Bagozzi Theory Construction Marketing PDF

A Prospectus for Theory Construction in Marketing
Author(s): Richard P. Bagozzi

Source: Journal of Marketing, Vol. 48, No. 1 (Winter, 1984), pp. 11-29
Published by: American Marketing Association
Stable URL: http://www.jstor.org/stable/1251307 .
Accessed: 29/04/2014 07:16
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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 support@jstor.org.
American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing.
http://www.jstor.org
This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

All use subject to JSTOR Terms and Conditions
RichardP. Bagozzi
This articleaddresses the question, "How should A Prospectus

theories be constructed?"In doing so, two ap-
proaches are considered: the classic positivist
paradigmand an emerging realistperspective.An
for Theory
attemptis made to develop criteriafor represent-
ing theories so that they can be more rigorously
developed, tested, and evaluated.
Construction
in
Marketing
Introduction the intermediary role of how theoriesshouldbe formed.

is perhaps most often scrutinizedin In general, we might construe theory construction
RESEARCH
termsof the inputs that produceit and the outputs in two ways: as a process and as a structure.The pro-
that characterizeit. Typically, inputs run the gamut cess of theoryconstructioninvolves the applicationof
from the ideas and energies of researchers to the principlesof logic, the implementationof methodsand
procedures(e.g., the experimentalmethod), and the
methods, prior findings, and new data of disciplines. observance of standardsof conduct and evaluation.
Outputs are additions to scientific and practical To these more formal processes we must add such in-
knowledge. These consist of new ways of thinking formalprocesses as creativityand decision making by
about and approachingproblems of interest.
The present article addresses one way that the in- individualresearchers;conflict, debate, and give-and-
take between researchers;and social and political pro-
puts and outputsare linked. The premise is that theory cesses among groups and institutions(Bagozzi 1976).
construction serves an intervening role between the
researcheron the one hand and knowledge production The process of theory constructionis a historicalone,
on the other hand. That is to say, the ways that the- too, in that new theories are forged from older ones
and are judged in relation to how well they subsume
ories are representedand tested are thought to entail
social constructionsconnecting individualresearchers older theories, accountfor currentanomalies,and lead
to the body of knowledge thatconstitutes a discipline. to futurediscoveries.
It is maintainedthat the natureand quality of knowl- At the same time, theory constructioncan be char-
acterized by its structure.By structurewe mean the
edge are formally dependent on theory construction.
If we are to effectively influencethe generationof new concepts in a theory, the hypotheses made by the the-
knowledge, we must purposely give considerationto ory, the observations and measurementsincluded in
the theory, and the formal organization of all these
elements in an overall representation.Any structure
will, therefore, have semantic and syntactic dimen-
RichardP.Bagozzi is Associate Graduate
Professor, Schoolof Business, sions.
StanfordUniversity.Theauthorwishesto thankProfessors
ClaesFor-
nellandJohnLittleandthe JMreviewers
In the present essay, discussion will focus on the
forcommentsmadeon an
earlyversionof thisarticle. structuralaspects of theory construction.We will be-
gin with an outline of one of the most highly devel-
Journal of Marketing
Vol. 48 (Winter 1984), 11-29. A Prospectus
forTheory
Construction
in Marketing
/ 11

oped and widely used frameworksin the philosophy
of science: the so-called Received View, also known FIGURE1
as the standardconception and more broadly as log- Elements of the CanonicalFormof the
ical empiricism(Suppe 1977). Our goal will be to de- Received View of ScientificTheories
scribe and critique the Received View with an aim
toward showing how it can serve as a startingpoint Language L ---- Logical Calculas K
for a new perspectivemore operationallyorientedthan
earlier, abstractmodels. The new approachis termed
the Holistic Construal, but it should be stressed that K----TheoreticalLanguage
LT Observation
Language
L-- K0
it is not a philosophy of science. Rather, it is a meth-
Theoretical Terms VT Observation TermsV0
odological programmedesigned to bridge the gap be-
tween the new philosophies of science (e.g., Brown
[Correspondence Rules]
1979, Suppe 1977) and the concepts and procedures C
needed to implementand evaluate the conduct of re-
search. Its philosophicalroots lie primarilyin the real-
ist theory of science (e.g., Bagozzi 1980, especially
intendedto tackle but one piece of the researchpuz-
pp. 19-20, 25-27, 39-40; Manicasand Secord 1983),
zle, albeit in a modest way.
yet it is broad enough to accommodate other view-
points, as developed hereafter.Overall, our goal will
be to lay the conceptualgroundworkneeded for sound Foundations: The Received View
theory constructionin marketing. To this end, ideas
from the philosophy of science are integrated with The CanonicalFormulation'
emerging views in marketing (e.g., Bagozzi 1980, In abbreviatedform, we may characterizethe logical
1983; Bagozzi and Fornell 1982; Bagozzi and Phillips empiricistmodel of the structureof theory as follows
1982; Hunt 1976; Zaltman, LeMasters, and Heffring (see Figure 1). Scientific theories are formulated in
1982; Zaltman, Pinson, and Angelmar 1973). By the terms of a language L and a logical calculus K. The
same token, we will suggest some new ways of rep- former consists of a theoretical language LT and an
resentingand testing theories. observationlanguage Lo. The latter is comprised of
Before consideration of the Received View and formalrules of reasoningand inferenceand exists also
Holistic Construal,we should mention a caveat. The in separate theoretical (KT) and observational (Ko)
process and structureof theory constructioninteract modes. Within LT are theoretical (i.e., nonobserva-
with each other and jointly determine the content of tional) terms VT; and within Lo are empirical (i.e.,
both scientific and practicalknowledge. Thus, in one observational)terms Vo. The VTare used to represent
sense it is artificial to separatethem. We do so here abstractconcepts in a theory, whereas Vo are used to
largely for expositional purposes and because treat- representempiricalconcepts. The L of a theory is ex-
ment of the process of theory constructionis beyond pressed in sentence form, with LT capturingthe cen-
the scope of this article. In addition, well-developed tral nonobservationalpropositions in the theory and
theories of the process are generally lacking. Toul- Lo reflecting empirical generalizations.The conjunc-
min's (1972) instrumentalistapproach,Kuhn's (1970) tion of LT, VT, and KT may be termed the theoretical
notion of scientific revolutions,Hanson's (1971) ideas postulatesT of the theory;whereas the conjunctionof
on the logic of discovery, Feyerabend's(1965, 1970) Lo, Vo, and Ko are called the implied empirical ob-
radicalempiricism, Lakatos' (1970) researchprogram servationsO of the theory. The empirical meaning of
methodology, and Shapere's (1969, 1974) work on the theory is said to occur through the specification
scientific domains and rationalitygo a long way to- of the relationshipsbetween VTand Vo. This is ac-
ward articulating the processural aspects of theory complished through correspondence rules C which
construction, but are not sufficiently formulated to be formally link the two. We designate the entire struc-
integrated with the structural dimensions developed tureof any theory-i.e., the conjunctionof T, C, and
herein. Indeed, each of the processural views contains O-by "TCO."
serious shortcomings, leading one commentator to
conclude that "today no single analysis . . . enjoys
general acceptance among most philosophers of sci- 'For a slightly different, but more detailed, description of the Re-
ence" (Suppe 1977, p. 119). If we can push our ceived View, see Suppe (1977, pp. 6-61). Except for a few additions
noted hereafter, the notation follows Suppe. Our interpretationof the
knowledge of a small portion of theory construction received view, however, departs from Suppe's in that we develop the
(i.e., its structure) one step further, then much will network of implied empirical observations and integrate these with
be gained along the way toward better formulation of theoretical postulates and correspondence rules. Suppe's treatment fo-
cuses relatively more on the logical and nonempirical aspects of the-
theories. Our explication of the structure of theory is ories.
12 / Journal
of Marketing,
Winter
1984

An Illustration whereasattitudemight be conceptualizedas particular
To make the canonical formulationmore concrete, let affective responses toward the political issue in the
us illustrateit with a simple example drawnfrom con- communication. All together, VT,, VT2, ,t KT,,
sumer research. Consider the effect of source credi- KT2, and the primitivetermsdefining VT,and VT2con-
bility in a persuasive communicationon attitudesto- stituteT, i.e., the theoreticalpostulate(s)of the effect
ward a political issue. Classic work in the commun- of source credibility on attitude. In a full exposition
ication literaturesuggests that greater source credi- of T, we might have primitiveterms, definitions, ax-
bility leads to more positive attitudes. Figure 2 pre- ioms, theorems, and nonobservationalpropositions.
sents one way to pictoriallyrepresentthe structureof The VT are given empirical meaning through C.
the theory. The most common form for correspondencerules ad-
The central hypothesis of the theory is expressed vocated by logical empiricists is
in sentence form and depicted as et, a nonobserva- Ex - [P(x) R(x)]
tional proposition. Generally, nonobservationalprop-
ositions are intendedto be statementsabout actual or which in words reads, "If x is subjected to experi-
potential lawlike generalizations. Two theoretical mental test procedure E, it will exhibit theoretical
terms-VT, and VT2-are containedin et and standfor propertyP if and only if it yields resultR" (e.g., Car-
source credibilityand attitude,respectively. Although nap 1956). We will discuss the implications of this
not shown in Figure 2, VT, and VT2would each typ- form of C shortly. For now, notice thatP corresponds
ically be defined with primitive terms, perhaps or- to VT, and E and R correspondto Vo. In Figure 2,
ganized in a particular way according to KT, and we see that VT,has, in fact, two observationalterms
KT2,respectively.Forexample, sourcecredibilitymight attached to it, Vo, and Vo02.The first might be the
be defined in terms of expertise and trustworthiness, experimentalmanipulation(s)used to create the im-
ages of different levels of credibility. For instance, a
common procedureis to employ different spokesper-
sons for experimentaland controlgroupssuch thateach
FIGURE 2 person possesses varying levels of knowledge, skill,
Representationof the Effects of Source experience, etc. with the issue at hand. The second
Credibilityon Attitude under the observational term might be a manipulation check
CanonicalFormulation consisting of measures of perceived expertise and
trustworthiness.Similarly, VT2 is shown connected to
V0 0V t Vo3, where the latter might be physiological, self-re-
T
port, or other measures of attitude. Next, notice that
the Vo are shown interconnectedby e, i.e., empirical
propositions.These are statementslinking the respec-
C C2 C3 c tive observationalterms. For Vo, and V02, the prop-
osition (i.e., fe,) is expressed as an empirical asso-
ciation, to reflect the fact that the concepts should
VV, 2 covary as a consequence of their common content in
VO2, meaning. The remaining propositions (i.e., fe2 and
I0
fe3) are representedas empirical generalizations, to
indicate that covariation here occurs as an observa-
3 l tional consequence of underlyinglaws linking differ-
Plane of Observation ent concepts. Finally, notice that each Vo is tied to
the plane of observationwith dashed lines. This is in-
Key tended to stand for the actual physical sensations or
VT,, source credibility sense extendingdata throughwhich the empiricalcon-
attitude
VT2 cepts implied by Vo are monitored. This might be a
Vol experimental manipulation of source credibility visual movement noted by an experimenter,a check
VO2 manipulation check
markplaced on a questionnaire,a deflection on a strip
Vo3 operationalization of attitude
chart recorder,or a whole host of other types of ob-
et theoretical (i.e., nonobservational) proposition: the
greater the source credibility, the more favorable servationalrecordingprocedures.
the attitude
c1, c2, c3 correspondence rules Status of the Canonical Formulation
e,e proposition of empirical association in Marketing
eo2, Eo3 proposition of empirical generalization
The canonicalformulationrepresentsan ideal which,
if implemented,would go far towardmaking our the-
A Prospectus
forTheoryConstruction
in Marketing
/ 13

ories more explicit and subject to detailed evaluation. (1971) discusses importantparadoxesof confirmation
In practice, however, the canonical formulation sel- and recent attemptsto solve them.
dom has been operationalizedin marketing.A number A second issue is empirical. In additionto the de-
of exemplarsdo exist that approachthe ideal in terms gree of correspondencebetween theoretical and ob-
of thoroughness of theoretical development and/or servationalterms, our ability to assess and interpret
methodologicalrigor. A partiallist of these might in- theoriesrests on the fidelity of our measurements.The
clude Bass (1974), Howard and Sheth (1969), Little practicein marketingand the social sciences has been
(1975), Phillips (1982), and Silk and Urban (1978). to focus upon e as a representationof ft (see Figure
However, in general, most research in marketing is 2). Unfortunately,to the extent thatmeasurementsare
built upon theories that are expressed in incomplete unreliableand/or that systematicmethodologicalcon-
forms. Overall, the practice is to address only a por- founds or other biases exist, e will deviate from ft.
tion of the elements comprising the canonical for- Notice that randomand systematic errorsin our mea-
mulation. Notice that we should be concerned with surementscan occur even if the latter are valid op-
two issues. One is scope, which refers to the range erationalizationsof theoreticalconcepts, and we have
of the elements of the canonical formulationthat are carefully specified the correspondence rules. It has
specified in any particulartheory. The second is qual- frequently been noted in marketing that the assess-
ity, which signifies the adequacyto which each of the ment of the reliability of measurements within any
criteriaincluded in the scope is attained. particularstudy is crucial to its evaluation, yet is sel-
Considerationof the quality of theory construction dom performed(Peter 1979, Ray 1979). To this com-
in marketingis beyond the scope of this essay, and mentary,we must add thatconsiderationof systematic
no firm generalizationscan be made. With respect to erroris also a rareevent.2 Both obscure our ability to
scope, however, at least five potential shortcomings understandphenomena.
in contemporarymarketing practice may be identi- The specification of the rationale underlying a
fied. nonobservationalpropositionin a theory is a thirdim-
The first issue concerns the relationships C be- portantissue. No theory can be considered complete
tween theoretical and empirical concepts. Typically, without an explication of why or how the terms in a
these either are not considered at all or are taken for propositionare linked. A rationaleis usually based on
granted. Seldom are correspondencerules specified, cause and effect or functionalarguments.In Figure 2,
and even less seldom are observational implications the rationalefor ft was not stated explicitly. Rather,
of the correspondence rules investigated. Without a psychological argument was implied, such as the
considerationof the relationshipsbetween theoretical expected effect of perceived expertise on the believe-
and empiricalconcepts, it is not possible to assess the ability of message content. Depending on the level of
meaning of one's terms in a theory. Closely related specificity of the theory, the rationaleas to why high
to this is the problem of construct validity (Bagozzi source credibility should lead to favorable attitudes
1980, Ch. 5; Peter 1981). Construct validity is the would either be a premise of the theory or else for-
extent to which an operationalizationmeasures the mally integratedinto the theoreticalpostulates (e.g.,
theoreticalconcept which it is intendedto measure. It through axioms, theorems, or intermediateproposi-
thus depends on the definitionaland theoreticalmean- tions).
ing of the concept, the behavior of its measurements, This aspectof theoryconstructionsuggestsone way
and the semantic and syntactic relationshipsbetween how new knowledge is generated. At any one point
the two. Our ability to detect true empirical general- in time, any theory may provide an explanationfor a
izations and interpretthem as evidence of underlying phenomenonbased on the content and structureof its
nonobservational propositions is dependent funda- TCO. But as we learn more about the specific TCO
mentally on construct validity and the relationships and gain new knowledge from other TCOs and areas
between theoreticaland empiricalconcepts. Failureto relevant to, but separatefrom, the TCO in question,
specify correspondencerules and employ valid mea- a basis may develop for deepening the rationale or
surementsof concepts in our theories not only creates revising the TCO in some other way. Often, the im-
a lacuna between theory constructionand hypothesis petus for change will spring from failures of the TCO
testing but prevents one from addressing the degree in predictionor the emergence of disconfirmingevi-
of confirmabilityor falsifiability of theories. From an
epistemologicalperspective, testability is an essential
propertythatmust be formallyintegratedwith our the-
2Bagozzi and Phillips (1982), Bagozzi and Silk (1983), and Phillips
ories if they are to have meaning (Popper 1959). This (1981) consider limited aspects of systematic error: namely, that re-
shouldbe done withoutinvoking a neutralobservation flected in correlated measurement residuals and in methods factors
without explicit measurements. But as we develop hereafter under the
language. That is, data do not exist independent of Holistic Construal, more formal methods and models are needed to
the theories we use to account for them. Swinburne interpret and correct for systematic error in many cases.
14 / Journal
of Marketing,
Winter
1984

FIGURE 3
ExpandedCanonicalRepresentationof the Effects of Source Credibilityon Attitudes
(presented in abbreviatedform for simplicity)
Initial Position
Stimulus of Audience Cognitive Responses Ultimate Response
Favorable
Counter
Arguments
Attitude
Source Neither or
nor
Behavior
Credibility Unfavorable
,Favoroble Change
Support
Arguments
Unfavorable
C
0
SC CAi
FU A
SC2 SAj
U
dence. In any case, any particularTCO will generally vationalpropositionbetween source credibilityand at-
serve a transitionrole over time. It begins as an un- titude, many are proposed. Moreover, the rationaleis
provenanswerto a problemor researchquestion.Then deepened. The impact of source credibilityis hypoth-
it achieves a certain degree of success and is recog- esized to be contingenton whetherthe audienceis ini-
nized as a valid explanationof some phenomenon.But tially favorable, unfavorable,or neitherfavorablenor
sooner or later it will become inadequateto the tasks unfavorabletoward the issue at hand. Depending on
of explanation,prediction,and control, and must either the audience'sinitialreceptivity,sourcecredibilitywill
be discarded,revised, or subsumedundera more gen- differentiallyaffect cognitive responses. Cognitive re-
eral TCO. The transitionmight be gradualor radical, sponses, in turn, will be integratedwith evaluativeand
such as is reflected in Kuhn's (1970) idea of paradigm decision-makingprocessesand influenceattitudesand/
shifts and scientific revolutions. or behavior. One set of predictionsmight be:
To take an example, consider how a cognitive re-
sponse rationalefor the effect of source credibilityon 1. Highly credible sources will be more persua-
attitudemight look underthe canonical formulationof sive than less credible sources when the re-
the Received View. Figure 3 presents an outline of a ceiver is initially unfavorable toward a mes-
cognitive responseschema which supplantsthe classic sage.
interpretationof Figure 2. Instead of one nonobser- 2. Highly credible sources will be less persuasive
A Prospectus
forTheory in Marketing
Construction / 15

than less credible sources when the receiver is formedundercontrolledconditions, will be subjectto
initiallyfavorable toward a message. alternativeinterpretations(Popper1959, p. 50). Some-
3. Highly crediblesourcesand less crediblesources times this will stem from methodological inadequa-
will not differ in their persuasive effects when cies that either allow natural(but unknown) external
the receiver is initially neither favorable nor forces to contaminaterelationshipsof interest or else
unfavorabletoward a message.3 introduce artificial confounds (e.g., demand charac-
The rationalefor these theoreticalpredictionsmight teristics, response sets). At other times theories may
be provided as follows. Consider first the prediction be ambiguous, incomplete, or less powerful than al-
that highly credible sources will be more persuasive ternatives. Generally, it is not possible to eliminate
thanless credible sources when the receiver is initially all rival hypotheses because our methodologies are
unfavorabletowardthe message. Under these circum- imperfect and our theories are embryonic. But at a
minimum, we can search and test for alternativeex-
stances, high credibility is hypothesized to inhibit
whereas low or moderatecred- planationsfor the phenomena we study. Platt (1964)
counterargumentation, was an early person to advocatethe systematicpursuit
is
ibility hypothesized to facilitate it. No change in of rival explanationsand suggested the following pro-
supportargumentsis expected, since the receiver is cedure:
unfavorableat the start. As a consequence, those re-
ceivers exposed to highly credible sources will have 1. Devise alternativehypotheses.
fewer negativeevaluationsto integrateand shouldhave 2. Formulatecrucial experiments, each with mul-
more favorable attitudes than those exposed to less tiple possibleoutcomesthatexclude one or more
credible sources. In contrast, highly credible sources of the hypotheses.
are expected to be less persuasive than less credible 3. Execute the crucial experiments.
sources when the receiver is initially favorabletoward 4. Recycle the sequenceas needed to considerstill
a message. Here it is maintainedthat low credibility other rival hypotheses for the original hypoth-
leads to the generationof more supportargumentsthan esis and/or rival hypotheses.
high credibility. This is thoughtto occur because low It is importantto stress that tests of rival hypotheses
credibility threatensone's initial position, and to re- should not be reservedfor separatestudies but should
lieve the tension, one generates positive support for
the message. However, no difference in counterar- be performedwheneverpossible within the context of
an on-going study. In this way, because subjects, set-
gumentationbetween those exposed to high versus low
credible sources is expected because the receivers are tings, instruments,etc., are held constant,we will have
initially favorable. Hence, those receivers exposed to greaterconfidence in the internalvalidity of the rival
high versus low credible sources will have fewer pos- hypotheses. The additionaltest of rival hypotheses in
itive evaluationsto integrateand should have less fa- other contexts is also advocated herein and provides
vorable attitudes. Finally, for receivers neither favor- evidence for externalvalidity. For examples of the use
able nor unfavorable toward a message, source of tests of rival hypotheses as supportfor inferences,
see Bagozzi (1981a) and Phillips (1982).
credibilityis hypothesized to have no discernableim- A final shortcomingof theory constructionin mar-
pact, as no difference in the balance of counter ar-
guments and supportargumentsshould arise. keting can be seen throughreference to Figure 3. In
Thus, returningto the issue of knowledge produc- practice, within the purview of any specific study,
marketershave tended to emphasize either the theo-
tion, we can see that there are two ways in which new
retical postulates T or the implied empirical obser-
knowledgeis generated.One occurs as outlinedabove. vations 0.
A given theory is refined through expansion of the
numberof nonobservationalpropositionsbetween in- Examples of the former can be found in some
dependentand dependentvariables and/or deepening managementscience models in marketing.Here elab-
of the rationalebehind the propositions. Or secondly, orate networks of nonobservationalpropositions are
an older theory is replaced with a newer one. In both specified, and while these are generally internally
cases, new knowledge is not produced until the re- consistent and well-formed, tests of the theory are
fined or new theoryis found to providea more general typically performed either on limited subsets of the
networkor else only on operationalizationsof single
explanationand/or more accuratepredictions.
The fourth shortcoming in theory construction in theoreticalvariables in the network. For instance, a
common practiceis to use a logit regressionto predict
marketingis the failure to systematically consider ri-
val hypotheses. Any test of a theory, even if per- choice, where the latteris a dependentvariable at the
end of a long chain of premises and nonobservational
propositions.The problem here is that the correspon-
'These propositions are drawn from Sternthal, Dholakia, and Leavitt dence between T and O is often a loose one, and the
(1978). concatenationof assumptions and hypotheses is not
16 / Journal
of Marketing,
Winter
1984

really tested. It is difficult, if not impossible, to as- Therefore,it is again recommendedthat emphasis be
certainwhethersuccessful or unsuccessful predictions placed on a TCO. Although this will not guarantee
of the theory are meaningful. Successful predictions thatall biases and confoundshave been accountedfor
at best suggest that the subset of the network tested or that a theoretical hypothesis is validly operation-
might be valid. But it also makes those assumptions alized, it does allow one to estimate and correct for
and propositionsnot directly tested ambiguous, since random and systematic errors and test for observa-
many other frameworkscould interface and be con- tional implicationsof correspondencerules and non-
sistent with the tested subset. Unsuccessful predic- observational propositions. The Holistic Construal
tionsalso produceambiguity,in thatone does not know presentedhereafterattemptsto accomplishthese goals,
whether(a) the subnetworkis flawed, (b) a portionof while circumventingcertainproblemsin the Received
the remaining network is invalid, (c) the entire re- View. Let us consider these problems now before
maining network is inadequate, or (d) both the sub- turningto the Holistic Construal.
network and remaining network are at fault. To de-
velop more valid theories, one must either reduce the Some Fundamental Problems
size of the theoreticalnetworkto bettermatch the ob- Perhapsbecause the question of what is a scientific
servationalframework,or else expand the numberof theory is so complex and difficult to answer, consid-
observationsto betterconformwith the theory.In either erable ambiguityexists throughoutthe many facets of
case, there is a need to better tie theoreticalconcepts the Received View. We will consider only a few of
to observationsin a formal TCO. Chainsof theoretical the more obvious problemsherein.4
concepts with more than two uninterpretedconcepts Theoretical and Empirical Terms. Many propo-
in a row are to be avoided. Indeed, most theoretical nents of the Received View have tended to draw a
concepts should be directly tied to one or more ob- sharp line between theoretical and empirical terms.
servations each. Focus on a full TCO would shift Theoreticaltermsrefer to nonobservablepropertiesor
scrutinyfrom sterile tautologies to testable theories. objects, whereas empirical terms refer to observable
In contrastto managementscience models in mar- propertiesor objects. However, as othershave pointed
keting which overemphasizeT at the expense of C and out (e.g., Achinstein 1968), none of the proponents
O, consumer behavior research tends to overempha- has specifiedsoundcriteriafor the distinction,andthese
size O at the expense of T and C. That is, ratherthan may well not exist. Carnap(1966, pp. 225-226) takes
developing an overall TCO framework, theories are the position that theoretical and empirical terms are
typicallytested at the level of O only. The dangerhere on end points of a continuumand that differences are
is that researchcould gravitatetoward a type of raw a matterof degree. He does not indicatehow one might
empiricism rather than achieving a balance between differentiatebetween theoretical and empirical terms
sound theory constructionand empirical verification. in any one theory, however. It is necessary to have a
Takingthe source credibilitymodel of Figure 3 as clear meaning for unobservableand observableterms
an example, the practice has been to provide a loose in order to determineconstructvalidity and differen-
theoretical discussion of T but to investigate hy- tiate between lawlike generalizationson the one hand
potheses strictly at the level of O. Because corre- and accidental or spurious associations on the other
spondence rules are not explicitly considered, mea- hand. Although no satisfactorysolution seems as yet
surementerroris implicitly assumed to be negligible, to exist in the philosophy of science, an attemptwill
and focus is placed strictly upon association among be made later in this article to refine the meaning of
measurements(i.e., on fe), it is not possible to in- terms in any theory.
terpretfindingsunambiguously.Positive findings(e.g., CorrespondenceRules. Correspondencerules serve
significantF tests) do not rule out the possibility of to provide empiricalsignificance to theoreticalterms,
interpretational confounding,i.e., O could correspond and in so doing, help us understandand validate the
with many alternativeTs, and it would not be clear meaningof our theories. At least three interpretations
what positive findings signify. Similarly, positive of correspondencerules have been given under the
findings could occur because of systematic method- Received View. One is that a correspondencerule is
ological confounds. Negative findings, in contrast, an analytic propositionbetween V, and Vo. Thus, it
mightindicaterejectionof the (ambiguous)theoretical is believed to be true or false and to depend on the
hypotheses,excessive randommeasurementerror,lack logical form of the sentence in which it is expressed
of constructvalidity, methodologicalbiases, or com-
binations of these. In any case, unless (a) measure-
ment errorin independentand dependentvariables is
truly low, (b) the specification and test of O closely
correspondsto T, and (c) methodological confounds 4For another perspective, see Suppe (1977, pp. 62-118).
are absent, the research will be uninterpretable. 5Some authors use the idea of an "empirical term" in a broader sense
to also encompass unobservables.
A Prospectus
forTheory
Construction
in Marketing
/ 17

and on the meaning of the terms in it. A second inter- "An empiricallaw helps to explain a fact that has
pretationof a correspondencerule is that it is a con- been observedandto predicta fact not yet observed.
vention. It functions much in the same way as a def- In similarfashion, the theoreticallaw helps to ex-
plain empiricallaws alreadyformulated,and to per-
inition, wherein the meaning of a theoretical term is mit the derivationof new empiricallaws" (p. 229).
specified to be entailed in a given empirical concept
througha priori agreementamong "authorities."The Nevertheless, Carnap seems to place primary impor-
problem with both the analytical and conventional tance on empirical laws:
interpretationsof correspondence rules is that they
"Theconfirmationof a theoreticallaw is indirectbe-
precludea role for empirical testing and historicalde- cause it takesplace only throughthe confirmationof
velopment of meaning for theoretical terms. Indeed, empiricallaws derivedfrom the theory.
the analyticinterpretation implies that observationsand
are The supremevalue of a new theory is its power to
experimentalprocedures logical implications of a predictnew empiricallaws" (p. 231).
prioripremises, ratherthan contingentassertionsabout
the world of experience. Similarly, the conventional Because of the need to search for fundamental
interpretationplaces empirical meaning in the realm generalizationsand causal relations and the inherent
of expert opinion. Thus, either perspective tends to contaminationpresentin observations(due to random
exclude empiricalmeaningfrom theoreticalterms, and and systematic error), we see a need for a shift in
this, in turn, leads to ambiguity as to constructvalid- emphasis away from empirical laws to theoretical
ity and testability. A third interpretationis that a cor- "laws." The Holistic Construaltries to present a new
respondencerule is a synthetic proposition. Its truth formulationof the structureof theory to accomplish
or falsity can be determinedby factual informationin this goal by focusing on the causal propertiesof struc-
the worldof experience,ratherthanby the logical truth tures.
implied by the form in which it is expressed. This The Image of Science. The Received View sees
overcomes the neglect of empirical meaning by the science in a particularlight. Knowledge is thought to
other two approaches,but at the same time, makes it begin with observation. This leads to empirical gen-
more difficultto arriveat lawlike generalizations(i.e., eralizationsamong observable entities. As our ideas
it) which occur between unobservableterms. In short, progress, theories are formulated deductively to ex-
it appearsthat correspondencerules are neither ana- plain the generalizations,and new evidence is used to
lytic, conventional, nor synthetic statements. Under confirm or disconfirm the theories. Throughout the
the Holistic Construal, we will attempt to provide a process, data are given precedence. Indeed, the entire
more formal and less ambiguousinterpretationof cor- process is viewed as essentially an inductiveone. Sci-
respondencerules. A final point to note here is that ence in general and knowledge in particularare be-
correspondencerules under the Received View are lieved to occur in an upward fashion: from data to
generally taken to supply all of the empirical (i.e., theory to understanding.Feigl (1970, p. 7) terms this
observable)meaning that theoreticalterms have. The "an 'upwardseepage' of meaning from the observa-
possibility for additional sources of empirical mean- tional termsto the theoreticalconcepts," and it is con-
ing will be introducedunder the Holistic Construal. struedin a similarway by Hempel (1952, p. 36), Car-
Laws and Regularities. Proponents of the Re- nap (1939, p. 65), and others identified with the
ceived View have been unclear as to the meaning of Received View.
laws and how scientific theories relate to them. For At least two problems can be identified with this
many logical empiricists, the purposeof theories is to outlook. First, it does not conform well with how sci-
arrive at empirical laws, i.e., regularitiesamong ob- ence is actually conducted. Science is influenced by
servations.Focus is upon ee. In fact, some authorsgo the characteristicsof individualscientists (Fisch 1977,
so far as to state that the use of theoreticalterms and
Mahoney 1976) and social forces (Merton 1973, Mi-
nonobservational propositions is superfluous, and one troff 1974), and begins just as frequently with theory
should attempt to eliminate and replace theoretical terms as it does with observations. Second, our presuppo-
with observable ones (Craig 1956; cf. Hempel 1965, sitions, biases, and cognitive faculties influence and
pp. 210-215; Nagel 1961, pp. 134-137). One prob- constrain the so-called immutable data (Nisbett and
lem with this policy is that our theories would not be Ross 1980, Thorndyke and Hayes-Roth 1979). Rather
extendable to new contexts and time periods. than a data-driven process, science is more and more
A slightly different position is offered by Carnap being described as a theory-driven activity (Brown
(1966, pp. 226-231). While acknowledging the ex- 1979, Chaps. 6 and 7; Suppe 1977, p. 125-221).
istence of empirical laws, he notes that theoretical
Nevertheless, it is probably more accurate to regard
laws-i.e., regularities between theoretical terms- theory and data in an inseparable and reflexive way.
are also possible. These are thought to be more gen- In this spirit, the Holistic Construal strives to more
eral than empirical laws and to explain the latter:
accurately serve as a model for the structure of theory.
18 / Journal
of Marketing,
Winter
1984

FIGURE 4
The Elements and Structureof the HolisticConstrual
D D DC
A F
t2
y I
3 X4 x5 t3 a, t4
S\ "a2
f C
a, a2 a3 f2 I C2
el e2 e3 e4 e5 X6 7 X8 9 10
w
I Cl C2 c3 c4 c5
'I 'e28 e9 e10
XII/ x12
el, el2
Definition of Symbols Used in Figure 4
Theoretical Concepts EmpiricalConcepts
A Antecedent a, Measurements of A
F Focal Concept fj Measurements of F
C Consequent ck Measurements of C
M Method Factor m, Measurements of M (if they exist)
DA, DF, Dc, DM Definitions of Theoretical Concepts (expressed as attributional, structural, and/or dispositional definitions)
Hf, Nonobservational propositions relating antecedent and consequent to focal concept, respectively
Rf, Hfc
Rkf Rationales for Hafand respectively
y, 1 Inferred(i.e., estimated)Hfc,
representations for Haf and Hf, respectively
Xm Inferred(i.e., estimated) implications of correspondence rules relating theoretical concepts to empirical concepts
a2 Inferred(i.e., estimated) derivative "definitions" for subdimensions of C
a•, rn
Inferred(i.e., estimated) contribution of M (i.e., that due to systematic error)
w, Correlated error terms across constructs
W2 Correlated error terms within constructs
?o Errors in equations or conceptual error (i.e., unexplained variation in theoretical variables)
e, Errors in variables or measurement error (i.e., random error)
A Prospectus
forTheory
Construction
in Marketing
/ 19

An Alternative: many, must be possessed by a concept (Achinstein
The Holistic Construal6 1968, Chaps. 1 and 2). That is, none of the attributes
in the cluster is requiredto be a necessary property
Figure4 presentsa diagramof the elements and struc- (althoughone or a subset may be), and selected sub-
ture of the Holistic Construal. For purposes of dis-
sets are sufficient to define the focal concept. For in-
cussion, we will considerthe HolisticConstrualin three stance, consider the following definition of attitude:
parts, but it should be stressed that the entire network an attitudeis a learnedpredispositionto respondto an
is needed for a full representationand interpretation
object in a preferentialor evaluative way. Depending
of any theory. Our pedagogical subdivision of the
on one's purposes, the predispositions defining atti-
Holistic Construalis based on the ways of ascertain-
tude might include physiological reactions, expres-
ing meaning in any theory: conceptual meaning, em- sions of affect (e.g., like or dislike), implicit evalu-
piricalmeaning, and spuriousmeaning. Together, the ative responses, beliefs, intentions, and/or approach/
three types of meaning combine in an integratedway
avoidance actions. In fact, various authorshave em-
to form the basis for theory constructionand its use
in explanation, prediction, and control. ployed one or more of these as being consistent with
the ideas of "a learned predispositionto respond."
Conceptual Meaning A special case of the cluster attributionaldefini-
To grasp the idea of conceptual meaning, let us con- tion is the law cluster concept (Putnam 1962). A law
sider the focal concept F shown at the top center of cluster concept consists of a group of laws which de-
terminethe identity of the concept. Again, not every
Figure 4. In general, the conceptual meaning of F is law is needed to specify the extension of the concept,
obtainedthrough a specification of (1) its definition,
or more broadlywhat it is, and (2) its relationto other only a subset. Laws are regularitieswith a high degree
of corroboration.Putnam (1962, p. 379) asserts that
concepts in a larger theoretical network, that is, its "most of the terms in highly developed science are
relation to A and C. Notice that the meaning of F
law-cluster concepts." In physics, energy is an ex-
depends in part on other concepts in the theory.
The definition of a concept is a linguistic opera- ample; whereas in marketing,we might stretchthings
a bit and include the concept of perceived risk as an
tion(s) that establisheda relationship(s)between a fo-
cal term and one or more other terms. Although the example because it is defined through "laws" of per-
ception, informationprocessing, and motivation.
relationship(s)might be one of equivalence or iden-
tity, it may also be one of refinement, expansion, or A structuraldefinitionof a concept specifies a set
of elements (e.g., properties)and relationsamong ele-
partial specification, under certain conditions noted
hereafter.Typically, three types of definitions can be ments such that the concept is given meaning through
seen to contributeto the conceptual meaning of a the- the entire network. The elements might be organized
oretical concept: attributional,structural,or disposi- in hierarchical,associative, interactive, or other pat-
tional definitions. terns. An example of a structuraldefinition can be
An attributionaldefinition specifies the attributes, found in Bagozzi's (1981la) representationof attitude
toward donating blood. Here attitude was hypothe-
characteristics,or propertiesof a concept. An atom-
istic attributionaldefinition states all of the charac- sized to exist as a unidimensionalglobal affective re-
teristicsthat a concept possesses. For example, an ex- action connected to a three dimensional expectancy-
value response consisting of perceived consequences
pectancy-valuemodel of attitudetoward an act might
be defined as the sum of the productsof beliefs about of the act and their evaluation. Until recently, the di-
the consequences of the act times the evaluations of mensionality of concepts is a topic that has largely
those consequences. Each attribute(i.e., beliefs and been ignored in marketing, as well as in the social
sciences.
evaluations) is a necessary component of the defini-
A third type of definition is the dispositional def-
tion, and all attributestogether sufficiently define at-
titude. Let us contrastthe atomistic attributionaldefinition. The dispositional definition describes the ca-
inition to the cluster attributionaldefinition which pacitiesor tendenciesof a concept. An attemptis made
consists of a group of propertieswherein not all, but to capture the intrinsic nature of a concept through
specificationof either its characteristicsor its internal
structure,and the power and/or liability it affords for
6Forbackground on the conceptual development of the Holistic Con- change or for influencing or being influenced by an-
strual, see Bagozzi (1979, 1980) and Bagozzi and Fornell (1982). other concept. Notice that the dispositional definition
More methodologically oriented expositions can be found in Bagozzi is in one sense similarto the attributionalor structural
(1983) and Bagozzi and Phillips (1982). Illustrations of the Holistic
Construal in substantive contexts can be found in Bagozzi (1981a, b; definitions. However, there is one importantdiffer-
1982), Phillips (1981, 1982), and Bagozzi and Silk (1983). Sheth (1967, ence. The dispositional definition carries with it the
pp. 740-742) was an early scholar in marketing who recognized the implicationfor undergoingtransformationor entering
need for a holistic framework based on the philosophy of science.
a causal relation under certain conditions. Nicosia's
20 / Journal
of Marketing,
Winter
1984

(1978) notion of an organization of percepts-in-ten- other cooperative and conflictual relations connected
sion can be regardedas a dispositional specification to the parties. Associative meaning frequently arises
of cognitive structureand its emotive potential. in complex systems of concepts where equilibrating
We thus can see thatconceptualmeaningof a con- or disequilibratingforces are at work. In consumer
cept is provided through its definition: i.e., through behavior, for instance, deprivationmight produce a
the semanticcontent of the terms in the definition and tension or imbalanceamong needs, drives, and goals.
the syntacticsignificance of their organizationand re- Priorto deprivation,the needs, drives, and goals would
lation to the definiendum(i.e., the focal concept). It be associated in a network and achieve an equilib-
is importantto note that this meaning is distinct from rium. After deprivation, internal arousal spreads
the empirical meaning provided through correspon- throughoutthe network as an instigatorfor action.
dence rules. Furthermore,conceptual meaning is also
providedby the relationshipsa focal concept has with EmpiricalMeaning
otherconceptswithinthe theoreticalpostulatesin which Empiricalmeaning refers herein to the observational
it is embedded. content associated with theoreticalterms. This is ac-
In general, in addition to its definition, the con- complished formally through correspondencerules.7
ceptualmeaningof a focal conceptF is obtainedthrough Although correspondencerules also supply empirical
specification of (1) the antecedents, determinants,or meaning in the Received View, they function differ-
causes of F, (2) the consequences, implications, or ently in the Holistic Construal.Recall that in the Re-
results of F, and (3) the associative (i.e., nonfunc- ceived View there is a one-way and upwardflow of
tional, noncausal) links to F. Let us briefly consider meaningfrom observationalterms to theoreticalterms.
these. In the Holistic Construal,the flow of meaning is ba-
As shown in Figure 4, some conceptual meaning sically downwardfrom theoreticaltermsto observable
of F arises through its antecedents. Whereas a defi- terms. That is, theoreticalterms are hypothesized to
nition specifies what a concept is and perhapswhat it imply particular observations. Nevertheless, under
is capableof becoming and doing, its antecedentssup- certainconditions,eitherupwardor nondirectional(i.e.,
ply informationas to where it has been (that is, its associative) empirical meaning is possible. Let us
history and development)and/or how it is formed or consider all of these alternatives.
influenced. The meaning is supplied throughthe con- Table 1 presentsthe most common ways that cor-
tent of the hypothesis (Haf) linking A to F and its ra- respondencerules are represented.We will look at the
tionale (Raf). The content of Haf consists of a state- point forms first, which reflect the connections be-
ment of the nature of the relationshipof A to F and tween experimentaltest operationsand individualre-
is expressed in propositionform. This might entail a sponses.8In the operationaldefinitionmodel, "we mean
relatively nonspecific statement such as "the greater by any concept nothing more than a set of operations;
the magnitudeor level of A, the less the magnitude the concept is synonymouswith the correspondingset
or level of F," or it might entail a more specific state- of operations"(Bridgman1927, p. 5). One shortcom-
ment as to the functional form of the relationshipor ing of this model is that any new measurementop-
even the amountof change expected in F as a function erationimplies a new theory. Not only does this result
of A. The rationale for the hypothesis is needed to in a proliferationof theories and difficulty in estab-
completethe meaningof F providedby A. In general,
a rationalefor a hypothesis can be obtained through
specificationof the mechanismby which A influences 7We might informally allow some empirical content to be attributed
F. Typically, this will be expressed through theoret- to the meaning of theoretical terms through their definitions. That is,
ical laws and an explicationof how A produceschange the primitive terms or dispositional implications of the definitions might
convey empirical meaning. This arises through convention and is a
in F (e.g., a causal explanation). historical development of intersubjective agreement among scientists
In a parallel fashion, the meaning of F is also de- concerned with a particular theory. When we draw on past research
termined through its relations to consequences C (see in theory construction, we often implicitly assign some empirical
meaning to theoretical concepts or at least their attributes. Another
Figure 4). That is, the implications of F supply in- way that empirical meaning arises (in addition to definitional and cor-
formation as to where a phenomenon is going, what respondence rule sources) is through a putative iconic relation to a
it can lead to, and/or what influence it has. Again, model (e.g., Hesse 1970; Nagel 1961, pp. 90-117). Here a physical
analogy is used to represent the theory or part of it. Notice that em-
the meaning arises through delineation of the form of pirical meaning in the Holistic Construal is not restricted to that pro-
Hfc and its rationale Rfc. vided by correspondence rules as it is in the Received View. Our
Finally, conceptual meaning of F occurs, at times, expansion of the sense of empirical meaning allows one to represent
the effect on the structure of theory of historical and political pro-
simply through associative or correlative links to other cesses among scientists.
concepts (not shown in Figure 4). For example, we 8For a discussion of the operational definition and partial interpre-
tation point forms, see Petrie (1971). Keat and Urry (1975) discuss
might interpret part of the meaning of an exchange
the causal indicator point form. To this author's knowledge, first con-
relationship between wholesaler and retailer through sideration of the structural form was by Bagozzi and Fornell (1982).
A Prospectus
forTheory
Construction
in Marketing
/ 21

TABLE1
Formal Models of Correspondence Rules in Scientific Inquiry
Model Point Form StructuralForm
The Operational P(x) (E(x) -- R(x)) "x has theoretical P*(x) = (E(x) -- R(x)) & S(x) "x has
Definition Model property P by definition, if and only if, theoretical property P* by definition, if
when x is subjected to experimental test and only if experimental test procedure E
E, it yields result R." is applied and result R occurs and is
organized in structure S."
The Partial E(x)- (P(x) R(x)) "If x is subjected to P*(y) - (E(x) -- (P(x) R(x)) & S(x) "y will
Interpretation experimental = test procedure E, it will have theoretical property P* if (1)
Model exhibit theoretical property P, if and only experimental test procedure E is applied
if it yields result R." to x and its result R is defined to have
theoretical property P and (2) R is
The Causal Indicator P(x) -- (E(x) -- R(x)) "If x has theoretical P*(y) -- (P(x) -- (E(x) -- R(x)) & S(x) "y
Model property P, then if experimental test will have theoretical property P* if (1) x
procedure E is applied, it will yield result having theoretical property P implies that
R." when experimental test procedure E is
applied, it will yield result R and (2) R is
lishing generalities, but it precludes multiple opera- term (Petrie 1971; Suppe 1977, pp. 102-104). More-
tionalization of theoretical concepts. This, in turn, over, although the form of the partial interpretation
prevents the determination of construct validity model does not necessarily imply a directionalitybe-
(Campbell1969). For these and otherreasons, the op- tween theoreticaland empirical terms, proponentsof
erationaldefinition model is no longer advocated by the Received View generally assert that an upward
philosophersof science. However, in marketingand process holds.
the social sciences, the operationaldefinition model As an alternative, the causal indicator model de-
has been perhaps the most frequently applied ap- serves scrutiny (e.g., Keat and Urry 1975). Here a
proach. Whenever a theory is tested on observations, causal link is specified between a theoreticalterm (or
and each observationis defined as an operationaliza- network of terms) and a test operation(s) and its re-
tion of a single theoretical concept (and vice versa), sult(s). A phenomenonor state representedby a the-
the operational definition model is implicitly being oretical term is thought to imply or explain observa-
employed. The aforementionedshortcomings of the tions. The correspondencerule, then, functions as a
ReceivedView with respectto reliability,validity, and scientific law linking theoreticalterm to experimental
testing issues also apply here. test procedure to observed results (Schaffner 1969,
The partial interpretation model of correspon- Sellars 1961). Notice that the correspondencerule is
dence rules is the second point form worth consider- not part of the theory or the observationsto which it
ation (see Table 1). This should be recognized as the is linked. Rather, it is an auxiliary hypothesis con-
preferredmodel of correspondence rules under the cerning theoreticalmechanismsexisting between the-
Received View (Carnap 1956). It is called a partial oretical terms and observations. Suppes (1962) sug-
interpretationbecausethe meaningof a theoreticalterm gests thata hierarchyof theorieslinks theoreticalterms
is only specified under particulartest (i.e., measure- and implied observations:a physical theory (e.g., of
ment) conditions. In the absence of test conditions, the instrumentation),the theory of the experiment,the
the theoretical term has no meaning. One important theoryof experimentaldesign, the theory of data, and
implicationof the model is that any theoretical term ceteris paribusconditions. One propertyof the causal
is permittedto have multipleoperationalizations. Thus, indicatormodel is that it betterconforms to the image
it is meaningful to consider internalconsistency reli- of science as a theory-drivenprocess. It is not so much
ability and construct validity, both of which require thatdata are ignored as it is that one's theory implies,
multiple measurements for their execution. Among throughcausal laws, certainobservations.In addition,
other problems, however, the partial interpretation the causal indicatormodel permits an independentse-
model does not allow theoretical terms to have se- manticinterpretation to exist for theoreticalterms(e.g.,
mantic content over and above that provided by the through definitions, analogies, iconic models), and
rule, and any change in a measurementprocedurewill multipleoperationalizationsare allowed as well. One
change the meaning of the correspondingtheoretical drawbackof the approach is that some observations
22 / Journalof Marketing,
Winter1984

can be produced by different or multiple theories it rests on a definitionalrelationshipbetween theoret-
and/or experimentalprocedures. Thus, a certain de- ical and empirical terms. The latter characteristicin-
gree of ambiguity exists in the correspondencerule. troducesa conventionaland analyticalpropertyto em-
One way to lessen the ambiguity is to examine the pirical meaning rather than specifying a synthetic
convergent, discriminant,concurrent,predictive, and contingency. Thus, the structuralform of the opera-
nomological validities of measures of concepts (Ba- tional definition model, as with its point form, is
gozzi 1980, Chap. 5; 1981a). flawed.
Before we considerthe structuralform of the three The structuralform of the partial interpretation
models of correspondencerules, we should point out model and the structuralform of the causal indicator
that the rigid dichotomy between unobservable and model both hypothesizethatmultiple measurescan be
observableterms advocated in the Received View is organizedin a set of subdimensions.For example, in
rejectedherein. Rather, it is proposedthat three kinds Figure 4, consequent C is shown to have two sub-
of unobservablesand two kinds of observablescan be dimensions C1 and C2, and these, in turn, have two
identified. One unobservableV~' is unobservablein and three measurements associated with them, re-
principle and includes certain primitives and theoret- spectively. Of course, the two models sharethe afore-
ical terms not subject to observation in even an in- mentionedcharacteristics noted for the respectivepoint
direct, inferential way. Nonreductive or emergent forms.
phenomena or purely analytical concepts are exam- Correspondencerules supply empiricalmeaningto
ples. A second unobservableV(' is also unobservable theoreticalterms and imply that the correspondence
in principle but either implies empirical concepts or can be representedas a matterof degree. In Figure 4,
can be inferred from observations. For instance, a the empiricalimplicationsof correspondencerules are
consumer's attitude might be reflected in his or her depictedas Xs. It is importantto stress that we do not
evaluationsor choices. The thirdunobservableV(' is observe X but rathermust infer it from data. For ex-
defined in termsof observables. For example, the mo- ample, X might be a parameterderived from infer-
rale of a salesforce might be defined as the average ential statistics. When we assess the adequacy of X,
of the observed morale scores of the individual sales- we must not only rely on statistical criteria but also
persons.A definitionspecifies conditionsthat are both on the conceptualmeaningof the correspondencerule
necessary and sufficient. The operational definition employed.
model essentially provides meaning to a V(' through We have drawn the arrows correspondingto the
its equationto observationsin this sense. The partial Xs in Figure 4 from the theoretical concepts to em-
interpretationmodel provides a necessary condition pirical concepts in order to be consistent with the
and a sufficient condition for theoreticalterms. Either meaning of the causal indicatormodel. Factor anal-
a V~' or a V(' can be addressedwith a partial inter- ysis, structuralequation models, and causal models
pretationmodel. The causal indicatormodel supplies mostclosely conformto this usage (e.g., Bagozzi 1980,
a sufficient, but not necessary, condition for theoret- Bentler 1980, Joreskogand Sorbom 1982). Neverthe-
ical terms. As with the partial interpretationmodel, less, under certain conditions, the arrowheadscould
either a V(' or a V(' may be addressed. The opera- have been reversedor omitted, dependingon the the-
tional definition model is a definition, whereas the oretical interpretationand methodology used to op-
partial interpretationand causal models are not true erationalizethe theory(e.g., Bagozzi andFomell 1982).
definitions. The operationaldefinition model assumes that empir-
Two types of observables are proposed. One ical conceptsdeterminethe meaningof theoreticalterms
Vg~can be directly observed with one or more of the and is most congruent with principal components
senses, whereas the other Vb2 can only be observed analysis and possibly partial least squares and can-
with sense extending instruments(e.g., physiological
onical correlationanalyses. The partial interpretation
recording instruments, questionnaires). Either Vg) or model is perhapsthe most flexible in terms of its re-
Vo can serve as observations in any of the corre- lations to observations.Althoughit is used most often
spondence rule models. when empirical concepts determine theoretical con-
Let us now consider the structural forms of the cepts, this does not appearto be necessary, and it may
correspondence rule models (see Table 1). We see that be applicablewhen the opposite sequence is assumed
the operational definition model defines a theoretical or none at all. However, it still suffers the problems
term as a function of its observations. Principal com- noted for the point form. A final caveat is that various
ponents analysis is one way to implement this rule methodologies are not mutually exclusive and some-
because each component is a linear function of its times can be used to solve similarproblems(e.g., Ba-
measurements. The shortcomings of this approach are gozzi, Fornell, and Larcker 1981; Fornell and Book-
that it assumes an upward accrual of knowledge while stein 1982). The relationshipsbetween correspondence
neglecting the possibility of a priori hypotheses, and rules and dataanalyticproceduresare still little under-
A Prospectus
in Marketing
/ 23

stood, however, and should be approachedwith cau- F is relatedto antecedentA throughnonobservational
tion. propositionHg and to consequent C through nonob-
servationalpropositionHfc. The inferredestimates for
Spurious Meaning these propositionsare shown as y and 3, respectively.
It would be ideal if our theoriescould be characterized Random errors in observations for A, F, and C will
solely through the representationof conceptual and generally suppress the values of y and P. However,
empirical meaning. However, theory and measure- procedures(e.g., structuralequation models with la-
ments must be constructed in an imperfect world by tent variables) exist for the correction of the delete-
imperfecthuman beings, and the conceptual and em- rious effects of randomerror (Bagozzi 1980, Bentler
pirical meaning of our theoreticalterms (and our non- 1980, Joreskogand Sorbom 1982). Nevertheless, ran-
observationalpropositions)are, by necessity, contam- dom error in measures beyond a certain point (i.e.,
inated by errorsin observations. greaterthanabout50%)generallyleadsto a pooroverall
In general, this contaminationor "spuriousmean- goodness of fit.
ing" occurs in at least two ways: as random mea- Systematic Error. In addition to random error,
surement error and as systematic error. Let us ex- spuriousmeaning occurs as systematic error. System-
amine both. atic error reflects patterned biases arising from re-
RandomError. To grasp the meaning of random sponse sets, evaluation apprehension, demand char-
error, consider again Figure 4. Each measurementof acteristics,methodsfactors,or omittedvariables.These
an empiricalconcept (shown as lower case letters en- external contaminators induce regular or irregular
closed in boxes) can be representedas the sum of a changes in the means, variances, and/or covariances
true, theoreticalconcept plus randomerror: of observations. This especially becomes a problem
when a subset of measurements is so affected. By
y=T+e
suppressingor enhancing variation in selected obser-
wherey observedmeasurement,T = true-score(i.e.,
= vations in an orderlyway, sources of systematicerror
theoreticalconcept), and e = randomerror(Lord and may influence key phenomena of interest over and
Novick 1968). Intuitively, what this equation means above the true causes and randomerror.
is that our measurementsof theoretical concepts re- What are the specific implications of systematic
flect not only the process, attributes, or phenomena error?As with randomerror, systematic errorcan in-
underlyingthe theoreticalconcept but also a deviation fluence the values of the Xs or y or 3 shown in Figure
aroundthe true theoretical concept. Each attempt at 4. However, ratherthan a general suppressioneffect
measurementwill typically reflect a slight over- or un- upon these parameters,systematicsources of errorcan
derestimateof the true value of the underlying con- either inflate or deflate parameters,depending upon
cept, although in the long run, the mean of these er- their pattern and magnitude. Whereas random error
rorswill tendto cancel out. Randommeasurementerror tends to make it more difficult to ascertainconstruct
can be considered the outcome of the sum of many validity or detect causal paths, systematic error can
counteractingforces including, among others, chance cause us to conclude either that a phenomenon(e.g.,
variation,haphazardfluctuationsin data gatheringand a causal relation) is present, when in fact it is not, or
instrumentationprocedures, and transientchanges in that a phenomenonis not present, when in fact it is.
the phenomenabeing studied (e.g., random fatigue, From a practicalmodeling standpoint,there are at
memory, and motor vacillations). least three ways to detect and take into account sys-
What are the consequences of random error? In tematic error. The simplest, but least satisfactory, is
general, it may have two effects. The most direct ef- to merely representthe presence and effects of sys-
fect is that randomerrorwill deflate the value of the tematic errorthroughcorrelatedresiduals in measure-
parameterX relating a measurementto its respective ments. As shown in Figure 4, there are two classes
theoreticalconcept. That is, the empirically inferred of correlatedresiduals: correlated errors across con-
implicationof a correspondencerule will indicate less structs(w1)and correlatederrorswithinconstructs(w2).
agreementbetween concept and measurementto the Correlatederrorsmean that variationin the respective
extent that random error is present. Notice in Figure measurementsis produced by something in addition
4 that each measurementwill have a separateamount to the underlyingtheoreticalconcepts and randomer-
of randomerrorassociatedwith it (shown as ei). Thus, ror. Unfortunately, not only is the additional cause
to the extent that random error exists, we will expe- unknown, but one or more causes could account for
rience lower reliability and greater uncertaintyas to the correlatederrors. Within the context of structural
the validity of what we are measuring. equation modeling, the use of correlated errors can
The other effect of randomerroris upon our abil- correct for systematic biases. However, this will al-
to
ity infer relationshipsamong theoretical concepts. ways weaken the interpretationof the meaning of the
For the simple model of Figure 4, the focal concept model underinvestigation.The correlatederrorsacross
24 / Journal
of Marketing,
Winter
1984

constructsshown in Figure 4, for example, indicate ation in measurementsin additionto that arising from
either the presence of a common method bias in a3 randomerrorand the contributiondue to a theoretical
and f2 or an omitted variable that serves as an ante- concept. Figure 4 illustratesa method factor imping-
cedent(e.g., suppressoror augmentorvariable)to both ing upon the first measurementof each theoretical
A and F. In either case, our confidence that y rep- concept in the model. A common type of method fac-
resents the true relation between A and F will be tor is systematicbias from a measurementprocedure.
weakenedto the extent of the magnitudeof wi. Sim- For example, in Figure 4, it is possible that a,, fl, cl,
ilarly, correlatederrors within constructs such as W2 and c3 were all obtained from a key informantwho
in Figure4 suggest that either c4 and c5 share a com- exhibited a characteristicbias in each measurement.
mon method bias, or C2 is not unidimensionalas hy- The remainingmeasurementsare assumedto have been
pothesized. Again, this weakens the interpretabilityof collected by othermethodsand to not sharea common
the model. systematic bias. We might then write the causes of
Ideally, one would preferthatno systematicerrors total variationin a measurementas follows:
occur in a model. In reality, the potential exists for
at least some systematicerrorin most studies. The use Yi -= iTj + ITiMk + ei
of correlatederrors might then representone way to where yi = measurementi for true score Tj, Mk = a
correctfor this errorbut should be used with caution. hypothesized method factor, ei = random error in
We suggest that correlatederrorsbe investigated un- measurementyi, and hi and reflect the relative con-
der the following conditions. First, if possible, one 7"i
tributionsof Tj and Mk, respectively. This equation
should have a plausible hypothesis as to why error proposes that the variation in a measurementcan be
terms should be correlated. Correlated errors are divided into that due to an underlyingtheoreticalcon-
sometimesjustifiable when the same variable is mea- cept, a methodfactor (i.e., systematicerror),and ran-
suredat differentpoints in time, when a common pro- dom error(see Bagozzi and Phillips 1982). This is a
cedure is used to record data, or when other system- hypothesis that can be tested.
atic biases can be identified. Second, it is desirable Methods factors should also be used with caution
that correlatederrorsbe small in magnitudeand have and only as a last resort. One should have conceptual
no or a small impacton parameterestimatesin a model and/or methodologicalreasonsfor introducinga method
(i.e., on the Xs, y-, and p in Figure 4). This would factor, its influence should be small, few should be
indicatethat the importanceof the unknown contam- employed, and at least 50% variation in measure-
inator(s)is small. Along this line, it would be desir- ments should be accounted for by underlying theo-
able also that the explained variancein measurements retical concepts. Otherwise, as with correlatederrors,
accountedfor by a hypothesized underlyingtheoreti- a danger exists in overfitting or getting ambiguous
cal concept be greater than about 50% (Fornell and findings.
Larcker1981). This implies that less than 50% of the The use of methodsfactors goes somewhat farther
variationis due to randomand systematicerrors.Third, and is somewhatmore meaningfulthanthe use of cor-
the numberof significantly correlatederrors, no mat- relatederrors.In Figure 4, we have drawnthe arrows
ter how small in magnitude, should be kept to a min- associated with XA,and X12, respectively, as dashed
imum. Correlatederrorschange the substantivemean- lines to indicate that two versions of methods factors
ing of a model, so too many correlationswould result exist: one without and one with separate measure-
in a vastly different model. Finally, correlatederrors ments for the factors. The version without separate
shouldbe employed only as a last resortand after one measurementsrepresentsthe effects of systematic er-
has explored a pure model without correlatederrors ror but does not explicitly identify the source of the
and/or a model with unmeasuredor measuredmeth- error. The version with separate measurementsrep-
ods factors, topics to which we will turnshortly. Even resentsboth the source and effects of systematicerror.
under the above conditions, correlatederrors detract For example, if we measured organizationalproper-
from the theoreticalelegance and empirical interpret- ties of wholesale distributorsand used key informants
ability of a model. Nevertheless, correlatederrorscan as sources of information (e.g., the chief executive
be a useful tool in exploratoryresearch if the above officer, a vice president,and a middle managerwithin
safeguardsare followed. Of course, cross-validation, each organization),then we might expect systematic
examinationof rival hypotheses, and an ongoing re- biases from each source. To account for the effects of
search program should also be employed whenever the bias, we might merely introducea method factor
feasible to reduce the probabilityof false inferences. for each type of respondent.But to more specifically
A slight improvementover the use of correlated and accuratelymodel the systematic error, we might
errorsas a means for addressing systematic error is use position in the organization, knowledgeability,
the introductionof a method factor or methods fac- education,tenure, or other measuresto operationalize
tors. A methodfactor is a hypothesizedcause of vari- the methods factors. Similarly, in consumer behavior
A Prospectus
in Marketing
/ 25

research,we might employ different methods to mea- correspondencerules are not interpretedas analytical,
sure phenomenaand thereforeexpect systematicbiases conventional,or syntheticstatements.Rather,they are
associated with each. taken as auxiliaryhypotheses about the causal impli-
In sum, we see that systematic error can be rep- cations that theoretical postulates have for observa-
resented on a continuum ranging from a partial actions. Second, theoreticaltermsand observationalterms
counting of contamination(i.e., correlatederrors) to are regardedneither as mutually exclusive categories
a relatively complete accounting (i.e., methods fac- nor as end points on a continuum.Instead, threetypes
tors with explicit operationalizations). It should be of theoreticaltermsand two types of observationterms
stressed again that the modeling of systematic error are proposed. Third, empiricalmeaning of theoretical
shouldbe done cautiouslyand in the light of the issues terms is not restricted to that provided through cor-
noted heretofore. Moreover, because we can never respondencerules, but ratheroccursin additionthrough
know the true state of affairs, any test of a model is definitions, analogies, and iconic models. Fourth,
at best a tentative and imperfect picture of reality. It scrutiny in the Holistic Construalis upon theoretical
is perhapsbest to construe individualtests as a means laws. Empiricallaws are regardedas error-riddenand
to eliminate false hypotheses or to fail to disconfirm biased reflections of theoreticallaws. Finally, the im-
plausibleones. Only throughresearchprogramsor ac- age of science suggested by the Holistic Construalis
cumulatedknowledge over time do we approach an one of a downward,theory-drivenprocess. However,
understandingof the workings of a phenomenon. althoughtheoriesare formulateddeductively and prior
knowledge and world views shape the hypotheses and
Final Comments on the Holistic outcomes, parameterestimation and hypothesis test-
ing are generallyaccomplishedthroughinductivepro-
Construal cedures. Thus, we see a need for greatersynergy be-
The Holistic Construalhas its roots in both the phi- tween theory constructionand data analysis. Overall,
losophy of science and multivariatestatistics. How- then, upwardfeedback from observationsmust be in-
ever, a numberof importantdifferences deserve men- tegratedwith the downward influence of theory con-
tion. First, at a global level, the Holistic Construalis struction. Researchers have long advocated such a
broaderand more integrative than either philosophy philosophy, but we have lacked a formal means to
of science or statistical approaches. Philosophers of accomplish this until recently.
science have tended to focus on logical and episte- It is beyond the scope of this article to consider
mological issues in general and to neglect specific the statistical aspects of the Holistic Construal (cf.
questions as to operationalizationand empirical cri- Bagozzi 1980, 1983; Bentler 1980; Bentler and Bo-
teria. This is understandable,given their aims and the nett 1980). But a numberof differences from classical
orientationin philosophy for pursuit of general prin- procedurescommonly employed in marketingshould
ciples of knowledge and broad normativecriteria. By be noted. First, structuralequationmodels with latent
the same token, statisticians have emphasized tech- variablespermitthe researcherthe opportunityto rep-
nical issues and ignored formal criteriaof theory con- resentboth errorsin variablesand errorsin equations.
struction. They have taken a more mechanical ori- The errorterms in traditionaleconometric models re-
entationand been less concernedwith the philosophical flect an amalgam of all errors without providing for
bases and substantive interpretationof research. The a separationand identificationof the sources. In Fig-
Holistic Construalconstitutes an effort to bridge the ure 4, the e's are errorsin variables (i.e., randomer-
philosophical and statistical traditionsby being con- ror), and the 9'sare errors in equations (i.e., unex-
cerned with conceptual and empirical issues as they plained variation in latent dependent variables). Not
relate to a particularresearch problem and body of only does this modeling permit a direct assessment of
research. random error, but the structuralparameters(e.g., y
The purposeof the Holistic Construalis to provide and p13 in Figure 4) are purged of random error in a
the foundationfor rigorous methodological thinking. manneranalogous to that provided by correction for
Following Kaplan(1964) we construemethodology in attenuationprocedures.Second, the effects of system-
its broadest sense to mean concern with "such pro- atic errorcan also be accountedfor undercertaincon-
cedures as forming concepts and hypotheses, making ditions throughthe incorporationof correlatederrors
observations and measurements, performing experi- and/or methods factors. This is a consequence of the
ments, building models and theories, providing ex- additionalinformationprovidedby multiple measures
planations, and making predictions" (p. 23). Thus, of latent variables, a feature not possible with classic
stress is placed on the formulation of a sound TCO regressionanalyses. Third, structuralequationmodels
and its implementation,test, and interpretation. with latent variables give the researchera powerful
In contrast to the Received View, however, a and straightforwardmeans to examine reliability and
number of differences should be pointed out. First, various forms of validity. Heretofore, these diagnos-
26 / Journal
of Marketing,
Winter
1984

tics were based on more error ridden and arbitrary lem is due to a failure to meld togetherthe theoretical
correlationalprocedures. Finally, the statistical tests domain with the empirical. Up until now, at least, we
appliedto models formulatedunderthe Holistic Con- have suffered from a lack of guidelines and formal
strualallow one to investigate a variety of individual means to integratesound theory with rigorous meth-
level and multigrouppredictions in either controlled odology. This has been compounded by a tendency
laboratoryor naturalisticfield settings. Of course, de- toward specialization. Substantively oriented re-
pending on whether one uses maximum likelihood, searchers emphasize conceptual aspects of their re-
partial least squares, or other methods to operation- search but sometimes place relatively less stress on
alize the Holistic Construal,one must be wary of the methods and measures. Methodologicallyorientedre-
assumptionsand limitations of the techniques. searcherstend to focus relatively more on techniques
A final point to note with respect to the Holistic than on theory development. At the same time, both
Construalis the importanceof making any piece of camps spend more of theirenergies readingliteratures
researchan integralpartof a body of knowledge. This and interactingwith others with similar leanings. This
means drawing upon prior learning in the literature, tends to perpetuatethe split. Nevertheless, probably
making special efforts to develop sound measure- the most importantstumbling block has been the ab-
ments (e.g., Churchill 1979), and generally engaging sence of a formal language to link substance with
in a researchprogramof interlockingstudiesover time method and data.
so that one may get as close to a phenomenonas pos- Eitherthe Received View or the new realist model
sible and benefit from the efficiencies and serendipity provides a way to more purposefully conduct re-
that often accompanies such endeavors. Examples of search. The focus of both is upon how theories can
earlyapplicationsof the HolisticConstrualcan be found and should be constructed. This authorhas a prefer-
in Phillips(1981, 1982) and Bagozzi (1981a, b; 1982). ence for the realist model and has attempted,through
the Holistic Construal, to develop a programmefor
its operationalization.By addressingthe content and
Summary and Conclusions structureof our theories in more depth, we can make
the science and art of marketingless haphazardand
In recent years we have witnessed a considerablede- more subject to evaluation and control. Marketingis
gree of dissatisfactionand controversy with the way enteringa new era that will demandof its researchers
researchhas been done in the discipline. This is es- excellence in both theory and method. The irony is
pecially evident in the subareas of management sci- that as we become more explicit in our theorizingand
ence (e.g., Bass 1982 and Ehrenberg1970) and con- modeling, so too will we become more vulnerableto
sumerbehavior(e.g., Jacoby 1978) and touches upon criticism. But perhaps this is the price that must be
the adequacyand relevancyof our theories, methodol- paid if we are to renew ourselves and innovate. Rather
ogies, measurements,and findings. What accountsfor than playing it safe and hiding behind a well-devel-
this turmoil?It certainlydoes not appearto stem from oped theory that is never really tested, or a sophisti-
a lack of thoughtfulnessby our leaders, for many fine cated methodologythat is applied without sound con-
conceptualizationsand models exist and continue to ceptual groundwork,we should demand of ourselves
be generated.Nor does it seem to arise predominantly the best that can be offered in both substance and
from a lack of sophisticationin methodology or anal- method. Only then will we push our knowledge to a
ysis, for the field is no less rigorous than the disci- plane truly reflective of the interplaybetween theory
plines it draws upon. and data, which is, after all, our only valid window
This authorbelieves that at least part of the prob- on reality in the marketplace.
REFERENCES
Achinstein, P. (1968), Concepts of Science, Baltimore, MD: cago: American Marketing, 15-33.
Johns Hopkins Press.
(1980), Causal Models in Marketing, New York:
Bagozzi, R. P. (1976), "Science, Politics, and the Social Con- JohnWiley.
struction of Marketing," in Marketing: 1776-1976 and Be-
(1981a), "An Examination of the Validity of Two
yond, K. L. Bernhardt,ed., Chicago: American Marketing, Models of Attitude," Multivariate Behavioral Research, 16
586-592.
(July), 323-359.
(1979), "The Role of Measurement in Theory Con- (1981b), "Attitudes,Intentions, and Behavior: A Test
structionand Hypothesis Testing: Toward a Holistic Model,"
of Some Key Hypotheses," Journal of Personality and So-
in Conceptual and Theoretical Developments in
Marketing, cial Psychology, 39 (October), 607-627.
0. C. Ferrell, S. W. Brown, and C. W. Lamb, eds., Chi-
(1982), "A Field Investigation of Causal Relations
A Prospectus
forTheory
Construction
in Marketing
/ 27

among Cognitions, Affect, Intentions, and Behavior," Spiegel-Rosing and D. deSolla Price, eds., London: Sage.
Journal of Marketing Research, 19 (November), 562-583. Fornell, C. and F. L. Bookstein (1982), "A Comparative
(1983), "A Holistic Methodology for Modeling Analysis of Two StructuralEquation Models: LISREL and
Consumer Response to Innovation," Operations Research, PLS Applied to Market Data," Journal of Marketing Re-
31 (January-February), 128-176. search, 19 (November), 440-452.
and C. Fornell (1982), "Theoretical Concepts, and D. F. Larcker (1981), "Evaluating Structural
Measurements, and Meaning," in A Second Generation of Equation Models with Unobservable Variables and Mea-
Multivariate Analysis, Vol. II, Measurement and Evalua- surement Error," Journal of Marketing Research, 18 (Feb-
tion, C. Fornell, ed., New York: Praeger. ruary), 39-50.
, and D. F. Larcker (1981), "Canonical Hanson, N. R. (1971), Observation and Explanation: A Guide
Correlation Analysis as a Special Case of a Structural Re- to Philosophy of Science, New York: Harper and Row.
lations Model," Multivariate Behaviorial Research, 16 Hempel, C. G. (1952), Fundamentals of Concept Formation
(October), 437-454. in Empirical Science, Chicago: University of Chicago Press.
and L. W. Phillips (1982), "Representing and Test- (1965), Aspects of Scientific Explanation, New York:
ing Organizational Theories: A Holistic Construal," Ad- Free Press.
ministrative Science Quarterly, 27 (September), 459-489. Hesse, M. B. (1970), Models and Analogies in Science, Notre
and A. J. Silk (1983), "Recall, Recognition, and Dame, IN: University of Notre Dame Press.
the Measurement of Memory for Print Advertisements," Howard, J. A. and J. N. Sheth (1969), The Theory of Buyer
Marketing Science, 2 (Spring), 95-134. Behavior, New York: John Wiley.
Bass, F. M. (1974), "The Theory of Stochastic Preference and Hunt, S. D. (1976), Marketing Theory: Conceptual Founda-
Brand Switching," Journal of Marketing Research, 11 tions of Research in Marketing, Columbus, OH: Grid.
(February), 1-20. Jacoby, J. (1978), "Consumer Research: A State of the Art
(1982), "A Discussion of Different Philosophies in Review," Journal of Marketing, 42 (April), 87-96.
the Analysis of Advertising-Sales Relationships," Dallas: Joreskog, K. G. and D. Sorbom (1982), Analysis of Linear
University of Texas, working paper. Structural Relationships by the Method of Maximum Like-
Bentler, P. M. (1980), "Multivariate Analysis with Latent lihood, Chicago: National Educational Resources.
Variables: Causal Modeling," Annual Review of Psychol- Kaplan, A. (1964), The Conduct of Inquiry, Scranton, PA:
ogy, 31, 419-456. Chandler Publishing Company.
and D. G. Bonett (1980), "Significance Tests and Keat, R. and J. Urry (1975), Social Theory as Science, Lon-
Goodness of Fit in the Analysis of Covariance Structures," don: Routledge and Kegan Paul.
Psychological Bulletin, 88 (No. 3), 588-606. Kuhn, T. S. (1970), The Structure of Scientific Revolutions,
Bridgman, P. W. (1927), The Logic of Modern Physics, New enlarged ed., Chicago: University of Chicago Press.
York: MacMillan. Lakatos, T. S. (1970), "Falsification and the Methodology of
Brown, H. I. (1979), Perception, Theory, and Commitment: Scientific Research Programmes," in Criticism and the
The New Philosophy of Science, Chicago: University of Growth of Knowledge, I. Lakatos and A. Musgrave, eds.,
Chicago Press. Cambridge: Cambridge University Press, 91-196.
Campbell, D. T. (1969), "Definitional Versus Multiple Op- Little, J. D. C. (1975), "BRANDAID: A Marketing-Mix
erationalism," Et Al, 2, 14-17. Model," Parts I & II, Operations Research, 23 (July-Au-
Carnap, R. (1939), Foundations of Logic and Mathematics, gust), 628-673.
Chicago, IL: University of Chicago Press. Lord, F. C. and M. R. Novick (1968), Statistical Theories of
(1956), "The Methodological Character of Theo- Mental Test Scores, Reading, MA: Addison-Wesley.
retical Concepts," in Minnesota Studies in the Philosophy Mahoney, M. J. (1976), Scientist as Subject, Cambridge, MA:
of Science, Vol. I, H. Feigl and M. Scriven, eds., Min- Ballinger.
neapolis: University of Minnesota Press, 33-76. Manicas, P. T. and P. F. Secord (1983), "Implications for
(1966), An Introduction to the Philosophy of Sci- Psychology of the New Philosophy of Science," American
ence, New York: Basic Books. Psychologist, 38 (April), 399-413.
Churchill,G. A., Jr. (1979), "A Paradigmfor Developing Better Merton, R. K. (1973), The Sociology of Science: Theoretical
Measures of Marketing Constructs," Journal of Marketing and Empirical Investigations, N. W. Storer, ed., Chicago:
Research, 16 (February), 64-73. University of Chicago Press.
Craig, W. (1956), "Replacement of Auxiliary Expressions," Mitroff, I. (1974), "Norms and Counter-normsin a Select Group
Philosophical Review, 65, 38-55. of the Apollo Moon Scientists: A Case Study of the Am-
Ehrenberg, A. S. C. (1970), "Models of Fact: Examples from bivalence of Scientists," American Sociological Review, 39
Marketing," Management Science, 16 (March), 435-445. (August), 579-595.
Feigl, H. (1970), "The 'Orthodox' View of Theories: Remarks Nagel, E. (1961), The Structure of Science, New York: Har-
in Defense as Well as Critique," in Minnesota Studies in court, Brace.
the Philosophy of Science, Vol. 4, M. Radner and S. Wi- Nicosia, F. M. (1978), "Brand Choice: Toward Behavioral-
nokur, eds., Minneapolis: University of Minnesota Press, Behavioristic Models," in Behavioral and Management
3-16. Science in Marketing, H. L. Davis and A. J. Silk, eds.,
Feyerabend, P. K. (1965), "Problems of Empiricism," in Be- New York: Ronald Press, 12-55.
yond the Edge of Certainty, R. Colodny, ed., Englewood Nisbett, R. E. and L. Ross (1980), Human Inference: Strat-
Cliffs, NJ: Prentice-Hall, 145-260. egies and Shortcomings of Social Judgment, Englewood
(1970), "Against Method: Outline of an Anarchistic Cliffs, NJ: Prentice-Hall.
Theory of Knowledge," in Minnesota Studies in the Phi- Peter, J. P. (1979), "Reliability: A Review of Psychometric
losophy of Science, Vol. 4, M. Radner and S. Winokur, Basics and Recent Marketing Practices," Journal of Mar-
eds., Minneapolis: University of Minnesota Press, 17-130. keting Research, 16 (February), 6-17.
Fisch, R. (1977), "Psychology of Science," in Science, Tech- (1981), "Construct Validity: A Review of Basic Is-
nology, and Society: A Cross-disciplinary Perspective, I. sues and Marketing Practices," Journal of Marketing Re-
28 / Journal
of Marketing,
Winter
1984

search, 18 (May), 133-145. Science: A Perspective in the Philosophy of Science," in
Petrie, H. G. (1971), "A Dogma of Operationalism in the So- PSA 1972: Proceedings of the 1972 Biennial Meeting of the
cial Sciences," Philosophy of the Social Sciences, 1 (May), Philosophy of Science Association. Boston Studies in the
145-160. Philosophy of Science, Vol. 20, K. F. Schaffner and R.
Phillips, L. W. (1981), "Assessing Measurement Errorin Key Cohen, eds., Dordrecht, Holland: Reidel, 407-419.
Informant Reports: A Methodological Note on Organiza- Sheth, J. N. (1967), "A Review of Buyer Behavior," Man-
tional Analysis in Marketing," Journal of Marketing Re- agement Science, 13 (August), B718-756.
search, 18 (November), 395-415. Silk, A. J. and G. L. Urban (1978), "Pre-Test Market Eval-
(1982), "Explaining Control Losses in Corporate uation of New Packaged Goods: A Model and Measure-
Marketing Channels: An Organizational Analysis," Journal ment Methodology," Journal of Marketing Research, 15
of Marketing Research, 19 (November), 525-549. (May), 171-191.
Platt, J. R. (1964), "Strong Inference," Science, 146 (Octo- Sternthal, B., R. Dholakia, and C. Leavitt (1978), "The Per-
ber), 347-353. suasive Effect of Source Credibility: Tests of Cognitive Re-
Popper, K. R. (1959), The Logic of Scientific Discovery, New sponse," Journal of Consumer Research, 4 (March), 252-
York: Basic Books. 260.
Putnam, H. (1962), "The Analytic and the Synthetic," in Min- Suppe, F. (1977), The Structure of Scientific Theories, 2nd
nesota Studies in the Philosophy of Science, Vol. 3, H. ed., Urbana, IL: University of Illinois Press.
Feigl and G. Maxwell, eds., Minneapolis: University of Suppes, P. (1962), "Models of Data," in Logic, Methodology,
Minnesota Press, 350-397. and Philosophy of Science: Proceedings of the 1960 Inter-
Ray, M. L. (1979), "Introductionto the Special Section: Mea- national Congress, E. Nagel, P. Suppes, and A. Tarski,
surement and Marketing Research-Is the Flirtation Going eds., Stanford, CA: Stanford University Press, 252-261.
to Lead to a Romance?," Journal of Marketing Research, Swinburne, R. G. (1971), "The Paradoxes of Confirmation-
16 (February), 1-6. A Survey," American Philosophical Quarterly, 8 (Octo-
Schaffner, K. F. (1969), "CorrespondenceRules," Philosophy ber), 318-330.
of Science, 36 (September), 280-290. Thorndyke, P. W. and B. Hayes-Roth (1979), "The Use of
Sellars, W. (1961), "The Language of Theories," in Current Schemata in the Acquisition and Transfer of Knowledge,"
Issues in the Philosophy of Science, H. Feigl and G. Max- Cognitive Psychology, 11 (January), 82-106.
well, eds., New York: Holt, Rinehart, and Winston, 57- Toulmin, S. (1972), Human Understanding, Princeton, NJ:
77. Princeton University Press.
Shapere, D. (1969), "Notes Toward a Post-Positivistic Inter- Zaltman, G., K. LeMasters, and M. Heffring (1982), Theory
pretation of Science," in The Legacy of Logical Positivism, Construction in Marketing, New York: John Wiley & Sons.
P. Achinstein and S. Barker, eds., Baltimore, MD: Johns Zaltman, G., C. R. A. Pinson, and R. Angelmar (1973),
Hopkins Press, 115-160. Metatheory and Consumer Research, New York: Holt, Ri-
(1974), "Discovery, Rationality, and Progress in nehart and Winston.
NOW AVAILABLE:
The ONE Book You'l Need for Understanding Marketing and the Law
LEGALASPECTS OFMARKETINGwaysthecourtviewsandactsuponthe Marketing
Professionals,toordera copy
STRATEGY:Antitrust
and areasofproductdesign,branding,
prod- foryourownpersonal/professional
uctadditions
viamergers-acquisitions
or reference
libraryona15-day examination
ConsumerProtectionIssues jointventures,
pricing, chan- basis,pleasespecify
marketing titleandwriteto:
LouisW.SternandThomasL. nels,advertising,
andsalespromotion. Stephen Landis, Prentice-Hall,
Inc.,Book
Eovaldi,bothofNorthwestern University
Alsoincluded arethemostrecent Distribution
Center,Route 59atBrook Hill
merger Drive,WestNyack, NY10994.
Co-authoredbyLouisW.Sternand guidelinesissuedbytheJusticeDepart-
Thomas L.Eovaldi,twowell-respectedmentandtheFTC; definitions
oftrouble- Afteryou'vehada chance toexamine the
namesinthefieldsofmarketing andlaw,
someconcepts, suchas HHI andpreda- book,eithersendtheindicated amount,
thisnewbookcoversall
respectively, torypricing;anddetailed coverage ofthe includingpostage andhandling charges,
majorantitrust
andconsumer issuessurrounding
protection productliability, orreturnthebookandowenothing. Ifyou
issuesontheFederallevelandfocuseson
warranties,deception,andunfairness. includepayment withyourorder,
thepub-
themanylawssurrounding (andoften Anexcellent tool lisherwillpayallpostageandhandling
learning forstudents, charges.Samefullrefund, samereturn
keydecisions
affecting) madebymar- SternandEovaldi's LegalAspects ofMar-
keting
executives. keting is alsoaninvaluable ref- privilege
as guaranteedabove.
Strategy
16ActualCourtCases erenceworkforprofessional marketers. For"SUPER-QUICK" Service,dial
Toillustrate
theflavor
andreasoning
of AvailableMarch 1984 TOLLFREE (800)526-0485*
between
thecourttobothtoday's
andtomorrow's 512pp.(est.) Cloth$28.95 9:00a.m.-4:30
p.m.,EST,Mon.-Fri.
marketing 16separate
executives, court *notapplicable
inNewJersey, Hawaii
Alaska, or
decisionsareexamined.
Detailed
arethe University
Professors, to receive
your Puerto
Rico;intheseareas,pleasewrite
complimentary copyforpossible class- tous.
roomadoption, pleasewriteonyour
schoolletterheadto:Mr.Robert Jordan,
Dept.J-278,Prentice-Hall,Inc.,Engle-
woodCliffs,NJ07632.
fl Prentice-Hall
A Prospectus
forTheory
Construction
in Marketing
/ 29


Bagozzi Theory Construction Marketing PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bagozzi Theory Construction Marketing PDF

Uploaded by

Copyright:

Available Formats

A Prospectus for Theory Construction in Marketing

Author(s): Richard P. Bagozzi

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This articleaddresses the question, "How should A Prospectus

Introduction the intermediary role of how theoriesshouldbe formed.

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

'I 'e28 e9 e10

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

This content downloaded from 117.211.92.90 on Tue, 29 Apr 2014 07:16:16 AM

You might also like