a definition of theory: research guidelines for different theory-building … · journal of...

Ž .Journal of Operations Management 16 1998 361–385

A definition of theory: research guidelines for differenttheory-building research methods in operations management

John G. Wacker )

Department of Management, Iowa State UniÕersity, Ames, IA 50011-2065, USA

Abstract

This study examines the definition of theory and the implications it has for the theory-building research. By definition,theory must have four basic criteria: conceptual definitions, domain limitations, relationship-building, and predictions.Theory-building is important because it provides a framework for analysis, facilitates the efficient development of the field,and is needed for the applicability to practical real world problems. To be good theory, a theory must follow the virtuesŽ .criteria for ‘good’ theory, including uniqueness, parsimony, conservation, generalizability, fecundity, internal consistency,empirical riskiness, and abstraction, which apply to all research methods. Theory-building research seeks to find similaritiesacross many different domains to increase its abstraction level and its importance. The procedure for good theory-buildingresearch follows the definition of theory: it defines the variables, specifies the domain, builds internally consistentrelationships, and makes specific predictions. If operations management theory is to become integrative, the procedure forgood theory-building research should have similar research procedures, regardless of the research methodology used. The

Ž .empirical results from a study of operations management over the last 5 years 1991–1995 indicate imbalances in researchmethodologies for theory-building. The analytical mathematical research methodology is by far the most popular methodol-ogy and appears to be over-researched. On the other hand, the integrative research areas of analytical statistical and theestablishment of causal relationships are under-researched. This leads to the conclusion that theory-building in operationsmanagement is not developing evenly across all methodologies. Last, this study offers specific guidelines for theory-buildersto increase the theory’s level of abstraction and the theory’s significance for operations managers. q 1998 Elsevier ScienceB.V. All rights reserved.

Keywords: Theory; Research guidelines; Operations management

1. Introduction

Although many business professionals, social sci-entists, and other academics have very similar beliefson the definition of theory, there are still somedifferences of opinion in a theory’s exact nature. Forexample, some practitioners and academics believe

) Tel.: q1-515-294-8111; e-mail: [email protected]

that theory and its application are very limited and,therefore, not very useful in the real world of busi-ness. Others feel that very little theory exists in theacademic world. For example, consider the followingdiscussions of theory and scientific investigation:

Theory, for theory’s sake, can easily degenerate intoan uninteresting art form. Yet, practice without the-ory can quickly become a dull and dangerous occu-pation. Unfortunately, the world is a complicatedplace and complicated solutions and processes are

0272-6963r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved.Ž .PII S0272-6963 98 00019-9

( )J.G. WackerrJournal of Operations Management 16 1998 361–385362

often required to make complex organizations run.The ability to live with uncertainty and the insightinto both one’s professional powers and limitations

Žis the sign of a mature management science Shubik,.1987 .

. . . Of all our valid knowledge of the social world,most of it seems to have been the product of layrather than professional inquiry . . . A typical situa-tion in social science is that scientific inquiry onlymodestly raises the validity of a lay proposition by

Ž .qualifying it Lindblom, 1987, p. 517 .

Ž .. . . N o theory of inventory exists unless the theorydictates the manner in which the measurements ofthe parameters of the theory are to be made. Thisamounts to saying that today we have no theory ofinventory. We will only have such a theory when wecan specify the information necessary to test the

Žtheory and justify this specification Churchman,.1961, 132 .

The implication of the first statement is that the-ory does not necessarily require application. Thisstatement means that theory is abstract and does nothave to be applied or tested to be a ‘good’ theory. Ifthis is the case, then theory can remain totally ab-stract and non-applied. The crux of whether theorycan be totally abstract and non-applied depends onthe definition of theory.

The second statement implies that scientific inves-tigation and theory-building are not necessarily use-ful for the development of social science or manage-rial decision-making fields of academic study. Thisstatement implies that ‘good’ theory is ‘found’through trial and error rather than through scientificinvestigation in a systematic matter. Further, thisstatement implies that ‘good’ theory-building in so-cial sciences and managerial decision-making is de-rived from lay investigation rather than by scientificinvestigation. The validity of this statement dependsupon how theory is defined.

The third statement implies that unless theoryexplicitly indicates how it is measured, it is not a

Ž .theory. Churchman 1961 states that since inventorytheories do not indicate how they are to be measured,they are not theories. The statement infers that atheory should offer how it is to be measured for

empirical testing, and without this testing, a theorycannot be ‘good’ theory. Churchman’s conclusionthat no inventory theory exists in management alsodepends upon the definition of theory.

All three of the above statements cause concernsfor academics since they infer three common criti-

Ž .cisms of theory: 1 theory does not have to beŽ .applied, 2 it does not make significant improve-

Ž .ments in the external world, and 3 theory does notexist due to lack of measurement of definitions. Eachconcern has some measure of truth. Unless opera-tions management research addresses relevant practi-cal problems to explain complex phenomena, it can-not develop into a theory-building discipline. Each ofthe above concerns depends upon the definition oftheory and, more importantly, what criteria is used todevelop ‘good’ theory. It does not seem logical tocall any theory a ‘good’ theory without definingtheory or the criteria for ‘good’ theory.

Besides these concerns, there are additional rea-sons why theory is important for researches and

Ž .practitioners: 1 it provides a framework for analy-Ž .sis; 2 it provides an efficient method for field

Ž .development; and 3 it provides clear explanationsfor the pragmatic world. The first reason is that goodtheory provides a framework of analysis for opera-tions management since it provides structure forwhere differences of opinion exist. For example, a‘theory of internationally competitive manufactur-ing’ would provide a structure to evaluate factories’manufacturing international competitiveness. Thistheory enables academics to enumerate the exactconditions to classify firms into degrees of interna-tional competitiveness. Since it is unlikely that allacademics would agree on which factors are mostimportant in a hierarchy of factors, a framework of‘good’ theory procedures provides an understandingof where these differences of opinion lie.

The second reason for developing theory is effi-ciency. Theory development reduces errors in prob-lem-solving by building upon current theory. Build-ing upon current theory is equivalent to incorporat-ing all that is known from the current literatureŽtheoretical, mathematical, empirical, and practi-

.tioner research into a single, integrated consistentbody of knowledge. For researchers, using a singleintegrated body of knowledge for analytical andempirical testing gives the results a deeper theoreti-

( )J.G. WackerrJournal of Operations Management 16 1998 361–385 363

cal meaning by differentiating between the compet-ing theories. An integrated body of knowledge canonly be pursued efficiently if integrated theory isdeveloped through consistent theory-buildingmethodologies.

A third reason why theory is important is itsapplicability. Consider this example: a group of man-agers hires a consultant to improve their firm’s com-petitiveness. The consultant provides a set of sugges-tions. If the consultant does not provide anythingelse, there is little reason to be surprised if themanagers disregard these suggestions, since man-agers need more rationale than a set of rules. Morespecifically, they want to know why this set of rules

Žhold for their specific manufacturing facility what,.who, where, when and how and why these specific

Žrelationships could, would, or should improve pre-.diction claim their manufacturing facility. Although

managers seek advice, it is essential for them to beskeptical of any suggestions since they are directlyresponsible for both good and bad results. Conse-quently, they must logically be convinced that noth-ing important is overlooked. They want the analyti-cal reassurance that these suggestions are logicallyand practically compatible with each other. Addition-ally, they want to know specific instances where

Ž .these rules have been successful empirical support .Therefore, even in the most practical of instances,use of the formal definition of theory is important forall managers wishing to achieve measurable results.For this reason, many academics believe that: ‘‘There

Žis nothing as practical as a good theory’’ Hunt,.1991; Van de Ven, 1989 . ‘‘Good theory is practical

precisely because it advances knowledge in a scien-tific discipline, guides research toward crucial ques-tions, and enlightens the profession of management’’Ž .Van de Ven, 1989 .

Because of the need for ‘good’ theory, the pur-pose of this study is to set guidelines for integratingtheory-building research. In order to achieve thislofty goal, this study first determines what a ‘good’theory is. Next, it closely ties ‘good’ theory tospecific research procedures. After the good-theory-building procedures are outlined, this study catego-rizes different research methodologies by their the-ory-building purpose and depicts how all theory-building procedures are similar. Then this study em-pirically investigates the prevalence of different re-

search methodologies in operations management todetermine where operations management is weak intheory-building research. Last, the article concludeswith the guidelines for theory-building research. Asare all articles suggesting theory-building and re-search investigation, this article is an opinion articlesince ‘‘the value laden nature of theory can never be

Ž .eliminated’’ Bacharach, 1989 . Therefore, the opin-ions expressed here should be evaluated on theirlogical appeal and internal consistency.

2. A definition of theory

Before beginning any discussion on theory, thisstudy must differentiate between the common notionof ‘theory’ and a formal definition of theory. In thisarticle, the term theory is interpreted as following theformal definition and operationalization of theory.This operationalization of the definition of theoryshould directly be tied to the necessary componentsof theory. Generally, academics point to a theory as

Ž . 1being made up of four components, 1 definitionsŽ . 2of terms or variables, 2 a domain where the

Ž .theory applies, 3 a set of relationships of variables,Ž . Ž . Žand 4 specific predictions factual claims Hunt,

.1991; Bunge, 1967; Reynolds, 1971 . Theories care-fully outline the precise definitions in a specificdomain to explain why and how the relationships are

1 Theoretical definitions are not observable as such. Rather,their existence and properties are asserted in order to account forwhat is observable. Theoretical definitions are conceptual in na-ture. Even a relatively simple term such as manufacturing leadtime has conceptual foundations that transcend its measurement.By definition, manufacturing lead time is ‘‘The total time requiredto manufacture an item, exclusive of lower level purchasing time’’Ž .Crawford, 1987 . Conceptually, manufacturing lead time repre-sents the internal time a manufacturing system takes to manufac-ture an item. This concept is not observable because the precisemoments from when all materials become available to when theorder is completed cannot be exactly specified.

2 Domain of a theory: the domain of the theory is the exactsetting or circumstances where the theory can be applied. Forexample, a just-in-time theory domain may be manufacturingfacilities that focus on few products and compete on cost withacceptable quality. The just-in-time theory domain would giveextensive factors that limit the instances of when and where it isapplied.


logically tied so that the theory gives specific predic-tions. Therefore, the precision of good theory causesa theory to be very exacting for all the key compo-

Ž .nents of a theory, or as Poole 1989 and Van de VenŽ .1989 state: ‘‘A good theory is, by definition, alimited and fairly precise picture.’’ A theory’s preci-sion and limitations are founded in the definitions ofterms, the domain of the theory, the explanation ofrelationships, and the specific predictions.

Authors usually agree that the goal of ‘good’theory is a clear explanation of how and why spe-cific relationships lead to specific events. Conse-quently, these explanations of relationships are criti-cal for ‘good’ theory-building. Other authors’ state-ments on theory indicate the importance of relation-ship-building:

Ž .Theory is . . . an ordered set of assertions about ageneric behavior or structure assumed to holdthroughout a significantly broad range of specific

Ž .instances Sutherland, 1976: 9 .

Researchers can define theory as a statement ofrelationships between units observed or approxi-mated in the empirical world. Approximated unitsmean constructs, which by their very nature cannotbe observed directly. . . . A theory may be viewed asa system of constructs and variables in which theconstructs are related to each other by propositionsand the variables are related to each other by hy-

Ž .potheses Bacharach, 1989 .

These statements indicate the importance of rela-tionship-building in explaining how and why specificphenomena will occur. Sometimes how and why andspecific predictions are condensed into the expres-sion ‘adequate explanation’, which implies that un-less an explanation can predict, it is not considered

Ž .adequate Hunt, 1991 .A very important aspect of a theory definition is

phrased in the common questions that researchersrequire to exactly specify a theory. Consider thisstatement: ‘‘The primary goal of a theory is to

Ž .answer the questions of how, when or where , andwhy . . . unlike the goal of description, which is to

Ž . Žanswer the question of what or who ’’ Bacharach,.1989 . In short, any definition of theory should

answer common questions that researchers face. First,theory defines all variables by answering the com-

mon questions of who and what. The domain speci-fies the conditions where the theory is expected tohold by using the common questions of when andwhere. The relationship-building stage specifies thereasoning by explaining how and why variables arerelated. And last, the predictive claims specify thewhether ‘‘Could a specific event occur?’’, ‘‘Should aspecific event occur?’’, and ‘‘Would a specific eventoccur?’’ In short, the definition of theory providesguidelines to answer the common questions thatoccur in natural language. From the pragmatic per-spective of operations managers, the predictive claimsfrom theory answer the could, should, and wouldquestions which are quite critical for managers’ fu-ture success. Consequently, the should, could andwould questions are very important for theory to beconsidered useful to managers.

In summary, the definition of theory suggested bythis study has these four components: definitions,domain, relationships, and predictive claims to an-swer the natural language questions of who, what,when, where, how, why, should, could and would.

3. Virtues of ‘good’ theory

The purpose of this section is to limit the domainof theory-building to the domain of what PopperŽ .1957 calls ‘good’ theory. Theories in the domain of‘good’ theory are superior to other theories becausethey possess what philosophers of science call the

Žtheory’s virtues Popper, 1957; Kuhn, 1980; Quine.and Ullian, 1980 . ‘Good’ theory’s superiority is

important because researchers must evaluate the rela-tive significance of opposing theories. Although thereis no general agreement between philosophers ofscience concerning the relative importance of eachvirtue of ‘good’ theory, there seems to be a fairlywidespread agreement as to what they are. A com-mon set of virtues of a theory is: uniqueness, parsi-mony, conservatism, generalizability, fecundity, in-ternal consistency, empirical riskiness, and abstrac-

Ž .tion Quine and Ullian, 1980 .Table 1 provides an overview of the virtues of

‘good’ theory. The first five virtues are fairlystraightforward: each theory must be differentiated

Ž .from other theories uniqueness : new theories can-not replace existing theories unless they are better


Table 1The virtues of ‘good’ theory: key features and why they are important to ‘good’ theory development

Virtue Key feature Why important for ‘good’ theory and for the develop-ment of the field

Uniqueness The uniqueness virtue means that one theory must If two theories are identical, they should be consideredbe differentiated from another. a single theory. Although it applies to all criteria for

theory, this virtue directly applies to definitions sincedefinitions are the most elemental of building blocksfor theory.

Conservatism A current theory cannot be replaced unless the new Therefore, current theory is not rejected for the sake oftheory is superior in its virtues. change. This criteria is needed so that when a new

theory is proposed, there is a good reason to believe allŽother theories are lacking in some virtue Quine and

.Ullian, 1980; Kuhn, 1980; Popper, 1957 .Generalizability The more areas that a theory can be applied to If one theory can be applied to one type of environ-

makes the theory a better theory. ment and another theory can be applied to manyenvironments, then the second theory is a morevirtuous theory since it can be more widely applied.Some authors call this virtue the utility of the theorysince those theories that have wider application havemore importance.

Fecundity A theory which is more fertile in generating new Theories which expand the area of investigation intomodels and hypotheses is better than a theory that new conceptual areas are considered superior to theo-has fewer hypotheses ries which investigate established research areas. This

meansTheory parsimony theory The parsimony virtue states, other things being If two theories are equal in all other aspects, the onesimplicity, theory effi- equal, the fewer the assumptions the better. with fewer assumptions and the fewer definitions isciency Occum’s razor more virtuous. This virtue also includes the notion that

the simpler the explanation, the better the theory. Thisvirtue keeps theories from becoming too complex andincomprehensible.

Internal consistency Internal consistency means the theory has identified Internal consistency refutation means that the theoryall relationships and gives adequate explanation. logically explains the relationships between variables.

The more logically the theory explains the variablesand predicts the subsequent event, the better the theoryis. This internal consistency virtue means that thetheory’s entities and relationships must be internallycompatible using symbolic logic or mathematics. Thisinternal consistency means that the concepts andrelationships are logically compatible with each other.

Empirical riskiness Any empirical test of a theory should be risky. If there are two competing theories, the theory thatRefutation must be very possible if theory is to be predicts the most unlikely event is considered theconsidered a ‘good’ theory. superior theory. In the opposite case, if the theory

predicts a very likely event then it is not seen as beinga very valuable theory. This criteria is sometimes putin a different way: ‘‘Every good theory has at least oneprohibition; it prohibits certain things from happening’’Ž .Popper, 1957 .

Abstraction The abstraction level of theory means it is indepen- The abstraction level means it is better to integratedent of time and space. It achieves this indepen- many relationships and variables into a larger theory. Ifdence by including more relationships. one of two competing theories integrates more inter-

nally consistent concepts, it is more virtuous than atheory that integrates fewer internally consistent rela-tionships.


Žtheories evaluated in the light of their virtues con-.servation : new theories can be introduced that have

Žrich new areas of investigation fecundity virtue,some academics would call this area grounds for a

.paradigm shift ; and last, any theory should not beŽ .overly complex theory parsimony .

However, the remaining virtues are worthy ofadditional discussion since they are the focus of mostof the academic discussion. There are two theoryrefutation virtues: internal consistency and empiricalriskiness. The internal consistency virtue implies thatthe theory is internally consistent using either mathe-

Ž .matics or symbolic logic Bacharach, 1989 . Withoutinternal consistency, each relationship is independentfrom every other relationships causing the theory tonot be integrated within itself. This virtue is differentfrom face validity because each relationship in atheory may be explained individually, and thereforethe relationship may have face validity. However,each relationship may not be consistent with otherrelationships in the theory, thus, causing the entiretheory to be internally inconsistent. Since theories tietogether many concepts, it usually is not easy toidentify internal inconsistency. Therefore, it is neces-sary to develop the theory using mathematics orsymbolic logic.

The criterion of empirical riskiness has been thefocus of most of critical evaluators of ‘good’ theory.Most academics believe that empirical tests of theoryshould be risky so that there is a good chance of thetheory being refuted. On the other hand, if an empiri-cal test of theory supports the occurrence of a likelyevent, then the theory is deemed to be a ‘weak’theory. One empirical riskiness criteria is frequentlycalled the ‘single event refutation’ criteria whichstates that it takes only one sample to question atheory’s legitimacy. Put another way, every legiti-mate empirical test is designed to disprove the theory

Ž .and should be risky Popper, 1957 .Sometimes the two refutation criteria are not com-

patible. The incompatibility between the two refuta-tion criteria typically occurs when empirical evi-dence does not support an existing theory. What arethe deciding criteria to determine which evidence ismore important, empirical or logical? Generally, thelogical evidence is considered superior to the empiri-cal evidence since there is less chance of error. In thephilosophy of science the phrase used is the ‘power

of deduction rules’. The ‘power of deduction rules’provides the underlying structure for theory-building

Ž .research Popper, 1957 . A brief example may proveuseful to clarify why the ‘power of deduction rules’.Suppose a researcher gathered a large set of data ona bank and statistically found that when the arrivalrate is greater than the service rate, the line isshorter. To most operations management academics,there is no amount of empirical evidence that couldconvince them that having greater arrival rates thanservice rates will lead to shorter lines. This conclu-sion is contrary to queuing theory’s deductive con-clusion. In short, the ‘‘power of deduction rules’’.

Yet, most operations management academicswould investigate the specific circumstances to de-termine which queuing model should be applied. Oralternatively, academics would develop a new theoryto explain the empirical evidence. Therefore, basedupon the empirical evidence, operations managementacademics may develop a ‘new’ internally consistent

Ž . Ž .theory model to replace the existing theory model .The last virtue is the theory’s abstraction level,

which is usually classified into three levels: high,middle, and low. High abstraction level theoriesŽ .general or grand theories have an almost unlimitedscope, middle abstraction level theories explain lim-

Žited sets of phenomena which serve as the rawmaterials for the construction of more general theo-

. Žries , and lower level theories called empirical gen-.eralizations of very limited scope serve as simple

Žrelationship identifications Bluedorn and Evered,.1960 . These three levels of abstraction are points on

an abstraction continuum. A low abstraction leveltheory can be directly applied to a specific instancesince a low abstraction level theory completely de-

Ž .fines the variables, domain, relationship s , and thenstates the factual claim for this specific instance.However, the usefulness of low abstraction leveltheories to academics and practitioners is limitedbecause in the majority of instances, the theorycannot be applied since the definitions and domainare so narrow that the theory only applies to a fewinstances. Yet, low abstraction level theories areused to build middle abstraction level theories which,in turn, lead to high abstraction level theories. At thehighest abstraction level, the application of a theoryis the broadest since the domains are broad. Highlevel abstraction theory, therefore, has wide applica-


tion. Consequently, an important goal of theory-building is to advance lower abstraction level theo-ries to middle level theories, and then to high ab-straction level theories. This leads to the generaliza-tion that as a field progresses, it tends to have theoryin higher abstraction levels. This progression fromlow abstraction to high abstraction is called ladder-

Ž .climbing Osigweh, 1989 .In operations management, ladder climbing is be-

coming evident by several of the empirical general-Ž .izations in just-in-time JIT theories. For example,

the following are three low-level abstraction theo-Ž .ries: 1 shorter set-up times facilitate smaller lotŽ .sizes, 2 smaller lot sizes cause lower work-in-pro-

Ž .cess, and 3 lower work-in-process causes increasedquality. These three different smaller theories arebeing integrated into a large theory through a morecareful explanation as to how they are related to eachother. When addressed individually, each of thesetheories is a low abstraction level theory. Yet, whencombined into single integrated theory, it becomes a

Žmid-range theory Miller and Roth, 1995; Demeyer.and Arnoud, 1989; Wacker, 1987, 1996 . Still, the

integration into high abstraction level theories seemsto be somewhat in the future for operations manage-ment.

Though all of these theory virtues are highlysignificant for theory-building, the relative weightgiven to each virtue becomes important for compar-ing competing theories, since virtues trade off witheach other. For example, a new theory may explain

Ž .phenomena in many domains generalizability , butŽbe relatively complex hampering the parsimony cri-

.teria . Or a theory may be rich in hypotheses devel-Ž .opment fecundity , but can only be applied to a very

Žlimited set of conditions hindering the generalizabil-.ity criteria . Or theories that are generalizable, may

Žnot be internally consistent hindering the internal.consistency virtue . To some degree, each ‘good’

theory weighs each virtue against other virtues. Thisevaluation requires value judgments to decide whichtheory is superior to other theories based on eachtheory’s virtues. Although there are always trade-offsamong virtues, the internal consistency and empiricalriskiness virtues are generally considered the mostimportant virtues since without refutation being pos-sible or likely, the theory becomes tautologicalŽ .Popper, 1957 .

4. When does a theory become a ‘good’ theory?

One of the most difficult questions to answerabout theory is: when does a theory become a ‘good’theory? The answer to that question is inherentlycontroversial since it involves the degree to whichindividuals believe in adhering to the formal defini-tion of theory and follow the virtues of ‘good’theory. For the purpose of this article, a ‘good’theory should meet the definitional criteria of theoryas well as follow the virtues of ‘good’ theory. There-

Žfore, ‘good’ theory must first be a theory i.e., havedefinitions, have a domain, have relationships, and

.make predictions and must meet each ‘good’ theoryvirtue to some degree.

Any theory which adheres to both the definitionof theory and the virtues of a ‘good’ theory is a‘good’ theory. This adherence, however, does notmean that the ‘good’ theory is valid since ‘good’theories can be ‘just plain wrong’. Yet, ‘good’ theo-ries which are wrong are more quickly identified as

Žbeing wrong since they are more easily refuted in-.ternally inconsistent or empirically invalid . Still,

‘good’ incorrect theories serve a very important pur-pose for field development, since ‘good’ theory is abeginning point to determine why the theory iswrong. Therefore, because ‘good’ theory is easilyrefuted and is a beginning point for future investiga-tion, using ‘good’ theory for empirical tests seems tobe a laudable objective for the development of anacademic field.

On the other hand, a theory that violates one ofthe virtues of ‘good’ theory is difficult to refute.Since the virtues of theory apply to the definitionalcriteria of theory, there are innumerable ways a‘bad’ theory can be developed. Consequently, it ispossible to give myriad examples of how a ‘bad’theory hinders the development of an academic fieldby detracting research efforts away from ‘good’ the-ories.

5. The general procedure for ‘good’ theory-build-ing research

The purpose of this section is to demonstrate ageneral research procedure that satisfies the follow-ing specifications: theory’s definition, ‘good’ theory’s


virtues, and answers to the natural language ques-tions of who, what, when, where, why, how, should,could, and would. From the logical perspective, theseguidelines are all necessary conditions for theory-

Ž .building Whetten, 1989; Van de Ven, 1989 . Theprocedure must state the purpose of each step.

The academic literature suggests many differentŽprocedures for theory-building Eisenhardt, 1989;

.Swamidass, 1986; Bacharach, 1989 . Generallyspeaking, these procedures suggest how to opera-tionalize specific types of research projects. Theresearch procedures suggested here are similar tothose operationalizations of research. However, thisstudy identifies a minimal procedure for ensuringthat all guidelines for ‘good’ theory-building are metregardless of the type of research. Each step reflectsa necessary condition to fulfill ‘good’ theory-build-ing guidelines, since all theory-building requires def-initions, domain, relationships, and predictions. Con-sequently, the same procedure is required regardlessof the methodology used for any theory-buildingresearch. Therefore, from the viewpoint of theory’svirtues, this study suggests that the simplest, mostgeneralizable procedure be adopted. The requirementthat theory-building research follows the definitionof theory leads to similarities in theory-building re-search for both analytical and statistical procedures.

More about these two basic research methods arediscussed below.

Although the stages suggested here are listedsequentially, they are not necessarily sequential sincethey interact with each other. For example, when atheory is visualized, a new definition may be neededto identify a new concept. Or when a prediction orfactual claim is made, it may require a differentdomain or a different relationship. In addition, theoperationalization of variables for empirical testsmay require some modifications of selected defini-tions. Consequently, although this procedure is listedin stages, a research project may not necessarilyproceed inexorably sequentially through each stageto a ‘good’ theory.

For all stages of theory-building, the role of theliterature search in the research procedure is ex-tremely important since it provides the accepteddefinitions, domains of where a theory applies, pre-

Žviously identified relationships along with empirical.tests , and specific predictions of other theories.

Therefore, to assure that all theory-building condi-tions are fulfilled, an extensive literature search ofthe academic as well as the practitioner articles isrequired.

Table 2 outlines the general procedures for devel-oping theory. The precise definitions of variables are

Table 2A general procedure for theory-building and the empirical support for theory

Purpose of this step Common question ‘Good’ theory virtuesemphasized

Definitions of variables Defines who and what are included and Who? What? Uniqueness, conserva-what is specifically excluded in the tiondefinition.

Limiting the domain Observes and limits the conditions by When? Where? GeneralizabilityŽ .when antecedent event and where the

subsequent event are expected to occur.Ž .Relationship model Logically assembles the reasoning for Why? How? Parsimony, fecundity,

building each relationship for internal consis- internal consistency,tency. abstractness

Theory predictions and Gives specific predictions. Important for Could the event occur? Should the event Empirical testsempirical support setting conditions where a theory pre- occur? Would the event occur? refutability

dicts. Tests model by criteria to giveempirical verification for the theory.The riskiness of the test is an importantconsideration.


needed to limit the area of investigation by definingwho and what. Generally, the literature provides abase for defining the variables. New definitions mustdemonstrate why current definitions are not ade-quate, otherwise the conservation virtue is violated.Ordinarily, currently defined concepts should be usedto avoid violating the conservation virtue. Althoughthere are many difficulties in precisely defining andmeasuring variables, and the complete discussion ofthe implications are beyond the scope this paper,there are several concerns which bear mentioning.One important difficulty with definitions is what hascome to be known as ‘concept stretching’ whichoccurs:

Ž .. . . W hen concepts are broadened in order toextend their range of application, they may be so

Ž .broadly defined or stretched that they verge onbeing too all-embracing to be meaningful in therealm of empirical observation and professional prac-tice. . . . Because of this broadness, it tends toconfuse more than help the development of the field:. . . Because concept stretching in organizational sci-ence results in amorphous unclear conceptualiza-tions, what appears to have been a gain in exten-

Ž .sional coverage breadth often has been matchedand even surpassed by losses in connotative preci-

Ž . Ž .sion depth Osigweh, 1989 .

One common source of ‘concept stretching’ isnatural language which is too broad for precisemeasurement, causing a research project to build anill-defined theory, which in turn, gives misinforma-

Ž .tion Osigweh, 1989 . Because of the difficultieswith natural language, artificial languages and defini-

Ž .tions sometimes called formal language must bedeveloped to avoid the confusions caused by naturallanguage. Some feel that theory-building researchcannot be effectively achieved unless the academic

Žfield has a precise artificial language Teas and.Palan, 1997 . To develop an artificial language and

avoid ‘concept stretching’, some theorists suggestthat the definitions be examined by the negationprinciple. That is, definitions must be examined bywhat they specifically exclude. Although it is nottrue in the extreme, generally, the more a conceptualdefinition excludes, the more precise it is and the

Ž .more likely it will be unique Osigweh, 1989 .

Once the precise definitions of variables are estab-lished, the domain is established to limit when andwhere the theory hold. The domain of the theorydirectly limits the its generalizability since the morespecific the domain of the theory is, the lower thegeneralizability. Theory-building research extendsdomains to new broader areas by testing the theoryin a new environment or a different time period. Inshort, theory-building research extends the domain ofthe theory.

Typically, after both the definitions are clarifiedŽ .and the domain is specified, relationship model

building begins. This step is necessary to establishwhich variables have logical connections to othervariables. Technically, every variable used shouldexplicitly state how and why it is related or unrelatedto each and every variable in the model. The summa-tion of the related plus the unrelated variables shouldbe equal to the total number of variables in themodel. Hence, in any given theory, the relationshipbetween any two variables in the theory must beexplicitly stated, or else the theory cannot be shownto be internally consistent. The third stage is quite

Žcomplex since the theory and the model by which.the theory is to be tested should be fully developed

and internally consistent. For this step, the academicliterature suggests which relationships are importantfor the development of the theory. The more care-fully a researcher builds the relationships from otherresearch, the more theoretically important the re-search is, since it is integrating theory to raise itsabstraction level.

In the relationship-building step, there are fourtypes of theory-building relationships: those relation-

Žships that are assumed to be true fundamental laws.or axioms ; those laws that are derived from the

Ž .fundamental laws derivative laws or theorems ; thoselaws that span the gap from the theoretical to the

Ž .empirical world bridge laws or guiding hypotheses ,Žand the relationships that are being investigated re-

. Ž .search or theoretical hypotheses Hunt, 1991 . Forexample, research on cost and small lot sizes analyti-cally assumes that the number of orders that areplaced in a year is equal to demand divided by the

Ž .fixed order size a fundamental law . Additionally,the average inventory is equal to maximum plus

Žminimum inventory divided by two a second funda-.mental law . The total cost of inventory is equal to


the order cost times the number of orders plus theŽholding cost times the average inventory a derived

.law or theorem . The cost optimization solution leadsŽ . Žto the economic order quantity EQQ a second

.derived law . A bridge law is necessary to tie theanalytical results to the empirical world. For exam-ple, given the annual demand, annual holding cost,and ordering cost, firms should have a predicted

Žorder quantity the research or theoretical hypothe-.sis . One bridge law that would facilitate this empiri-

cal test is: rational firms act to minimize their totalinventory costs. Note that without a bridge law, theEQQ derived law does not span into empirical world.The area of bridge laws is important for integrationof theory in operations management. It is the lack of

Ž .bridge laws that Churchman 1961 referred to thebeginning of the study. Consequently, Churchman’scriticism cannot be summarily dismissed, since manyanalytical research studies do not give any guidanceas to how they are to be tested in the externalempirical world.

In the above example, fundamental and derivedlaws are not tested, but are assumed to be trueanalytically. If these relationships were tested alongwith other relationships, the statistical complexitymay be too complex for meaningful investigation.Because of this complexity, it is important that re-searchers limit their investigation using the literature,since the parsimony virtue may be violated. Theliterature provides the best guidelines as to whichrelationships are theoretically important for investi-gation and which relationships may be consideredfundamental or derived laws that do not need consid-eration in the investigation. Consequently, theory-building research uses the literature as a guideline todecide which relationships are important for investi-gation.

The last stage is theory prediction. In order for atheory to meet the minimal requirements for thedefinition of theory, it is technically not necessarythat a theory offer evidence to support predictive

Ž .claims only predictions are necessary . Yet, from apractical perspective, it is possible that a huge num-ber of theories could be proposed. Therefore, manyacademics prefer that empirical evidence be pre-sented to verify that a proposed theory has somemerit in the empirical world. Different methodolo-gies use different empirical evidence to verify their

predictive validity. More will be discussed about thispredictive stage below when discussing proceduresfor analytical and empirical research.

Though each stage of theory-building needs tofollow all the virtues of ‘good’ theory, each stage ismore closely related to certain specific virtues. Forexample, in the definition step, uniqueness and con-servation are key virtues since they limit theory-building to developing current concepts before de-veloping new unique concepts. Some researchersargue that many definitions such as total qualitymanagement, continuous improvement, and just-in-

Žtime overlap and are neither unique, nor new West-.brook, 1987; Robinson, 1990 . They argue that these

concepts do not limit the definition and are notnecessarily unique from each other which makesthem malformed thus causing data-gathering difficul-

Ž .ties as well as misinformation Osigweh, 1989 . Inthe domain specification stage, the generalizabilityvirtue is important because the more domains inwhich a theory can be applied, the more importantthe theory is. In the relationship-building step, parsi-mony, fecundity, and abstraction virtues enhance thetheory by using only necessary relationships, offer-ing new areas for investigation, and integrating rela-tionships for a higher abstraction level. Also in thisstage, internal consistency is important to verifywhich relationships are logically compatible witheach other. Generally, as more internally consistentrelationships are integrated into a theory, the theorycan explain more, therefore raising the theory’s ab-straction level. In the theory prediction step, theimportance of internal consistency and empiricalriskiness are both needed for the theory to makepredictions.

Table 2 summarizes the procedure for ‘good’theory-building research. Each of the stages are re-quired for a theory to be considered a ‘good’ theory.Unless a proposed theory has all these stages, it doesnot meet the criteria of the formal definition oftheory. The first column shows the components oftheory. The second column states why these compo-nents are necessary. The third column gives thecommon question that each stage addresses. Whilethe first three columns use the definitional criteriafor theory, the last column gives the most relevantvirtues for that stage to ensure that any theory devel-oped is a ‘good’ theory.


‘Good’ theory-building research requires the ful-fillment of the formal definition of theory. Addition-ally, a ‘good’ theory should not violate any ‘good’theory virtue. The stages of a ‘good’ theory-buildingmirrors the four criteria for theory and meet thevirtues of ‘good’ theory. However, since it is rela-tively easy to propose many ‘good’ theories thatmeet the theory criteria, usually some empirical evi-dence is needed for ‘good’ theory to be ‘good’theory-building. Therefore, empirical evidence usu-ally is included in the prediction stage to support the

Žtheory it could be argued that this is an additional.stage of theory-building . Consequently, ‘good’ the-

ory-building should have empirical evidence to sug-gest why the theory has some legitimacy in theempirical world.

ŽAs a side note, the statements by Shubik that.theory can be totally impractical cannot be summar-

ily dismissed since a ‘good’ theory does not have tobe tested to be ‘good’ theory. In short, it is possiblethat a new ‘good’ theory may not yet have externalempirical evidence to support it. For this reason,many academics believe that empirical evidenceshould be offered before any theory is considered a

Ž‘good’ theory Whetten, 1989; Poole and Van De.Ven, 1989 .

6. The two objectives of research: some importantobservations

The two general objectives of research aretheory-building and fact-finding. The difference be-tween these two objectives is grounded in the pur-pose of the research. ‘Good’ theory-building re-search’s purpose is to build an integrated body ofknowledge to be applied to many instances by ex-plaining who, what, when, where, how and whycertain phenomena will occur. On the other hand,‘good’ fact-finding research’s purpose is to build alexicon of facts that are gathered under specifiedconditions. The division line between the two typesof research is quite fine since both types of researchusually include data gathering and empirical estima-tion. However, the contrast between the two pur-poses comes from two areas: how and when the apriori predictions are made; and how the results areintegrated with other studies.

First, theory-building research carefully definesconcepts, states the domain, explains how and whyrelationships exist, and then predicts the occurrenceof specific phenomena. After the prediction, it typi-cally gathers evidence to see if the phenomena oc-curs. ‘Good’ fact-finding research also carefully de-fines concepts and states domains. Then, ‘good’fact-finding research uses evidence to discover ifrelationships exist. Next, ‘good’ fact-finding researchthen explains how and why specific phenomena oc-curred. In the formal sense of the definition oftheory, fact-finding research is not theory-buildingresearch since the evidence is gathered before therelationships are explained and before the relation-ships are predicted. Therefore, fact-finding researchdoes not meet the conditions of ‘good’ theory-build-ing, since it lacks a priori explanations and predic-tions before the data are gathered.

Even though fact-finding research is not theory-building research, it serves an important purpose byproviding facts that can later be integrated into atheory. ‘Good’ fact-finding research serves to pro-vide fertile ground for subsequent new theory-build-ing. Fact-finding research is not constrained by exist-ing theory-based relationships and, therefore, canmore easily investigate new relationships that canlater be integrated into a theory. However, to be ofvalue for ‘good’ theory-building, ‘good’ fact-findingresearch must follow the first two conditions fortheory, since without precise definitions and well-de-fined domains, facts are not meaningful for anyfuture use. Because fact-finding research investigatesrelationships and then offers explanations as to whythe results occurred, fact-finding research suggestsnew relationships for new theory development. Dueto the freedom from the constraints placed ontheory-building research, fact-finding research fre-quently provides the basis for new theory develop-ment. Consequently, there is a large measure of truth

Ž . Žin the statement of Lindblom 1987 statement made.at the beginning of this study that many new theo-

ries come from fact-finding investigation.A second contrast between theory-building and

fact-finding research is the degree of integration ofthe results with other studies. This degree of integra-tion has two dimensions: the internal integration ofall the empirical data and the integration of theresults with other studies. Empirical evidence used in


theory-building research stresses subtle systematicsimilarities between all data to raise the theory’sabstraction level. These systematic similarities pro-vide additional factors to explain all data in oneintegrated framework applied to diverse environ-ments. By using one theory across many differentenvironments, theory-building research raises the ab-straction level to explain how and why the theorycan be applied to predict events. When the empiricaldata do not support the theory, the reason for thelack of support is important for the further develop-ment of the theory. If the theory is logically inter-nally consistent, then the reason for the lack of

Ž .support must be a missing explanatory factor s . ThisŽ .explanatory factor s is introduced into the theory

and then the theory is examined for internal consis-tency. Once the ‘new’ theory is internally consistent,then the theory can be empirically tested. This se-quence of empirical tests contradicting the theory,

Ž .introducing new factor s , building a ‘new’ internallyconsistent theory, and empirical testing of the ‘new’theory is a common theory-building sequence fortheory building. This sequence has the goal of build-ing a single integrated theory to explain all data. Inshort, when empirical tests do not support a theory,theory-building research examines the theory to inte-grate new factors to raise the abstraction level. Con-sequently, ‘good’ theory-building research is alwaysstriving to find integrating factors to expand thetheory to apply to diverse environments, therebyraising the theory’s abstraction level.

In contrast, fact-finding research typically stressesthe descriptive differences in data. These descriptivedifferences exist because all data, to some degree,are different. Fact-finding research attempts to dis-cover differences in data and explains these differ-ences. Unfortunately, without a theory to be tested inthe empirical world, many times the explanations ofdescriptive differences are not specifically tested.Even more unfortunate, is that they may not beexplanations at all. Consider a case where a re-searcher finds that Japanese manufacturers whencompared to USA manufacturers, have statisticallysignificantly higher manufacturing performance onsome performance measures. The basic fact indicatesthat Japanese manufacturing managers are superiorto USA manufacturing managers on these perfor-mance measures. However, this fact does not offer

Ž .adequate or any explanation of how or why thisresult occurred since the source of the difference

Žcould be better planning forecasting, and resource. Žscheduling ; control forecast control, production ac-. Žtivity control ; advanced technology better informa-

. Žtion technology, CAD, FMS, etc. ; culture work.ethic, managerial behavior, etc. or numerous other

causes; all of which were not specifically tested. Inshort, fact-finding research frequently explains de-scriptive differences, but because the explanation isnot specifically tested, any inferences andror con-clusions are deceptive. For this reason, these expla-nations are deceptive descriptive differences.

One type of integration used in theory-buildingresearch is the identification of subtle systematicsimilarities across different studies. For example,

Ž .Ettlie 1995 noted the empirical similarities betweenseveral new product development research studiesŽMadique and Hayes, 1984; Crawford, 1987; Kekreand Srinivasan, 1990; Klemschmidt and Cooper,

.1991 . The similarities he deduced were between thetwo extremes on product development. At one ex-treme, firms which introduce variants of existingproducts reap significant benefits from increasedmarket share and profits. At the other extreme, henoticed that firms which introduce totally new prod-ucts are also reap significant benefits from increasedmarket share and profits. Yet, firms that attemptedan intermediate strategy were not successful. Hence,

Ž .he concluded, ‘‘ T he relationship between productinnovativeness and success in the market place isu-shaped.’’ His integration of the three studies intoone theory is accomplished by identifying the subtlesimilarities between different studies. This integra-tion across several studies raises the abstraction level,therefore increasing the importance of his study’sfindings.

In short, researchers are faced with whether tofind subtle systematic similarities or to explain de-ceptive descriptive differences between individuals,organizations, businesses, industries, and countries.Fact-finding research focuses on descriptive differ-ences among data, while theory-building researchconcentrates on the underlying factors for similari-ties. Theory-building research raises the abstractionlevel by integrating subtle systematic similaritiesacross the descriptive dimensions of individuals, or-ganizations, businesses, industries, and countries.


Consequently, from the standpoint of ‘good’ theory-building, it seems that systematic similarities aremore important than descriptive differences.

7. Types of theory-building research

The purpose of this section is to classify thetheory-building types of operations management re-search by their theory-building purpose. There are

Žtwo major classifications of research: analytical for-.mal and empirical. These differences are best ex-

Ž .pressed by Sax 1968 :

When an organized body is based primarily on de-ductive rules, it is called formal science, to distin-guish it from those areas of knowledge that dependprimarily upon empiricism and induction . . . calledempirical science. Considered from this point ofview, mathematics, logic, and library science areprimarily formal sciences, whereas chemistry, psy-chology, and education are primarily empirical sci-ences.

Ž .This statement means that analytical formal sci-ences use deductive methods to arrive at theorieswhile empirical sciences use induction methods toarrive at theories.

This section uses these traditional major classifi-cations of analytical and empirical theory-buildingresearch and further divides them into three sub-cate-gories for each major classification. The categoriza-tion of the research types serves two purposes: itillustrates the need for different research methodolo-gies to raise abstraction levels of theory and, itillustrates that the ‘good’ theory-building procedures

Žare applicable to different types of research dis-.cussed in the Section 8 . The first purpose for the

classification is based on the goal of the ‘good’theory-building research. This study argues that thereare six different types of theory-building research,and each type serves to develop and verify theoryusing different research methodology. This theoryverification usually is called triangulation and isextremely important for the final verification of the-

Ž .ory Meredith et al., 1989 . Ideally, if a theory istested by all six methods with supporting results,then these results would be compelling evidence tobelieve that the theory is confirmed.

7.1. Analytical research

The analytical research method uses deductiveŽ .methods to arrive at conclusions Swamidass, 1986 .

Analytical research methods primarily use logical,mathematical, andror mathematical–statistical meth-ods. The three different sub-categories of analyticalresearch have subtle differences in how they uselogic and mathematics for the development of thetheory. Additionally, the analytical sub-categorieshave different theory-building purposes.

7.1.1. Analytical conceptual researchFrom a theory-building perspective, the purpose

of analytical conceptual research is to add new in-sights into traditional problems through logical rela-tionship-building. This research methodology com-prises new insights through logically developing re-lationships between carefully defined concepts intoan internally consistent theory. These studies usuallyuse case study examples to illustrate these conceptu-alizations. There are several types of research in thissub-category. One example is called introspectiveresearch which uses the researcher’s experience toformulate concepts. Consequently, it describes andexplains relationships from past experience to de-velop theory. A second example in this sub-categoryis conceptual modeling. This modeling is where amental model of deduced relationships is posited,which may then be evaluated using a framework thatcaptures the essence of the systems under investiga-tion. A third example in this sub-category ishermeneutics research which deduces facts from what

Ž .is being observed Meredith et al., 1989 .A representative article of analytical conceptual

Ž .research is the Kim and Lee 1993 article where theresearchers logically tied together manufacturingstrategies and production systems. In their study,they logically integrated technical flexibility, techni-cal complexity, and production system type withdifferentiation and cost efficiency strategies. Thistype of research falls into the analytical conceptualsub-category since the methodology used is logicaland does not use empirical data for theory develop-ment.

7.1.2. Analytical mathematical researchThe theory-building purpose of this research sub-

category is to develop sophisticated relationships


between narrowly defined concepts through develop-ing new mathematical relationships to study how themodels behave under different conditions. These arti-cles mathematically develop the relationships andgive numerical examples from their derivations orcomputations. Analytical mathematical research doesnot use external data to test the theory, but insteaduses deterministic or simulated data to draw conclu-sions. The research in this sub-category is sometimescalled ‘operations research’ andror ‘managementscience’.

There are many types of research in this sub-cate-gory: reasonrlogical deductive theorem proving;normative analytical modeling research; descriptiveanalytical modeling; proto-typing and physical mod-eling research methods; experimentation; and mathe-matical simulation. In all these methods, the modelsusually are built using formal logic and tested usingartificially developed data. Since the data is derivedartificially, all these methods are classified under the

Žanalytical mathematical research method Meredith.et al., 1989 .

7.1.3. Analytical statistical researchThis research sub-category integrates

logicalrmathematical models from analytical re-search and statistical models from empirical research

Ž .into a single integrated theory Moorthy, 1993 . Theanalytical statistical research is different from theanalytical mathematical method since its models areexplicitly developed for future empirical statisticaltests. This methodology uses large bodies of knowl-edge and integrates them into a single model forfuture empirical tests. Typically, the variables usedfor investigating relationships have measurement er-rors due to random variability caused by the externalworld. In sum, the purpose of analytical statisticalresearch is to provide larger, more integrated modelsfor empirical statistical testing.

7.2. Empirical research

In the empirical research major classification, themethodology must use data from external organiza-tions or businesses to test if relationships hold in theexternal world. Empirical research methods could beclassified more correctly as ‘real world’ empiricalmethodologies. However, since this is a fairly longphrase, it is abbreviated to empirical methods.

7.2.1. Empirical experimental researchThe theory-building purpose of this sub-category

is to investigate relationships by manipulating con-trolled treatments to determine the exact effect onspecific dependent variables. The research designuses treatment variables that are manipulated to de-

Ž .termine their effect on the dependent variable s .Because direct manipulation of the treatment vari-ables causes direct effects on the dependent vari-ables, this research sub-category comes the closest todemonstrating causality between variables. Thissub-category of empirical research is also called

Ž .‘field experiments’ Meredith et al., 1989 .The empirical experimental research methodology

is difficult to implement in operations managementsince the environment must be closed to contamina-tion effects. In operations management, the systemfrequently is an open system and therefore can besubject to contamination effects. Yet, it seems possi-ble that controls could be placed on some experi-ments to determine if one treatment causes a certainresult. For example, a researcher may find an organi-

Ž .zation that is about to install a just-in-time JITsystem. If the facility has two different lines ormanufacturing areas that are separate, the researchermay experiment with the amount and type of trainingto determine the effect of training on selected perfor-

Ž .mance measures throughput time, defect rates, etc.This type of experiment could offer convincing evi-dence of the necessity or non-necessity for specificamounts or types of training needed for JIT imple-mentation.

This research type is not to be confused with theexperimental design in mathematical simulationfound in the analytical mathematical sub-category

Žabove where the data are developed by the re-.searcher in a closed simulated environment .

7.2.2. Empirical statistical researchThis research sub-category’s purpose is to empiri-

cally verify theoretical relationships in larger sam-ples from actual businesses. Generally, the morecomplex the research issues are, the more likely thestudy will use this methodology. A typical topicwould be: manufacturing strategy’s effect on manu-facturing performance.

There are many types of research that fall into thisŽsub-category. Structured and unstructured elite and


.intensive interviewing processes which gather datafor statistical analyses are examples of this sub-cate-gory of research. Other representative examples aresurveying and historicalrarchival research, expertpanels, and Delphi techniques. Each of these meth-ods has the goal to statistically analyze data fromrelatively large external samples. Therefore, from atheory-building perspective, this methodology offersempirical support for theoretical relationships in

Ž .larger samples in real world Meredith et al., 1989 .

7.2.3. Empirical case studyThe purpose of this type of research is to develop

insightful relationships within a limited set of com-panies. By limiting the number of companies investi-gated, this research method investigates small sam-ples using a large number of variables to identifynew empirical relationships. Frequently, this researchmethod analyzes organizations across time and pro-vides the dynamic dimension to theory to elevate thetheory’s abstraction level. This sub-category includes

Žfield studies and action research Meredith et al.,.1989 .

A typical research article of this sub-category isŽ .the article of Marucheck et al. 1990 on manufactur-

ing strategic practice, which examines the strategicpractices in six firms. Using the practices of the sixfirms, they found that firms follow the general con-ceptual models developed in the academic literature.

Ž .Additionally, they stated ‘‘.. T he real benefits ofmanufacturing strategy come from implementing the

Ž .manufacturing strategy’’ p. 121 . The research pro-Ž .ject of Marucheck et al. 1990 analyzed data from a

limited number of firms to identify manufacturingstrategy procedures. Since a limited number of firmsare utilized to identify possible theories, this projectis classified as empirical case study. Hence, the keydifference between the empirical case study methodand the analytical conceptual method is that theempirical case study method uses data to form thetheory and the analytical conceptual method usesdeduction to form theories.

7.3. An important conclusion

From the standpoint of good theory, one impor-tant conclusion drawn is that no single researchcategory or sub-category is superior to any other

research category or sub-category. Each methodserves a very different, but important, purpose fortheory development in operations management. Thebasic purpose of all three types of analytical method-ologies is to develop logically internally consistenttheories and models. First, the analytical conceptualresearch type serves as a forum for expressing newconceptual perspectives on theory to better explainand integrate underlying relationships. The sub-cate-gory of analytical mathematical research serves tologically evaluate the internal consistency of com-plex relationships. The analytical statistical researchserves to integrate both the analytical mathematicalresults and the empirical statistical results into alarger theoretical body of knowledge for statisticalestimation. In short, all these analytical researchsub-categories serve to develop internally consistenttheories through logical analyses.

The empirical methodologies provide empiricalverification of models, while offering evidence forthe development of new theory. The empirical exper-imental research uses experimental design to verifythe causality of a specific theory while elevatingrelationships from a testable hypothesis to an empiri-cally verified theory. A verified theory may not needto be tested in future research projects and may beassumed to be a fundamental law. The empiricalstatistical research methodologies verify models fortheir empirical validity in larger populations to re-duce the number of relationships in future research.The empirical case studies provide new conceptualinsights by empirically investigating individual casesfor an in-depth understanding of the complex exter-nal world.

Table 3 provides a summary of the classificationof research along with its relative importance tooperations management. Basically, theories devel-oped using the analytical research methodologies arerefuted by empirical evidence, while theories devel-oped using the empirical research methodologies arerefuted by internal consistency. The last row of thetable states the theory-building purpose of the re-search method. It would be difficult to disagree withthe importance of any of these six sub-categories,since it would be paramount to stating that opera-tions management does not need: new conceptualtheory; internal mathematical consistent theory; inte-grative theory; causal verification of theory; large

()

J.G.W

ackerrJournalof

Operations

Managem

ent161998

361–

385376

Table 3Specific research sub-category, refutation methods, and importance to operations management theory-building

Analytical Empirical

Conceptual Mathematical Statistical Experimental design Statistical sampling Case studies

Types of research in- Futures research sce- Reasonrlogical theo- Mathematical statisti- Empirical experimen- Action research struc- Field studies, casecluded narios, introspective rem proving, norma- cal modeling tal design, descriptive tured and unstruc- studies

reflection, hermeneu- tive analytical model- analytical modeling tured research, sur-tics, conceptual mod- ing, descriptive ana- veying, historicaleling lytical m odeling, analysis, expert pan-

proto-typing, physical elsmodeling, laboratoryexperiments, mathe-matical simulation

Refutation methods Empirical data from Empirical data from Empirical data from Analyticalrlogical in- Analyticalrlogical in- Analyticalrlogical in-empirical methods empirical methods empirical methods consistency consistency consistency

Importance to opera- Develops new logical Explores the mathe- Integrates the other Tests and verifies Tests the theory by Tests and developstions management relationships for con- matical conditions un- five methods into a causal relationships investigating statisti- complex relationshipstheory-building ceptual models of the- derlying the relation- single theory for em- between variables cal relationships to between variables to

ory ships used in theory- pirical investigation verify their existence suggest new theorybuilding in larger populations


population verification of theory; and new empiricalrelationship exploration for theory development.Consequently, all these methods are extremely im-portant for the complete development of theory-building in operations management.

8. The need for similarity in theory-building re-search procedures

This section suggests that general research proce-dures should apply to all research methodologies iftheory-building is to be important for operationsmanagement. If all research methodologies followcommon definitions, common domains, common re-lationships, and common predictions, the integrationacross the six different sub-categories of theory-building methodologies would be more likely.

Table 4 provides an outline of the procedures forthe different theory-building methodologies. All sixmethodologies have the same four stages in theprocedure, since these stages map directly to thedefinition of theory. In the first stage, it is importantthat the conceptual definitions used in the analyticalmethodologies are the same definitions that are oper-ationalized in the empirical methods. It does notseem logical for analytical methods to use concep-tual definitions for their mathematical convenience ifthese definitions have no hope of ever being opera-tionalized for measurement in empirical studies.Therefore, analytical researchers should give mea-surement guidelines to empirical researchers. Con-versely, it is important for empirical methods not toredefine an existing conceptual definition for thesake of measurability. From the theory-building per-spective, ideally, all six methodologies would use thesame conceptual definitions for the same concept sothat analytical and empirical evidence are buildingon the same theory. Therefore, for theory-building,the first step in theory-building research should beidentical regardless of the sub-category of researchpursued.

In the second stage of theory-building research,the domain of the theory is defined. In this stage, allthree types of analytical methodologies should spec-ify where their theories apply. Generally, thesemethodologies have broader domains than empiricalmethodologies. Yet a common challenge to the ana-

lytical methodologies is a cautious domain specifica-tion to identify exactly where the theory is to beapplied. Without this specification, it is assumed thatall domains are included, which causes these theories

Ž .to be questioned or refuted by a single empiricalobservation from any domain. For empirical method-ologies, the domain usually is carefully specified by

Ž .the research design empirical experimental design ,Ž .sampling method empirical statistical methodology ,

Žor by the specific case studied empirical case study.methodology . From the perspective of theory-build-

ing, ideally, the domains suggested by the analyticalmethods would be investigated by the empiricalmethods to verify relationships to raise the abstrac-tion level.

In the third stage of theory-building research pro-cedures, relationships are suggested for theory-build-ing. In all three of the analytical methodologies,logic is used to determine the relationships. Theanalytical methodologies use fundamental laws todeduce derived laws to suggest which relationships

Ž .are logically compatible internally consistent witheach other. Empirical methodologies should test ana-lytically developed relationships before offering newrelationships since new relationships may not beinternally consistent. Empirical methodologies shouldexhibit great care before offering new relationshipssince some relationships may have already beenidentified using analytical methods. If an empiricalstudy fails to examine the analytically derived rela-tionships and the study results are contrary to theanalytical relationships, the empirical results are con-sidered artifacts, since the ‘power of deduction rules’Ž .Hunt, 1991 .

The last stage of theory-building research is theprediction and verification. ‘Good’ theory-buildingsuggests that operations management theory shouldoffer internally consistent predictions in the ‘realworld’ empirical realm. This suggestion means thatanalytical methodologies should use bridge laws aspart of their methodology to provide the means totest their theories’ empirical predictions in the ‘real’world. This suggestion is not to be interpreted asmathematical models are not needed in operationsmanagement, but rather, these models should offersome hope of improving operations management asit is practiced. To develop practical models, it seemsthat offering ‘how’ these results will be tested in the

()

J.G.W

ackerrJournalof

Operations

Managem

ent161998

361–

385378

Table 4The theory-building procedure for different theory-building types of research

Theory-building pro- Analytical Empiricalcedure Conceptual Mathematical Statistical Experimental design Statistical sampling Case studies

Definitions develop- Conceptual defini- Conceptual defini - Conceptual defini - Conceptual defini - Conceptual defini - Conceptual Defini -ment tions. However, many tions from the litera- tions from the litera- tions. However, fre- tions from the litera- tions from the litera-

times, new definitions ture. ture. quently the method ture. However, many ture. However, manyare offered. may require new, times, new constructs times, new relation-

more measurable con- are developed to rep- ships require new def-cepts. resent the theoretical initions.

concept.Domain limitations Logically developed Mathematically de - Mathematically and Experimental design From analytical statis- Developed from cases

veloped statistically developed with a narrow con- tical models or devel- studied.trolled domain. oped experimentally

Ž .Relationship model Usually, relationships Usually, relationships Usually, other studies Usually developed Usually using other The combining of thebuilding are logically devel- are mathematically are used to develop with limited relation- statistical studies’ relationships discov-

oped. developed without mathematical statisti- ships between vari- suggested theories. ered from the case.stochastic error terms. cal models with error ables.

terms.Theory predictions: Usually, predictions Mathematically de - Mathematically and Prediction from the Prediction from other Supported by caseevidence of predic- come from logical duced predictions. logically derived pre- experimental design. studies. Results from studiestion analyses. Empirical Examples from math- dictions. Uses empiri- Statistical signifi- sample’s statistical

evidence comes from ematical calculations cal evidence from cance of the tests tests for significance.case studies. or simulated results other studies.


empirical realm is a valuable aid for ‘real world’empirical testing.

In summary, for each stage of the theory-buildingprocedure, there is a need for integration of theresults from other research methods to raise theabstraction level of operations management theory.To be a theory-building study, each research projectshould not ignore results from other methodologies,since ignoring other studies lowers the abstractionlevel and reduces the significance of any findings.

9. The current state of theory-building method-ologies in operations management

One question that should be answered is: how arethe different research methods in operation manage-ment being utilized? Table 5 illustrates the currentstate of theory-building in operations management

Ž .over 5 years 1991–1995 inclusively . The top eightacademic and practitioner journals in operationsmanagement provide data for investigating the cur-

Žrent state of operations management research seeŽ ..Barman et al. 1993 . These journals are the Interna-

tional Journal of Operations and Production Manage-Ž . Ž .ment IJOPM ; Decision Sciences DS ; Journal of

Ž .Operations Management JOM ; Management Sci-Ž .ence MS ; Production and Operations Management

Ž .Journal POM ; International Journal of ProductionŽ . Ž .Research IJPR ; Harvard Business Review HBR ;

Ž .and Production Inventory Management PIM . Onecaveat is needed before any further discussion: thisarticle did not investigate the relative degree oftheory-building of each article but rather concen-trated its efforts on classifying each article into thesub-categories of research methodologies.

There are two deliberative ‘judgment calls’ inthese classifications. The ‘first judgment call’ ismade when classifying each article in terms ofwhether its primary contribution is to the operationsmanagement literature, or to another academic field.This study uses a broad perspective to include abroad spectrum of topics from the traditional list of

Žoperations management topics see Hahn et al.Ž ..1982 . Of the 2817 articles reviewed, only 2002 areclassified as operations management. Almost all ofthe articles in the operations management journalsŽInternational Journal of Operations and Production

Ž .Management IJOPM , Journal of Operations Man-Ž .agement JOM , Production and Operations Manage-

Ž .ment POM , International Journal of ProductionŽ .Research IJPR , and Production Inventory Manage-

Ž .ment PIM are designated as contributions to theŽoperations management literature It is not surprising

to operations management academics that DecisionŽ . Ž .Sciences 43% , Management Science 35% , and

Ž .the Harvard Business Review 17% all have signifi-cantly less than 50% of the articles related to tradi-

.tional POM topics. See Table A.1 in Appendix A. .The ‘second judgment call’ is the sorting of arti-

cles into the six major categories. This sorting re-quired considerable judgment since studies fre-quently use multiple research methodologies. In thisstudy, the classifying procedure concentrated on thepredominant methodology used in the study. For

Žexample, the productivity article not included since. Ž .pre-1991 of Hayes and Clark 1989 would be

considered a cross between empirical and analyticalresearch since it uses statistical samples along within-depth case studies to logically derive a conceptualproductivity theory. However, when considered moreclosely, the article’s primary methodology uses sev-

Table 5Overall classification of the articles by research sub-category

Classification 1991 1992 1993 1994 1995 Total

Ž . Ž . Ž . Ž . Ž . ( )Analytical conceptual 103 27.39% 110 26.63% 100 23.58% 114 27.01% 63 17.17% 490 24.48%Ž . Ž . Ž . Ž . Ž . ( )Analytical mathematical 221 58.78% 214 51.82% 246 58.02% 218 51.66% 206 56.13% 1105 55.19%Ž . Ž . Ž . Ž . Ž . ( )Analytical statistical 2 0.53% 5 1.21% 5 1.18% 3 0.71% 7 1.91% 22 1.10%Ž . Ž . Ž . Ž . Ž . ( )Empirical experimental 2 0.53% 4 0.97% 5 1.18% 2 0.47% 2 0.54% 15 0.75%Ž . Ž . Ž . Ž . Ž . ( )Empirical statistical 28 7.45% 41 9.93% 34 8.02% 45 10.66% 56 15.26% 204 10.19%Ž . Ž . Ž . Ž . Ž . ( )Empirical case study 20 5.32% 39 9.44% 34 8.02% 40 9.48% 33 8.99% 166 8.29%

Total 2002


eral case studies over time to develop the factoryproductivity theory. Consequently, it would havebeen classified in the empirical case study sub-cate-gory.

Classifying articles into the six research sub-cate-gories addresses the question of: ‘‘How are thearticles distributed into research sub-categories overthe last 5 years?’’ Using Table 5, it is apparent thatover the last 5 years over 55% of all operationsmanagement’s published articles use the analytical

Žmathematical method see Table A.2 in Appendix A.for the breakdown by journal . The next most popu-

lar methodology is analytical conceptual, followedby empirical statistical, and empirical case study.Both empirical experimental and analytical statisticalare not popular methodologies in the operationsmanagement literature. For the academic journals,

Žfive of the journals Management Science, Interna-tional Journal of Production Research, Decision Sci-ences, Production Operations Management Journal

.and the Journal of Operations Management had amajority of articles following the analytical mathe-matical methodology. Only the International Journalof Operations and Production Management has itslargest percent of articles in any other methodologyŽ .analytical conceptual was the most popular . As one

Žmight expect, the broader based journal Harvard.Business Review and the practitioner-oriented jour-

Ž .nal Production and Inventory Management had ahigher percentage of their articles classified intoanalytical conceptual and empirical case studies.However, the fact remains, that in the respectedacademic journals of operations management, thepredominant research methodology is analyticalmathematical.

One important implication of these results is thatempirical experimental design and analytical statisti-cal methodologies are not being used to verifycausality and integrate theory. The implications ofthis result leads to two important constraints fortheory-building in operations management. First, be-cause experimental design is not being used, there isa lack of studying causality in the empirical worldwhich hinders the development of verified relation-ships that are assumed to be true. This lack ofverified relationships means that all relationshipsmust be tested through empirical testing, causingstatistical models to be too complex for meaningful

Ž .investigation Hunt, 1991 . Second, because analyti-cal statistical methodology is under-researched, thereis a lack of an integrated internally consistent theoryacross all methodologies used in operations manage-ment. Since analytical statistical methods’ basic pur-pose is to develop integrated models across all re-search methods, it provides a critical linkage toelevate theory to higher-level abstraction levels. Inshort, the evidence presented here suggests that oper-ations management is not fully utilizing all researchmethods to verify old relationships and to buildintegrated theory.

10. How this article follows the criteria for ‘good’theory-building

This article developed a definition of theory fromthe academic literature. The definition refined theconditions for what constitutes a theory. Next, itlimited the domain to ‘good’ theory using the tradi-tional virtues of ‘good’ theory. Third, it suggested aprocedure for all types of theory-building researchwhich fulfills the definition of theory. Additionally,this study classified research into six categories anddemonstrated that all sub-categories of theory-build-ing research in operations management use the samestages. Finally, it presented evidence that the sixpredominant types of theory-building research in op-erations management are not all equally distributed.This unequal distribution led to the conclusion thatresearch on causality and theory integration are hin-dering the elevation of the abstraction level of opera-tions management theory.

11. Guidelines for theory-builders and conclusions

The purpose of this section is to offer theory-building guidelines for theory-builders. It is impor-tant for the academic field to follow the formaldefinition of theory for theory-building research andincorporate these elements: definitions, domain, rela-tionships, and predictions. Consequently, the follow-ing is a set of questions to provide guidelines fortheory-builders.

Ž .1 For definitions: are the terms used in the studyŽ .standard artificial terms agreed upon by aca-

demics? Are the terms unique? Are new terms care-


fully differentiated from the standard terms in exis-Ž .tence conservation criteria ? Do all the new and

standard terms avoid ‘concept stretching’? Do all theterms used carefully exclude other concepts? Aredifficulties with measurement of definitions high-lighted for future researchers?

Ž .2 For domains: are the specific conditions ofwhen and where the data were gathered carefullyenumerated? Are conditions for when and where theresults apply prudently articulated? Is the domain ofthe results wide enough to be of value for re-searchers using other research methodologies?

Ž .3 For relationship and model building: are allthe variables listed in the estimates necessaryŽ .parsimony criteria ? Do all the variables used

Žspecifically state how and why they are related or.unrelated to each and every other variable in the

Ž .study internal consistency criteria ? Is the model asŽsmall as possible to explain all the results parsimony

.criteria ? Do the relationships offer new areas forŽ .research fecundity criteria ? Are all the relationships

specified before the data is gathered and before theŽrelationships are estimated internal consistency and

.empirical riskiness criteria ? Does the model elevatethe level of abstraction by adding variables and

Ž .inter-relating theories abstraction criteria ?Ž .4 For theory prediction: does the model used

make specific predictions which could likely be fal-Ž .sified empirical riskiness criteria ? Do the theory

predictions prohibit some events from happening?Does the theory discuss specifically how the theoryis to be used and tested in the external empirical

Žworld i.e., specifying bridge laws which are needed.for empirical riskiness ?

Although all these guidelines are necessary for alltheory-building articles, there are four specific con-cerns for the full development of operations manage-ment theory. First, there is a need for cautiousattention to the measurement of the defined terms,since without exacting definitions, all theory is tenu-ous at best. Second, the specific domain of when andwhere the theory is to be applied is needed for thetheory to be empirically tested. Third, the relation-

ship-building stage is most important for empiricalresearchers since all variables should specific

a definition of theory: research guidelines for different theory-building … · journal of...

Documents