1

European Journal of Information Systems 7 (2), 1998, pp. 90-107

Diversity in Information Systems Action Research Methods

Richard BaskervilleDepartment of Computer Information SystemsGeorgia State UniversityPost Office Box 4015Atlanta, Georgia 30302-4015, USATel. +1 (404) 651-3880, Fax +1 (404) 651-3842Internet baskerville@acm.org

A. Trevor Wood-HarperInformation Research CentreUniversity of Salford, Salford, Manchester, M5 4WT, England

AbstractDiscussions of action research in information systems often proceed as if there were one definitive action researchmethod. This paper describes and analyzes the different frameworks, assumptions and goals that characterise thediverse forms of action research. A more inclusive action research paradigm is delineated that offers a basis forvalidating a wider range of IS research. Acceptance of the full range of the diverse forms of action research may enablethe information systems field to be more explicit about its research methodology, thereby enabling criteria to beimproved and applied to a broader range of information systems research.

Introduction

Discussions of qualitative research methods ininformation systems (IS) are typically illustrated withcase study research methods (Benbasat 1987, Lee1989), ethnographic research methods (Agar 1986,Tanton & Fox 1987, Fetterman 1989), and actionresearch methods (Kaiser & Bostrom 1982, Wood-Harper 1985). Discussions of these methods oftenproceed as if there were one definitive case method,ethnography method, or action research method.However, each of these terms reference a genre ofresearch methods. While different forms of case studyapproaches have been explored (Yin 1989), anddifferent forms of ethnography have been discussed(Agar 1986) we are only beginning to develop similarcomparative studies of the various forms of actionresearch methods (Lau 1997).

The discipline of information systems seems to bea very appropriate field for the use of action researchmethods. IS is a highly applied field, almost vocationalin nature (Banville & Landry 1989). Action research

methods are highly clinical in nature, and place ISresearchers in an "helping role" within theorganizations that are being studied (cf. Schein 1987, p.11). It should not be surprising that action research isthe "touchstone of most good organizationaldevelopment practice" and "remains the primarymethodology for the practice of organizationaldevelopment" (Van Eynde & Bledsoe 1990, p. 27).Action research merges research and praxis thusproducing exceedingly relevant research findings. Suchrelevance is an important measure of the significance ofIS research (Keen 1991).

There is disagreement about the action researchparadigm for information systems. Authorities variouslydefine patterns or models of action research that aremore or less exclusive. This disagreement not onlyreflects a broader social scientific disagreement over theparadigm, but also confusion about the IS actionresearch tradition. This confusion arises in concernsabout excluding certain IS research (e.g., consulting)that "falsely" claims to be action research (Jonsson1991), along with concerns about including traditional

2

action-based IS research (such as socio-technicalinquiry) that is "implicitly" action research, but doesnot explicitly lay claim to fit the action researchparadigm (Baskerville, et al. 1997). The issue revolvesaround definitional details about the term "actionresearch" that may eventually help determine tworelated issues: agreement about establishing the validityof certain forms of action-based research; and the futureexclusivity or inclusivity of the scholarly action-basedliterature in IS.

The two issues are related because the validityissues help determine whether research methodologyand findings are sound enough for publication inscholarly journals. If researchers are able to presentsound and convincing arguments for the validity oftheir findings, the chances for circulation of their ideasis greatly improved. By improving the inclusivity of theaction research paradigm, we believe a broader range ofresearch activities will fall under its validity criteria,and the openness of the IS scholarly literature will beimproved. We will show how different forms of actionresearch have different models, different structures anddifferent sets of goals. In the past, overly-exclusivedefinitions of action research created a problem incomparing the rigor and results of different actionresearch studies which have adopted different actionresearch approaches. The goals and success of aresearch project could be misinterpreted because thereader is unfamiliar with the particular form of actionresearch being conveyed. Furthermore, highly practicalsegments of published research in the field ofinformation systems simply ignore the researchmethods issue. Some of this research falls within theboundaries of an inclusive action research paradigm,and implicitly meets the criteria by which such researchwould be considered valid.

However, it is not just the readers who wouldbenefit from an inclusive action research paradigm forIS. It is likely that many researchers conduct validaction research without fully understanding how theirwork fits one of the action research forms, or thephilosophical assumptions that underlie the technique.This lack of understanding limits their ability todemonstrate the validity of their findings, andconsequently limits the wider distribution of theirknowledge.

The purpose of this paper is to describe and analyzethe different models, structures and goals in the variousforms of action research. This description extends theaction research paradigm to embrace the diverse forms

of research that involve purposeful researcherinterventions within the uncontrolled social arena of theinvestigated phenomena. The contribution of thisextension is the ability to classify and validate researchfindings from a segment of the IS literature that oftenarises from IS practice, the validity of which may beerroneously assumed to be unprovable.

The paper is organised into five sections. Followingthis introduction, the second section clearly defines theboundaries of research methods that may be consideredpart of the action research paradigm. Section threedescribes the intellectual genealogy of various streamsof action research thinking, and positions informationsystems action research within this genealogy. Thefourth section describes the essential characteristics ofthe various forms of IS action research. The fifthsection discusses the implications of this description forresearch in the IS field.

Boundaries of an IS action research paradigm

We recognised that action research has particularlyimportant implications for IS research. We alsorecognise that an inclusive boundary for action researchmay benefit the field of IS. In this section we willdevelop an action research paradigm or model througha set of characteristics that defines a more inclusiveboundary for IS action research. This set differs in someimportant ways from the boundaries as defined by someauthors in the social sciences. In particular, thecharacteristics below are oriented toward the process ofaction research, rather than the goals of the researcher(see the appendix for a detailed discussion of thedifferences between these characteristics and keycharacteristics from the social science literature).

Action research is a cognitive process that dependson the social interaction between the observers andthose in their surroundings. The reactive process ofstimulus-response provides a filter by which some of themyriad social actions can be connected into a causalmodel. When a certain action is taken in a particularsocial setting, a response is recorded. In its broadestsense, action research resembles the act of researchersconducting a highly unstructured field experiment onthemselves together with others.

The term "participant observation" is being used intwo senses in the literature. In one sense, it is claimedas a complete research method (discussed later in thispaper). In a second sense, the term regards onlyresearch techniques for participatory data collection

3

within a qualitative research method like a case study oran ethnography. To avoid confusion, we will use theterm "participatory observation" for the data collectiontechnique, and "participant observation" for theresearch method.

The stimulus-action sequence is an alternative topassive observation because the filter for critical dataabout the event proceeds naturally from the event itself,and is not imposed a priori or a posteriori by theresearchers. Passive observation of social settings placesgreat demands on the observer because of the enormouscomplexity of social events. In any particular socialsetting, there are an enormous number of speech acts,gestures, movements and other clues with which theevent might be interpreted. Some structure is necessaryin order to filter these myriad clues to discover thosethat are relevant to the development of a theoreticalframework. Passive observation filtering either requiresan a priori framework, such as a classification schemefor speech acts, or an a posteriori framework, such asgrounded theory categories. With action research, thefilter is defined by the state change represented in thestimulus-reaction pairs. A certain action is taken in asocial setting and the social setting changes state.Action research observes the social setting in motionafter a defined event. This motion provides the filter forcritical data in action research: things that changedafter the event.

Unlike other experimental methods, however,action research rests in an interpretive philosophicalframework (Susman & Evered 1978). The state-changedata is only relevant to the particular social setting. Thesame stimulus in a different social setting may yield adifferent response because of the multivariate nature ofhuman social interaction. Consequently the state-change response still must be interpreted in the socialcontext. For the knowledge to be generalised in a broadtheory, the impact of different social contexts must beconsidered. From a practical viewpoint, this means thatthe reader must have enough information about theexact social setting in which the stimulus-response wasobserved such that the stimulus may be revised oradjusted in order to make sense in the reader's ownsocial setting.

In practice, the essence of action research is asimple two stage process Blum (1955). First, thediagnostic stage involves an analysis of the socialsituation. Hypotheses are formulated concerning thenature of the research domain. Second, the therapeuticstage involves change experiments. In this stagechanges are introduced and the effects are studied.

To summarise, action research, in all its variousforms, is characterised by (1) its multivariate socialsetting, (2) its highly interpretive assumptions aboutobservation, (3) intervention by the researcher, (4)participatory observation and (5) the study of change inthe social setting. Although the various forms of actionresearch may distinguish themselves by imposing otheradditional characteristics, each form shares these fivecharacteristics.

Historical Background of IS Action Research

The history of IS action research can be traced throughfour major periods. The first two periods, the originsand fragmentation periods, represent the early socialscience foundations of action research and are commonto IS and other disciplines. It is in the latter twoperiods, the fragmentation and diffusion periods, that IShas begun to discover and apply the method. Theperiods overlap, but approximate dates for each periodare assigned. Since various writings about actionresearch may appear outside of their period, these datesare not firmly related to publication dates.

First Period: Origins (1940 - 1960)The origins of the action research method developedwhen the calamities of World War II precipitatedmassive social changes in the research arena of thesocial sciences. Kurt Lewin (1947) is credited withdeveloping the method at the Research Centre forGroup Dynamics (University of Michigan) in order tostudy social psychology within the framework of fieldtheory. However, another group working independentlyat the Tavistock Clinic (later the Tavistock Institute)developed a similar method as a sort of psychosocialequivalent of operational research (see Trist 1976 orWarmington 1980).

The Tavistock Institute dealt with psychologicaland social disorders caused by battlefields and prisoner-of-war camps. Previously to this war, thesepsychological syndromes had not been identified insuch a large population of patients. Scientists did notunderstand enough about the complex causes of such`social illnesses' to formulate any confidence in anyuniversal treatments. Each case appeared somehow"different". Hence, the idea of social action arose.Scientists intervened in each experimental case bychanging some aspect of the patient's being orsurroundings. Since scientist and therapist were one,the scientists were participants in their own research.

4

The effects of the actions were recorded and studied. Inthis manner, a body of knowledge was developed aboutsuccessful therapy for the illnesses.

Lewin's work sought a general theory of how socialchange could be facilitated. His original model of actionresearch included iteration of six phased stages, ratherthan the five now commonly assumed. The six stageswere (1) analysis, (2) fact finding, (3)conceptualisation, (4) planning, (5) implementation ofaction, and (6) evaluation. A foundational stream ofrelated work in action research follows Lewin and theTavistock experience. This work generally arises in thefield of organizational sociology and social psychology,although Blum (1955) published a discussion on theimplications of action research for the more generalscientific arena.

Second Period: Disputes (1960 - 1975)After the initial period, the approach seems to havegone through a period of experimentation that disclosedsome essential problems. The refereed material onaction research of this period is dominated by the socialscience researchers who recognised problems andlimitations with action research. The major problemsidentified with the use of the technique in social scienceconcerned ethical dilemmas and the nature of researchsponsorship in the post-war period.

Rapoport (1970) first explained the three dilemmasthat severely inhibit the ethical use of the technique inpractice. These include a goal dilemma between thepractical problem-at-hand and the research question,bringing an ethical conflict to the research because thisdual-goal environment sometimes conflicts. There isalso a second dilemma between the roles of researcherand consultant in which one individual must serve.Sometimes these roles conflict, for example whenconsulting fees are paid to the researcher or editorialcontrol over research reports passes to the client. Athird ethical conflict is found in the concomitant valuedilemma. The values that inhabit the client culture mayconflict with those of the researcher. For example,clients may place a high value on quick, decisive action,whereas the researcher's academic culture may valuelengthy abstract reflection before considering action.

A second major problem regarded the fundingstructure of social science research in the 1950s and1960s. Peter Clark (1972) reasoned that research wasbeing increasingly sponsored by public money. Inresponse, leading researchers tended to seek projectsthat relied on "hard" quantitative data: projects thatfeatured computer analysis and thereby attracted

government attention. This post-war emphasis onprofessionalism and precise data collection methods ledto a general decline in qualitative research skills. As aresult, action research methods were seldom applied,and when these were used, they were often of marginalscientific quality due to the limited resources. Clarkfinds action research among the methodological"orphans" in post-World War II science.

A third problem arose in the increasing associationwith action research and organizational consulting.Edgar Schein was exposed to the technique in 1957 atthe National Training Laboratories (Schein 1987) andbegan exploring the use of action research fordeveloping solutions to group problems inorganizations. His Process Consultation (1969) had animmense and positive impact on the field oforganizational consulting. However, this may have onlyincreased the perception among social scientists thataction research was purely a practical method, and hadlittle scientific significance.

The decline of action research appears to have beennearly complete by the mid-1970s. In Anton Clark's(1976) compilation of action research papers inorganizational work, Sanford (1976) asks "What everhappened to action research?".

Third Period: Fragmentation (1975 - 1990)In the late 1970s, action research regained some of itsrespectability, but also became more clearly fragmentedin its forms and its applications. The main stream ofaction research work became more strongly related tothe organizational sciences, Susman and Evered (1978)presented an encouraging evaluation of the scientificmerits of action research in terms of post-positivistscience. This was followed by a meticulously-developeddefinition of the important aims of the approach by Hultand Lennung (1980).

In this period, the work becomes more clearlyfragmented along defined streams of ideas within theorganizational sciences. The consulting stream thatevolved out of Schein's early work continued throughothers, e.g., the Lippitt brothers (Lippitt & Lippitt1978—although Ron Lippitt was a student of Lewinand had first-hand knowledge of the original actionresearch techniques), or Kubr (1986). In the field ofmedical practice, action learning (Burnard 1991, Pedler1991) is discovered and becomes increasingly importantas a mechanism to solve the educational crisis.

Two streams develop during this period thatemphasize the relationship between reflection andaction. One stream evolves from work by Argyris and

5

Diffusion

Origins

Disputes

Burnard 1991Pedler 1991Action Learning

Kurt LewinGroup Dynamics1947

Argyris & SchönOrg Learning1978

SchönReflective Practice1983

Argyris, Putnam &SmithAction Science1985

Blum: "Action Research"Philosophy of Science1955

Rapoport3 Dilemmas (HR)1970

Peter ClarkOrganizational Change1972

Anton ClarkOrg. Experiments1976

Susman & EveredScientific Merit (ASQ)1978

Hult & LennungDefinition (JMS)1980

LauIS AR Classes1997

ScheinProcessConsultation1969

Lippit & LippitConsultingProcess in Action1978

KubrManagementConsulting 1986

ScheinClinicalFieldwork 1987

StringerHandbook1996

WhyteParticipatoryAR 1991

ChecklandSystems Thinking1981, 1990

JacksonCritical SystemsTheory 1985

Wood-HarperAR in IS1985

Baskerville &Wood-HarperDomains1996

SystemsScience

MumfordETHICS1979

Fragmenta-tion

Figure 1. Genealogy of IS Action Research.

Schön (1978) in double-loop organizational learning,which also uses some of Lewin's basic concepts.Consequently, these ideas are quite coherent with theaction research cycle. This stream continues withSchön's (1983) work in reflective practice andeventually reconnects with action research in a book byArgyris, Putnam and Smith (1985). Schein (1987)builds on both action science and his experience withprocess consultation in formulating his view of clinicalfield work. The second of these streams arises inCheckland's merger of action research and systemsscience in creating the concepts of systems thinking,and soft systems methodology (Checkland 1981).

During this period the information systems actionresearch originates as a distinct application area foraction research. Pioneering work includes Mumford'sETHICS, defined as an action-oriented socio-technicalinformation systems development method (Mumford &Weir 1979, Mumford 1983a), and Wood-Harper's useof action research for the study of information systemsdevelopment (Wood-Harper 1985).

Fourth Period: Diffusion (1990 - )The various fragmented forms of action research nowappear to be diffusing broadly across practical fields ofstudy. The various forms are not merging so much asthey are being applied appropriately for various usefulresearch purposes. Action research publications areincreasingly concentrating upon the results of actionresearch studies rather than the abstract philosophy ofaction research. Handbooks (Stringer 1996) andresearch anthologies (Whyte 1991) are importantmilestones. A similar trend is noticeable in theinformation systems field. Checkland, for example,developed a more refined soft systems methodologycouched among exemplars (Checkland & Scholes1990). The practical domains of information systemsaction research are explored (Baskerville & Wood-Harper 1996), and many of the information systemsaction research journal publications are inventoried andclassified (Lau 1997).

Figure 1 diagrams the IS-relevant streams ofaction research thinking through their four periods. Thestreams are shown vertically progressing from theorigins at the top of the diagram downward throughlater periods toward the bottom. Five fragmentedstreams of development are leading to the presentdiffusion in the IS field (lower right). From left-to-right, the five streams are social and organizationalscience, organizational learning, process consultation,systems science, and information systems action

research. The linkage down the diagram from Lewindoes not imply exclusion of the influence of largebodies of other literature, nor does it always define acitation trail. These are simply chronological streams ofrelated work. The systems science and consultationliterature is not intended to represent a completelinkage of ideas, but are examples selected from themyriad important works. Somewhat distinctly in thesystems and information systems developments, bothJackson's (1985) later versions of critical systems theoryand Wood-Harper, Antill and Avison's (1985)multiview approach suggest methodological pluralismis more appropriate than action research alone. Theaction science work, perhaps because of its strongerrigor (Argyris & Schön 1991), is also beginning todirectly influence information systems practice andresearch. Current consulting has been heavilyinfluenced by the Schein concepts, but there is someoverlap with current action research (see Bledsoe &Van Eynde 1990).

For the researcher and practitioner in informationsystems, action research appears to arise from these fourdistinct streams: the original "canonical" form of actionresearch, management consulting, soft systems

6

methodology, and organizational learning. The termaction research is frequently mentioned in each stream,but in each, the term has slightly different contextualmeaning. In addition, there are some methods that arequite similar to action research, such as prototyping andparticipant observation. Although these techniques didnot explicitly develop from the original action researchstreams, they are sometimes called action researchbecause of the strong similarities. As a practicalconsequence of these undeniable similarities, thesetechniques will be admitted in the following discussionas forms of IS action research.

Characteristics of action research forms

Each form of action research is distinguished by a widevariety of characteristics. At least four types of thesecharacteristics are conducive to a comparative analysis.These are process model, structure, typical researcherinvolvement, and primary goals. To illustrate how eachof these forms can be discovered "in the wild" in theaction research literature, each form will be illustratedby one or more examples. Explicit forms are examplesthat openly identify the applicable action research form.Implicit forms are examples that do not openly identifythe applicable action research form, indicating that theresearch approach falls within the boundary of actionresearch (defined earlier) and possesses thecharacteristics described by the form. In these implicitexamples, the authors could have (but have not) madeclaims to the action research paradigm as a foundationfor the validity of their work.

Process model. Three distinct process models characterise differentaction research forms. An iterative process modelinvolves a repeating sequence of activities, typicallycycling between action activities and problem diagnosisactivities (Blum 1955). A reflective process model isnecessarily iterative, but focuses less on problemdiagnosis than on reflective analysis of theory-in-useversus espoused-theory. An espoused theory is one thatan individual claims to be following. A theory-in-use isone that can be inferred from action (Argyris & Schön1978). A reflective process model concentrates on thediscovery of differences between the two theories. Alinear process model does not involve iteration, but asingle sequence of activities such as (1) engage, (2)diagnose, (3) unfreeze, (4) change, (5) freeze, (6)disengage.

This characteristic can be used as a taxonomy toorganise the diverse forms of IS action research intothree categories. We will use this taxonomy to organisethe discussion of these forms in the paper below:iterative, reflective and linear IS action researchmethods.

StructureTwo distinctive forms of structural guidance arecharacteristic in different forms of action research.Rigorous structure is characterised by delineated stages,steps or activities carried out in a sequence or cycle, orselected according to rules or heuristics. Fluid structuredefines activities very loosely, allowing substantialsimultaneity or leaving the temporal location of variousactivities relatively undefined.

Typical researcher involvement.Individual researchers can choose to be involved withtheir study subjects in highly independent and differentways. (As we have defined the boundaries of actionresearch, the researcher must be involved as aparticipatory observer and in determining theinterventions.) One of the following typical researcherinvolvements is characteristic of the literature on eachform of action research. Collaborative involvementimplies that the researcher is an equal co-worker withthe study subjects. The study tasks are shared withoutdistinction and the participants' backgrounds areassumed to be equally valuable. A facilitativeinvolvement distinguishes the researcher as an expertamong the study subjects. While the work is stillcooperative, the tasks of the researcher and the subjectsare quite distinct. The burden of solving the immediateproblem setting rests with the study subjects. The taskof the researcher is to facilitate or help the subjects withexpert advice, technical knowledge or an independentviewpoint. However, the subjects are responsible fordetermining exactly what interventions will be created.An expert involvement also distinguishes the researcheras an expert among the study subjects, and still involvescooperation and distinct tasks. However, the burden ofsolving the immediate problem setting rests with theresearcher. The researcher's decisions will determine toa large degree what interventions will be created.Choosing facilitative versus expert involvementinvolves reversing the tasks of the subjects and theresearcher with regard to the immediate problem.

Some forms of action research allow the researcherto adopt different involvement roles depending on theproblem setting. An example is Multiview, which

7

permits researcher to select their role as either doctor,facilitator, emancipator and social warrior dependingon the setting (Avison & Wood-Harper 1990). Anotherexample is prototyping, which exists in both facilitative(Connell & Shafer 1989) or collaborative forms(Gronbaek 1989).

Primary goalsThe different forms of action research can be used withmany different goals in mind. However, each form ischaracterised by certain primary goals in their commonformulations. Organizational development involves aprimary goal of improving the human organization. Inthis viewpoint, we consider the human organization tobe embodied in the regular patterns of social interactionamong the persons involved in the organization. Thusorganizational development implies the development ofsocial conditions of the organization. These conditionsmay include higher morale, structural efficiency,structural effectiveness, or better information flows.System design involves a primary goal of creating ormodifying organizational systems. These systems arethe structural artifacts that people create as aconsequence of the human organization. For ourpurposes, system design usually involves a computer-based information system. But it might also broadlyrefer to an accounting system or an inventory system.Scientific knowledge implies a primary goal ofcontributing a generalisable understanding of theproblem setting to the scientific literature in the field.The generalisability is necessary for practitioners to usethe understanding in different settings, or for scientiststo build further studies on the knowledge. Traininginvolves a primary goal of individual learning from thestudy. This is an educational goal, where the primaryreward is the improved understanding certain types ofproblem situations by a single researcher (or group ofresearchers) who undertake the study.

Forms of action research

Table 1 illustrates an analysis of the various forms ofaction research that appear in the IS literature. Thecolumns in this table represent the distinguishingcharacteristics of these forms as described above. Eachrow in the table specifies the characteristics of one ofthe forms of action research. The left-hand column liststhe name of the action research forms: canonical actionresearch, information system prototyping, soft systemsmethodology, action science, participant observation,action learning, multiview, ETHICS, clinical field

work, and process consultation. The right-hand columnlists one example of IS published work that illustratesthese characteristics. The marks in the cells of the tabledesignate the presence of one of the characteristics inone of the forms of action research. Each of these formsis discussed below.

Iterative IS action researchThe distinguishing characteristic of iterative actionresearch is the overall repeating sequence of majoractivities such as diagnosis, action and learning. Mostforms of action research will involve iteration at somelevel in the activities. However, iterative IS actionresearch uses iteration as its primary organisingprinciple, and the entire set of research activities isrepeated until the practical problem is resolved. Formsof IS action research that exhibit this characteristicinclude canonical action research, soft systems, andprototyping.

Canonical action researchAction research, as it was originally formed, has beenused with particular frequency to advance ourunderstanding of information systems development(Baskerville & Wood-Harper 1996). Lewin's originalsix-stage form of action research involved distinct,iterative phases. Hult and Lennung (1980) define thisform of action research concisely:

"Action research simultaneously assists inpractical problem-solving and expandsscientific knowledge, as well as enhances thecompetencies of the respective actors, beingperformed collaboratively in an immediatesituation using data feed back in a cyclicalprocess aiming at an increased understandingof a given social situation, primarilyapplicable for the understanding of changeprocesses in social systems and undertakenwithin a mutually acceptable ethicalframework."

8

Process Model Structure Typical Involvement Primary Goals Published IS

ExamplesItera-tive

Reflec-tive

Linear Rigor-ous

Fluid Colla-borative

Facili-tative

Expert OrgztnlDevlpmt

SystemDesign

ScintfcKnwldge

Train-ing

CanonicalAction Research

! ! ! x x Baskerville 1992

InformationSystemsPrototyping

! ! + + ! Kyng 1991

Soft Systems ! ! ! x x Checkland & Scholes 1990

Action Science ! ! ! x x Reponen 1992

ParticipantObservation

! ! ! ! Jepsen et al 1989

Action Learning ! ! ! ! Naur 1983

Multiview ! ! + + + ! Avison & Wood-Harper1990

ETHICS ! ! ! x x Mumford 1983

Clinical FieldWork

! ! ! x x Hammer & Champy 1993

ProcessConsultation

! ! ! ! Coad & Yourdon 1991

Key: ! signifies a dominant characteristic, + (or) signifies characteristics that will dominate in different studies, x (and) signifies characteristics that may occurtogether in the same study.

Table 1. Characteristics analysis of action research forms.

9

Diagnosing

ActionPlanning

ActionTakingEvaluating

SpecifyingLearning

Client-SystemInfrastructure

Figure 2. Canonical action research process model(Susman 1983).

Susman (1983) clearly describes a slightly revisedversion of this form as the five stages illustrated inFigure 2. Each stage is briefly described below.

The client-system infrastructure is the specificationand agreement that constitutes the researchenvironment. It provides the authority, or sanctions,under which the researchers and host practitioners mayspecify actions. It also legitimates those actions with theexpress expectation that eventually these will provebeneficial to the client or host organization.Considerations found within the agreement may includethe boundaries of the research domain, and the entryand exit of the scientists. It may also patently recognisethe latitude of the researchers to disseminate thelearning that is gained in the research. Thisinfrastructure should also define the responsibilities ofthe client and the researchers to each other. Forexample, the infrastructure will probably assume thatthe researchers will not purposely specify actions thatare harmful to the organization.

A key aspect of the infrastructure is thecollaborative nature of the undertaking, and this definesthe typical involvement of the researcher (see Table 1).The research scientists work closely with practitionerswho are located within the client-system. Theseindividuals provide the subject system knowledge andinsight necessary to understand the anomalies beingstudied. Peter Clark describes these practitioners thus:

"For convenience it is useful to think of thepractitioner as part of a set of actors who areoriented to solution of practical problems, whoare essentially organizational scientists ratherthan academic scientists." (Clark 1972, p.65)

Diagnosing corresponds to the identification of theprimary problems that are the underlying causes of theorganization's desire for change. This involves self-interpretation of the complex organizational problem,not through reduction and simplification, but rather ina holistic fashion. This diagnosis will develop certaintheoretical assumptions (i.e., a working hypothesis)about the nature of the organization and its problemdomain.

Researchers and practitioners then collaborate inthe next activity, action planning. This activity specifiesorganizational actions that should relieve or improvethese primary problems. The discovery of the plannedactions is guided by the theoretical framework, whichindicates both some desired future state for theorganization, and the changes that would achieve sucha state. The plan establishes the target for change andthe approach to change.

Action taking then implements the planned action.The researchers and practitioners collaborate in theactive intervention into the client organization, causingcertain changes to be made. Several forms ofintervention strategy can be adopted. For example, theintervention might be directive, in which the research"directs" the change, or non-directive, in which thechange is sought indirectly. Intervention tactics can alsobe adopted, such as the recruiting of intelligentlaypersons as change catalysts and pacemakers. Theprocess can also draw its steps from social psychology,e.g., engagement, unfreezing, learning and reframing.

After the actions are completed, the collaborativeresearchers and practitioners undertake the evaluatingof the outcomes. This includes a determination ofwhether the theoretical effects of the action wererealised, and whether these effects relieved theproblems. Where the change was successful, theevaluation must critically question whether theundertaken action, among the myriad routine and non-routine organizational actions, was the sole cause ofsuccess. Where the change was unsuccessful, someframework for the next iteration of the action researchcycle (including the adjustment of the hypotheses)should be established.

While the activity of specifying learning is formallyundertaken last, it is usually an ongoing process. Theknowledge gained in the action research (whether theaction was successful or unsuccessful) can be directedto three audiences. First, what Argyris and Schön(1978) call "double-loop learning," the restructuring oforganizational norms to reflect the new knowledgegained by the organization during the research. Second,

10

Would-beimprovers ofthe problemsituation

history

Real-worldproblemsituation

tasks,issues

relevantsystems

models situation

the situation isa culture

differences betweenmodels and the realworld

changes: systematicallydesirable, culturallyfeasible

action to improvethe situation

analysis of theintervention

"social system analysis

"political system" analysis

STREAM OFCULTURALANALYSIS

LOGIC-BASED STREAMOF ANALYSIS

compare

Figure 3. Contemporary soft systems methodology(Checkland and Scholes 1990)

where the change was unsuccessful, the additionalknowledge may provide foundations for diagnosing inpreparation for further action research intervention.Finally, the success or failure of the theoreticalframework will provide important knowledge to thescientific community for dealing with future researchsettings.

The action research cycle can continue, whether theaction proved successful or not, to develop furtherknowledge about the organization and the validity ofrelevant theoretical frameworks. As a result of thestudies, the organization thus learns more about itsnature and environment, and the constellation oftheoretical elements of the scientific communitycontinues to benefit and evolve. Thus this original formof action research addresses two primary goals:organizational development and scientific knowledge(see Table 1).

The clearly defined nature of each stage in theprocess is characteristic of a rigorously structured actionresearch form. Likewise, the form is clearly an iterativeprocess model, and lacks the concern with actiontheories (theory-in-use versus espoused-theory) thatcharacterises reflective process models. See Table 1.

An explicit example of canonical action research inthe field of information systems can be found inBaskerville (1993). This paper describes thedevelopment of a theory of semantic databaseprototyping within a canonical action research project.The paper explicitly discusses the various canonicalcomponents and the theory evolution during actionresearch iterative cycles.

Soft systems methodologyAction research has been linked closely to systemstheory from its inception, although Susman and Evered(1978) make the seminal connections. These ideasrecognise that human activities are systematic, and thataction researchers are intervening in social systems.Peter Checkland's (1981) use of action research inconnection with systems analysis is a landmark for thetechnique in the information systems researchcommunity. Checkland drew heavily from both systemsscience and action research, and his ideas have broughtconsiderable attention to action research in the IScommunity (although his work is not as widely followedin the United States). Checkland not only used anaction research approach extensively in developing his

soft systems methodology, but action research conceptsfor gaining professional knowledge permeate the softsystems approach itself.

Although the term "reflection" inhabits theliterature on soft systems methodology, this term is notbeing used in the narrow sense of theories-in-action.Rather, the term regards a period of abstraction, inwhich the essential components of social reality in thetarget systems world are captured and modelled.Reflection regards the ability of the abstract systemsmodels to reflect the essential social reality, andembody the strong partition between the systemsthinking elements of problem-diagnosis and real-worldelements of action-taking.

The process model of soft systems is essentiallyiterative (see Table 1), although the description of theprocess is rather linear: "Changes implemented as aresult of the use of SSM of course change the problemsituation as originally perceived, and in the newsituation the cycle of learning stimulated by themethodology can begin again. . . . [sic] It is in principlenever ending, and ending a systems study is an arbitraryact." (Checkland & Scholes 1990, p. 30).

Soft systems methodology is also characterised bya well-defined structure. This structure consisted ofseven iterative stages in the 1975 version, which weredivided between real-world activities and systemsthinking activities. This was replaced with a two-stream, iterative process model in the 1990 version,divided between a logic-based stream and a culturalanalysis stream (see Figure 3). The logic streamconsists of the comparison between relevant systems,models and the situation; determination of changes; andaction to improve the situation. The cultural analysis

11

RapidAnalysis

Database Creation

Menus

Functions

User Experimentation

Iteration

UserApproval

Testing

Tuning

Operation & Maintenance

Plan

Figure 4. The Connell and Shafer (1989)evolutionary prototyping process.

consists of analysis of the intervention, the socialsystem and the political system. The determination ofchanges is dependent on systematic and culturalfeasibility, and this stage interacts with the culturalanalysis.

The typical involvement is characterized by acollaborative role, a "joint insider-outsider problem-solving team" tackling a messy problem situation (p.60). There is at least one example of a facilitative softsystems study (the ICI case in Checkland and Scholes1990), but this is regarded as not quite fitting thepattern.

The primary goal of soft systems is system design,but the term is used in the sense of human activitysystems. Information systems (data manipulationsystems) design and creation is one area in which softsystems can contribute models and instances ofpurposeful human activity systems. Consequently, theprimary goals of soft systems methodology arecharacterised as both organizational development(human activity systems) and system design.

Checkland and Scholes (1990) detail severalexplicit examples of soft systems methods. Althoughsome of these systems are broader than just IS, there areclear relationships drawn to the IS domain. Since thepurpose of the detailed description is to illustrate softsystems methodology, the cases are explicit examples,and generally follow the defined form of this research.

Information system prototypingThe prototyping approach to systems analysis anddesign bears a striking resemblance to action research,although there is no strong intellectual heritage ofaction research in the development of prototyping. Eventhough some authorities may exclude prototyping as aform of action research, the technique exhibits qualitiesthat fall within the boundaries we have established forforms of action research: (1) it moves the designprocess into the user's multivariate social setting, (2) itpermits highly interpretive assumptions aboutobservation allowing highly qualitative data, (3) itrepresents an intervention by the designer into the userwork setting, (4) the designer is conductingparticipatory observation about the suitability of thedesign and (5) the designer is studying the impact ofdesign changes in the user's social setting. It is difficultto exclude action research as an information systemsdevelopment method (illustrated by prototyping,ETHICS, and soft systems), without excluding thewidely accepted use of action research as anorganizational development method. The arguments for

including this form of action research in this taxonomyare compelling.

Unlike the fairly singular form of canonical actionresearch, prototyping regards a fairly broad class ofinformation system development techniques. Connelland Shafer (1989) identify several categories ofprototypes within this class. There are "throw-away"design prototypes, e.g. mock-ups and user interfaceprototypes, that have limited functionality and precedethe specification process. There are specificationprototypes that provide a throw-away working model ofan entire system prior to specification and construction.There are design-driven prototypes that provide a pre-finalisation "test drive" of a traditionally developedsystem. The classic prototyping approach is embodiedin evolutionary prototypes that begin as designprototypes and cycle through iterative phases ofprototype reconstruction and user evaluation until fullfunctionality is achieved. The characteristic primarygoal of prototyping as a form of action research issystem design (see Table 1). In most versions, actiontheories are not considered, and this form of actionresearch is characteristically iterative (see Table 1).

In most cases, the prototyping form is also quiterigorous with defined stages or phases. For example,the Connell and Shafer version of the process (seeFigure 4) involves ten clearly defined, iterative stagesfrom project planning to operation and maintenance.

As a framework for research into applicationcontexts, prototypes are usually a mechanism thatallows designers to facilitate user validation of systemdesigns (cf. Naumann & Jenkins 1982, Boar 1984, orIivari & Karjalainen 1989). Prototypes surmount theesoteric nature of system design descriptions bypresenting a working model of a specification, andallow the user to understand and comment on the

12

Unstructuredproblem setting

Implement changes

Debate onchanges

Comparison oflogical model withanalysed situation

Logical model

Relevantsystems androotdefinitions

Situation analysed

rich pictureprimary tasksissues

REALWORLD

ABSTRACTSYSTEMSWORLD

iteration sequence

logical sequence

Figure 5. Multiview abstract systems thinking(Avison and Wood-Harper 1990).

design. The tasks of the designer and the user areclearly defined by their backgrounds, and theresponsibility for the usability of the design shifts to theuser through the process.

In other forms of prototyping, the roles of the usersand designers are less distinct. Prototypes are preciselyintended to improve user-designer communications(Mason & Carey 1983), and they are successful becauseprototypes are more effective linguistic artifacts (PelleEhn 1989). In some forms, this communication throughprototyping has enabled fully collaborative involvementby the designers and users (Gronbaek 1989). As aconsequence, the typical involvement of the researchermay be seen to be either facilitative or collaborative.This relationship is denoted in Table 1 by the presenceof a "+" in the collaborative and facilitativecharacteristics.

While there are many instances of prototyping ininformation systems practice, the impact of these arerarely described in terms of explicit action research. Foran example of research that implicitly fulfils thesecharacteristics, while making no specific claims onaction research for its validity, see Kyng (1991). Kyngprovides a concise description of two collaborativeprototyping projects involving group support systemsfor a large-scale civil engineering project.

MultiviewWood-Harper, Antill and Avison (1985) describe acontingency method for information systemsdevelopment that is expanded in Avison andWood-Harper (1990). Multiview is a flexible frameworkwhich provides an alternative to choosing betweendifferent methodologies. The techniques and toolsavailable within the framework are chosen and adjustedaccording to the particular problem situation. It is ablended methodology drawing from a number of majormethodologies, but contrasts because not all steps applyto all situations.

Multiview has a rigorous structure, despite itscontingent components. There are five stages in theframework: (1) human activity analysis, (2) informationanalysis, (3) socio-technical analysis and design, (4)human-computer interface design, and (5) technicaldesign. The overall framework is linear in nature.

Multiview is included among the action researchforms because Checkland's soft systems methodologystrongly influences the human activity analysis stage ofmultiview, and no alternative tools for this stage aresuggested. Within this multiview human activity stage,Checkland's original seven-stage iterative actionresearch technique used to build the conceptual, orhuman activity model (see Figure 5).

The typified involvement of the researcher inmultiview is suggested to be variable, and Burrell andMorgan's (1979) framework is used to define fourresearcher roles: (1) technical expert (functionalist,doctor), (2) facilitator (interpretative, teacher), (3) agentfor social progress (radical structuralist, warrior), and(4) change catalyst (radical humanist, emancipator).Consequently the typical involvement of the researcherin multiview could be selected from among the typicalcollaborative, facilitative or expert roles (this isdesignated in Table 1 as "+" in each column). Theprimary goal of multiview is systems design.

Wood-Harper, Antill and Avison (1985) andAvison and Wood-Harper (1990) describe explicitexamples of IS development projects that use this formof action research. Similarly to the soft systemsexamples above, the purpose of the descriptions is toillustrate the method, so the various characteristics aregenerally followed and explicitly described.

Reflective IS action research

The important characteristic of reflective IS actionresearch is its focus on the distinction between theory-in-use versus espoused-theory. All action research is

13

reflective in the sense that action is typically followedby a stage in which the researchers reflect on theevaluation of the results and learning from the effects ofthe action. However, for our purposes, the term"reflective" is applied here in a more specific sense:What is critical in reflective action research is theactors' discovery of where their behaviour isunexplained by their own understanding. Diagnosis anditeration are implied by this search for anunderstanding of this distinction, but iteration is nolonger an end in itself. There are three forms of ISaction research which meet this criterion. These formsare all three widely used outside of information systems,and all three are generally fluid in their structure. Theseforms include action science, participant observation,and action learning.

Action scienceArgyris, Putnam and Smith (1985) openly seek todistance their work from earlier, "corrupt" forms ofaction research by using the term action science. Theseauthors believe that during the disputes period actionresearch progressively became separated from theorybuilding and testing, and that the method becamefurther corrupted by positivist attempts to impose therigor of more traditional scientific experiments intoaction research projects. From their viewpoint, theeffect disconnected theory from reality, and renderedthe resultant findings irrelevant.

Action science builds on work by Argyris andSchön (1978) in double-loop organizational learningand Schön's (1983) work in reflective practice. Thetheories-in-action perspective is characteristic of actionscience, however, the technique is also characterised bya fluid structure, compensating for the "corrupting"effects of positivist rigorous structures. Instead, actionscience seeks rigor in high standards of practicalknowledge, which is defined as contextual precisionwithin social reality. "Contextual" means that practicalknowledge is relevant to its purpose and norms.Technical knowledge without a chosen end is notpractical, and consequently has undefined rigor. Actionscience is defined by the elements "epistemology ofpractice" and "empirical testing of practicalknowledge".

Epistemology of practice involves three concepts.The first concept regards the problem setting. Theproblem always involves a means-end deliberation, anda situational "frame" for practical knowledge. Thesecond concept is tacit knowledge, which involves theunconscious categories, schemas, patterns and other

rules that form knowledge. Often tacit knowledge canonly be inferred from skilful action. The third conceptinvolves making tacit knowledge explicit by a cycle ofreflecting and acting. This cycle involves the double-loop learning process: a single loop of means-endrationality (epistemology of practice) and a double loopin which this rationality is framed by a problem setting(action science epistemology). This double loop may,for example, set a different problem frame for the singleloop.

Empirical testing is the second defining element ofaction science. This embraces the experimentationthrough intervention in order to test rule-governedhuman iteration and the discovery of tacit knowledge.Data collection focuses on talk as a window on practicalreasoning, and its validity is threatened by contextualproblems (like laboratory-style controls, ambiguity ininterpretations or a variety of defensive routines such asface-saving behaviour). Validity is improved byintersubjective agreement. Action science alsorecognises two types of claims, dispositionalattributions (e.g., tendencies of certain individualstoward certain forms of behaviour), and causalresponsibilities (e.g., taking a particular action will leadto a particular result).

The domination of concepts like epistemology andempirical testing imply that the primary goal of actionscience is scientific knowledge. However, the emphasison the double loop is aimed at breaking away andreplacing old cognitive frames. Consequently the dualaims of organizational development and scientificknowledge are intertwined as goals in action science(see Table 1).

The researcher is involved clearly in a "helping"mode in action science. The aim of the researcher is tofacilitate the explication of tacit knowledge (theories-in-use) and thus enable the subjects to break out ofinappropriate frames (associated with espoused-theories). Both the researchers and the subjects areinvolved in the learning processes, but the role of theresearcher in action science is quite distinct from therole of the subject (see Table 1).

An explicit example of the use of action science inthe study of information systems is found in Reponen(1992). This paper describes an action science caseinvolving the development of a framework for ISstrategy formulation. The research is carefullydeveloped from the action science framework, and thereflective principles are generally followed.

14

Participant observationClassical natural science typically places a high valueon uninvolved and dispassionate observations. Inmodern social science such detached views aresometimes clouded with a thick cultural bias thatignores key elements of social reality. For example,early Western European anthropologists historicallycharacterised other cultures as "primitive". Participatoryobservation is a key element in removing such culturalbias. Participatory observation requires that theresearcher become directly involved as a participant inpeoples' daily lives. This involvement is a strategy forgaining access to the interior, seemingly subjectiveaspects of these lives by becoming a member or insiderin their world (Jorgensen 1989, p. 21).

Participatory observation can be used as the centralorganizing principle for a research method (i.e.Jorgensen's "participant observation", 1989) or as a datacollection technique for possible use within other typesresearch designs, for example, case study methods,ethnographic methods, or action research methods. Inculturally immersive research methods like actionresearch and ethnography, participatory observation is"crucial to effective fieldwork" (Fetterman 1989, p. 45).In research methods that admit detached observations(like case studies), participatory observation is viewedas an optional data collection technique (Yin 1989).

Participant observation as a methodology is fluid instructure, chiefly because it is difficult to control humaninteraction in social settings. The process of participantobservation "may seem unsystematic; in the beginningit is somewhat uncontrolled and haphazard" (Fetterman1989). It is characterised by a reflective process model,which values the explication of both the researcher'sand subjects' cognitive frames, for example when aresearcher's encounter with the subjects cannot beunderstood with the researcher's frame (Agar 1986).See Table 1.

Not all research involving participatory observationcan be regarded as action research. When theparticipating researcher intervenes to precipitate changeand then studies the effects of the intervention, thesimilarities are too strong to be ignored. Participantobservation is so broadly defined, and can be structuredin so many different ways that it is difficult tocharacterise the primary goals and typical involvementof the actual study. However, the general tone of moststudies that report the results of participant observationmethods, and most discussions of the participatoryobservation technique, reflect a primary interest in thescientific knowledge that is gained from the study. To

this degree, the researcher is typically seen as anauthoritative expert, immersed among the subjects inorder to create a reliable window onto their culture. Anycollaborative or facilitative role by the researcher withregard to the subjects' immediate problem setting is nottypically featured in these studies. The primary goalappears to be scientific knowledge, and the typicalinvolvement is that of an expert (see Table 1).

An example of published IS research that implicitlyadopts an action-oriented participant observationmethod is found in (Jepsen, Mathiassen & Nielsen1989). This paper describes the use of group diaries inthe analysis phase of a systems development project.While the authors do not claim to be following anyspecific research method, the research they describefalls within the boundaries of action research (definedearlier) and exhibits the characteristics of this form ofresearch.

Action learningAction learning, also called experiential learning(Burnard 1991), is a pedagogical technique used forimproving student learning in highly applied fieldswhich involve social settings. Action learning has beenparticularly important in fields with clinical settings.For example, action learning has been used for trainingpurposes in nursing and psychology. Action learning isreal-world apprentice-style experience typicallyinterspersed with more passive educational pedagogy,such as classroom or laboratory work.

Pedler (1991) describes action learning asessentially a two-phase process of theory-in-actionfollowed by reflection on the experience. To a largedegree, the techniques represent theories-in-actionconcepts applied with the primary goal of individuallearning rather than scientific knowledge ororganizational development. Like action science, actionlearning involves a reflective process model, operatingwith very fluid structures, but for the primary purposeof training (see Table 1). The rigorous attention to datais another feature shared characteristically by actionscience and action learning. The researcher's reflectionand understanding is improved by writing (withincumbent structuring of the knowledge) about theexperience. However, the experience does notnecessarily involve learning on the part of the subjects,the involvement of the action learner is one of an expertwho is using their theoretical knowledge to improve thesubject's condition while gaining empirical experiencethat exposes the researcher's misperceptions aboutcarrying the theory into practice (see Table 1).

15

As an example of IS research that explicitly usesthis form, see Wood-Harper and Flynn (1983). Thiswork establishes the importance of methodology indefining objective reality for the field of IS developmentby focusing on the learning that arises when analystsuse different methods in practical cases. Anotherexample of IS research that uses implicit actionlearning is found in Naur (1983). Naur describes aresearch exercise that can be characterised as actionlearning. He studied the software design anddevelopment process by engaging in such activities askeeping a detailed diary describing exactly how hecarried out the project. Naur makes no explicit claimsto action learning for validity, however, this work fallswithin our action research boundaries and could bevalidated using this form of research.

Linear IS action researchLinear forms of IS action research are characterised byan overall process that is not planned to be iterative,i.e., the activities are not programmed to be repeateduntil a result becomes settled. This linear process doesnot necessarily imply that a specific set of steps isdelineated in the method, but only that the process willprogress steadily from initiation to conclusion.Examples of IS methods that represent linear actionresearch include ETHICS and multiview. Other formsthat are represented only implicitly in informationsystems include clinical field work and processconsultation.

ETHICSMumford's ETHICS (Effective Technical and HumanImplementation of Computer Systems) is a strongparticipatory methodology intended to produceeffective, socially compatible information systems(Mumford 1983a). The ETHICS methodology enablesresearchers to place correctly balanced emphasis ontechnical requirements and human needs. ETHICSmaintains a perspective that includes job satisfactionand quality-of-work-life issues through the strictmechanism of end-user participation in design. Majorelements of the ETHICS approach include the use of afacilitator in design teams to explain technical issuesand elicit human needs impacts from the people whowill be involved in operating the system. The approachis organised into seven phases: (1) diagnose user needsand problems, (2) delineate efficiency, effectiveness, jobsatisfaction and quality goals and objectives, (3)develop alternative designs that meet the objectives, (4)select the most appropriate design, (5) design detailed

hardware and software requirements, (6)implementation, (7) evaluation.

ETHICS traces its roots to the Tavistock work inaction research and socio-technical design through thegroundbreaking work of its founder, Enid Mumford.The ETHICS process generally follows a linear modelfrom diagnosis to evaluation, although certain phases(such as diagnosing user needs and delineating goalsand objectives) may iterate across several design teamsin order to reach agreement on needs and compatibilityin objectives. The phases provide a rigorous structurefor organizing the design activities. The researcher istypically engaged as a facilitator, particularly in usergroup meetings where there may be a strong need toovercome the intimidation of technology andmisperceptions about the importance of social needs inthe workplace. ETHICS is oriented toward developingboth the information system and the workplace designin balance, although some design studies provideexcellent sources for developing an understanding aboutthe importance of social issues in IS design. An explicitexample of the latter is a book length description of theeffectiveness of engaging secretaries in the design of anew office system (Mumford 1983b).

Clinical field workEdgar Schein (1987) contrasts the clinical perspectivein field work from other ethnographic techniques. Thisclinical perspective regards very highly trainedprofessionals who get involved in a helping role withindividuals, groups, communities or organizations. Thishighly trained "helping" role implies that the typicalinvolvement of the researcher is facilitative (see Table1).

The study subjects typically initiate and drive aclinical enquiry, seeking help with an immediateproblem. The client expects to pay fees. An action-change study model then operates which is conceptuallynormative (improve the problem situation) andnarrowly focused on certain detailed data regarding aparticular problem. The study is validated by animprovement in the problem situation as a result of theactions. The achievement of this validity means that thestudy is primarily motivated by both organizationaldevelopment and scientific knowledge (see Table 1).

Schein contrasts this type of study withethnography, which is typically initiated and driven bythe researcher without fees. It operates with a studymodel that seeks to understand the client systemwithout perturbing it, and achieve a completeness ofdescription such that the total situation can be

16

understood by the reader. Validation in ethnographyrests on replicability and internal consistency in thedata. The roles of clinician and ethnographer maybecome intertwined in settings where the situationdemands the researcher switch roles. For example, anethnographer may become drawn into action (thusbecoming a clinician) or the clinician may be unable toeffect action (thus becoming an ethnographer).

A clinical method of inquiry is highly situational,and a concrete set of steps or stages is not prescribed.The structure therefore seems very fluid. The idealprocess model is typically linear (see Table 1), derivedfrom the ethical obligation only to suggest actions thatimprove the client's problem situation. Scheinemphasizes the importance of researcher training, skillsand sense of responsibility (p. 21), and brings focus onthe ethical and legal responsibility to avoid malpractice.There is no framework for iterative experimentationand review, nor is a theories-in-action model found inthe description.

A segment of the IS published literature could bevalidated as action research on the basis that it meetsthe criteria we have defined for the clinical field workform. However, rarely will this work make explicitclaims for validity on this basis. An example of the kindof work that implicitly meets this definition may befound in Hammer and Champy's (1993) reengineeringstudies. This work reports the discovery of the majorfacets of reengineering through fieldwork in a clinicalmode of enquiry. The work arose across a number ofcases in which the authors were actively engaged inhelping develop new techniques that would allowcompanies to survive in a harsh competitive climate.Their work falls within the defined boundary of actionresearch, and exhibits the characteristics of this form.

Process consultationThis clearly defined approach to organizationaldevelopment by the temporary introduction of anoutside consultant has had an important impact in thepractical realm of management consulting. EdgarSchein (1969) drew heavily on his training in socialpsychology and his experience as a consultant indeveloping the approach. There are seven stages,defined less as a temporal sequence than as logicalareas in which the consultant must work (p. 78). Thesestages are (1) initial contact with the clientorganization; (2) defining the relationship, formalcontract, and psychological contract; (3) selecting asetting and a method of work; (4) data gathering and

diagnosis; (5) intervention; (6) reducing involvement;and (7) termination.

The structure is relatively rigorous, and the processmodel is, in general, linear. (The six-stage iterativemodel of action research is preserved as the primarymodel for group problem solving—one component ofthe diagnosis stage.) On the surface, the typicalinvolvement might seem facilitative, much like thehelping role of the clinician above. However, a deeperanalysis reveals a subtle, but important difference. Theprocess consultant must transfer values and skills to theclient in order to enable the client to accomplishcontinued self-helping problem solving. The valuesinclude increasing attention for human concerns overtask concerns, for human processes over organizationalstructures, for long-term effectiveness over short-runoutput, and for perpetual diagnosis over generalisationsand principles (p. 123). The client must also be helpedto develop the skills necessary to diagnose and resolveinterpersonal, group and organizational problems. Theprimary goal is not merely helping the client(facilitative), but teaching the client how to self-help(expert). For process consulting, the typicalinvolvement is essentially that of an expert, and theprimary goal is organizational development. See Table1.

There do not appear to be many explicit examplesof this kind of this kind of work published in thescholarly IS literature. However, the Coad and Yourdonwork on object-oriented analysis (1991) describes howthey helped several organizations develop the personaland group skills necessary to convert to object-orientedthinking in their software projects. These vignettesapproximate the sort of linear, expert, self-help processconsulting projects that Schein suggests. This workoffers an implicit example that could have beenvalidated as action research assuming a processconsultation form of enquiry.

Discussion

In the introduction, we have raised the promise of amore inclusive boundary for IS action research. Wethen defined boundaries for the IS action researchparadigm, discussed the common genealogicalfoundations of the various forms of action research, andthen discussed the distinguishing characteristics of tendifferent forms, pointing to explicit or implicitexamples in the published IS literature. This workraises six related arguments.

17

1. IS rights to an action paradigm. Is it necessary for IS to adopt action research exactly asdefined in the older social science literature? We argueagainst this restriction, and instead propose adaptingthis paradigm in order to make it more suitable for theIS discipline. The IS discipline, whether labelled asinformation systems, informatics, datology, computing,computer science, or systems science is a highlyempirical, professional research setting. IS researchersare expected to ground their models, algorithms andtheories in professional practice. An inclusive paradigmfor action research improves our selection of researchvehicles for applying our theories to practice. Anexclusive paradigm is less suitable and less rewardingfor our field. In particular, including more practitioner-oriented forms such as process consultation,prototyping, and clinical field work, opens vehicles bywhich enlightened practitioners can participate in thescholarly literature. This inclusion addresses thedifficulty of scholarly editorial boards who currentlymaintain a dual standard review system that separatesacademic authors and practitioner authors.

2. Validity in IS action research.How do we validate work in the IS action researchparadigm? We have claimed that an inclusive paradigmmakes action research validity measures available towider forms of action research. A detailed discussion ofaction research validity measures is beyond the scope ofthis paper. There are some indications from the broadersocial science literature, particularly in Susman andEvered (1978), Rapoport (1970), and Gummesson(1988). While further work will be needed to determinedetailed criteria for the IS paradigm, seven criteria arelikely to surface. The boundary characteristics aid inidentifying five possible validity criteria: (1) Theresearch should be set in a multivariate social situation.(2) The observations are recorded and analysed in aninterpretive frame. (3) There was researcher action thatintervened in the research setting. (4) The method ofdata collection included participatory observation. (5)Changes in the social setting were studied. In additionto these five characteristic criteria, at least two moreimportant criteria arise from the basic premises of theLewin and Tavistock experience: (6) The immediateproblem in the social setting must have been resolvedduring the research. (7) The research should illuminatea theoretical framework that explains how the actionsled to the favourable outcome.

The first five criteria follow the arguments in theboundaries section above. The last two criteria,

however, correspond to criteria arising in other forms ofqualitative research. Criterion 6 corresponds to internalvalidity in action research. Unlike other forms ofqualitative research, such as case studies orethnography, action research is primarily validated byits relief of the immediate social problem setting.Action research is a goal-directed field researchmethodology, and its goal achievement is central to thevalidity of actions directed toward this goal. Criterion7 corresponds to external validity in action research.This criterion is shared with case research, in the sensethat the understanding or learning developed by theresearchers from the action results must be generalisedto a theory that has potential for use in other problemsettings. This "generalisation to theory" (Yin 1989) canbe seen as a substitute for the statistical generalisationfound in quantitative studies. The theory should helpexplain why certain actions resolved the problemsetting and why certain actions failed to resolve theproblem setting. This theory component has importantvalue to practitioners who may face problem settingswith similar characteristics; and to researchers who canuse the theory as a linking tool between existing actionresearch evidence and further empirical results. Theoryenables action research to advance both IS practice andIS research.

3. Necessity for intellectual heritage.Is it necessary for all forms of action research to sharethe same intellectual genealogy? Of the ten formsdiscussed in this paper, nine can be traced, directly orindirectly, back to the roots of Lewin's action researchand the Tavistock Institute. One form, prototyping,emerged along totally separate lines, from the roots ofmanufacturing product design and industrialengineering. Prototyping shares all of thedistinguishing characteristics of action research, and isa frequent cohabitant of IS action research (e.g.,Baskerville & Stage 1996, Iivari & Karjalainen 1989).Genealogy alone is not likely to define an actionresearch form, but is there something about prototypingthat makes it seem less acceptable than ETHICS, SoftSystems, or Multiview (which can draw their roots backto early action research)?

4. Research-practice distinctions.The fourth issue is closely aligned with this prototypingquestion. How do we distinguish between "ordinary" ISdevelopment practice and action research? Whatprevents authors from developing systems and then"faking" the results to the scholarly community as

18

action research. This issue is an old problem amongaction researchers and the validity criteria above arecertainly helpful in making the distinction. However,part of the problem lies in misperception. Enlightenedpractice often seeks to be scientific, and it is desirablethat professionals, intent on high-quality IS design, mayengage in the same sort of investigative practices asthose used by scholars intent on understanding situatedIS phenomena. In other words, quality IS analysts anddesigners may properly execute "internal" actionresearch to meet the ends of the development project. Inmost cases, however, the designers will be toounfamiliar with action research to properly execute aproject with the necessary rigor, and will fail to meet allof the action research criteria suggested above.

In a normative sense, however, practitioner actionresearch ought to be encouraged. If the rigor of internalpractitioner research projects, undertaken in support ofsystems design, can be improved to the degreenecessary to satisfy action research criteria, then thefrequency of practitioner publications in scholarlyoutlets may also improve. The IS discipline's positionshould not exclude prototyping as a research method,but rather include those prototyping projects thatqualify (under the criteria suggested above) as actionresearch. Similarly, qualified practitioner work usingprocess consultation, clinical fieldwork, and other formsof validated action research would become publishable.However, the path to this Utopian future is blocked bycertain practicalities, including graduate-levelpractitioner education and a revised priority scheme inmany IS development organizations. It is much morelikely that publishable research studies will continue toresult from the collaboration of trained researchers andenlightened practitioners, rather than from practitionersalone. Under the supervision of trained researchers,properly executed action research may take the form ofprototyping, clinical fieldwork, process consultation,etc. and meet the criteria suggested above.

5. Selection of proper formHow do we decide which form of action research iscorrect for a particular situation? While a detailednormative framework is beyond the scope of this paper,early indications can be guided by the characteristicsdetailed in Table I. The researcher will need todetermine the primary goals of the research project:organizational development, system design, scientificknowledge or training. The researcher involvement willbe determined by the knowledge and skills of theresearcher and potential research team collaborators in

the problem setting. Where the knowledge and skills ofthe team collaborators is low, the researcher will needto act as an expert or facilitator (assuming theknowledge and skills of the researcher are adequate).Where the knowledge and skills of the collaborators ishigh (or the knowledge and skills of the researcher arelow), the involvement may shift to collaborative. Theresearch structure will depend on the social setting. Forexample, if a legal contract is involved in the researchinfrastructure, a more rigorous structure with cleardeliverables may be desirable. The process model maysimilarly depend on the social setting. If the problem ortask is clear at the outset, a linear process may workbest. If the problem is ill defined, an iterative processcould be desirable. If the social-setting focus is onlearning-oriented outcomes, a reflective model may bedetermined.

6. Explicit methodologyIs it necessary for authors to explicitly declare theirresearch method in every scholarly publication? This isa particular problem for IS action research, becauseempirical articles are frequently published in IS that areunclear about their research method. Many articles areparticipatory observation and action based, especiallywith regard to IS development, yet authors couch theirresults as case based. This obscurity seems to suggestthat many researchers are confused about actionresearch, especially about what qualifies as actionresearch and the degree to which their work meets thecriteria of action research. This obscurity may alsoreflect the orientation of journal referee panels towarda relaxed view of case research, while enforcing a strictview of action research. It might also reflect a degree ofignorance of research methodology on the part ofreferee panels. A substantial portion of the actionresearch published in the IS field may have beenmischaracterised as case research, and critical aspectsof the validity of this research thereby left unpublished.In many cases, a wider boundary on the IS actionresearch paradigm would have enabled these authors tobuild a stronger claim for the validity andgeneralisability of the results of their research. Thisimproved claim could have increased the impact onpractice and further research. By adopting such a widerboundary at this time, future research may thus beenabled to strengthen validity claims and improvedissemination of the ideas to other researchers andpractitioners.

19

Appendix: How IS action research boundaries differThe boundaries of the IS action research paradigmdefined in this paper parallels the publishedcharacteristics of action research in the social scienceliterature. However, this literature is dominated by thecanonical form of action research, and tends toemphasize action research characteristics based ongoals and objectives rather than characteristics based onthe process. Our boundaries differ in the followingways, following Hult and Lennung's (1980) six majorcharacteristics of action research:

1 Action research aims at an increasedunderstanding of an immediate social situation.This goal is very similar to our first process-oriented characteristic, although we emphasize thecomplex and multivariate nature of this socialsetting in the IS domain.

2 Action research simultaneously assists in practicalproblem solving and expands scientific knowledge.We extend this goal into two important processcharacteristics: first, there are highly interpretiveassumptions being made about scientificobservation; second, the researcher intervenes inthe problem setting.

3 Action research is performed collaboratively andenhances the competencies of the respective actors.We focus on the narrower process of participatoryobservation implied by the collaborative goal.Enhanced competencies (an inevitable result ofcollaboration) is relative to the previouscompetencies of the researchers and subjects, andthe degree to which this is a goal and its balancebetween the actors will depend upon the setting.The competencies goal may help to determine theform of action research: seen as a primary goal insome settings (e.g., canonical or clinical forms);seen as a byproduct in other forms (e.g., processconsultation and prototyping).

4 Action research is primarily applicable for theunderstanding of change processes in socialsystems. This characteristic is adopted explicitly indefining the IS action research boundaries.

5 Action research uses data feed back in a cyclicalprocess. We did not adopt this characteristic. Theempirical nature of action research data is impliedby participatory observation. The cyclical process,

while characteristic of some action research forms,cannot be justified as a critical definingcharacteristic of all action-based research. It isfeasible that the outcome of the first (and only)iteration will be satisfactory. Some action-basedresearch forms may assume this first outcome willusually be satisfactory.

6 Action research is undertaken within a mutuallyacceptable ethical framework. We did not adoptthis characteristic. We agree completely with thisplatitude, and strongly feel that all research shouldadopt a mutually acceptable ethical frameworkregarding human subjects. Accordingly, we notethat this characteristic does not distinguish actionresearch from any other form of acceptable socialresearch.

References

Agar M (1986) Speaking of Ethnography. Sage,Newbury Park, California.

Argyris C and Schön D (1978) OrganizationalLearning: A Theory of Action Perspective.Addison-Wesley, Reading.

Argyris C and Schön D (1991) Participatory actionresearch and action science compared. InParticipatory Action Research (Whyte W Ed), pp85-96, Sage, Newbury Park New Jersey.

Argyris C, Putnam R and Smith D (1985) ActionScience: Concepts Methods and Skills forResearch and Intervention. Jossey-Bass, SanFrancisco.

Avison D and Wood-Harper A (1990) Multiview: AnExploration in Information Systems Development.McGraw Hill, London.

Banville C and Landry M (1989) Can the field of MISbe disciplined?. Communications of the ACM 32(1) 48-61.

Baskerville R (1993) Semantic database prototypes.Journal of Information Systems 3 (2), 119-144.

Baskerville R and Stage J (1996) Controlling prototypedevelopment through risk analysis. MIS Quarterly20 (4), 481-504.

Baskerville R and Wood-Harper AT (1996) A criticalperspective on action research as a method forinformation systems research. Journal ofInformation Technology 11, 235-246.

Baskerville R, Meyers M, Nielsen PA and Wood-Harper T (1997) Panel—The impact of actionresearch on information systems. In Information

20

Systems and Qualitative Research (Lee A,Liebenau J and DeGross J Eds), p 69, Chapman &Hall, London.

Benbasat I, Goldstein D. and Mead M (1987) The CaseResearch Strategy in Studies of InformationSystems. MIS Quarterly 11 (3), 369-386.

Bledsoe J and Van Eynde D (1990) The changingpractice of organisation development. Leadership& Organization Development Journal 11 (2), 25-30.

Blum F (1955) Action research—A scientific approach?Philosophy of Science 22 (1), 1-7.

Boar B (1984) Application Prototyping: ARequirements Definition Strategy for the 80's.Wiley, New York.

Burnard P (1991) Experiential Learning in Action.Avebury, Aldershot.

Burrell G and Morgan G (1979) SociologicalParadigms and Organisational Analysis.Heinemann, London.

Checkland P (1981) Systems Thinking SystemsPractice. Wiley, Chichester.

Checkland P and Scholes J (1990) Soft SystemsMethodology in Practice. Wiley, Chichester.

Clark A (1976) Experimenting with OrganizationalLife: The Action Research Approach. Plenum, NewYork.

Clark P (1972) Action Research and OrganizationalChange. Harper & Row, London.

Coad P and Yourdon E (1991) Object-OrientedAnalysis 2nd Ed. Yourdon, Englewood Cliffs.

Connell J and Shafer L (1989) Structured RapidPrototyping: An Evolutionary Approach toSoftware Development. Yourdon, EnglewoodCliffs.

Ehn P (1989) Playing in reality. In Proceedings of the12th IRIS: Information Systems Research SeminarIn Scandinavia Part I (Bodker, S Ed), pp 129-147,Aarhus University DAIMI PB-296-I, ArhusDenmark.

Fetterman D (1989) Ethnography Step by Step. Sage,Newbury Park.

Gronbaek K (1989) Extending the boundaries ofprototyping: Towards cooperative prototyping. InProceedings of the 12th IRIS: Information SystemsResearch Seminar In Scandinavia Part I (Bodker,S Ed), pp 219-238, Aarhus University DAIMI PB-296-I, Arhus Denmark.

Gummesson E (1988) Qualitative Methods inManagement Research. StudentlitteratureChartwell-Bratt, Lund Sweden.

Hammer M and Champy J (1993) Reengineering theCorporation: A Manifesto for Business Revolution.Harper-Collins, New York.

Hult M and Lennung S-Å (1980) Towards a definitionof action research: A note and bibliography.Journal of Management Studies 17 (May), 241-250.

Iivari J and Karjalainen M (1989) Impact ofprototyping on user information satisfaction duringthe is specification phase. Information &Management 17 (August), 31-45.

Jackson M (1985) Social systems theory and practice:The need for a critical approach. InternationalJournal of General Systems 10, 135-151.

Jepsen L, Mathiassen L and Nielsen P (1989) Back tothe thinking mode: diaries for the management ofinformation systems development projects.Behaviour and Information Technology 8 (3),207-217.

Jonsson S (1991) Action research. In InformationSystems Research: Contemporary Approaches andEmergent Traditions (Nissen H-E, Klein HK,Hirschheim R Eds), pp 371-396, North-Holland,Amsterdam.

Jorgensen D (1989) Participant Observation: AMethodology for Human Studies. Sage, NewburyPark.

Kaiser K and Bostrom R (1987) Personalitycharacteristics of MIS project teams: An empiricalstudy and action research design. MIS Quarterly 6(4), 43-60.

Keen P (1991) Relevance and rigor in informationsystems research: Improving quality confidencecohesion and impact. In Information SystemsResearch: Contemporary Approaches & EmergentTraditions (Nissen H-E, Klein H and HirschheimR Eds), pp 27-49, North-Holland, Amsterdam.

Kubr M (Ed) (1986) Management Consulting: A Guideto the Profession 2nd Ed. International LabourOffice, Geneva.

Kyng M (1991) Designing for cooperation: Cooperatingin design. Communications of the ACM 34 (12),65-73.

Lau F (1997) A review on the use of action research ininformation systems studies. In InformationSystems and Qualitative Research (Lee A;Liebenau J and DeGross J, Eds), pp 31-68.Chapman & Hall, London.

Lee A (1989) A scientific methodology for MIS casestudies. MIS Quarterly 13 (1), 33-50.

21

Lewin K (1947) Frontiers in group dynamics II. HumanRelations I (2), 143-153.

Lippitt G and Lippitt R (1978). The Consulting Processin Action, University Associates, San DiegoCalifornia.

Mason R and Carey T (1983) Prototyping interactiveinformation systems. Communications of the ACM26 (5), 347-354.

Mumford E (1983a) Designing Human Systems ForNew Technology: The ETHICS Method.Manchester Business School, Manchester.

Mumford E (1983b) Designing Secretaries ManchesterBusiness School, Manchester.

Mumford E and Weir M (1979) Computer Systems inWork Design: The ETHICS Method. AssociatedBusiness Press, London.

Naumann J and Jenkins A (1982) Prototyping: The newparadigm for systems development. MIS Quarterly6 (3), 29-44.

Naur P (1983) Program development studies based ondiaries. In Psychology of Computer Use (Green T,Payne S and van der Veer G Eds), pp 159-170,Academic Press, London.

Pedler M (1991) Action Learning in Practice 2nd Ed.Gower, Brookfield Vermont.

Rapoport R (1970) Three dilemmas of action research.Human Relations 23 499-513.

Reponen Tapio (1992) Action research in informationsystems strategy formulation and implementation.In Action Research in Management InformationSystems Studies (von Hellens Liisa Ed). TurkuSchool of Economics and Business AdministrationSeries A-3, Turku Finland.

Sanford N (1976) Whatever happened to actionresearch? In Experimenting with OrganizationalLife: The Action Research Approach (Clark A Ed),pp 11-18, Plenum, New York.

Schein E (1969) Process Consultation: Its Role inOrganizational Development. Addison-Wesley,Reading.

Schein E (1987) The Clinical Perspective in Fieldwork.Sage, Newbury Park.

Schön D (1983) The Reflective Practitioner: HowProfessionals Think in Action. Basic, New York.

Stringer E (1996) Action Research: A Handbook forPractitioners. Sage, Newbury Park New Jersey.

Susman G (1983) Action research: a sociotechnicalsystems perspective In Beyond Method: Strategiesfor Social Research (Morgan G, Ed), pp 95-113.Sage, Newbury Park.

Susman G and Evered R (1978) An assessment of thescientific merits of action research. AdministrativeScience Quarterly 23 (4), 582-603.

Tanton M and Fox S (1987) The evaluation ofmanagement education and development:participant satisfaction and ethnographicmethodology. Personnel Review 16 (4), 33-40.

Trist E (1976) Engaging with large-scale systems. InExperimenting with Organizational Life: TheAction Research Approach (A Clark Ed), pp 43-75. Plenum, New York.

Van Eynde D and Bledsoe J (1990) The changingpractice of organization development. Leadership& Organization Development Journal 11 (2),25-30.

Warmington A (1980) Action research: Its methods andits implications. Journal of Applied SystemsAnalysis 7, 23-39.

Whyte W F (1991) Participatory Action Research.Sage, Newbury Park New Jersey.

Wood-Harper T (1985) Research methods ininformation systems: using action research. InResearch Methods in Information Systems(Mumford E, Hirschheim R, Fitzgerald G andWood-Harper T, Eds), pp 169-191, North-Holland,Amsterdam.

Wood-Harper T, Antill L and Avison D (1985)Information Systems Definition: The MultiviewApproach. Blackwell Scientific, Oxford.

Wood-Harper AT and Flynn DJ (1983) Action learningfor teaching information systems. The ComputerJournal 26 (1), 79-82.

Yin R (1989) Case Study Research: Design andMethods. Sage, Newbury Park.