, ~,
Marketing Letters 13:3, 195-205, 2002
~. @ 2002 Kluwer AcademicPublishers.Manufacturedin The Netherlands.
Context Dependence and Aggregation in Disaggregate
Choice Analysis
JOFFRESWAIT
Advanis Inc. and University ofFlorida
WIKTOR ADAMOWICZ
UniversityofAlberta
MICHAEL HANEMANN
University ofColifornia at Berkeley
ADELE DIEDERICH
Universitat Oldenburg
JON KROSNICK
Ohio State University
DAVID LAYTON
University ofWashington
WILLIAM PROVENCHER
University ofWISconsin.Madison
DAVID SCHKADE
University of Texasat Austin
ROGER TOURANGEAU
University ofMaryland
AcceptedFebruary 7, 2002
Abstract
There is an emergingconsensusamongdisciplinesdealing with humandecisionmaking that the context in which
a decision is madeis an important determinantof outcomes.This consensushasbeen slow in the making because
much of what is known about context effectshas evolved from a desireto demonstratethe untenability of certain
common assumptionsupon which tractable models of behavior have generally been built. This paper seeks
to bring disparatedisciplinary perspectivesto bear on the relation between context and choice, to formulate
(1) recommendationsfor improvementsto the state-of-the-practiceof RandomUtility Models (RUMs) of choice
behavior, and (2) a future researchagendato guide the further incorporation of context into these models of
choice behavior.
Key words: choice, context, random utility
196
SWArT ET AL
1. Introduction
Overthe pastthirty years,the contextdependence
of choicebehaviorhasbecomethe focus
of a large literature in psychology,marketing and economics.The literaturestartedwith
Lichtenstein and Slovic's (1971) discoveryof preferencereversals-in comparing two
gambles,subjects' preferencesvaried dependingon whetherthe task involved choice or
matching.Also influentialwas Tversky and Kahneman's(1981) demonstrationof framing
effects.In their "Asian disease"experiment,subjects,including physicians,reversedtheir
preference for treatment strategieswhen the outcomeswere framed in terms of the
probability of surviving or the probability of dying.
Since those initial demonstrations,the phenomenaof preference reversal, framing
effects, and loss aversionhave been documentedin a wide variety of settings, both
experimentaland real-world, and their scopehas beenextendedconsiderably(see Roth
1995and Camerer1995 for surveys).Many forms of context-dependent
preferenceshave
beenidentified (spaceconsiderationsprecludereferences):(1) habit or experiencedependenceeffects,(2) social interdependence,
(3) accountabilityeffects,(4) menu-dependence,
(5) chooser-dependence,
(6) mental accounting, (7) choice bracketing, (8) motivation
effects, (9) decoyeffects, (10)referenceprices,and (11) complexityeffects.This list is by
no meansexhaustive!
Whenpreferencereversalsand framing effectswere first demonstrated,
the discussionin
the literatureoften focusedon the implied violation of individual rationality (Tversky and
Kahneman 1981, 1986). However,as repeatedefforts to discreditthese effectsfailed, it
becameapparentthat they were not mere artifacts but robustfeaturesof actual behavior
that need to be taken seriously.Even in economicsthere is now a growing attempt to
respondto this weight of evidenceby lesseningthe dependenceon idealizedmodels of
hyper-rationalityand turning instead to more behavioralmodels of individual decision
making (e.g. McFadden2001).
Lack of knowledge about context's linkages to choice behavior may leave policy
analysts blind to potential/unexpectedeffects on choice set formation, constraints,
evaluationrules and decision rules. Hence, one should seek to explicitly incorporate
appropriateaspectsof context in model developmentby explicit manipulationof context
(payne,Bettrnanand Schkade1999),whether in a samplingsense(e.g. stratificationover
respondentsin different contexts) or in an experimentalsense(e.g. placing randomly
selectedrespondentsin different informationenvironmentsin an SPchoicetask). A priori,
it would seemsensiblenot only that contextmustbe incorporatedin theseways,but also
that some attemptmustbe made by the analystto matchthe preferenceelicitation context
with the prediction context (payne,Bettrnanand Schkade1999).The end result of such
matching should be more robust models of choice behavior,better able to predict to a
varietyof contextsas well aspermitting "averaging" overcontextsto removeconditioning
by context(more on this subsequently).
Someintrospectionon the issueof contexteffectsand contextmanipulationgivesrise to
a Lucas-like "context critique" (Lucas 1976). That is to say,predictive models which
ignore contextmay give rise to biasedpredictionsbecausepolicy actionsmay also impact
context,which in turn may impact any or all of the sub-processes
of choice (to be defined
CONTEXT DEPENDENCE AND AGGREGATION IN DISAGGREGATE CHOICE ANALYSIS
197
below). Hence,the impetus for context manipulationand context inclusion in models of
choice behavior is also to permit context itself to be policy sensitive.Ultimately, then,
developmentof predictive models should aim not just at the incorporationof context
effects,but the very modeling of contextitself.
In choice behavior there exist distinguishablestages,including the initial cognitive
representationof the decisionproblem, information acquisitionand interpretation,information combinationleading to an evaluation,and expressionor mappingof the evaluation
onto a response,which opens up the possibility of contextimpacting processesat each
stage(payne,Bettmanand Schkade1999).Theseclosely parallel the stagesidentified by
Tourangeau(1984), who used this model to account for many of the context effects
documentedin survey researchby showing how the context of a survey item can affect
eachstageof the responseprocess.The insight that contexteffectscanbe mapped,andthis
knowledgecan be used constructivelyto obtainan improvedunderstandingof individual
behavior,hashad a powerfulinfluenceon the field of surveyresearchwhere it providedthe
conceptualunderpinning for an interdisciplinary movement on Cognitive Aspects of
Survey Methodology (CASM).
In choice modeling, we believe a shift in emphasisto mapping the psychological
processesof an individual in comprehendingand respondingto the decisiontask at hand
could bring some significantgains in terms of a richer insight into individual preferences,
improved accuracyin predicting individual behavioralresponsesto economicchangesor
policy interventions,and a more realistic assessment
of the impact of policy interventions
on individualwelfare.To realizethesegains,oneneedsa conceptualframeworkto identify
the sourcesof contextdependencein individual decisionprocesses,mathematicalmodels
to representthe context-dependence
of choice behavior,empiricalproceduresto measure
context-dependence
in the field and econometrictechniquesto estimatethesemodels from
data.This is the focus of our paper-we offer suggestionsfor building context explicitly
into models of individual choice behavior,and propose an initial researchagenda for
bringing this ambition to fruition.
2. A Model of Choice in Context
It is useful to begin by relating our approachto a framework commonin the literatureon
psychologicalmeasurementand testing: observedresponse(B) is representedas a linear
function of someunobserved"true" response(say,preferenceP) and measurement
error B
as in B = P + B. To allow context to affect choice behavior,it is natural to extend the
modelby writing B = P + 0 + '1,where 0 is context.This decomposes
the responseerror
B into a context-dependentcomponent,0, and a componentthat it is independentof
context,'1thus: B = 0 + '1. Commonassumptionsarethat (i) Bhasa zero mean,and Band
0 are independentand uncorrelated,or (ii) '1 has a zero mean, and '1, 0 and P are
independentand uncorrelated.The implicationsof (i) and (ii) are that contextis "noise,"
and its impact on choice behavior washes out in the aggregate.Given the empirical
evidenceof the abundanceand persistenceof context-dependence
effects,we suggestan
alternativestochasticspecificationin which '1 = P .0 + u .0 + u; thus,
B
198
SWAIT ET AL.
=P+O+p.O+v.O+v,
(1)
where (iii) D has zero mean, and D,P and .Q are independentand uncorrelated.The
implications of (1) and (iii) are that context can have a systematicimpact on choice
behavior,it can interactwith preferences,and the effectsof contextneed not wash out in
the aggregate.In principle, contextcanbe disentangledfrom preferenceon the basisof(l).
In practice, however,(1) is likely to be too simple a formulation, so we now developa
more articulatedmodel of how contextmay interactwith preferenceto determinechoice
behavior.
2.1. A Formal Model of Choice
To keepthings simple,we will focus on a purely discretechoiceas in McFadden(1974), as
opposedto mixed discrete-continuous
choicesof the sortconsideredby Hanemann(1984).
In a stylized model of decisionmaking, let D representthe decisionstrategy,C represent
the choice set,j index alternatives,n indexindividualsand t index time. Let V be an index
of strengthof preference(or utility) and e an error processthat is assumedto arise from
elementsof the processunobservedby the researcher.The stylized decisionstructurecan
be framed as producing the chosenalternativei:t accordingto
i:t +- Dn{~nt' BjnJ.
jeC.,
(2)
We now examinethe many ways in which contextcan affect this decisionstructure.
Choice Set Formation. Contextualfactors affect the set of alternativesthat are being
actively considered,or those that can possiblybe considered.The processof choice set
fonnation can be consideredone of initially identifying all known alternatives,then
reducing the setto the final choice setused in decisionmaking. Part of this processwill
involve infonnation on structuralrestrictionson the choice set,while anotherpart will be
basedon contextualfactors(e.g. public vs. private consumption).At the extreme,people
may consideronly one or two options that happento be salientbecauseof context.
Constraints. Constraintstraditionally consideredto be involved in the decisionprocess
include income,time and other(e.g. cognitive capabilities)resources.Contextmay affect
the perceptionof availableresources:if conceptsof relative income are more important
than absolute income,then income constraintsare clearly affected by social group and
related contextual factors (Frank and Sunstein2000). The time horizon, or temporal
bracketing,that individuals employdiffers by contextand will also affectthe definition of
income and time constraints(Read et al. 1999).
Evaluation Rules. A common economic evaluation rule involves deternlining an
aggregateindex that allows compensatoryevaluationof an alternative.In (2), V implies
2.2.
CONTEXT DEPENDENCE AND AGGREGAllON IN DISAGGREGATE CHOICE ANALYSIS
199
a utility function approachto evaluation.However,compensatoryapproachescan require
significant cognitive effort to employ (Johnsonand Payne 1985). Evaluation rules in
contextsof perceived inconsequentialdecisions(e.g. arguably,purchasingketchup) may
be very simplified: "purchase what was purchasedin the past." In such cases,utility
arising from attributes,while embeddedin previous decisions,has little impact on the
currentdecisions.Context,in terms of choice history,also plays a role here:if thereis no
history,thenthe processcannotbe basedon previouschoicesandrequiressomeevaluation
of attributes.Thus, initial conditionsand learningare important considerationsfor certain
segmentsof a market, but not necessarilyfor others.
Decision Rules. Optimizationover all attributesand alternativesusing a compensatory
decision rule is a common assumptionin economic analysis. Researchin economics,
psychology,marketing and other disciplines shows that a wide variety of decisionrules
appearto be used by individuals, often dependenton context.Increasingchoice complexity, for example,appearsto result in increasedprocessingby dimensionalreduction, or
even avoiding making choices (payne 1976; Swait and Adamowicz 2OO1b).Context
variation may result in changesfrom near-utility optimizationto elimination-by-aspects,
to
simplified dimensionalreductionor majority-of-confirming-decisionheuristics,amongan
infinity of other possibilities.
Contexteffects,then,canbe categorizedasthose that influencechoice sets,constraints,
evaluationrules and decision strategies,any subsetof which may be simultaneously
affectedby a specific contextfactor.As before,let n representchoice context,and rewrite
(2) thus,
i:t +- D(Q)n{V(Q)jnt' e(Q)jnt}.
jEC(n).,
(3)
This expansionof the decisionframeworkillustratesthe consensusof the workshopthat
contexteffectsmustbe understoodby examiningacrosscasesthathavevariation in Q and
where effectson the elementsof choice (D, V;C and e)canbe identified.Identificationwill
require formal model structuresthat provide for isolation of elementsD, V; C and e and
sufficient variation in Q to identify effectsof context.
Below we review prior work in two topic areas,choice set formation [C(Q)] and
decisioncontextcomplexity[D(Q)l to demonstratethat incorporationof contextin choice
modelsis an areawhere researchhas alreadybegun,but where muchremainsto be done.
Examples of ContextModeling
The choice set formationliteraturebuilds uponthe theory that choicefollows a two-stage
process: (1) selectionof a subsetof goods C from the universal set of alternativesM,
followed by (2) choice of the mostpreferredgood in C. In this literature (see,e.g. Swait
and Ben-Akiva 1987and Ben-Akiva and Boccara 1995), structuralmodels of the twostageprocesshave been shownsuccessfulin capturing non-compensatorybehavior,but
3.
~
200
SWAIT ET AL.
suffer from the combinatorialcomplexityarising from the latentnatureof the choice set
formationprocess.More recently,reducedform representations
of the two-stageprocessin
a single-stagemodel have beendone: e.g., classesin a latent class model that partially
capturedifferent choice setstructures(Swait 1994),or the incorporationof personaland
contextual constraints affecting choice set formation in a pseudo-Lagrangianutility
function that summarizesboth stagesof choice (Swait2001).
The psychology and consumerbehaviorliteratureshave identified complexity of the
decisioncontextas a major factoraffectingthe useof decisionheuristics,and thus,a major
factor in establishingdecisionaccuracyand effectiveness.Swaitand Adamowicz (200la)
extend the usual view in these literatures that complexity arises from number of
alternatives,number of attributesand interattributecorrelation,and proposean extension
to the MNL model in which the varianceof the error term is an endogenousfunction of
entropy,which they showto be a useful uni-dimensionalsummaryof contextualcomplexity. Swait and Adamowicz (200lb) also use entropy as a complexity measureto show,
through an ordered latent class choice model, that decision difficulty and cumulative
cognitive load lead respondentsto switch from a full-information, compensatoryevaluation strategyat the beginning of a SP task, to a simplerstrategyignoring someattributes
towardsthe end of the task(seeLouviere et al. 2002 for additionaldiscussionof modelsof
variance,task design and complexity).
Context,Choiceand Their Measurement
3.1. Exploratory Techniques
for Measurementof Context
Psychologistshaveuseda variety of tactics in trying to uncoverthe conceptualstructures
and cognitive processesthat underlie many different behaviors,including choice behaviors. In this sectionwe describesome of thesemethodsand show how they might be
applied to the discovery of the choice sets, attributes, decision rules, history and
expectationsthat affect choicebehavior.We emphasizethat thesetechniquesare exploratory in characterand would typically be used to generatehypothesesrather than to test
them. Nonetheless,we believe that manyof themcan be fruitfully appliedto the problem
of characterizingand measuringchoice context.
Identifying Choice Sets. Cognitive psychologistshaveused severaltechniquesto get a
betterunderstandingof how everydaycategoriesarestructured;manyof thesetechniques
seem especially relevant to discovering how people construe choice sets. One simple
method---cardsorting-asks respondentsto sortthe possiblemembersof a choice set into
whatevergroupsmake senseto them; in addition,the respondentsare oftenaskedto label
the groups they form. The aim of this procedureis to discoverthe classesthat people
spontaneouslyuse when given little guidance. The resulting data can be examined
informally or analyzed rigorously via clustering algorithms. Related techniques ask
respondentsto rate the similarity of pairs of membersof the choice set (ketchup and
~
201
3.2.
CONTEXT DEPENDENCE AND AGGREGATION IN DISAGGREGATE CHOICE ANALYSIS
mustard,barbecuesauceand steaksauce);scalingprocedures(e.g. INDSCAL) are useful
to determinethe dimensionsunderlying theseratings.
Decision Rules and Attribute Use. Another exploratorytool is the retrospectiveprobe:
respondentsare asked what they were thinking about as they made some judgment or
solved someproblem. For example,respondentsmight be askedwhat attributescame to
mind as they made a decision. Suchprobescan be very broad ("Tell me everything you
were thinking about as you made that decision") or more focused ("Did you consider
price?"). Once a set of attributeshas beenidentified in this way,the choicesthemselves
can be examinedto seewhetherthey reflect consistentpreferencesoverthe attributes.A
related strategyis to askpeople how they arrived at their decisions.Again, eitherbroad,
open-endedprobes ("How did you arrive at your choice?") or more focusedones("Did
you focus mostly on one thing as you consideredthe alternatives?")can be used to
discoverdecisionrules governinga setof choices.
PastHistory. A crucialattributeof the decision-makeris his or her pasthistoryof similar
choices,his or her personalcontext. Specialprobesmight be cmftedto assesswhetherthe
decisionmakerseesthe currentchoiceas relatively novelandunfamiliar or as familiar and
routine ("Does this productremind you of similar products you bought in the past?"). If
the choice is a recurring one, it may be possibleto ask the personwhether they have a
regular policy or rule that governstheir choices("Do you alwaysbuy the samebrand?Do
you buy the cheapestbrand? Do you delibemtely vary your choices for the sake of
variety?").
Information Gathering and Processing. Thesesametactics can be applied to discover
what informationpeople seekas theymakechoicesand how theyuse thatinformation. For
example, for important choices (car purchases)decision-makersmay seek information
from severalsources(ConsumerReports,the Internet),and they mayprocessand organize
it in different ways (writing it down, discussingit with friends,putting it into a spreadsheet).Straightforwardprobing can identify the sourcesof information they tappedand
how they processedit and thus identifies actions associatedwith the decision-making
context.
While theseideasare intendedsimply as guidesand suggestionsto researchers,
and are
themselves subject to potential problems (Nisbett and Wilson, 1977), the important
messageto take awayis that it is feasibleto measurethe likely shapeof contexteffects
on the componentsof choice. At that point, the choice model analystfacesthe task of
deciding which contexteffectsmustbe included in the model frameworkbeing developed.
Choice Model Development,Testingand Prediction
Dynamics and Decision Sequencing. An essentialfeatureof any model of choice is
whetherthe choice is dynamic, in the sensethat currentchoicesaffect future payoffs or
opportunities,and the decision-makeris awareof this relationship. Many decisionsare
202
SWAIT ET AL.
clearlydynamic-the decisionto retire, for instance,or decisionsaboutmedicaltreatment.
We are also led to distinguish between dynamic and sequentialdecisions: whereas
dynamic decision making is more concemed with controlling the process over time,
sequential decision making describes a situation where the decision-makermakes
successiveobservationsof a processbeforea final decisionis made(e.g. price searching).
Life is replete with decisionsthat may be dynamic but are nonethelessmodeledby social
scientistsas static.We suspecttwo reasonsfor this discrepancy:(1) estimationof structural
dynamic decisionsremainsa difficult task-the pioneeringwork of John Rustand others
(see Rust 1994)providessometractablestructuraldynamicmodelsof choice,but evenin
these cases estimation remains a formidable undertaking; (2) economists have not
seriouslyconsideredthe possibility that contextualvariablesmay causeapparentlystatic
decisionsto be dynamic. For instance,in their discussionof choicebracketing,Read et al.
(1999)discuss "mental" budgeting and cite experimentsindicating that householdsare
often quite strict about keeping money within distinct budget categories,which has
implications for whetherbehavioris dynamic or static.
Casting behavior as static when it is dynamic will generate inaccurate forecasts
and biasedwelfare estimates.Nonetheless,the difficulty of estimating suchmodels suggeststhe need for a modest initial researchagendawith the following three elements:
(1) theoreticalanalysis that sheds light on how context induces or suppressesdynamic
behavior; (2) the developmentof extended decision-level panel data conducive to
exploring the existence and role of dynamics in the decision process; and (3) the
developmentof simple diagnostics for detecting dynamic behavior. This last task is
potentially difficult becausebehavioralmodelscharacterizedby sequentialstatic optimization often may fit observedbehavioras well or betterthan their "true" dynamic counterparts. Perhapsdirect querying of individuals in surveys about dynamic aspectsof their
choiceswill initially prove the bestavailableand most sensiblediagnostic.
Econometric Estimation and Model Comparisons. Consideragainthe choice model in
equation(3): decisionrules,choice sets,the indirect utility function,andthe errorstructure
are all (potentially) a function of context. Some overarchingversion of a choice model
would include all of theseeffectssimultaneously.While this shouldbe the eventualapplied
objective,we believe that substantialprogresscanbe madeby first consideringsub-setsof
the problem. Considerimplementationof equation (4), which presentsa simpler model
where context may affect perceptions of attributes, prefer/:nce weights, and error
components.
Ujnn = fn(~n. Pnn)+ 8n.
(4)
Note that eventhis simpler model makes a distinction betweencontext affecting error
components(see Louviere et al. 2002),and contextaffecting structuralaspectsof choice
(preferenceweights and attribute perceptions).The sophisticatedeconometricianmight
attemptto model the entire impactof contextthroughthe error termvia a rich setof error
componentsthat could be a function of unobservedpreferenceheterogeneityand other
unobservedfactors, both evolving temporally. This would result in a Random Utility
CONTEXT DEPENDENCE AND AGGREGATION IN DISAGGREGATE CHOICE ANALYSIS
modelwith a multivariateerrorstructure(e.g.Keane1997).It is importantto determinethe
extentto which this error componentsstrategywill succeedin capturingcontexteffects.In
a manneranalogousto howignoring statedependencewill likely overstatethe importance
of unobservedpreferenceheterogeneity(and vice versa)in discretepanel data,we expect
that ignoring structuralcontext effectswill overstatethe importanceof error components,
and conversely,ignoring context dependencyin the error componentswill overstatethe
importanceof structuralcontexteffects.Still, formal approachesneedto be developedand
applied studies examining these issues must be conducted.We describe one possible
approachnext.
To be specific, imagine a model with observablecontext effects and an lID Type I
ExtremeValue error structure,which we will term a ContextMultinomial Logit (CMNL)
model. To testmodels,conceptuallyit is a simple matterto add a multivariatenormal error
structure to a CMNL model. With this adjustmentwe would arrive at a Context MNP
(CMNP) model nesting both the CMNL model (at least approximately)and the standard
MNP model, thus allowing one to test for the presenceof both contextand unobserved
heterogeneity.We suspectthat both contextand unobservedheterogeneitywill be present
in most applications.Modeling only the unobservedheterogeneitywhile ignoring context
raises deep questionsregarding the interpretationof MNP covariancematrices,not to
mentionmeans.
Moving from econometricsto prediction,one can predict conditionalon given contexts,
or averageover the relevant contextsto form unconditionalpredictions. When context
matters,averagingshould result in bettermodel fits and betterpredictions.Averagingover
many contextswill also allow the researcher
and consumersof researchto betterassessthe
robustnessand relevanceof the model.
4. Recommendations
for Future Research
Two guiding principlesanimatedour workshopdiscussions:(I) we had a clearobjectiveof
producing a researchagendafor improving empirical modeling of disaggregatechoice
behaviorbasedon richer models of consumerbehavior,more sophisticatedeconometrics
and better data collectionmethodsto yield more information about consumermotivation
and perceptions;and (2) we worked within the flexible framework of random utility
models to help us through the (potentially)unendingmazeof contexteffects on choice.
During the course of the paper, we have mentioned a number of researchand
implementationinitiatives that we believe are important. We list below the ones we
considermost important.
Data Collection. The developmentof improved models of choice that incorporate
context effects only makes sensewhen allied to supporting data collection methods.A
rich literature from the CASM movementin surveyresearch(see,e.g., Tourangeauet al.
2000) is availableas a starting point to guide researchon data collection and on their
practical implementation.Specifically,supportingthe developmentof contextmodeling
will require collection of data on (I) attitudes/perceptions,(2) dynamics (history,
204
SWAIT ET AL.
expectations), (3) mental models, (4) task and context complexity, and (5) context
manipulationchecksand debriefingprotocols.The goal hereshouldbe to producereliable
and easy-to-usecontextmeasurementtools, as well asto characterizepotentialbiasesin
tools.
Context Modeling. Our goal here is particularly to urge the incorporationof what is
already known about context effectsinto choicemodeling practice,as well as calling for
the continueddevelopmentof knowledgeaboutcontexteffectsand their incorporationinto
choice models. Some relevant topics for advancing choice models based on current
knowledgeare (I) referencedependenceof preferences,(2) choice set formation, (3) taste
heterogeneity,error componentsand heteroskedasticity,(4) choice dynamicsand sequential decision making, and (5) prediction with context-sensitivemodels. Prior work has
shown that severalextensionsto the RU frameworkare likely to prove fruitful in these
endeavors:choice setformation(e.g. SwaitandBen-Akiva 1987),latentclass(e.g. Dayton
and Macready 1988)and error-components
(see Louviere et al. 2002)models areamong
potential starting points for future researchin this area. In addition, explicit manipulation
of contextsin controlled experimentsand econometricanalysisusing RU methodsshould
be employedto assesssystematicdifferencesin context.
References
Ben-Akiva, Moshe, and Bruno Boccara. (1995). "Discrete Choice Models with Latent Choice Sets," International Journal ofResearchin Marketing, 12(1),9-24.
Camerer,Colin. (1995). "Individual decisionmaking," in The Handbook ofExperimental Economics,Editors,
John H. Kagel and Alvin E. Roth, Princeton University Press,New York.
Dayton,C. Mitchell, and GeorgeMacready.(1988). "Concomitant-VariableLatent-ClassModels," Journal ofthe
American Statistical Association,83 (March), 173-178.
Frank, R. H., and C. R. Sunstein. (2000). "Cost Benefit Analysis and Relative Position," The University of
Chicago John M. Olin Law and Economics WorlcingPaperNo. 102.
Hanemann,w: Michael. (1984). "Discrete/ContinuousModels of ConsumerDemand,"Econometrica,52,541-62.
Johnson,E. J., and J. w: Payne.(1985). "Effort and Accuracy in Choice," ManagementScience,30,395-414.
Keane, Michael. (1997). "Modeling Heterogeneity and State Dependencein Consumer Choice Behavior,"
Journal ofBusinessand Economics Statistics,15(3),310-317.
Lichtenstein, S., and P. Siovic. (1971). "Reversal of Preferencesbetween Bids and Choices in Gambling
Decisions," Journal ofExperimental Psychology,89, 46-55.
Louviere, Jordan et al. (2002). "Dissecting the Random Componentof Utility," Marketing Letters,this issue.
Lucas,R. E. Jr. (1976). "Econometric Policy Evaluation: A Critique," In K. Meltzer and A. H. Meltzer (eds.),
Supplementto the Journal ofMonetary Economics,The Phillips Curve and Labor Markets.
McFadden,Daniel. (1974). "Conditional Logit Analysis of QualitativeChoiceBehavior." In PaulZarembka(ed.),
Frontiers in Econometrics, Academic Press,New York.
McFadden,Daniel. (2001). "Economic Choices (Nobel Lecture)," American Economic Review,91(3), 351-378.
Nisbett,R. E., andT. D. Wilson. (1977). "Telling More than we CanKnow: VerbalReportson MentalProcesses,"
Psychological Review,84(3), 231-259.
Payne,J. W. (1976). "Task Complexity and Contingent Processingin Decision Making: An Information Search
and Protocol Analysis," Drg. Behavior and Human Performance,22, 17-44.
Payne,J. W.,J. R. Beltman, andD. A. Schkade.(1999). "Measuring ConstructedPreferences:Towarda Building
Code," Journal of Risk and Uncertainty,19, 243-270.
CONTEXT DEPENDENCE AND AGGREGATION IN DISAGGREGATE CHOICE ANALYSIS
205
Read,D., G. Loewenstein,andM. Rabin. (1999). "Choice Bracketing," Journol ofRiskand Uncertainty,19(1-3),
171-198.
Roth, Alvin E. (1995). "Introduction to ExperimentalEconomics," in TheHandbook ofExperimental Economics,
EditolS, John H. Kagel and Alvin E. Roth, PrincetonUniversity Press,New York.
Rust, John. (1994). "StructumI Estimation of Markov Decision Processes."In R. F. Engle and D. L. McFadden
(eds.), Handbook of Econometrics,Vol 4, New York, Elsevier-North Holland, 4, 3081-3143.
Swait, Joffre. (1994). "A Structural Equation Model of Latent Segmentationand Product Choice for CrossSectional, RevealedPreferenceChoice Data," Journal ofRetailing and ConsumerServices,1(2),77-89.
Swait, Joffre. (2001). "A Non-CompensatoryChoice Model Incorporating Attribute Cutoffs," Transportation
ResearchPart B, 35, 903-928.
Swait, Joffre, and Wiktor Adamowicz. (2001a). "Choice Environment, Market Complexity, and Consumer
Behavior: A Theoretical and Empirical Approach for Incorporating Decision Complexity into Models of
ConsumerChoice," Organizational Behavior and Human Decision Processes,86(2), 141-167.
Swait,Joffre, and Wiktor Adamowicz. (200lb). "Choice Complexity and Decision StrategySelection," Journal
of ConsumerResearch,28 (June), 135-148.
Swait,Joffre, and Moshe Ben-Akiva. (1987). "Incorporating RandomConstraintsin Discrete Choice Models of
Choice Set Generation," TransportationResearchB, 21B, 91-102.
Tourangeau,Roger. (1984). "Cognitive Scienceand Survey Methods." In T. Jabine,M. Straf, J. Tanur, and R.
Tourangeau(eds.), Cognitive Aspects ofSurveyDesign: Building a Bridge BetweenDisciplines, Washington,
DC: National Academy Press,73-100.
Tourangeau,Roger,Lance Rips, and KennethRasinski.(2000). ThePsychology ofSurvey Response,Cambridge:
Cambridge UnivelSity Press.
Tversky, Amos, and Daniel Kabneman. (1981). "The Framing of Decision and the Psychology of Choice,"
Science,211, 453-458.
Tversky, Amos, and Daniel Kahneman.(1986). "Rational Choice and the Framing of Decisions," Journal of
Business,59, 251-S278.
© Copyright 2026 Paperzz