, integration of informa-

tion. Building on earlier ideas about con-structed dominance by Montgomery and

Svenson in the 1980s, Holyoak & Simon (1999)and Russo and colleagues (2000) posit that

choices are speeded up and made with mini-mal regret by distorting the value of options

to support early-emerging favorites. The exis-tence of an early favorite leads to subsequent

information being interpreted in a waythat sup-ports that favorite, bolstering its chances of be-

ing chosen (Simon et al. 2004), even for a single

option (Bond et al. 2007). Simply being listed asthe first option can cause this distortion of val-

ues and increase in choice (Russo et al. 2008),showing the influence of attentional focus on

subsequent evaluation and choice.



10/36

Inferences from External Search

In contrast to mechanisms such as availability,which posit that biases in inference result from

biased representations produced by recall, sev-eralresearchershavearguedthatsuchbiasescan

result from biased sampling of external infor-mation, either as a function of how the informa-

tion is presented by the environment or by bi-asesinasearchonthepartofthedecisionmaker

(Fiedler 2000). For example, the observer of a

conversation, which provides a sampling of thebeliefs of the two conversing parties, may get a

biased sample of what the participants believebecause a range of Gricean conversational rules

apply restrictions (e.g., not repeating what wasjust said). As a result, Fiedler argues, the ob-

server may well conclude that the conversationis more hostile than it really is. By arguing that

the observer is insensitive to the bias in the ob-served sample of beliefs, Fiedler (2000) moves

the origins of observed bias from the decisionmakers memory (as in availability) to the en-

vironment, aided by the decision makers lack

of understanding the biased origin of the sam-ple. Juslin et al. (2007) have applied very similar

ideas to confidence judgments.

Goal and Framing Effects

McKenzie & Nelson (2008) suggest that dif-ferent semantic frames that might be seen as

logically equivalent (e.g., a glass being half fullor half empty) linguistically transmit different

information because different frames elicit dif-ferent semantic associates. Fischer et al. (1999)

similarly suggest that different response modeshave different goals and that evaluation dif-

fers to accommodate those goals. For exam-ple, prominent attributes receive more weight

in tasks whose goal is to differentiate among

options than in tasks whose goal is to equateoptions.

MEMORY PROCESSES

Making decisions without recourse to relevant

prior memories is a difficult task and is a topic

that has long fascinated writers and filmma

Memory is necessary for our ability to learnto draw on past experience to predict futur

sires, events, or responses to outcomes. Ye

connection between properties of memoryjudgment and choice has previously been

derexplored. During the past decade, memconsiderations have played a more promi

role in explanations of JDM phenomenatemptingtoleveragewhatweknowaboutm

ory to provide insight into the processes unlying known decision phenomena (Reyna

2003, Schneider 2003), but this is still a tively underdeveloped area of behavioral

sion research.

Memory Storage and Retrieval

Memory accessibility and priming. See

stimulus results in a transient increase in

cessibility of the representation of that stlus and related concepts, a phenomenon c

priming, with effects on subsequent mory access, i.e., shorter reaction times

greaterlikelihood of retrieval.Priming is wused in social cognition, where primed

tudes and values shape behavior. Extendingparadigm, Mandel & Johnson (2002) dem

strated priming effects in multiattribute chIn a consumer choice task, their selective p

ing of product attributes with appropriate wpaper on the initial page of an online

affected not only choice but also informasearch and use.

Memory is reactive. Unlike computer mory, human memory is changed by attem

at retrieval. Accessing memory both incrshort-term accessibility and changes the l

term content of memory.

Short-term effects. Studies of anchoring sugthat priming memory accessibility, and co

quently preference, can be changed by askprior question, even if the answer to this q

tion should be irrelevant to subsequent tsuch as using the last four digits of a social s

rity number as an anchor for pricing a gam

62 Weber Johnson


11/36

(Chapman & Johnson 1999). This effect was

replicated with fine wine by Ariely et al. (2003),who also show that such accessibility-mediated

anchoring effects are strong and robust and

persist in the presence of significant accuracyincentives, experience, and market feedback.

The selective accessibility model provides sim-ilar mechanisms and provides evidence that an-

chors make some information more accessi-ble as measured by reaction times (Mussweiler

& Strack 2001), though accessibility may notbe sufficient to explain all anchoring effects

(Epley & Gilovich 2001).

Long-term effects. Accessing information aboutpossible choice options not only generates

short-term changes in the accessibility of re-lated information but also changes memory in

a more permanent fashion, a phenomenon longrecognized in social cognition. In the context

of consumer choice and a line of research thatgoes back to the work on the self-correcting na-

ture of errorsof prediction, measuringthe long-

term effects of purchase intentions on memoryhasbeenshowntochangesubsequentpurchases

(Chandon et al. 2004).

Retrieval and preference construction. Arecent perspective on preference construction,

query theory (QT; Johnson et al. 2007), sug-

gests that decision makers consult their mem-ory (or external sources) with queries about the

choice alternatives, in particular their merits orliabilities. QT assumes that most tasks suggest

a natural way to the order in which queriesare posed. When one class of components of

a memory structure is queried, the accessibil-ity of other components that could be response

competitors is temporarily suppressed to mini-mize intrusions, but with consequences for the

success of subsequent queries for which thesecomponents are legitimate responses. Memory

inhibition as the result of prior recall of re-

lated and competing material is one of the old-est and most developed memory phenomena

(Anderson & Neely 1996). Johnson et al. (2007)show that QT accounts for the endowment

effect, under the assumption that sellers and

Query theory (Qa process model ofvaluation describinhow the order ofretrievals frommemory (queries

play a role in judgithe value of objectemphasizing outpuinterference

buyers have different query orders, and they

demonstrate the causal involvement of queryorder andmemory inhibition by making the en-

dowment effect disappear by switching the nat-

ural order of queries. Extending this paradigm,Weber et al. (2007) show that queries about

reasons supporting immediate versus delayedconsumption are issued in reverse order for in-

tertemporal decisions about accelerating or de-laying consumption, explaining the well-known

result that people are much more impatientwhen delaying than when accelerating con-

sumption. Explicitly prompting queries in theorder opposite to the naturally occurring one

again eliminates the effect. The task- and goal-specific distortions in balance of support that is

generated by QT-predicted and empirically ob-

served memory retrieval interference presum-ably have the same function (i.e., faster deci-

sions with less postdecision regret) in decisionsbased on internal search that predecisional dis-

tortions (discussed in the previous section) havein decisions based on external search. Both pre-

decisional distortion of external informationand QT-related biased memory retrieval sug-

gest that the process of preference or infer-enceconstructionis characterized by systematic

path dependency, contrary to the assumptionsof most mathematical models of judgment and

choice.

Consistent with a memory interference ac-count, Danner et al. (2007) show that three or

more retrievals of a specific means towards agoal will succeedin inhibitingcompeting means

for the same goal. It is worth noting that thisdiscovery in social cognition in the context

of habit formation and goals-means networkscoincides with experimental practice in proac-

tive interference studies (e.g., Dougherty &Sprenger 2006). Thus, memory retrieval is one

more way in which goals have been tied moreclosely to decision making over the past decade.

Memory and Inference

Memory and support theory. Support the-ory (ST), proposed by Tversky & Koehler

(1994), models probability judgments as a



12/36

Luces choice axiom:the probability ofselecting one item overanother from a pool ofmany items is afunction of only the

utilities of those twoitems and is notaffected by thepresence or absence ofother items in the pool

Support theory: amodel of inferenceabout the probabilityof an event that usesthe relative weight ofwhat we know and cangenerate about theevent in question (its

support) andcompares it to what weknow and can generateabout all otherpossible events

RH: recognitionheuristic

comparison of support for focal hypothesis A

(s(A)) with support for a set of alternative hy-potheses B (s(B)), in the form of a ratio fa-

miliar from Luces choice axiom: p(A,B) =

s(A)/(s(A) + s(B). Support theory is a rationalmodel in the sense that it assumes that set B

includes only relevant alternative hypotheses,i.e., hypotheses that have some probability of

occurring. Since competing hypotheses are of-ten generated by associative memory processes

from long-term memory(Dougherty & Hunter2003), irrelevant alternative hypotheses (that

have no possibility of occurring in the contextof interest) may well be generated and may af-

fect probability judgments by occupying valu-able slots in limited-capacity working memory

[referred to as inhibition failure by Dougherty

& Sprenger (2006)]. Irrelevant alternatives inworking memory may not be identified as ir-

relevant, referred to as discrimination failureby Dougherty & Sprenger (2006), who provide

evidence for such failures using a proactive in-terference paradigm. A negative correlation ex-

ists between individual differences in working-memory capacity and degree of subadditivity of

probability judgments. The judged probabilityof a focal event (e.g., rain) is larger when com-

pared to the implicit disjunction (not rain) thanwhen it is compared to the explicit disjunctions

(e.g., sunshine, snow, cloudy, all other), suggest-

ing that people with greater working-memorycapacity are able to include more alternativehy-

potheses in the implicit disjunction condition(Dougherty & Hunter 2003). In combination,

these and related studies suggest that augmen-tation of support theory with realistic assump-

tions about the retrieval and evaluation of al-ternative hypotheses can significantly increase

its predictive accuracy. Dougherty & Sprenger(2006)also illustratehowmeasuresof individual

differences can help distinguish among hypoth-esized judgmental processes.

Memory-based heuristics for inference. In1996, Gigerenzer and Goldstein suggested the

take-the-best (TTB) strategy as both an accu-rate and easy procedure for inferences based on

memory retrieval. TTB mimics what is known

as a lexicographic decision rule in choice,

gesting that good inferences can be made bing the most diagnostic cue(s) that disting

betweentwo alternatives. Knowledge aboudiagnosticity depends, of course, on metaco

tive insight about past inferential accuracy

tial simulations showed surprising levels offormance for a process that uses such lim

information. TTB performs particularlywhen the distribution of cue validities is h

skewed. However, TTBis nottheonly heuthat does well. Simulations show that he

tics that are even simpler than TTB caquite well in the same environments (Hog

& Karelaia 2007). Other simple heuristicswell or better (Chater et al. 2003) in other

ronments. Examinations of TTB as a destive model of memory-based inference sug

that it is not universally used, but also no

frequently employed, describing between and 72% of inferences (Broder & Gaissm

2007). More importantly, use of the strategpears to vary in a way that is adaptive given

environment, with more-intelligent decismakers being more adaptive (Broder 2

New developments are models that integTTB and full information use along a co

uum, specified by the amount of weight gto the comparison of different attributes (L

Cummins 2004), and generalizations that the assumption that decision makers know

exact cue weights (Bergert & Nosofsky 20

A similar story surrounds the recognheuristic (RH), posited as a powerful rul

inference in cases in which only one ofprovided comparison alternatives is recogn

and applied in tasks such as deciding whictwo cities is larger (Goldstein & Gigere

2002). Initialdemonstrations showed goodformance over a wide range of domains

subsequent studies have delineated bounconditions. In a paradigm that teases

recognition and cue validity, Newell & Sh(2004) show that RH is abandoned when re

nition is not the most reliable cue. Similthe recognition heuristic is not used w

recognition can be attributed to other ca

(Oppenheimer 2003). Although it is clear

64 Weber Johnson


13/36

recognitioncanbeausefultoolininference,the

debate now seems to be whether recognition isalways used as a first stage in inference (Pachur

&Hertwig2006)orwhetherrecognitionissim-

ply one cue in inference that can be integrated(Richter & Spath 2006) but has no special sta-

tus. In choice, recent work on decision modes(Weber et al. 2005a) identifies recognition as

a decision mode that uses identification of achoice situation as a member of a class of situ-

ations for which a prescribed best action exists,following in the tradition of image theory by

Lee Beach and work by James March in theearly 1990s.

Work in inference seems to be reaching aconclusion similar to that of previous work in

choice by Payne and colleagues (1992). The

number of processes in the adaptive toolbox islarge, and their use is adaptive to task character-

istics. The interesting questions are how pro-cessing strategies are selected and when they

succeed and fail. Answers to these questionswill come from explicit models of strategy selec-

tion (Rieskamp & Otto 2006) and more formaland detailed models of the role of memory and

forgetting in inference (Dougherty et al. 1999,Schooler & Hertwig 2005).

MULTIPLE INFORMATION

PROCESSESNormative JDM models have an appealing sim-

plicity. With an axiomatic foundation, they em-ploy a small number of primitives, abstract from

content and context, and give rise to consistentjudgments and decisions across situations. Ini-

tial attempts to make these models psycholog-ically plausible and better able to describe ob-

served judgment and choice patterns coincidedwith the cognitive revolution in psychology that

used the digital computer as its metaphor for

human information processing and contrastedalgorithmic with heuristic solutions. Norma-

tive model modifications thus focused oncognitive shortcuts taken by limited-capacity

information processors. This repertoire of al-ternative cognitive strategies was first investi-

gated in the context of preference by Payne

et al. (1992) and subsequently extended to in-

ference tasks (Goldstein & Gigerenzer 2002).In the context of preference, affective processes

have recently been added to the list of poten-

tially adaptive strategies (Finucane et al. 2000,Luce et al. 2000).

The Emotions Revolution

Though successful in many ways, the cognitive

revolution may have been too focused on an-alytic and computational processes. The emo-

tions revolution of the past decade or so hastried to correct this overemphasis by docu-

menting the prevalence of affective processes,depicting them as automatic and essentially

effort-free inputsthat orientand motivate adap-

tive behavior. Review articles that describe therole of emotions in risky choice and their

effort-reducing potential (Finucane et al. 2000,Loewenstein et al. 2001) incorporate prior work

on emotional priming by Johnson and Tverskyin 1983 and on psychological risk dimensions

(Slovic 1999). Following Peters et al. (2006a),we describe research on four functions of af-

fect: as spotlight (discussed under Attention),information, common currency, and motivator.

Affective Processes

Affect as information. Emotions experiencedwhile making a decision are incorporated as in-

formation into choices (Schwarz2002). Positiveand negative past associations with available

choice outcomes thus contribute to newdecisions. Loewenstein et al. (2001) distinguish

between immediate emotions and anticipated/expected emotions. Immediate emotions,

aroused either by task-relevant characteristicsor incidentally, and their effect on judgment

and choice are the topics of this section.

Choice-optionelicited immediate emo-tions are at the base of traditional economic

interpretations of utility as emotional carri-ers of value. Positive emotions increase value

and result in approach, whereas negative val-ues decrease value and result in avoidance (see

Affect as Motivator below). The Iowa gambling



14/36

task (Bechara et al. 1994) popularized the no-

tion of a somatic marker that carries memo-ries of the negative affect associated with losses

in high-risk gambles; these memories prevent

healthy respondents from choosing such gam-bles on subsequent trials. The absence of such

affective information [initially demonstrated infrontal lobe patients and since then in other pa-

tient populations, including substance abusers(Stout et al. 2004)] is associated with perfor-

mance deficits in the form of increased choicesof disadvantageous risky gambles.

Incidental emotions (i.e., emotions unre-lated to the judgment or decision at hand, typ-

ically elicited by a preceding event or activity)have also been shown to influence choice. Al-

ice Isens mood maintenance hypothesis from

1987assumes that people ina good mood wouldlike to maintain this pleasant state and thus

try to avoid hard, analytic work and use cog-nitive shortcuts instead. Consistent with this

hypothesis, Au et al. (2003) found that finan-cial market traders traded differently when in a

goodorbadincidentalmood(elicitedbymusic).Good mood resulted in inferior performance

andoverconfidence, bad mood resulted in moreaccurate decisions and more conservative trad-

ing. Chou et al. (2007) compared mood main-tenance to mood priming to explain patterns

of risk taking in either a positive, negative, or

neutral incidental mood, and found evidencemostly for mood priming (i.e., more risk taking

in a happy mood and less in a sad mood) forboth younger and older adults.

Incidental feelings influence judgmentsor choice also by being misattributed to

having been elicited by the task at hand.Misattribution, an old experimental paradigm

going back to Schwarz and Clore in 1983,is still in active use. Men were shown to

misattribute their arousal after viewing photosof attractive females to arousal generated by

the prospect of having to delay consumption

in a subsequent intertemporal financial-choicetask, and they therefore discounted future out-

comes more strongly (Wilson & Daly 2004).Misattributions of the absence of fluency, the

subjective feeling that forming a preference

for a specific option is easy, as the resu

incidental characteristics (a hard-to-read font) have been shown to affects consu

decisions (Novemsky et al. 2007). We sto have metacognitive awareness that t

misattributions can occur, as evidenced by

fact that we use knowledge of other peoincidental mood states in strategically co

ways (Andrade & Ho 2007).

Affect as common currency. Interpretaof utility as the pleasure or pain associated

the experience of outcomes (experienced ity) go back to Bentham, predating the cur

economic interpretation of utility as infefrom choice (decision utility). Contextua

fects on risky choice have been explainedecision affect theory as modifications o

emotional reactions to obtained outcom

the result of pleasure or displeasure inducerelative comparisons between the obtained

counterfactual alternative outcomes (Meet al. 1999). In this sense, experienced e

tions provide a common currency on whicheffects of both different outcome dimens

and variations in decision context can be grated. Decision affect theory provides a u

ing framework that incorporates special casemotional reactions to counterfactual outc

comparisons such as regret or disappointm(Connolly & Zeelenberg 2002) or loss aver

in its interpretationas affective reaction (Le

et al. 2004). To the extent that the outpumultiple processing channels needs to be c

bined, an affective common currency seembe a promising hypothesis.

Social psychological perspectives on Jalso rely on affect as a common currency. W

people make a risky decision in a mannerfits their self-regulatory orientation (e.g., a

motion or prevention focus, which can be echronic or situationally induced), they feel

about the process. This value from fit has shown to transfer to their evaluation of th

tained outcome (Higgins 2005).

Affect as motivator. Just as preference

constructed, so is affect. Affect construal th

66 Weber Johnson


15/36

(Ellsworth & Scherer 2003) shows that the ef-

fect of affective reactions cannot be satisfac-torily attributed to the emotions valence and

intensity, but rather is influenced by other sit-

uational appraisals. Emotions can be similar invalence and intensity (like fear versus anger) but

result in very different judgments or choicesbecause they are associated with different ac-

tiontendencies. Thus, Lerner & Keltner (2001)show that fear increases risk estimates and risk-

averse choices, whereas anger decreases riskestimates and increases risk-seeking choices.

Similar results were found in a natural exper-iment, conducted after the 9/11 terrorist attack

in the United States (Lerner et al. 2003). In anationally representative sample of Americans,

those who scored higher on an anxiety scale

(fear) had greater perceptions of risk, and thosewho scored higher on a desire-for-vengeance

scale (anger) had lower perceptions of risk upto 10 weeks after the attack. Gender differences

in risk perception, with men perceiving fewerrisks, were largely accounted for by gender dif-

ferences in self-reported emotions. Emotionsalso affected endorsement of different terror-

ism policies.

Dual-Process Explanations

Dual-process models have a long history in the

social sciences. Adam Smith argued that behav-ior was determined by the struggle between

passions and an impartial spectator (Ashrafet al. 2005). More recent psychological models

have distinguished between a rapid, automaticand effortless, associative, intuitive process

(System 1), and a slower, rule-governed, ana-lytic, deliberate and effortful process (System 2)

(Kahneman2003). Ferreira et al. (2006)provideexperimental evidence for this dichotomy by

varying processing goals, cognitive resources,

priming, and formal training of respondents,and show that the automatic and controlled

processes affected by these manipulations makeindependent contributions to judgments and

choices under uncertainty. There is debateabout the extent and way in which the

two systems interact (Evans 2008, Keysers

et al. 2008). Serial interventionist models put

System 2 into a supervisory role because Sys-tem 2 knows the analytic rules that the intuitive

System 1 is prone to violate and thus can inter-

vene to correct erroneous intuitive judgments(Kahneman 2003), but other relationships, in-

cluding parallel-competitive horse-race models(Sloman 1996), need to be considered.

Valuation of risky options. Both cognitive

( Johnson et al. 2007) and affective processes(Lerner et al. 2004) have been shown to in-

fluence peoples evaluative judgments. Hsee &Rottenstreich (2004) contrast valuation by feel-

ing and valuation by calculation. Emotional re-actions are assumed to be far more binary (i.e.,

elicited or not) than analytic assessments of ei-

ther value or likelihood, with the result that,for more emotionally charged choice options,

we observe both greater scope insensitivity anda more highly nonlinear probability-weighting

function.

Risk taking. Behavioral researchers have pro-vided psychological generalizations of the nor-

mative model of finance, which assumes thatthe prices of risky investment options reflect a

tradeoff between risk and return that are moreaffect based. In finance (e.g., the capital asset

pricing model), both risk and return are as-

sumed to be immutable statistical propertiesof the risky option, captured by the variance

and expected value of the outcome distribution.Psychophysical risk-return models assume that

perceptions of risk and return are psychologi-cal constructs that can vary between individu-

als and as a result of past experiences and de-cision content and context. Perceived benefits

are often well predicted by analytic consider-ations such as expected returns based on past

returns (Weber et al. 2005b), but they also varyas a function of interests or expertise (Hanoch

et al. 2006). However, perceived risk is less pre-dicted by analytic considerations (such as ex-

pected volatility as a function of past volatil-

ity) and more by affective reactions related tofamiliarity with the choice option (a domestic

stock with high name recognition) (Weber et al.



16/36

2005b) or decision domain (Weber et al. 2002).

Observed risk taking is the result of a long list ofcognitive and affective evaluation and integra-

tion processes. For example, payoff sensitivity

as well as health and social risk taking as mea-sured by a recent domain-specific risk-taking

scale (Weber et al. 2002) uniquely predict recre-ational drug use by college students (Pleskac

2008). Although some affective reactions andtheir effect on risk taking are objectively justi-

fiable [e.g., the cushioning effect of financiallysupportive networks found in more collectivist

cultures (Weber & Hsee 1998)], others are not(Slovic 1999).

Perceptions of risk and ambiguity also seemto mediate the effect of narrow versus broad

choice bracketing (Read et al. 1999) on risk

taking (Venkatraman et al. 2006). Two studiespresented choice options in a segregated way

(narrow bracketing) or aggregated way (broadbracketing). These studies found that perceived

riskiness [which loaded on affective variables,such as worry and loss, as also found by Weber

et al. (2005b)] and perceived ambiguity (whichloaded on cognitive variables, such as uncer-

tainty, lack of understanding, and informationneeds) were distinct factors that independently

mediated the effect of presentation format onpreference.

Iowa gambling task. TheIowa gambling task,mentioned above, assumes that somatic mark-

ers that carry memories of the negative affectassociated with losses in high-risk gambles pre-

vent normal respondents from choosing suchgambles on subsequent trials. Busemeyer and

Stout (2002), however, show that both cogni-tive and affective evaluation and learning pro-

cesses are needed to account for the choicesmade by normalandabnormalpopulations with

the Iowa gambling task.

Dynamic risk-taking tasks. Much real-world

risk taking (e.g., binge drinking) involves re-peated decisions where risk levels escalate as

the result of previous decisions. Estimates ofrisk taking assessed in static risky-choice tasks

do not predict risk taking in dynamic environ-

ments very well (Wallsten et al. 2005). Se

assessment instruments have attempted tthis gap. The initial tool was devised by S

in 1966 for use with children, who face thpeated choice between continuing in the g

by pulling one of a finite number of swit

that have a high (but decreasing) probabiliearningagain,orstoppingtoclaimtheaccu

lated rewards. One of the switches (the determinates the game, with a loss of all accu

lated rewards. Performance in this game dicts real-world risk taking of children w

crossing a street (Hoffrage et al. 2003).The Columbia Card Task (Figner e

2008) is like the devil task in its nonstatioriskiness, as an increasing number of cards

of 32) are turned over, but in addition, thevaries the number of loss cards that term

the game as well as the gain and loss per

and loss card. In addition, the task allownet losses, not just the elimination of prev

gains. Thus, the Columbia Card Task alfor an assessment of the sensitivity of resp

dents choicesacross conditions (i.e., the quof their information use) as well as their risk

ing. In the Balloon Analogue Risk Task (Leet al. 2002), points are gained with each

that incrementally inflates a balloon, with acreasing probability that the balloon may b

and all acquired gains will be lost. Althouis structurally equivalent to the devil task

Columbia Card Task in that the risk of b

ing increases with previous puffs, the BalAnalogue Risk Task does not explicitly in

decision makers of this nonstationarity,Wallsten et al. (2005) find that participants

construe the task as stationary. Pleskac (2focuses attention on the nonstationarity o

in his Angling Risk Task by specifying esampling with or without replacement (c

and release versus catch and keep) and by ving the clarity of the water and thus knowl

of remaining odds. Respondents are founuse cognitive strategies in contingent and a

tive ways in this domain of dynamic risk

ing, just as reported for choice task 25 yago (Payne et al. 1992) and for inference

more recently.

68 Weber Johnson


17/36

Intertemporal choice. Both cognitive and af-

fective mechanisms have been demonstrated togive rise to the discounting of future events.

The cognitive processes specified by QT, which

also explain theendowment effectandthe statusquobias, account forboth individual differences

in discounting and for the observed asymmetryin discounting when people accelerate or delay

consumption (Weber et al. 2007). An affect- orimpulse-based process for choices that allow for

immediate consumption is assumed to give riseto hyperbolic discounting in Laibsons (1997)

beta-delta model, with some neuroscience evi-dence corroborating the involvement of imme-

diate affect (beta regions) in only such decisions,with other more cognitive (delta) regions being

activated by all intertemporal tradeoff decisions

(McClure et al. 2004) but also some dissentingopinions (Glimcher et al. 2007).

More impatience for choices involving im-mediate consumption is not always found when

controlling for length of delay. Read (2001) al-ternatively explains hyperbolic discounting as a

form of subadditivity of discounting: People areless patient (per time unit) over shorter inter-

vals regardless of when they occur. Zaubermanet al. (2008)find that peoples subjectivepercep-

tions of prospective duration lengths are non-linear and concave in objective time and that

intertemporal choices reflect a relatively con-

stant rate of discounting relative to subjectivetime.

Self-other discrepancies. A dual-process

model also explains differences in the riskydecisions people make for themselves versus

those they predict others will make. Althoughones own emotional reactions to choice

options are very accessible and salient, thoseof others are not. Analytic considerations such

as differences in expected value, on the otherhand, can be assumed to apply equally to

oneself as well as to others. As a result, peoples

choices on the gain (Hsee & Weber 1997)and loss side (Faro & Rottenstreich 2006) are

further away from risk neutrality than are thepredictions they make about the choices of

others. Evidence that this discrepancy (and

HOW MANY PROCESSES?

Dual-process models have enjoyed great success and popularityperhaps in part because we seem to be drawn to dualities, both bi

ologically (with two eyes, ears, arms, andlegs) andphilosophicall(with point and counterpoint). Our review documents how dual

process models have accounted for many judgment and decisionmaking phenomena. A more global perspective suggests, however, that ultimately a single system needs to integrate input from

two or more subsystems to move from deliberation to action. Incontrast, a more local perspective suggests a need for more than

two systems since, in addition to the distinction between a reflective and reflexive system, reflexive processes engage multipl

mechanisms, including automatic emotional reactions, semantipriming, or automated action sequences (Evans 2008, Keyser

et al. 2008). Going into the future, computational modelinof these different subsystems and their reciprocal interconnec

tions will likely build on and possibly supersede dual-procesarguments.

misprediction) is due to a different mix ofaffective and analytic considerations comes

from the fact that the discrepancy is largerwhen predicting the decisions of abstract rather

than concrete others (Hsee & Weber 1997) andis moderated by self-reported empathy (Faro

& Rottenstreich 2006). Regardless of whetherdual-process explanations will be supported by

neuroscience evidence (see sidebar How ManyProcesses?), the distinction between affective

and cognitive processes has been very fruitfulat a conceptual level.

Dual-Representation Models

Knowledge representation is centrally con-

nected to the psychological cognitive processesthat make use of them. Fuzzy trace theory

(Reyna 2004) accounts for apparent inconsis-

tencies in inference and preference tasks by as-suming that different cognitive processes can

take advantage of different memory representa-tions of choice options, i.e., encodings at differ-

ent levels of precision, as a function of age andexpertise (Reyna & Adam 2003). Dehaene et al.

(2004) find evidence for an inbred rudimentary



18/36

number representation system, which presum-

ably complements more sophisticated repre-sentations, in single-cell recordings that iden-

tify number-count cells in the monkey parieto-

frontal cortex.

Decision modes. Multiple-process assump-tions underlie distinctions between qualita-

tively different modes of making decisions.Goals are chronic (personality-, gender-, and

culture-based) and domain-specific, and theyinfluence peoples choice of affective, analytic,

or rule-based processes because these decisionmodes differ in their effectiveness of satisfy-

ing material and nonmaterial goals (e.g., affil-iation versus autonomy; Weber et al. 2005a).

Social norms dictate the use of different deci-

sion principles in different domains (e.g., moralversus business decisions; Tetlock 2002). Peo-

ple seem to have metacognitive awareness thatthe mode in which a decision is made car-

ries diagnostic information about the decisionmakers motivation. Recipients of a requested

favor evaluated the favor and favor granter dif-ferently depending on whether they thought

that the favor granter had decided based onaffect, cost-benefit calculation, or role-based

obligation (Ames et al. 2004).

LEARNINGHomo sapiens needs to survive in stochastic

and often nonstationary environments that re-quire constant learning and updating. Although

learning is often vicarious and transmitted to usin summarized form (similar to the prospectus

of an investment option, providing a distribu-tion of past returns), learning from experience

still plays a powerful role in our judgments anddecisions. Learning, as a topic of JDM research,

may have been the proverbial baby that went

out with the bathwater when the cognitive rev-olution replaced behaviorism. Most choice the-

ories, including PT and DFT, do not includeany learning processes (Pleskac 2008).

Elwin et al. (2007), in a historical summaryof learning from feedback, go back to the argu-

ment made by Einhorn and Hogarth in

that selective and incomplete feedback vents us from accurate judgments and ch

in many decision environments. Addressin

important and understudied topic of peomental representation of feedback, they di

guish between positivist coding that reprewhat one sees and constructivist coding

represents what one believes, supplemenperception with knowledge and theory. T

present evidence consistent with theirconstivist representation that reinforces the vie

attention as an active process.Reinforcement-learning rules of the

originally suggested by Bush and Mostell1955 offer psychological process account

arriving at rational (Bayesian) learning as w

deviations. Reinforcement-learning rules recently been investigated in a variety of J

contexts. Fu & Anderson (2006)show thatforcement learning provides an integrativ

planation for a broad range of dependent msures in tasks from recurrent choice to com

skill acquisition.Erev (1998) revisits signal detection th

and replaces its ideal observer cutoff wcutoff reinforcement-learning process, al

ing him to account for phenomena from servatism to probability matching and the g

blers fallacy. Weber et al. (2004) show

reinforcement learning in risky decisionsare made from repeated personal experi

predicts risk sensitivity to be proportionthe coefficient of variation of the risky opt

rather than its variance, consistent with animal and human data. Following Ma

1996 simulations that demonstrate that rforcement learning in risky choice in

junction with adaptive sampling gives riPTs pattern of risk aversion for gains and

seeking for losses, Denrell (2007) formaadaptive sampling in risky choice, i.e., op

selection that utilizes the evaluations of choptions that are constantly being updated i

ongoing decision-by-experience process.

model predicts that apparent risk takingrisk avoidance can be the result of adap

70 Weber Johnson


19/36

sampling, even when the decision maker has a

risk-neutral value function and learning is op-timal, reinforcing the realization that the re-

lationship between risk attitudes and observed

risk taking is more complex than envisagedby expected utility (Weber & Johnson 2008).

Denrells (2007) model also predicts that in-formation about foregone payoffs will affect

risk taking, consistent with other attempts toincorporate counterfactual outcomes or ficti-

tious play into reinforcement-learning models(Camerer & Ho 1998). Finally, Erev & Barron

(2005) operationalize implicit decision-modeselection as a reinforcement-learning process,

where past success with different modes dictatestheir future use. They show that, in repeated

risky decisions from experience, their model ac-

counts for the observed effect of payoff variabil-ity, the underweighting of rare events, and loss

aversion.Practice ought to make perfect, and re-

searchers have continued to look for evi-dence of optimal performance. Recently, such

performance has been reported for humanmovement-planning tasks, where the tip of a

finger needs to be placed on a computer touchscreen so that gains will be incurred for hitting

indicated target areas and losses are avoided forindicated penalty areas (Trommershauser et al.

2006). People learn to execute such pointing

responses in ways that resemble expected-valuemaximization and are very accurate in select-

ing the higher expected-value option from apair of possible responses. These tasks can be

shown to be conceptually equivalent to choicesbetween money gambles, where people often

fail to achieve expected value or expected util-ity maximization (Erev & Barron 2005). More

research on the precise differences between thisparadigm and gambling choices is needed, but

some differences are apparent. There is cleargoal focus in the pointing task (hitting target

area and avoiding penalty area), the appearance

of a correct answer that can be found ratherthan a preference to be expressed, a continu-

ous space of response alternatives, and a largeamount of feedback.

Expected utility:average utility fromsome risky choice.Like expected valuexcept that outcomare nonlinearly

transformed intoutilities, usually widecreasing marginreturns

Predictive Accuracy

Future states/experiences. Most decisions

are forecasts of how options will make us feelin the future. This idea is captured by the dis-

tinction between decision utility (how we think

options will make us feel) and experience utility

(how experiencing those options actually feels).People tend to underestimate the ease of adapt-ing to lifetime changes such as a move from

California to Ohio, winning the lottery, or be-ing turned down for tenure (Kahneman 2000).

Other systematic mispredictions of subsequentexperiences have recently been reported for re-

gret (Sevdalis & Harvey 2007), loss (Kermeret al. 2006), and time slack and time savings

(Zauberman & Lynch 2005).Two mispredictions of time provide

cognitive-process explanations for intertempo-

ral inconsistencies (in contrast to the affectiveor dual-process explanation discussed above).

Zauberman & Lynch (2005) show that time-money tradeoffs change over time because

people have more (and overly) optimisticpredictions about future time availability than

about money availability. Greater discountingof costs in time than costs in money can lead

to housing/commuting time decisions that donot maximize experienced well being. Trope &

Liberman (2003) show that we often mispredictour preference among choice options that lie in

the future because we construe events that lie

in the future in more abstract and higher-levelterms than events in the near future or present.

Anticipation of negative emotional reactionssuch as regret or negative reactions to loss after

outcome feedback is received helps to motivatecareful analysis of choice options and their pos-

sible outcomes (Connolly & Zeelenberg 2002).It is also adaptive to have mechanisms in place

that minimize these negative feelings, ex-post,as they decrease outcome satisfaction and con-

sume processing capacity. The fact that peo-ple experience fewer negative emotions as they

get older (Mather & Carstensen 2003) suggests

that negative emotion regulation is an acquiredskill.



20/36

Events. Predicting future events is a challeng-

ing task, as documented by Tetlock (2005) ina longitudinal study of expert political predic-

tions. The accuracy of predictions of future

key political events is generally not much bet-ter than chance. However, experts who acquire

information broadly and on multiple topics,and who contingently apply different predic-

tion strategies (foxes, in Isaiah Berlins terms),are more successful in predicting future events

than are experts who specialize in a small fieldand apply a smaller number of strategies more

rigidly (hedgehogs).

CHARACTERISTICS OF THEDECISION MAKER

JDM research in psychology and economicshas been mostly interested in average or typ-

ical behavior. Exceptions to this are risky andintertemporal choice, where individual differ-

ences in behavior have been examined and in-corporated into normative models as parame-

ters that capture the individuals taste for riskandtime delay. Risk attitude in particular (rang-

ing from risk aversion to risk seeking) has some-times been treated as a trait, despite a long liter-

ature showing that risk attitudes as measured by

expected utility lack the cross-situational con-sistency required of traits. Personality theorys

insight that individual traits exist but interactwith situational variables explains existing re-

sults about the domain specificity of risk takingwithout giving up on stable traits (see Weber

& Johnson 2008). Recent statistical advancessuch as hierarchical linear modeling and related

Bayesian methods provide means to measureand explain individual differences in behavior

in these more sophisticated ways.Research over the past decade suggests that

individual and cultural differences in decisionmaking seem to be mediated by two classes

of variables: (a) chronic differences in values

and goals, presumably related to historical, ge-ographic, or biological determinants, that fo-

cus attention on different features of the taskenvironment and its opportunities and con-

straints; and (b) differences in reliance on differ-

ent automatic versus controlled processes

lated to cognitive capacity, education, operience. The review below is organize

predictor variable (what individual differdimension?), describing for each which

pendent measures (what behavior?) this

vidual difference moderates. Dependent msures for which individual differences

been reported include (a) observed judgmor choices, in particular reported percept

of risk, and risky and intertemporal cho(b) model-based parameters inferred from

served behavior, includingrisk aversion andaversion; (c) the accuracy of judgments or i

ences, as measured by their adherence tovalues; and (d) the consistency of judgm

or choices across situations/frames. In somstances, what we list as predictor variable

themselves shown to be predicted by other

dictor variables.

Gender

Women appear to be more risk aversmany contexts and situations (Byrnes e

1999, Jianakopolos & Bernasek 1998). Wthe sources of this observed gender differ

in risk taking are unpacked, women perc

the riskiness of choice options to be lain most domains (all but social risk; see

ber et al. 2002) rather than having a maverse attitude toward risk as they perceive

those (and only those) domains where theyceive the risks to be larger, they appear t

more risk averse. Slovic (1999) summarizeidence that observed gender differences in

taking are not essentialist (i.e., biological)rather the result of deep-seated affective c

fort (or discomfort) with risk (feeling thatcontrollable, or not) that comes with lowe

cial status in a society. Emotional discom

translates into larger perceptions of riskian affective mechanism that connects thes

dividual differences in risk taking to situateffects such as the home bias in investmen

cisions (Weber et al. 2005b) or gain/loss fing in medical informed-consent commu

tions (Schwartz & Hasnain 2002). In con

72 Weber Johnson


21/36

to these reliable gender differences in risk tak-

ing, no consistent gender differences have beenreported on loss aversion or time discounting.

Age

Because psychological processes have develop-mental trajectories, JDM research has shown

weber_mindful judgment and decisionmaking(1).pdf

Documents