The Structure of Consumer Choice Processes
Author(s): James R. Bettman
Source: Journal of Marketing Research, Vol. 8, No. 4 (Nov., 1971), pp. 465-471
Published by: American Marketing Association
Stable URL: http://www.jstor.org/stable/3150238
Accessed: 03-09-2016 00:06 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing Research
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
JAMES R. BETTMAN*
Comparing different information processing models of the same consumers'
decisions and utilizing findings from clinical judgment studies give information
about the structure of consumer choice processes. Cue consistency and conditional
models are discussed and a general paradigm for choice under uncertainty is
proposed.
The Structure of Consumer Choice Processes
BACKGROUND
as the reliability of the judgments. Adding intera
configural terms did not improve prediction. For
Clinical Judgment Research
ample, Hoffman's study on ulcer diagnoses, chosen
cifically
Certain findings in clinical judgments are relevant
for to yield configural judgments, found app
mately
90% of the reliable variance in response c
understanding consumer decision processes. The
experihave
been predicted by a simple linear model [10
mental paradigm underlying this research is as
follows:
summary
judges, provided with sets of cues or variables
for eachresults of other studies, see [8, 10, 19]. J
case, are asked to make quantitative judgments
ment
on many
researchers have made many attempts to ex
the
"linear vs. configural models" problem
hypothetical cases. The experimenter attempts to
fitso-called
the
lustrated
judges' responses by some mathematical function
of theby these findings. This article will exam
these
cues. Typical situations examined are diagnoses
of explanations.
benign or malignant gastric ulcers [10], where the cues
Consumer Decision Process Research
are diagnostic signs such as X-ray findings; diagnoses of
The Newell, Shaw, and Simon postulates for an inpsychotic or neurotic patients, where cues are Minnesota
Multiphasic Personality Inventory (MMPI) profile
scoresprocessing theory of problem solving [15]
formation
wereschool
used in an earlier article to construct models for
[8] and judgments of intelligence based on high
two individual consumers' choices of grocery products
grades, study habits, scores on English effectiveness
tests, and the like [19].
[3]. The models were complex discrimination nets, inAccording to the judges, their subjectively held
ferredmodfrom protocols obtained during several shopping
trips for each consumer. For one of these models, see
els used the cues nonlinearly, and, more importantly,
[3, p. 371].of
configurally; that is, attention is given to patterns
cues so that a cue's effect on judgment dependsAfter
on the
defining and attempting to treat problems of
value of another cue or set of cues. For example,
data codingafor the various cues specified at the nodes
of the models,
judge of intelligence would be using cues configurally
if the models were tested against two sets
of actual
data, the original set from which the models
he stated the rule that study habits influence rating
of a
werehigh,
inferred, and a validation set. Overall, the models
subject's intelligence only if high school grades are
and have no influence otherwise.
correctly matched over 87% of the consumers' accept/
reject decisions.
The surprising finding of many clinical judgment
Simpler
information processing models of these same
studies is that if a simple linear regression model,
using
two consumers'
the cues as predictor variables and the quantitative
decision mechanisms were then con-
Figure 1 shows the simple model corresponding
judgment as the dependent variable, is fit to sidered.
the data,
the level
complex model shown in [3, p. 371]. A comparithe model's accuracy is at approximately the to
same
son of the two figures shows that the simple model is
* James R. Bettman is Assistant Professor of Businessless
Ad-configural, employing a "buy the cheapest" decision
ministration, University of California, Los Angeles. A modified
rule for most cases. The simple and complex models for
version of this study was presented at the Association forthe
Con-second consumer are similar--one highly complex,
sumer Research's Annual Meeting, 1970.
and the other much simpler and less configural.
465
Journal of Marketing Research,
Vol. VIII (November 1971), 465-71
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
466 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1971
Complex model Simple model
Figure 1
A SIMPLE MODEL FOR CONSUMER C1'
Consumer 1 87.2 82.4
Consumer 2 87.5 70.8
INTERPRETING THE RESULTS
Since simpler models predict well,2 what guar
that all processes stated in the protocols are really u
xa
x,
N
a
N
X5 y x6 y X7 y A
X8 Y AA
R
X9y x10 AR R A
N
The findings are like those from the clinical jud
studies, that subjectively complex decision proces
be approximated fairly well by much simpler m
The simple consumer decision models are certain
linear. However, examination of the problem of
vs. configural models in clinical judgment may
solving the similar problem of comparing compl
simple models.
Three hypotheses have been advanced to explai
linear findings in clinical judgment studies [8, p.
1. Human judges behave in fact remarkably like
data processors, but somehow they believe that
are more complex than they really are.
2. Human judges behave in fact in a rather conf
fashion, but the power of the linear regres
model is so great that it serves to obscure the
x31 y A
KEY TO FIGURE 1
The major problem considered here is analy
Dictionary: A: Accept
R: Reject
AR: Associate risk (bad experience) with this
Y: Yes
N: No
configural processes in judgment.
3. Human judges behave in fact in a decidedly li
fashion on most judgmental tasks (their reports
withstanding), but for some kinds of tasks the
more complex judgmental processes.
product
consumer information processing models as it re
these hypotheses. After briefly discussing the tech
used in analyzing the models, inferences about d
process structure will be examined in some deta
X1: Is this meat or produce?
Analysis of the Models
X3: Is color satisfactory?
X4: Is this the biggest "okay" one?
X5: Is this eggs?
The simple and complex consumer information
essing models are represented as graphs or as a
X2: Is price below "justified level"?
X6: Is the price of extra large more than 50 more than the
price of large?
X7: Is this the large size?
X8: Is this the extra large size?
X9: Was this product (brand) bought the last time a purchase
in this class was made?
nodes (points) with arcs (lines) connecting p
nodes. Each node represents a test on a particu
(e.g., is perceived risk high?), and the arcs represen
processing sequence taken, depending on the valu
cues assume. For example, Figure 1 has nodes X
X3 and so forth, and arcs representing yes
X10: Was experience with it okay?
Xl1: Is risk associated with this product (bad experience)?
branches. The initial flow of processing is as foll
X31: Is this cheapest (that they have in stock)?
the product is meat or produce, first check pric
X41: Does this feel okay?
X42: Is this for a specific use?
X43: Is this size okay for that?
X44: Is this produce?
color, and so on. Otherwise check to see if the p
is eggs, and so forth.
The main idea developed in analyzing these mo
that of a conditional decision process graph. Sup
configuration or set of cues were given, with th
These simple choice models, when applied to the data, assuming certain given values (e.g., high risk and c
did surprisingly well. The percentages of correct pre- est price). For that particular set of cues, since t
dictions, compared to those of the complex models, to be taken out of the nodes representing thos
were:1
2 It is possible that the protocol methodology itself encourages
development
1 Certainly the predictive power of simpler models
needs of
tomodels of spurious complexity. Findings in the
be tested over more subjects and more shopping choices for
cognitive process models literature have not supported this contention, however.
each subject.
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
THE STRUCTURE OF CONSUMER CHOICE PROCESSES 467
would then be known, the graph conditional
on that
Figure 2
knowledge will be simpler than
the original
ABSTRACT DECISION
PROCESS WITH FIVEgraph.
CUES
As an example, postulate a decision process with just
two outcomes, 00 and 01. Also, let Xj represent a "yes
response to the test on cue i, and Xj a "no" response
Then Figure 2 is an abstract decision process graph wi
x
five cues. If X2 and X4 hold as a given cue configuration,
the graph of Figure 2 can be collapsed
to the simple
X2 2
Figure 3, because of the certainty of following the ar
corresponding to X2 and X4.
A formal mathematical technique can be developed
x,
for carrying out this type of collapsing for any particul
given cue configuration [2]. The process is trivial for the
simple case shown above, but for complex
decision pro
XS ,
ess graphs the mathematical technique is necessary.
With this idea of a decision process conditioned on
the values of a set of cues, it can be shown [2] th
both simple consumer models can be obtained from th
B *
complex models by conditioning those
complex model
on given cue configurations,awhere
the conditioning sets
0o and 01 represent outp
of cues are very plausible for the persons modeled.3 I
is important to note that a less configural model can
cur rather frequently
viewed as a complex model
conditioned by a given cu
expects to), the simple
configuration: thus complex models can be collapsed a
cisions, leading to a go
outlined above, given known values for cues.
figural
model.
If
cues
ar
Implications of the Model lem-solving
Analysis
processes a
are
in order.
The
confi
It can now be argued that perhaps
persons
often
per
sion
makers
in
their
p
ceive the external world in terms of cue patterns or con-
therefore,
since problem
figurations, rather than in
terms of separate
cues. An
of what
pro
if perceptual structure is sciousness
such that many
products
ar
This configuration
two-step process
characterized by a certain given
of cue
subprocess
app
then a simple conditional decision
model using
only a small
s
experimenters,
al
of cues may be invoked a other
large proportion
of the tim
techniques
have that
notpa
a
That is, the conditional decision
process given
cussed
several
clinical
jud
ticular frequently perceived cue configuration is simp
concluded
that of
"Su
and is used a large part ofand
the time.
The analysis
th
of cues,
consumer models bore out tial
this utilization
argument.
governed
by perhaps
the presen
Carrying the argument a
step further,
it is
cies
between
relevant
these configurations of conditioning cues which
are most
similar
experiments
expected by the decision out
maker.
Consistency
of cues
grade
point
ratings,
wh
may then be defined in terms
of
expectations.
Thos
rating
and
English
effe
cue inputs that are expected and are coded normal
differing
amounts
of a
i
into configurations are seen
as consistent;
those that
work
on impression
unexpected, and hence coded
differently,
are seen asfo
i
paradigm thus
is to
present
consistent. Inconsistent configurations
call
up more
describing
a
hypotheti
complex subprocesses and make the entire decisio
to judge
how
much h
process seem more configural.
Under this
interpretati
adjectives
can
be varie
of cue consistency as fulfilled
perceptual
expectation
consistent
among them
a two-step process is postulated:
cue consistency
is a
his
colleagues
ar
sessed and then the relevant subprocesses in
are this
invoke
ings of
interaction
If cues are consistent, habitual
simple
decision bet
rul
most of Anderson's wo
are invoked. Since such consistent cue combinations oc3 For the consumer represented in the complex and simple
the
adequacy
findings
add
models, these configurations represented product types that werethe consumer behavior models.
of
the
weight
seen as not risky or situations where no brand was preferred
by family members. These perceptions were very typical for Thus, where Goldberg concludes that his second hy-
this consumer.
pothesis is the most likely explanation of the linear-
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
l
to
468 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1971
Figure 3
A SIMPLER MODEL OF FIGURE 2
The paradigm for the general model is outlin
ure 4.
Aspects of the General Model
X X1
Block 2: Incongruity level as a goal. The goal of the
decision maker is to attain his incongruity adaptation
level through choice of behavior. An incongruity adapta-
tion level specific to a decision area (IAL) is considered
here, rather than a generalized level [6, p. 57].
Blocks 3 through 5: The cue manipulation process.
0 x cl
The inferences discussed earlier bear on this portion of
the scheme. In Block 4, it is further assumed that use
of
a dominant
decision mode tends
to reduce incongruconfigural problem [8, p. 491; 9] (and
it definitely
seems
ity. Continued
of a simple decision
rule may lead
to be a large contributing factor),
the use
present
analysis
a level below the
will set
in. This is
infers that perhaps a modifiedto version
of IAL-boredom
the third
hyX3 X3
seen later in Block 7.
pothesis is relevant. Of course, given the clinical nature
Block 6: Process dynamics.
represents learnof the models and data, such inferences
mustBlock
be 6 specuing, which was not considered in the consumer models
lative.
above. However,
speculations
about processes
condiAlso, it seems as though a more
complex
model
can
tional
upon
cue
configurations
lead
one
to
infer
be justified even if a simple model predicts nearly as that a
dynamic process
at work,above
with nodesperbecoming comwell for a given set of data. Indeed,
ifis the
bined into
learned
cue configurations
whichreinvoke subceptual arguments are valid, then
the
problem
with
processes.
placing a complex process model by its more simple
It is hypothesized that the more problem-solving
conditional model would be that one must assume there
will be no changes in the decision maker's perceptions processes are used, the more expected and hence more
of the distribution of coded configurations in the en- consistent are the associated sets of cues. At a level
vironment. Different sets of cues may become the ex- peculiar to the individual, these cues are learned as conpected sets, and the balance of habitual simple processes sistent, and the decision net collapses into a process
and more complex and configural problem-solving proc- conditional on that learned cue configuration. The process is no longer a problem-solving process, but is now a
esses may be altered.
simple standard decision rule. This learning process
would lead to a switch of decision processes from Block
IMPLICATIONS FOR A GENERAL DECISION
5 to Block 4.4 As an example of this process, consider
AND CHOICE MODEL
Figures 2 and 3. Figure 2's abstract decision process
five cues. After
One important use of the analyses for has
researchers
of this decision process has been used
several
times, suppose Cues 2 and 4 were combined into
consumer behavior would be in attempting
to formulate
a learned decision
configuration. Then the process would collapse
a general model of the structure of consumer
to
the
conditional
process shown in Figure 3.
processes. Such a model could provide a framework for
The idea manof a monitor process which controls the
research and also yield hypotheses about marketing
collapsing is
of made
decision process graphs has been impleagement applications. In this section an attempt
mented
with
varying
specificity by other researchers,
to integrate many strands of research on decision models
e.g.,
[4].
However,
growth
of decision nets has not been
seen as cue processing schemes and, in particular, the
considered
in thedegeneral scheme, although some inforpreceding speculations on the structure of
consumer
mation
processing
cision processes. At this stage of development
this
model models have considered decision net
growth
[4,
7]. This is definitely a shortcoming, since
will be very abstract.
obviously do not spring into existence in
One concept used in the general schemedecision
is an nets
integraall 59]:
their complexity. More research is needed on the
tion of cognitive motivation theory [6, p.
development of decision nets (perhaps similar to Sheth's
research on
foreign students [17]).
It is suggested that individuals develop general
adaptation levels concerning the amount of incongruity
Block 7: Exploratory
they
behavior or novelty seeking.
expect in their environments. The GIALThis
[general
inaspect of
decision processes embraces the consumer
congruity adaptation level] would represent
an wanted
opti- to try something new." As incongruity
who "just
mum incongruity level within an organism. It is pro'To optimum,
build an unambiguous model of consumer information
posed that as incongruity falls below the
processing,
one would need to explicitly model how the process
cognitive action will produce more input (e.g., exploraof collapsing based on cue configurations occurs. That process
tion, novelty seeking, etc.). As incongruity
increases
has not been formally modeled in this article, and further rebeyond the optimum, cognitive action will
attempt
toneeded to specify what conditions lead to
search
is certainly
reduce or simplify input (as in dissonance
reduction).
learning
of consistent cue configurations.
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
THE STRUCTURE OF CONSUMER CHOICE PROCESSES 469
Figure 4
A GENERAL DECISION AND CHOICE MODEL
Consider First
Decision Area
Encode
Environmental
Cues
<O
Perceived > Cue Coding yes
Incongruity: Cnitn
no
Simple Processes
_-Cue
Cue n Configurations
Problem Solving Configurations i+1 i+2 i .n
Confiaurations "I "' g .?-
Processes..,T
(Morea.,) e I Process Process Process Process
Configura") Process Process Process " i+1 i+2 ? -1?"( n
1
2
Cue Con- Cue Con- Cue Con-
sistency sistency sistency
Incongruity
Inconqruity
Activate Search
For New Cues
?
Search no Search
Successful oMore
Locally ? Nonlocally
yes
Cue Collapse Process To
tIncongruity 'Expected'
Consistency
At yes Conditional Process;
Level For Set Cue Coding To
A Consistent
Process?
Next
n
Consider
Are All
no
NDecsion no Decision Areas
Area Processed?
yes
feedback Decision
Experience
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
470 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1971
decreases from Block 4, eventually
it may
become
so
models to develop
predictions
of product
demand
on more behavioral
models
thanthis
those based
on time
low (below the IAL) that boredom
sets in.
At
point,
Block 2 would lead to Block 7,
and
a search
for
newmodels.
and However,
series
analysis
or stochastic
consumer
the would
mechanismsbegin.
which leadThis
to collapsing
complex modmore incongruous alternatives
search
would be local at first-that is,
for
consumer
decisions,
els into
simple
models are not
well understood. This
collapsing process
itself needs
be specified
for example, it might be for a different
brand
of to
the
same in much
detailnot
beforesuccessful,
such confident usemore
of simple models
type. However, if this search more
were
would be
justified. Oneor
cannot
draw the conclusion that
nonlocal search, such as for a new,
exciting,
unfamilconsumers' subjectively reported complexity of their
iar product might ensue.5
choice processes is spurious.
Relation to Other General Models
Also, changes in the decision maker's perceptions
The model, although developed entirely independ- about configurations may alter the balance of simple
ently, is in some aspects similar to Howard and Sheth'sand complex processes. If such perceptual changes are
theory of buyer behavior [11]. Their notions of thelikely, as in consumer choice processes, then the complex
"psychology of simplification" and the "psychology of model retains the flexibility to predict over more situacomplication" are mirrored in Blocks 3 to 5, Block 6, tions than a simple model tailored to one set of percepand Block 7. In addition, Howard and Sheth use Ber- tions about configurations.
At the present stage of abstraction of the general
lyne's theory of cognitive motivation, which is very
similar to Driver and Streufert's position (based heavilymodel for decision and choice, it would be gratuitous to
on Hunt's earlier integration [12], which in turn is based list several specific implications. Some specific implicaon Berlyne [6, pp. 42-6]). Finally, the decision proc-tions of the individual consumer choice models themesses imbedded in Blocks 3 to 5 are very similar toselves are given in [3]. However, the very abstractness
Howard and Sheth's plans with inhibitors [11, pp. 132-of the general model implies that it can only suggest
43]. However, the paradigm above was derived circui- broad approaches at this stage. For example, in chartously from cognitive process models, a very differentacterizing consumers more willing to try new products,
approach from that of Howard and Sheth. Finally, the Blocks 2 and 7 imply that search for new products may
paradigm also has close parallels with the framework be much more likely for people whose optimal level of
outlined by Cyert and March for organizational decisionincongruity is fairly high. By measuring this optimal
level for a sample of consumers using tests of stimulus
processes [5, p. 126].
These substantial parallels between paradigms imply seeking [16] and attempting to relate this level to other
that decision under uncertainty can be modeled in broadconsumer characteristics, one may be able to develop
strokes in a similar manner over many complex decisioncorrelates of high search activity that would allow differscenarios. This gives more evidence to support these ential promotion of new products to such consumers.
general schemes as representations of the structure of
REFERENCES
decision processes in general, and of the structure of
consumer choice processes in particular.
1. Anderson, Norman and Ann Jacobson. "Effect of Sti
Inconsistency and Discounting Instructions in Person
Impression Formation," Journal of Personality and S
Psychology, 2 (October 1965), 531-9.
Findings from both clinical judgment studies
and in2. Bettman,
James. "Applying a New Methodological Approach
to a Problem of Information Processing Model Valiformation processing models of consumer behavior
show
dation,"
Paper No. 152, Western Management
that complex decision process models can often
be Working
apScience Institute, University of California, Los Angeles,
CONCLUSIONS
proximated quite well by simple models. Using
the
1969.
notion of a conditional decision process for 3.
a .given
set Processing Models of Consumer Be"Information
of cues, a process of cue consistency assessment
and dehavior,"
Journal of Marketing Research, 7 (August 1970),
370-6.
cision subprocess use was postulated. Finally, an
attempt
4. Clarkson,
was made to develop an overall paradigm outlining
a Geoffrey. "Decision Making in Small Groups-A
Simulation Study," Behavioral Science, 13 (July 1968), 288structure for consumer choice processes.
305.
From these findings one might be tempted
to draw
5. Cyert,
Richard M. and James March. A Behavioral Theory
ofmodels
the Firm. Englewood Cliffs, N.J.: Prentice-Hall, 1963.
the conclusion that the approximation of complex
6. Driver,a Michael
by simple models is a general finding, and hence
col- and Siegfried Streufert. "The 'General InLevel' (GIAL) Hypothesis-An Analylection of rather simple decision process typescongruity
could Adaptation
be
sis and Integration of Cognitive Approaches to Motivation,"
used to model consumers' decisions. If this were
true,
Institute
Paper No. 114, Krannert School of Industrial Adthen market researchers could use such a set of
simplePurdue University, 1964.
ministration,
5 Although animal exploratory behavior has been well researched, much more work is needed on styles of incongruity
seeking as imbedded in human decision processes. For a stimulus variation-seeking scale, see [16].
7. Feigenbaum, Edward. "The Simulation of Verbal Learning
Behavior," in Edward Feigenbaum and Julian Feldman,
eds., Computers and Thought. New York: McGraw-Hill,
1963, 297-309.
8. Goldberg, Lewis. "Simple Models or Simple Processes?
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms
THE STRUCTURE OF CONSUMER CHOICE PROCESSES 471
Journal of PersonalityAmerican
and Social Psychology,
9 (May
Some Research on Clinical Judgments,"
Psychol1968), 1-6.
ogist, 23 (July 1968), 483-96.
15. Newell, Alan, J. Shaw, and Herbert Simon. "Elements of a
9. Green, Bert. "Descriptions and Explanations: A Comment
Theory of Human Problem Solving," Psychological Review,
on Papers by Hoffman and Edwards," in [13], 91-8.
65 (May 1958), 151-66.
10. Hoffman, Paul. "Cue Consistency and Configurality in Human Judgment," in [13], 53-90.
16. Penney, Ronald and Robert Reinehr. "Development of a
11. Howard, John and Jagdish Sheth. The Theory of Buyer BeStimulus-Variation Seeking Scale for Adults," Psychological
havior. New York: John Wiley & Sons, 1969.
Reports, 18 (April 1966), 631-8.
17. Sheth, Jagdish. "How Adults Learn Brand Preference,"
12. Hunt, Joseph McVicker. "Motivation Inherent in InformaJournal of Advertising Research, 8 (September 1968), 25-36.
tion Processing and Action," in O. J. Harvey, ed., Motivation and Social Interaction-Cognitive Determinants. New
18. Sidowski, Joseph and Norman Anderson. "Judgments of
York: Ronald Press, 1963, 35-94.
City-Occupation Combinations," Psychonomic Science, 7
(March 15, 1967), 279-80.
13. Kleinmuntz, Benjamin, ed. Formal Representation of Hu19. Slovic, Paul. "Cue-Consistency and Cue-Utilization in Judgman Judgment. New York: John Wiley & Sons, 1968.
ment," American Journal of Psychology, 79 (September
14. Lampel, Anita and Norman Anderson. "Combining Visual
1966), 427-34.
and Verbal Information in an Impression Formation Task,"
This content downloaded from 152.3.34.72 on Sat, 03 Sep 2016 00:06:38 UTC
All use subject to http://about.jstor.org/terms