The Lure of the Virtual

OrganizationScienceVol. 23, No. 5, September–October 2012, pp. 1485–1504ISSN 1047-7039 (print) � ISSN 1526-5455 (online) http://dx.doi.org/10.1287/orsc.1110.0703

© 2012 INFORMS

The Lure of the Virtual

Diane E. BaileySchool of Information, University of Texas at Austin, Austin, Texas 78701, [email protected]

Paul M. LeonardiDepartment of Communication Studies, Department of Industrial Engineering and Management Sciences,

Northwestern University, Evanston, Illinois 60208, [email protected]

Stephen R. BarleyCenter for Work, Technology, and Organization, Department of Management Science and Engineering,

Huang Engineering Center, Stanford University, Stanford, California 94305, [email protected]

Although organizational scholars have begun to study virtual work, they have yet to fully grapple with its diversity. Wedraw on semiotics to distinguish among three types of virtual work (virtual teams, remote control, and simulations)

based on what it is that a technology makes virtual and whether work is done with or on, through, or within representations.Of the three types, simulations have been least studied, yet they have the greatest potential to change work’s historicallytight coupling to physical objects. Through a case study of an automobile manufacturer, we show how digital simulationtechnologies prompted a shift from symbolic to iconic representation of vehicle performance. The increasing verisimilitudeof iconic simulation models altered workers’ dependence on each other and on physical objects, leading management toconfound operating within representations with operating with or on representations. With this mistaken understanding, andlured by the virtual, managers organized simulation work in virtual teams, thereby distancing workers from the physicalreferents of their models and making it difficult to empirically validate models. From this case study, we draw implicationsfor the study of virtual work by examining how changes to work organization vary by type of virtual work.

Key words : organizing for innovation; technological change; product design; virtual work; simulation; representation;digitization; semiotics

History : Published online in Articles in Advance December 2, 2011.

IntroductionSince the earliest days of the computer revolution, thelure of the virtual has seduced thinkers, writers, design-ers, and others with the idea that we might somedayaccomplish with computers that which we have histor-ically done only physically, thereby potentially allow-ing us to dispense with the physical. Although the ideaof virtuality still attains its zenith only in science fic-tion, computer scientists have long sought to programvirtual worlds, and since the 1990s, they have had con-siderable success, especially in video gaming (Tschang2007). Yet virtuality has moved well beyond games.Sophisticated graphical simulations are now used totrain aircraft pilots, soldiers, and emergency responders,among others (Kincaid et al. 2003, Macedonia 2002,Longridge et al. 2001). Teams now routinely communi-cate and coordinate via e-mail, instant messaging, andWeb-conferencing applications without ever meeting inperson (Maznevski and Chudoba 2000). Doctors oper-ate on patients and military personnel control drone air-craft from as much as a half a world away using digitalvideo images and other real-time data collected by sen-sors (Drew 2009, Marescaux and Rubino 2004, Satava1995). With continued innovation in digital technologies,the possibility of working virtually is moving from the

realm of science fiction to the everyday reality of theworkplace.

As these examples suggest, exactly what “workingvirtually” means varies widely. Although students ofwork and organizing have begun to study virtual work,they have yet to grapple with its diversity. The impli-cations of virtual work for organizing are likely to varywith the physical phenomena that technology digitizesas well as with the relation between those phenom-ena and the representations that technology creates. Thelatter are particularly important because virtuality typi-cally implies working with a representation of the phys-ical rather than with the physical itself. It is alongthese lines that we distinguish digitization from virtual-ity. Digitization involves the creation of computer-basedrepresentations of physical phenomena. Because theserepresentations can traverse long distances via com-puter and telecommunications links, digitization facil-itates separation between people and the representedphenomena (physical objects, physical processes, orother people). Virtuality occurs when digital represen-tations stand for, and in some cases completely substi-tute for, the physical objects, processes, or people theyrepresent.

1485

Bailey, Leonardi, and Barley: The Lure of the Virtual1486 Organization Science 23(5), pp. 1485–1504, © 2012 INFORMS

By representation, we mean what semioticians call asign: something that stands for something else withinthe bounds of a particular speech community, culture,or community of practice (Eco 1979). According toSaussure (1916/1966), signs are composed of two ele-ments: a signifier (such as a word, an image, a number,or a sound) and the signified (that which the sign denotesor connotes). Frege’s (1892/1980) distinction betweenmeaning and reference is one useful way to classifysigns. Some signs have both a meaning and a refer-ent: they enable interpretation among those who knowhow to read them (meaning) and also point to somethingdefinite in the world, a physical entity (referent). Forexample, a blip on a sonar screen signifies the presenceof a submarine, a fish, or another object in the water.If it did not, the blip would be instrumentally uselessbecause it would have no referent, although it could stillhave meaning, perhaps as the functional equivalent of alava lamp.

Other signs have meaning but no referents (Quine1961). Consider numbers in a spreadsheet signifying theworth of houses in a local market. As concepts withinthe context of the spreadsheet, the numbers clearly havemeaning. Residents of the community might see themas an estimate of the amount of money they can expectif they sold their homes. Sociologists might see themas evidence that a neighborhood is becoming tonier orless tony. Indeed, numerous meanings are plausible, butunder no circumstances do the numbers refer ostensivelyto houses. One could not pull a number from the spread-sheet, write it on a piece of paper, and know whichhouse had that worth, because the number itself couldstand for anything. One could, however, pull the addressof a house from the spreadsheet, write it on a piece ofpaper, and find the house. This is because an address isa signifier that refers.

A useful way of further differentiating among repre-sentations that have referents is to consider the natureof the signifier’s relationship to the signified. Peirce’s(1932) taxonomy of indices, icons, and symbols takesthis approach. Indices are signs that have a physical andexistential relationship to that which they signify. Anindex usually co-occurs with and is often produced bythat which it signifies. Classic examples include smokeas a sign of fire, vapor trails as evidence of a passingjet, or our ability to identify individuals by the sound oftheir voice.

Icons are signs that signify because they closelyresemble that which they signify. Typically, icons consistof visual images such as pictures and portraits. Archi-tects, engineers, scientists, and members of other occu-pations have long depicted the structure of objectsusing iconic representations. Sketches, blueprints, andcomputer-aided design (CAD) drawings are well-knownexamples. The desktop on your computer is likely tobe full of icons, albeit not very sophisticated ones. For

instance, you likely delete files by dragging and drop-ping them into an iconic trash can.

Unlike indices and icons, symbols signify entirely bycultural convention; the link between signifier and signi-fied is essentially arbitrary. On a flowchart of a produc-tion process, for example, a rectangle may represent amachine, but the machine itself may not be rectangularin shape. We come to understand that the rectangle rep-resents the machine because a legend on the flowcharttells us so or because standard practice in an industry isto represent machines by rectangles. The rectangle pro-vides no clue as to the appearance of the machine.

In this paper, we first draw on semiotics to develop ataxonomy of virtual work by distinguishing among threetypes of virtual work—virtual teams, remote control,and simulations—based on aspects of digitization, rep-resentation, and virtuality. Students of organizing haveattended to the first two types but have largely ignoredthe last (a few notable exceptions include Boland et al.2007, Dodgson et al. 2007, Thomke 2003). Yet simu-lations are incredibly important because it is with sim-ulation that virtuality comes closest to substituting forreality. Simulation, therefore, has the greatest potentialto change work’s historically tight coupling to the phys-ical and, with it, the work relations of people to objectsand to each other. Through a case study of an automobilemanufacturer, we show how innovation in digital tech-nologies and the increasing verisimilitude of iconic rep-resentations in simulation affected workers’ dependenceon physical objects and on each other. Ultimately, weshow how management’s failure to distinguish betweentypes of virtual work blinded it to new organizationaldependencies, thereby causing problems in the executionof work. From this case study, we draw larger implica-tions for the study of virtual work.

Types of Virtual WorkVirtual Teams: Operating with or onRepresentationsThe first and most widely discussed type of virtualwork in organization studies pertains to geographicallydistributed teams. Although students of “virtual teams”have used the term to gloss a host of phenomena, rang-ing from coordinating across time zones to managingthe cultural diversity of team members, all definitionspivot on the idea that teammates are spatially separatedfrom one another; thus, they must work together with-out face-to-face contact, typically via digital technolo-gies that mediate communication (Griffith et al. 2003,Gibson and Gibbs 2006). For this reason, virtual teamssubstantially change how work is organized.

From a semiotic point of view, e-mail, instant messag-ing, and other digital communication technologies facil-itate new forms of work organization by enabling people

Bailey, Leonardi, and Barley: The Lure of the VirtualOrganization Science 23(5), pp. 1485–1504, © 2012 INFORMS 1487

to interact with indices of each other. Voices on tele-phones, the text of an e-mail, or a cursor moving overan image displayed via a Web-conferencing tool standfor the other person because the voice, the typing, orthe moving cursor are indices of a person in preciselythe way that a vapor trail signifies the passing of a jetplane. Indexical representations thus render people ongeographically distributed teams virtual. Team membersoperate with these representations; for example, theyread e-mails and respond to them as a way of convers-ing with distant colleagues whom they cannot engage inface-to-face conversations.

Like many white-collar workers, members of virtualteams also operate on representations, in the sense ofcrafting or manipulating them. In these cases, the repre-sentations are typically not indices of people but sym-bolic or iconic representations that are themselves theobjects of work. Tasks that involve operating on rep-resentations are commonplace—for example, writing areport in a word processor, calculating a budget witha spreadsheet application, or drawing a building usingCAD software. In some occupations, the representationson which people operate have both a meaning and a ref-erent. Such is the case with CAD, where the drawingstands for an object that either already exists or that willbe built using the drawing as a guide. In other occu-pations and tasks, the representations on which peoplework have a meaning but no referent. When operating onrepresentations that have no referent, workers may craftmessages or change perceptions, but their manipulationsdo not directly affect physical phenomena. Consider, forexample, a distributed team of real estate analysts work-ing with spreadsheets that document the houses for salein local markets. The analysts can manipulate the num-bers in the spreadsheets to make the markets look hot-ter, which in theory should induce some homeowners tosell their homes. But no matter how the analysts changethe spreadsheets, they will not directly and immediatelycause “for sale” signs to appear in neighborhood yards.

Work that involves operating on digital representa-tions without referents is highly portable and, thus, lendsitself to execution by virtual teams: because the rep-resentations are digital, they can be easily transmittedfrom one location to another via e-mail or websites.Moreover, because the representations are not referen-tially tied to a physical entity, their meaning is free ofphysical (although perhaps not cultural) context. Virtualteams were not, however, the impetus for the creation ofsuch representations; these exist even when members ofcolocated teams work with computer applications. Theadvent of virtual teams, therefore, altered teammates’physical access to one another, but it did not change howteam members worked with many of the representationsgenerated in the course of everyday tasks.

Nor did the advent of virtual teams necessarilychange people’s roles. Managers often form virtual

teams because they require the expertise of a distantindividual (Boh et al. 2007, Sole and Edmondson 2002).For example, a team may turn to an individual in a dif-ferent country to acquire expertise with a particular mar-ket. In joining the team, the marketer would not usuallyassume new duties outside or inside the realm of mar-keting simply because she was now working virtually.Rather, she would carry out her role and its associatedtasks as before; the only difference would lie in her col-laborating with teammates via indices rather than in per-son.

Two decades of research on virtual teams tells usthat members experience a variety of problems preciselybecause digitally mediated contact forces them to inter-act via indices. Establishing trust through indices seemsparticularly difficult; absent trust, people are less likelyto share information (Kanawattanachai and Yoo 2002,Jarvenpaa and Leidner 1999). The problem, as Handy(1995, p. 46) contended, is that “trust needs touch.” Vir-tual teams also often struggle with the mechanics ofgetting work done, especially when tasks are interde-pendent. In general, task interdependence requires fre-quent coordination (Faraj and Sproull 2000), which isdifficult for virtual teams because members have trou-ble gaining access to the individuals and informationon which they depend. Numerous studies show that dis-tant others routinely fail to respond to team members’messages (Cramton 2001) and that sometimes they failto provide access to critical data (Levina and Vaast 2008,Metiu 2006). Coordination can also be difficult becauseteam members must work across time zones and cul-tures and because the communication technologies theyuse may foster incomplete messages, misunderstandings,and conflict (Maznevski and Chudoba 2000, Hinds andBailey 2003, Sproull and Kiesler 1991). Yet despitethese problems, organizations have continued to be luredby the idea that people can work just as effectively onvirtual teams as on colocated teams.

Remote Control: Operating ThroughRepresentationsA second type of virtual work involves digital tech-nologies that mediate our relations with objects ratherthan people. For example, operators in continuous pro-cess plants, such as paper mills and oil refineries, usedata collected from sensors located throughout the plantto issue commands from computer terminals to activateeffectors that change how the machines are working,all from a control room located away from the actualfactory floor. Similarly, engineers at the Jet PropulsionLaboratory monitored and operated the rover that NASAplaced on Mars via digital interfaces and commands.Firemen increasingly use robots to search for victims inburning buildings.

In such cases, people operate with data collected bya physical system and with digital representations of the


system’s functioning to change the system’s behavior. Aswith virtual teams, digitization has enabled spatial sepa-ration, but in this case, people are separated from objectsrather than other people. Instead of a digital communi-cation system as employed by virtual teams, here thetechnology is a complex cybernetic network of digitalsensors, digital control algorithms, digital state represen-tations, and digitally activated effectors. With this tech-nology, workers can remotely manipulate objects thatwere formerly amenable to only direct haptic control.

People whose work entails controlling objectsremotely are best thought of as operating through ratherthan with or on representations. In such work the goalis to directly and immediately affect the representation’sreferent. Historically, the idea that manipulating a signi-fier could affect the signified was tantamount to magic,as in the case of sticking pins in a voodoo doll to inca-pacitate an enemy. Today, remote control is becomingincreasingly prevalent as digital technologies for sens-ing and processing physical properties advance. The rep-resentations through which people operate remotely onphysical systems are likely to be either symbolic (ifinformation were displayed as, say, numbers, colors, orshapes) or iconic (if the information were displayed asreal-time pictures of processes).

Even though remote control is far from uncommon, ithas attracted less attention among students of work andorganizing than have virtual teams. One reason may bethat remote control typically occurs in technical or man-ufacturing contexts, and organizational studies of suchworkplaces have declined since the demise of indus-trial sociology. There is some evidence, however, thatmanipulating physical objects through digital interfacesprompts changes in the organization of work, alters theway people make sense of—and come to trust—theobjects with which they work, and transforms work-ers’ roles.

Studies of paper mills have documented this transfor-mation. Before the introduction of computerized infor-mation systems, operators relied on their senses wheninteracting with machines and materials to gain informa-tion about the production process (Vallas and Beck 1996,Zuboff 1988). For example, operators judged moisturecontent by running their hands over rolls of paper andlooked for weight variation by banging wooden stickson the finished product. They came to trust that whichthey could sense directly.

With the advent of remote control, workers were relo-cated to air-conditioned control rooms away from thetowering, loud machines that had populated their work-place. The isolation was figurative as well as literal: itplaced analytical distance between the operators and theobjects that had previously served as the source of theirknowledge and understanding. Now, operators neededto analyze information displayed on digital interfaces asdiagrams, maps, and graphs that served as symbols and

icons of the production process. In documenting oper-ators’ struggles to come to terms with the demands ofthis new “informated work,” Zuboff (1988) reported thatworkers had to resist the temptation to leave the controlroom to check production equipment. Operators had tolearn instead to trust the information displayed on theinterfaces as accurate representations of what was hap-pening on the floor.

These findings from paper mills resonate with stud-ies of nuclear power plants, a second context in whichwork with virtual objects has attracted scrutiny. Perrow(1983, 1999) and Hirschhorn (1984) argued that work-ing virtually with a complex, tightly coupled, technicalsystem not only increases an operator’s cognitive load,but also requires different forms of organizing preciselybecause complicated representational interfaces changethe nature of an operator’s work. For this reason, Perrow(1983) counseled that organizations should pay moreattention to the work of industrial engineers, psychol-ogists, and computer scientists who study the physical,cognitive, and social demands of working with digi-tized control systems. Unfortunately, Perrow’s counsel israrely heeded when organizations find themselves drawnto the lure of the virtual in the form of remote control.Indeed, studies of paper mills indicate that organizationsare rarely willing to consider how operating through rep-resentations might require further changes in the orga-nization of work and patterns of dependency—in part,because doing so would require managers to abdicateat least some of their power and authority to operators(Vallas and Beck 1996, Zuboff 1988).

Simulations: Operating Within RepresentationsA third type of virtual work also entails an altered rela-tionship between representations and physical entities,but it goes well beyond remote control. Rather thanmerely mediating relationships with objects or people,some new digital technologies promise, if only temporar-ily, to eliminate the need for a connection altogether.Importantly, most simulation technologies not only rep-resent physical entities, but they also emulate physicalprocesses and, for this reason, move us closer to thenotion of virtuality envisioned by science fiction writersand computer scientists.

Advanced simulations, which use computational, visu-alization, and modeling software to represent objects orpeople iconically, are often of this sort. Doctors in medi-cal schools increasingly use computer simulations of thebody to teach anatomy, dissection, and surgery in lieuof actual cadavers or patients (Prentice 2005, Csordas2001). Architect Frank Gehry used iconic representationsin design that were “complete digital prototype[s] of abuilding that [act] like the actual building” (Boland etal. 2007, p. 636). Fire engineers use simulations to studyhow fire and smoke will move through, and how peopleare likely to evacuate, a building (Dodgson et al. 2007).


In these cases, virtual no longer means working with dis-tant objects or people via representations that stand forthem: it means working solely with representations thatsubstitute for the object or person.

When workers substitute digital simulations forobjects or people, they no longer simply operate with,on, or even through representations. They begin tooperate within them. Operating within a representationmeans that the worker’s connection to the referent issuspended. The presumption is that what we learn fromsimulating is equivalent to what we would have learnedhad we experimented with the physical object or per-son being simulated. Another way of putting it is thatwe take a model of the phenomenon to be an adequatefacsimile of the phenomenon itself.

Writing before the widespread diffusion of comput-ers, the philosopher Max Black (1962, p. 222) defined amodel as a “material object, system or process designedto reproduce as faithfully as possible in some newmedium the structure or web of relationships in anoriginal.” Because models are referential, Black argued,they are useful for helping us develop hypotheses aboutthe phenomena that the model signifies. Yet preciselybecause models are referential, working with them car-ries an inherent danger, a danger that grows as themodel’s verisimilitude increases: mistaking the behav-ior of the model for the behavior of its referent. Blackcommented (1962),

The remarkable fact that the same pattern of relation-ships, the same structure, can be embodied in an end-less variety of different media makes a powerful and adangerous thing of the 0 0 0model. The risks of fallaciousinference from inevitable irrelevancies and distortions inthe model are now present in aggravated measure. Anywould-be scientific use of a 0 0 0model demands indepen-dent confirmation 0 0 0 0 Models furnish plausible hypothe-ses, not proof 0 0 0 0 The drastic simplifications demandedfor success of the mathematical analysis entail a seri-ous risk of confusing accuracy of the mathematics withstrength of empirical verification in the original field.Especially important is it to remember that the mathe-matical treatment furnishes no explanations. Mathematicscan be expected to do no more than draw consequencesfrom the original empirical assumptions. (pp. 222–225,italics in original)

Engineers who design and use simulations voice sim-ilar warnings when distinguishing between verificationand validation (Cunningham 2007, Kurowski 2008). Ver-ification entails checking the assumptions behind and theimplementation of the equations, parameters, and algo-rithms that comprise a mathematical model. By valida-tion, engineers mean checking a simulation’s predictionsempirically against reality—namely, the performance ofthe actual objects under the conditions being modeled.

Thornton (2010) interviewed numerous mechanicalengineers who specialized in finite element analy-sis (FEA), a sophisticated mathematical method for

simulating how objects respond to kinetic force, friction,wind shear, loads, and other types of stress or strain.Thornton’s engineers consistently reported that inade-quate validation is the most common problem with sim-ulation analysis. Thornton (2010, p. 42) wrote,

The biggest challenge in FEA is validation, carefullychosen and closely monitored physical tests that con-firm whether or not physical reality and virtual realityline up. A consensus among FEA analysts is that vali-dation ensures there are no hidden disconnects betweenthe model and the physical testing, that correct physi-cal properties are used and that properties are analyzedaccurately based on correct principles of physics.

The two most common reasons cited by Thornton’sengineers for inadequately validated models were engi-neers inexperienced with the intricacies of simulationand managers whose desire to cut costs leads them toassume that simulations are sufficient evidence. As oneof Thornton’s informants put it, “Too many engineeringmanagers do not understand the vast complexity of thephysical world that FEA addresses 0 0 0 0 Any mechanismthat allows developers to reduce the cost and time forproduct or process development is being seized upon byengineering managers” (2010, p. 41, emphasis added).

Trusting simulations to the point of mistaking themas complete substitutes for reality is not limited to engi-neering managers who are under pressure to cut costs.Turkle (2009) described how students believed too read-ily in the computer simulations of natural processesand experiments they viewed in physics and chemistryclasses: “When students claimed to be ‘seeing it actuallyhappen’ on a screen, their teachers were upset by how arepresentation had taken on unjustified authority” (p. 29,italics in original). The verisimilitude of iconic represen-tations on computer screens prompted one professor toconstantly remind his students that “simulations are notthe real world 0 0 0 0 There is no substitute for knowingwhat a kilogram feels like or knowing what a centimeteris or a one-meter beach ball” (Turkle 2009, p. 30). Suchreminders were important because students had to learnthat simulations could only match reality if the modelswere consistently validated against data obtained fromthe real world.

Table 1 summarizes the three types of virtual workin terms of digitization, representations, and virtuality.Table 2, which shows changes in work that arise in thecontext of virtual work for each of the three types, makesclear the gaps in our knowledge about changes in workorganization as well as tasks and roles under simulation.In addition, although Table 1 indicates that digitizationin simulation permits spatial separation between peopleand objects or other people, Table 2 points out problemsin model validation and excessive trust in models thatspeak against the logic of such separation.

In the remainder of this paper, we explore how thedevelopment of increasingly iconic simulations gradu-ally altered the organization of engineering in a large


Table 1 Digitization, Representations, and Virtuality by Type of Virtual Work

Type of virtual work

Virtual team Remote control Simulation

DigitizationPurpose of digitization Mediated communication Mediated operation Emulated operationTechnology Communication systems Complex cybernetic networks Computational, visualization, and

modeling softwareSpatial separation Between team members Between people and objects Between people and objects or

other people

RepresentationsWorkers’ relationship to

representationsOperating with or on Operating through Operating within

Common types ofrepresentations

With referents: Voices ontelephone, e-mail, CADdrawings

Without referents: Budgets,reports

With referents: Flowcharts,displays, diagrams, maps,graphs

With referents: Models,animations

Manipulation affectsreferent?

No Yes No

VirtualityPurpose of virtuality Distant collaboration Distant control Study and experimentationPhysical entity made

virtualPeople Object Object or people

Interaction betweenphysical and virtual

People interact with one anothervia indexical representationsthat stand for the teammembers

People interact with existingphysical objects via symbolicand iconic representations thatstand for the objects and theirassociated processes

People work with primarily iconicrepresentations that substitutefor physical entities (existing orfuture) and their associatedprocesses

Examples Distributed team of real estateagents

Paper mills, oil refineries, nuclearpower plants

Medical education, buildingdesign, new productdevelopment

automobile company, as well as the tasks and roles ofengineers. As the simulations’ representations becamemore iconic and realistic, managers’ failure to appreciatethe subtle relationship between the sign and referent ledthem to confuse the act of operating within representa-tions with the act of operating with or on representations.As a result, they came to believe that operating within

Table 2 Changes in Work Organization by Type of Virtual Work

Type of virtual work

Changes in work organization Virtual team Remote control Simulation

Work structure Interdependent workersplaced on teams withmembers distributedgeographically

Workers colocated with eachother, computers, andcontrols beyond theimmediate vicinity ofproduction machines

?

Tasks and roles Few or no changes Increased analytics,“informated” work, workersequipped for decisionmaking

?

Related problems Lack of trust among teammembers,misunderstanding ofindexical communication,coordination mishaps

Lack of trust inrepresentations, cognitiveoverload

Excessive trust in models,difficulty validating models

representations was a form of work amenable to virtualteams because they saw little need for access to physicalvehicles. By assigning simulation work to virtual teams,managers organized work and assigned tasks and rolesin ways that unwittingly broke the link between iconsand referents (and, hence, the ability to validate models),thereby creating difficulties.


MethodsResearch DesignInternational Automobile Corporation (IAC, a pseu-donym), headquartered in the United States, was thesetting for our study. Although the majority of IAC’sengineering workforce resided at its technical center inMichigan, IAC had long maintained engineering oper-ations abroad. In 2003, IAC opened a new center inBangalore, India to provide digital engineering servicesto IAC’s other engineering centers. Unlike IAC’s othercenters, Bangalore had no physical testing facilities;the workforce only provided math-based analyses andassisted with setting up simulations. In fact, manage-ment purposefully created the India center as a steptoward its goal of completely replacing physical testswith virtual ones. The decision to locate in Bangalorewas telling. With its large information technology infras-tructure, Bangalore was the offshoring, but not the auto-motive, capital of India (Chaminade and Vang 2008). Wefocused on IAC’s U.S. center and its interactions withthe India center. At the U.S. center, we studied engineersinvolved in product design, physical testing, and simula-tion. Because there were no engineers in product designor physical testing in India, we only studied Indian engi-neers who did simulations.

Data CollectionWe collected data between July 2003 and July 2006primarily through ethnographic observation of, andinformal interviews with, engineers at work. At the U.S.center, we observed 10 engineers in physical testing,7 engineers in product design, and 17 engineers in sim-ulation. In general, we observed each of these engineerson four separate occasions. In addition, we conducted 11semistructured interviews at the U.S. center with engi-neers, engineering managers, and individuals who man-aged, purchased, or developed the technologies that theengineers used. Because many of the American engi-neers and managers had considerable tenure with IAC,our conversations and interviews covered not only onthe way work was currently done but how it was doneand organized in the past. By combining such recol-lections with documents from earlier years, we wereable to construct a history of how design and per-formance analysis had evolved at IAC. We observed11 engineers in Bangalore. Because we observed mostIndian engineers on fewer occasions than we did engi-neers in the United States (typically twice), we con-ducted half-hour semistructured interviews with eachIndian engineer that we observed. We complementedthese interviews with 20 interviews of other Indian engi-neers and managers. Interviews at both sites were audio-taped and transcribed. Altogether, we observed engineersat work on 163 occasions.

Data collection spanned two countries and took threeyears to complete. Over this time, 18 individuals worked

on the research: 2 faculty members, 3 graduate stu-dents, 3 undergraduate students, and 10 IAC researchand development staff. Although single researchers carryout most qualitative field studies, some studies haveused teams and were, therefore, useful as we designedour research (Barley 1996, Miles 1979). Specifically,they taught us that we would need to develop specifictechniques to ensure consistency in how team membersrecorded data.

In addition to taking running notes on the engineer’sinteractions with technologies, people, and documents,we audiotaped conversations when we could not keeppace by writing notes or when discussions became verytechnical. When our informants worked at computers,we requested screenshots of software interfaces and dig-ital copies of the models, e-mail, and documents withwhich they worked. Each day, we photocopied key doc-uments that the engineers had used, including drawings,sketches, pages from brochures, handwritten calcula-tions, and scraps of paper on which they had scribblednotes.

Collecting so many types of data enabled us to pre-pare field notes that described actions, conversations,and visual images simultaneously and, thus, to producea record not only of what engineers did and said butalso of what they worked with and created. The firststep in assembling a day’s field notes was to expand therunning notes taken in the field into full narratives thatsomeone who had not been on-site could understand.We transcribed tapes directly into the body of this nar-rative at the point where the talk occurred. Similarly, weindexed screenshots and photocopies of documents atthe point in the field notes where they were used. Weav-ing together actions, conversations, and images allowedus to capture and better understand an engineer’s ges-tures, movements, and, especially, the referents of index-ical speech (e.g., “Here, you see 0 0 0 ” or “Right here,we need 0 0 0 ”). We wrote appendices for each day’s fieldnotes that described the documents we collected and, ifthe document was the result of the engineer’s work, howit evolved. These descriptions also covered how and whythe engineer used the document. Completing a full nar-rative of a day’s observation took between two and twoand a half days.

Data AnalysisWhereas the entire research team was engaged in datacollection activities, only the authors performed dataanalysis. Following Yin (1994), we organized our analy-sis around the motivation for our study—namely, deter-mining what happens when new digital technologiesfacilitate operating with, on, and within representationsof physical entities, as well as the transmission of theserepresentations across great distances.1 We paid partic-ular attention to how engineers understood and man-aged the relationship between digital representations and


their physical referents as well as to the organization ofwork both within and between the two sites we studiedover time.

Beginning with a within-case analysis (Miles andHuberman 1994), we first examined the U.S. centerand built separate narratives for each engineering role:product design engineers, physical testing engineers,and simulation engineers. For each role we documentedwork tasks and constructed a narrative of its history atIAC. We paid attention to how engineers in each roleinteracted with and used physical parts. Because sim-ulation represents the substitution of the representationfor the referent, we particularly noted engineers’ accessto, and dependence on, these objects as we consideredwork structure. We paid similar attention to engineers’interactions with, dependence on, and access to peers inother roles. We conducted multiple readings of our fieldnotes and interview transcripts, refining our narrativesin the face of supporting or disconfirming evidence forour developing ideas. We next built the same kind ofnarrative for the simulation engineers at the India center.

Finally, we turned to cross-case analysis, looking fordifferences across the engineering centers (Miles andHuberman 1994). Our focus in this step was primar-ily on the simulation engineers because the India centerdid not have the other roles. We paid particular atten-tion to how, and if, simulation engineers at both sitesvalidated their models. We contrasted and compared theengineers’ access to and dependence on physical objects.For the India center, we specifically explored how engi-neers handled their limited access to objects.

Because the historical narrative of the various occu-pations was strongly influenced by the adoption of newdigital technologies, we organized our results by thesequence in which engineering at IAC adopted CAD,FEA, and computer-mediated communication (CMC)technologies. We explain how work was divided acrossengineering roles in each period, how each new tech-nology offered new representations (symbolic, iconic,or indexical) of objects or people, how management’sbelief that one could separate simulation’s representa-tions from their referents led to the global distributionof this work, and how this change eventually led to anarray of problems.

Changes in Technology, Work Organization,and Representations at IACDistinct historical periods, each associated with thedeployment of a new technology, punctuated the pro-cess by which the organization of work at IAC changedover time. We begin by describing the situation beforethe arrival of CAD (Period 1). We then describe whatoccurred when IAC adopted CAD in engineering design(Period 2), what transpired when FEA was deployed forengineering analysis (Period 3), and finally what took

place after IAC used CMC technologies to offshore sim-ulation work to India (Period 4). For each period wedocument how roles, tasks, and engineers’ dependenceon and access to objects and people changed.

Period 1: Pre-DigitalBefore CAD arrived, IAC’s engineering divisiondesigned vehicle parts, produced blueprints of thoseparts, and analyzed how vehicles performed using phys-ical tests. Under the division of labor during this period,parts engineers designed parts and oversaw the build-ing of prototypes. Drafters drew the blueprints for theparts. Test engineers assessed the actual performance ofparts and vehicles. Modelers in the research and devel-opment (R&D) division were responsible for analyzingvehicle performance with primitive mathematical mod-els. Table 3 summarizes the allocation of tasks, the engi-neers’ dependence on and access to physical parts, andinterdependencies over time for engineering and R&D.

The top panel of Table 3 (Period 1) reveals thatbefore the arrival of CAD, all roles in engineeringdepended heavily on and had good access to vehicleparts. Test engineers had the best access. They split theirtime among teardown rooms for inspecting disassem-bled vehicles and experimental bays for studying vibra-tion and other phenomena. They also test drove vehi-cles on a racetrack at IAC’s proving grounds 30 milesaway. Parts engineers and drafters, all of whom oper-ated on blueprints of physical objects, were separatedfrom the test engineers in another building on IAC’smain campus. As a result, their access to parts and vehi-cles was slightly limited. Nevertheless, parts engineersand drafters kept key vehicle parts in their cubicles, andwhen they needed better access, they went to the testfacilities to witness physical tests or to inspect partsbefore or after testing.

The three engineering occupations not only dependedon access to parts, but they also depended heavily oneach other and were sufficiently close to each other tointeract face to face. Parts engineers needed drafters’drawings to negotiate with suppliers, manufacturingstaff, and other engineering groups; they used test engi-neers’ results to modify designs when parts failed.Drafters relied on parts engineers to set drawing speci-fications. Test engineers depended on parts engineers tohave parts built to specifications to ensure valid tests.

Figure 1 illustrates the physical objects and the repre-sentations that IAC engineers used to design parts andstudy performance over time. The top panel pertainsto designing vehicle parts; the bottom panel pertainsto analyzing vehicle performance. Said differently, thetop panel displays representations of structure, whereasthe bottom panel displays representations of functioningor process. Before CAD, parts engineers and draftersused physical vehicle parts (such as the actual engine


Table 3 Work Organization and Technology in Engineering and R&D at IAC by Period

Division Engineering R&D

Period 1: Pre-digitalRole Parts engineer Drafter Test engineer ModelerTask Design of parts Drawing of parts Performance analysis

(physical)Performance analysis (math)

Dependence onphysical parts

High High High Low

Access to physicalparts

Moderate Moderate High Low

Task interdependenceacross roles

High within engineering, low across engineering and R&D

Access to individualsin other roles

Moderate

Period 2: Computer-aided design arrivesRole Design engineer Test engineer ModelerTask Design and drawing of parts Performance analysis

(physical)Performance analysis (math)

Dependence on physical parts High High LowAccess to physical parts Moderate High Low

Task interdependence across roles High within engineering, low across engineering and R&DAccess to individuals in other roles Moderate

Period 3: Simulating performanceRole Design engineer Test engineer Simulation engineer Math tool developerTask Design and drawing

of partsPerformance analysis

(physical)Performance analysis

(math)Creation of new

math toolsDependence on physical parts High High High (but presumed to

be 0 by managers)Low

Access to physical parts Moderate High Moderate Low

Task interdependence across roles High within engineering, moderate across engineering and R&DAccess to individuals in other roles Moderate

Period 4: Offshoring analysisRole Design engineer Test engineer Simulation modeler

(in India)Simulation analyst

(in the UnitedStates)

Math tooldeveloper

Task Design and drawingof parts

Performance analysis(physical)

Construction ofperformancemodels (math)

Analysis ofperformancemodels (math)

Creation of newmath tools

Dependence on physicalparts

High High High (but presumedto be 0 bymanagers andsimulationanalysts)

High (butpresumed to be0 by managers)

Low

Access to physical parts Moderate High Low Moderate Low

Task interdependenceacross roles

High within engineering, moderate across engineering and R&D

Access to individuals inother roles

Low

in the photograph of Figure 1a) or formal engineer-ing drawings (such as the two-dimensional (2D) paperdrawing of engine components in Figure 1b). Like allengineering drawings, the drawing of the engine com-ponents in Figure 1b is an iconic representation of itsreferent: the drawing reflects the engine’s size and shapeas well as the position of its components. At this point,engineers had no technology that could iconically rep-resent how vehicles performed or how parts functioned.Test engineers relied solely on physical tests using real

parts and real vehicles, as shown in Figure 1d, a photo-graph of a frontal impact test.

Whereas the engineers worked with real vehicle partsand on their iconic representations, the R&D groupthat analyzed vehicle performance worked primarilywith symbolic representations of parts and vehicles. Inthe mid-1960s, the R&D group began using “lumped-parameter” models.2 An example of such a model isfound in Figure 1e. Lumped-parameter models repre-sented vehicles as masses connected by springs. The


Figure 1 Physical Parts/Tests and Their Representations

(f) 3D FEA simulation(d) Physical test

2 7

4 5

63

8

1

Bumper Engine

Cradle

Analyzing vehicle performance

(a) Physical part (b) 2D pencil drawing (c) 3D CAD rendering

Desiging vehicle parts

9

(e) 2D lumped-parameter model

representations were symbolic, with parts such as thebumper, engine, and cradle in Figure 1e represented byquadrilaterals that in no way mirrored the actual form ofthe parts. Likewise, one would not find the wiry springsof the lumped-parameter model under the hood of a realvehicle; the springs represented the parts’ physical prop-erties and enabled calculations of their performance onimpact. Located in a separate building on IAC’s maincampus, R&D personnel had limited access to physicalparts. This was not a problem, however, because R&Ddid not depend on parts for its work.

R&D modelers used mass-and-spring models’ simplemathematical relationships to describe a vehicle’s per-formance. Because lumped-parameter models were sym-bolic and bore no resemblance to real vehicles, theirdepiction of a vehicle’s performance was not very real-istic or accurate. As one simulation engineer who beganworking in R&D during the late 1960s recalled, sym-bolic analyses of this type were of little use for engi-neering work:

The [lumped-parameter] models were of good theoret-ical value, but they weren’t too practical. They repre-sented some abstract notion of a vehicle, but you couldn’tuse them to design a vehicle. They were more like athought experiment to see if we understood what hadjust happened in a [physical] test and if we could doit in math 0 0 0 0 That’s why we were in R&D and notengineering 0 0 0 0 We worked with ideas of vehicles, butour models didn’t really represent vehicles, at least thephysical manifestations of them.

That lumped-parameter models were of little use forengineering meant that mathematical analysis of vehi-cle performance played no role in design. Consequently,task dependence between the engineering groups and theR&D modelers was very low.

Table 4 summarizes the changes in the nature of howvehicle parts, vehicle performance, and people were rep-resented during each of the periods we consider. Thetable indicates whether the representations used duringa period were symbolic, iconic, or indexical. As therows for Period 1 indicate, 2D paper drawings wereiconic representations of the structure of parts, whereaslumped-parameter models were symbolic representationsof how vehicles performed. Because lumped-parametermodels were so symbolic and at best vaguely referential,engineering had no choice but to rely on physical partsand physical tests to analyze vehicle performance.

Table 4 Changes in the Nature of Representations by Period

Referent

Vehicle VehiclePeriod Representation part performance Person

1 2D paper drawing IconicLumped-parameter

modelSymbolic

2 3D CAD rendering Iconic3 3D FEA model Iconic

3D FEA simulation Iconic4 Voice, e-mail, other

communicationIndexical


Period 2: Computer-Aided Design ArrivesIAC brought CAD to engineering design in the late1970s. CAD initially allowed 2D and, eventually, 3Drepresentations of parts and vehicles. CAD renderings,like the paper drawings they replaced, iconically repre-sented the vehicle parts they referenced. As Figure 1cshows, a 3D CAD rendering is a highly detailed andaccurate representation of an object’s structure (in thiscase, an engine’s).

CAD replaced paper with digital drawings, but it didnot change engineers’ relationship with representations.With CAD, engineers continued to operate on repre-sentations. In other words, even though CAD allowedengineers to create and manipulate 3D images, it didnot alter the balance of symbolic and iconic representa-tion because paper drawings were already iconic. Con-sequently, the row for 3D CAD in Table 4 reiterates therow for 2D paper drawings.

CAD renderings did, however, blur the distinctionbetween the work of drafters and parts engineers. As inother settings that adopted CAD, managers realized thatwith CAD, IAC no longer required the drafters’ skills(Cooley 1980, Susman and Chase 1986). Accordingly,they combined the work of drafters and parts engineersinto the new role of design engineer. (See the secondpanel of Table 3.) Despite the shift to CAD, designengineers remained highly dependent on physical parts,as had parts engineers and drafters before them. Sim-ilarly, task dependence stayed the same (high amongengineering roles, low between engineering and R&Droles). There was simply one less player in the game,and the work of test engineers and R&D modelers wasunaffected.

Period 3: Simulating PerformanceAlthough CAD did not significantly alter the balanceof iconic and symbolic representation or how engineersworked with representations, the arrival of FEA did.Rather than model an entire vehicle as a handful ofmasses, as in lumped-parameter models, FEA modelsrequire that analysts decompose a vehicle into a large,finite number of small triangular or rectangular areasknown as “elements.” Adjacent elements are connectedat their nodes to define a system of connected elementsor a structure called a “mesh.” (See Figure 2 for anexample of a 3D FEA model with a mesh.) Analysts

Figure 2 3D FEA Vehicle Model with Mesh

represent the system of equations that define the meshas a very large matrix, which they use when simulatingvehicle performance.

Early FEA models were nearly as symbolic as thelumped-parameter models they challenged because itwas difficult to determine visually whether a simulatedvehicle had performed like a real vehicle in a physi-cal test from their 2D output. Therefore, throughout the1970s, IAC experimented with FEA models in R&D butnot engineering. Ultimately, the development of 3D FEAmodels paved the way for mathematical performanceanalysis to migrate from R&D to engineering because3D images made the modeling of performance moreiconic and more obviously referential. As one simulationengineer noted,

When [a 3D application] came out we knew it wouldn’tbe long before we could move simulation work into theengineering functions. The 3D capabilities let you visu-alize the vehicle. So the math you did actually gaveyou a visual representation that corresponded to theactual [physical] vehicle. You could see correspondencebetween what you were doing and what was happeningin the real world.

Nevertheless, as engineers who worked with early ver-sions of FEA observed, the verisimilitude of the repre-sentations was still far from adequate:

I started working with simulation tools in the early‘90s 0 0 0 0 They were cool because 0 0 0you could actuallylook at your screen and see the vehicle and how itcrashed. But they just weren’t that accurate. You had arepresentation of the vehicle, but it didn’t have the detailyou’d need to really do things. So you could sort of onlyuse it as a supplement to the physical tests.

[Simulation engineer]

Getting more accurate results required building morerefined FEA models with more and smaller elements.Using a finer mesh, however, required considerable com-putation. It was not until the mid-1990s, when comput-ers had become sufficiently fast and powerful, that IACengineers could refine their FEA models. The results ofthese new models looked increasing like the results ofphysical tests. Indeed, by the early 2000s, FEA’s iconsachieved such verisimilitude that they resembled digi-tized animations of physical crash tests. (Compare Fig-ure 1d with Figure 1f.) The strong resemblance betweenFEA simulations and physical tests convinced IAC man-agers that mathematical analysis could ultimately replacephysical analysis. They believed that working withinrepresentations would become sufficient and that inti-mate knowledge of the representations’ referents wouldbecome superfluous:

We want to move toward doing everything in math so wecan have a completely virtual [engineering process] 0 0 0 0It costs too much to run crash tests and it’s too slow tobuild the vehicles and to staff the testing facilities. Ulti-mately, we want simulations to replace physical testing.


We’re not there yet, but we’re pretty close 0 0 0 0 The [sim-ulations] are getting to be so good that our engineerscan rely on them and not have to go to the [physical]hardware. [Director of engineering, emphasis added]

As FEA moved into engineering, a new role emerged:the simulation engineer (see Period 3 in Table 3). Sim-ulation engineers spent most of their time in front of acomputer screen. They built meshes and models usingthe CAD files created by the design engineers and ana-lyzed performance digitally. From management’s per-spective, simulation engineers appeared to work withoutthe need to touch or see physical objects, but in reality,they did not. By moving FEA analysis to engineering,management had provided simulation engineers withde facto access to vehicle parts. Such access was crucialto the simulation engineers’ ability to do their work.

For example, one engineer we observed had spentseveral hours analyzing a simulated crash of a pickuptruck. Visually, the FEA model looked identical to thereal truck, but the simulated passenger dummy seemedto be hit by an expanding airbag with more force thanhad the real dummy in the physical test. Frustrated thathis model did not match reality, the engineer went tothe proving grounds where he sat inside the truck. Afterinspecting the seat anchors and the seat belt, he turnedhis attention to the dashboard. This excerpt from ourfield notes explains his discovery:

He lifts up the deflated airbag and inspects its connectionto the instrument panel. He rubs his finger on a blackimpression that looks like a smudge of shoe polish. Herotates the bag looking for other signs of the smudge,but sees none other than this one spot. He lets go of theairbag and returns to his exploration of the dash. He runshis hand up the instrument panel and grabs the passengerassist bar located on top of the airbag storage unit. Hesays, “Hmm,” and lifts the plastic flap that once coveredthe airbag compartment but is still attached, on its topedge, to the dashboard. When lifted, the flap seems tocome up about an inch over the top of the bar. He wipeshis finger on the inside of the flap and looks at his fingerto see that there is a black powdery residue on it. “It lookslike there is something going on here,” he says out loud.

Returning to his office, the engineer located a video ofthe physical test. As he watched the crash on his work-station, he noticed that the flap that once concealed theairbag flipped upwards and covered the passenger assistbar as the airbag deployed. He then positioned the sim-ulation in a window next to the video and watched themsimultaneously. At that point he noticed that he had notmodeled the flap covering the airbag, which meant thatthe virtual airbag became caught underneath the assistbar, delaying its expansion for 30 milliseconds. In thephysical test, the flap flipped up and covered the bar,so the airbag did not get caught underneath. (See thevideo still in Figure 3a and the simulation still in Fig-ure 3b.) Because the bar temporarily inhibited the virtual

airbag’s expansion, the virtual dummy had to travel far-ther to reach the airbag. Consequently, it experiencedmore force than it would have if the assist bar had notsnagged the airbag as it expanded. Had the engineer notnoticed the discrepancy between the model and realitywhile validating his model, the National Highway Trans-portation and Safety Administration would have giventhe vehicle a lower occupant protection rating.

The airbag incident reveals that simulation engi-neers treated FEA models as adequately referential onlybecause they worked tirelessly to validate their mathe-matical results against the results of physical tests, a pro-cess that often demanded that they inspect physical parts.Experienced engineers knew that discrepancies betweenreality and simulation usually meant that the simulationwas flawed. Note, however, that discrepancies could notbe discovered and rectified in the absence of the simula-tion’s physical referent. The verisimilitude of the iconsthat FEA produced could only be assessed relative to theicon’s referent, a point that was largely lost on managers.

The simulation engineers’ dependence on physicalobjects spawned a dependence on the engineers whocontrolled access to those objects. Test engineers con-trolled access to physical parts and tests that simula-tion engineers used to validate their models and results.Simulation engineers also depended on design engineersbecause the latter created and controlled the CAD filesfrom which they built their meshes. In theory, the designengineers also depended on simulation engineers. Man-agers had shifted FEA analysis to engineering preciselybecause they wanted design engineers to use the resultsof the simulations to avoid costly redesign later in theinnovation process. Despite management’s pressure totreat the simulations as adequate evidence, design engi-neers often distrusted the models until they saw the sameresults in physical tests. Regardless, task interdepen-dence among the occupations in engineering remainedhigh (see Table 3).

Because they no longer modeled vehicle performance,R&D modelers began devoting their energies to cre-ating better tools for mathematical analysis. To createnew tools, R&D personnel regularly needed access tothe design engineers’ CAD renderings and the simula-tion engineers’ FEA models for test cases. This need,combined with the fact that R&D now developed toolsthat engineers would use, reflected increased task inter-dependence, but not enough to make either functional orphysical separation problematic on a day-to-day basis.

Period 4: Offshoring AnalysisIconic representations of vehicle performance promptedmanagers to tout simulation as the way of the future,envisioning a day when the engineering process wouldbe “completely virtual.” Managers trusted simulationsmore than the simulation engineers who produced them


Figure 3 Comparison of Airbag Deployment Between Physical Test and 3D FEA Simulation

(normal airbag deployment) (airbag caught under bar)(a) Physical test: flap blocking bar (b) 3D FEA simulation: no flap modeled

did. The engineers recognized that it was only reason-able for them to work within representations so long asthey could validate the models against the actual partsand processes they were simulating. Consider, for exam-ple, the following exchange that occurred during a meet-ing to chart the design for a particular car program.Participants were looking at simulations projected on ascreen:

Engineer: What we keep getting here [pausing the ani-mation and pointing with his finger to thelocation of the engine mounts] is that theengine mounts break loose 0 0 0 0 So then whatwe did was move the location of the mountsby 12 millimeters 0 0 0 0 [pointing to the newlocation] And we get this—let me show you.[He plays a new animation with the enginemounts in the new position.] So you can seethey stay intact here [pointing to the location],which is what we want.

Manager: [calling on an engineer from another group]Eric, is that location going to affect the curvein the rail you have, because we want thefront end to be low from a design perspec-tive for aero, but also for styling? So will thischange that?

Eric: It doesn’t look like it. I don’t really think so.

Manager: OK, well I say we go with this plan then, ifthat’s the best direction we’ve got.

Engineer: [hesitantly] But we don’t know exactly if thisis going to work because we have troublemodeling the engine mounts.

Manager: Well, it [the simulation] is showing good per-formance, right?

Engineer: Yeah, but we’re not sure about the weld pat-terns there, whether they’re going to hold. Weneed to do some [physical] tests of this.

Manager: But the model doesn’t show any problems,right?

Engineer: Right.

Manager: Then do we really need the [physical] test?

IAC’s growing concerns over engineering costs coin-cided with the managers’ belief that simulation couldreplace physical tests and prompted senior managementto urge engineering managers to favor simulations overexpensive physical testing. The engineering managersdecided that the way to do more simulations withoutincreasing expensive U.S. headcount was to offshore thework to India, where labor was cheaper:

When we were thinking about how to scale our simula-tion work, we realized that we had high fidelity in ourmodels. So, you didn’t need to go to the proving groundseach day to map what you were seeing on the screenonto reality. We looked around and said, we’ve got lotsof bandwidth with our Internet backbone. So, it makessense that we can take things that we could only do inMichigan just a couple years ago and do them anywherein the world. Maybe even in places where our labor costsweren’t quite so steep. It seemed obvious to go to Indiabecause all the information anyone needed to do thework was in the model.

[Senior engineering manager, emphasis added]

The idea that models contained all the necessary infor-mation for analysis was equivalent to saying that onecould operate within iconic models without recourseto their referents. Managers, therefore, took what theythought was low dependence on the physical as a warrantto offshore simulation to the India center. Importantly,


Table 5 Nine Tasks in Building and Analyzing a VehiclePerformance Simulation Model

1. Build an FEA mesh from CAD files of parts supplied by designengineers.

2. Set the conditions for an FEA model to be run with the mesh.3. Run a simulation using the model.4. Analyze and interpret the simulation results.5. Correlate these results with the physical test to validate the

simulation; modify the FEA model if needed and repeat tasks2–5.

6. Develop improvement ideas for design based on validatedsimulation results.

7. Create case studies (additional simulations) to test theimprovement ideas.

8. Analyze and interpret the results of the case studies.9. Make recommendations to change the design of parts (and,

ultimately, the CAD files).

the India center had no teardown facilities where engi-neers could inspect parts and no proving grounds wherethey could observe physical tests. CMC technologies inthe form of FTP sites, high-bandwidth Internet connec-tions, and e-mail would allow IAC to rapidly transmitFEA models back and forth between the United Statesand India. In short, management thought IAC could dosimulation analysis effectively using virtual teams, thuscombining two types of virtual work.

Although the simulation engineers argued againstsending work to India, management insisted. Thus,the engineers divided their work process into the ninesequential tasks, displayed in Table 5. They reasonedthat tasks 1–3 (building a mesh and running a simula-tion model) could be sent to India, but subsequent tasks(analyzing and validating the results, making recommen-dations) should remain in Michigan. Simulation engi-neers viewed building a mesh as the least interesting partof their work. They also thought that engineers did notneed access to physical referents until they began analyz-ing and validating a simulation. A simulation engineerexplained,

We do lots of tasks. Most of what we do is analysis, but alot of our time is just model building. That’s routine, sortof standard stuff. It’s also not so detailed. What I meanis you’re working to build a mesh and you’re deciding onhow to shape and refine elements and connections. That’sall abstract stuff. So, you don’t need to look at partsor see tests because real objects don’t have elements—they’re not divided into boxes for computation. It’s justmore removed, at that point, from the actual vehicle.

Once offshoring commenced, simulation engineersroutinely sent the first three tasks to India, but theyalmost never sent later steps in the simulation process.The resulting division of labor effectively parsed thesimulation engineer’s job into two new roles: a simula-tion modeler residing in India and a simulation analystresiding in the United States (see the bottom panel inTable 3). Task interdependence in engineering remained

high (arguably, higher with the splitting of the simulationengineer’s role) with no change in engineering’s relationto R&D.

As it turned out, the simulation engineers underes-timated the importance of the knowledge they gainedthrough regular access to vehicles and parts. As a result,they were wrong in assuming that one could build mod-els without access to their referents. To understand whyaccess was important, we can explore what happenedeach time a simulation engineer began to build an FEAmodel from design engineers’ CAD files.

Inevitably, whenever a simulation engineer began cre-ating an FEA model from CAD files, some virtual partsjutted through others. Such overlaps occurred becausedesign engineers frequently altered their part’s design(and, hence, the CAD files) based on feedback fromphysical tests and simulations. Changes in the sizeor shape of one part typically had ramifications fornearby parts. However, because design engineers oftenneglected to tell each other about changes, the engineersresponsible for adjacent parts failed to alter their designs,resulting in an overlap.

Whenever two parts overlapped in the mesh—whatengineers called a “penetration”—the simulation engi-neer would have to decide how to remedy the problem.Because penetrations were numerous, most engineersopted to correct only those for parts that were impli-cated in the test they intended to run and ignored thosethat would not meaningfully affect the simulation. Sim-ulation engineers knew which parts of the vehicle wereimplicated from observing teardowns after physical tests.A simulation engineer explained how he decided whichpenetrations to correct:

Well, you just know from seeing the [physical] test. Yougo see the vehicle after the test and you look and seewhat areas of the vehicle were impacted—you see whatwas damaged and what wasn’t. You can even touch theparts to get a good sense of it. That gives you a sense ofthe basic area of the vehicle that will be your load path.That’s the area that you have to focus on. All the otherareas you can leave because they’re not so involved. Youjust look at test after test after test, and you eventuallylearn what your model should be like.

Although the Indian engineers were quite techni-cally competent, they immediately encountered prob-lems when fixing penetrations for three reasons. First,because automobiles are not nearly as common in Indiaas they are in the United States, the Indian engineerslacked the American engineers’ cultural familiarity withautomobiles. Second, because they also lacked accessto physical parts and the results of the physical teststhat they were asked to simulate, they could not checktheir models against referents. Finally, the Indian engi-neers confronted the coordination problems associatedwith most virtual teams, which, in turn, complicated theother two problems.


Most of the Indian engineers had never driven, muchless owned, an automobile. In fact, cars were so rarethat most Indian engineers lacked the kind of everydayknowledge of vehicles that Americans take for granted.For example, when asked to develop a model of a fuelsystem on a truck, an Indian engineer returned a modelwith a fuel fill pipe and fuel door on both sides ofthe vehicle. The engineer who developed the model hadsaved time by using the program’s mirror function toreflect the first half of his model to create the second.He did so because he failed to recognize that vehicleshave fuel intakes on only one side. The Indian engineersrealized that they did not have adequate familiarity withautomobiles and that this hampered their ability to fullyunderstand the models they were building and, in par-ticular, to deal with penetrations. To provide experience,managers arranged for the engineers to visit local cardealers, who hoisted vehicles in the air to allow the engi-neers to examine underbodies. The few Indian engineerswho owned vehicles received constant inquiries fromtheir colleagues about various aspects of the vehicle.

Furthermore, because Indian modelers had no accessto physical tests or teardowns, they were unsure whichpenetrations they should fix. Faced with such uncer-tainty, the modelers sometimes resorted to guessingwhich parts might be implicated in a particular analysis.Consider the following interaction between Suresh andVijay, as they examined Suresh’s computer screen onwhich an FEA model for a frontal impact simulation wasdisplayed:

Suresh: There are too many penetrations here [pointingto an area of the model on the screen] that needto be cleared up. Richard Danners [the U.S. sim-ulation analyst who requested the work] wantsto run a 216 [a specific type of frontal impacttest]. Do you think that the parts in the plenum3

need to be cleared up?

Vijay: I think maybe, but I’m not being so sure. I thinkfor the 208 [another frontal impact test] it isnecessary.

Suresh: Maybe these parts are involved only for occu-pant performance [hence, not the 216 test].

Vijay: I would say to fix that penetration only [pointingto the penetration in the plenum], just if you arenot sure, then it will be covered.

To compensate for his lack of knowledge of whichparts are implicated in the test, Vijay recommended fix-ing the penetration in the plenum, just in case. In otherinstances, Indian modelers chose not to fix penetrations.In general, because the choice of which penetrations tofix and which to leave was unguided by examination ofparts and physical tests, Indian modelers often left pen-etrations that would later derail analyses.

On other occasions, Indian engineers chose to fixall of the problems. The approach made the simulation

modelers vulnerable to the coordination and communi-cation problems that students of virtual teams repeat-edly describe. To build an FEA model, Indian modelersneeded the design engineers’ CAD files. But whereasthe U.S. analysts had access to the CAD repository, theIndian modelers only had access to the files that theU.S. analysts placed on their FTP site. Getting the mostrecent files required the modelers to ask the simula-tion analyst to locate and upload the files. Further-more, determining which part to alter when correctinga penetration required negotiation between the designengineers responsible for the parts. Because Indian mod-elers had no direct contact with design engineers, theyresorted to the guessing described above or routed theirqueries through the simulation analyst. In this manner,the discrepancy between the dependence on and accessto objects was amplified by mismatched task interdepen-dence created in part by the shift to virtual teams.

As Table 3 indicates, the distribution of work inPeriod 4 created mismatches between the Indian engi-neer’s high dependence on, but low access to, parts.Moreover, the organization of work into virtual teamsmeant that the Indian engineers lacked easy access toengineers on whom they depended for crucial infor-mation. Such large discrepancies had never appearedbefore: in Periods 1–3, when access was low, sowas dependence. The large discrepancies faced by theIndian modelers created significant problems in simula-tion work.

DiscussionBy turning to semiotics to analyze the representationsmade possible by digitization, and by tracing how newrepresentations prompted increased virtuality in work,our study demonstrates how simulation can engenderchanges in the work structure as well as in tasks androles. It also speaks to how problems encountered insimulation, particularly with respect to trust, may differfrom what has been documented in studies of the othertwo types of virtual work—namely, virtual teams andremote control. In so doing, our study helps to specifyhow the organization of work might change under sim-ulation (it begins to fill in, as it were, the empty boxesin Table 2). More broadly, our study provides the under-pinnings for building theories of virtual work througha focus on representations. In addition, and more prac-tically, our study highlights the dangers inherent whenmanagers confuse one type of virtual work with another.

Work StructureOn virtual teams, digitization permits separationbetween people who use indexical representations of oneanother to carry out mediated communication. In remotecontrol, digitization enables separation between peopleand objects; operators use symbolic and iconic repre-sentations of objects and physical processes to monitor,


manipulate, and alter the objects and processes from adistance. Working across distances (for the purposes ofcollaboration in the first case and control in the second)is a key element of work structure in both cases.

Simulation differs from virtual teams and remote con-trol in this respect. Although digitization allows for sep-aration between people and objects or other people, thepurpose of virtuality has nothing to do with distance.The purpose of virtuality in the case of simulation isstudy and experimentation. The very names of the threetypes of virtual work make clear this distinction. Theword “team” is preceded by “virtual,” the word “con-trol” is preceded by “remote,” but the word “simulation”lacks a modifier: we do not speak of virtual simulationor remote simulation. In short, with virtual teams andremote control, the entire point of organizing work vir-tually is to take advantage of digitization’s affordance ofspatial separation. The point of simulation is not to takeadvantage of spatial separation but to take advantage ofthe independence between people and objects or otherpeople, if only for short periods of time.

That the period of time is short speaks to the factthat this independence is bounded. Ultimately, the cre-ators of simulations must return to the objects or peoplethat they aim to represent to test and validate their mod-els. Each subsequent change in the model requires yetmore validation. Thus, with simulation, physical objectsor other people become deceptively distant but remainabsolutely vital, not because workers’ manipulations areintended to directly affect these entities, but because theentities are the referents of representations. Simulationswith high verisimilitude appear to substitute for objectsor other people because one can do within representa-tions what one could previously do only with physicalentities themselves. The adequacy of such simulations,however, still depends on validating their results againstphysical objects, people, and their associated processes.

Other organizational studies of digital technologieshave noted the tight coupling between representationsand referents. In the Dodgson et al. (2007) study of sim-ulation in fire engineering, engineers used fire drills tovalidate their models. Real fires were another source forcomparison and validation. One fire engineer remarked,“We always try to iterate between real cases of fireand our models”; another commented, “We are contin-uously interrogating our models 0 0 0 evidence from a firecan challenge our ideas” (Dodgson et al. 2007, p. 856).Similarly, Yoo et al. (2006) observed that Frank Gehry’sarchitectural firm maintained a tight coupling across 3Ddigital representations, 2D paper drawings and sketches,physical models, and actual buildings. The architecturalteam continually translated and validated Gehry’s designvision across these disparate media over the course ofa design process. Such tight coupling between represen-tations and referents provides clear implications for thestructure of simulation work.

As our study of automotive engineers shows, this tightcoupling in simulation means that the people who createrepresentations are highly dependent on physical refer-ents. Consequently, the creators of representations needa work structure that provides good access to the objectsor people represented. In the case of objects, the cre-ators also need ready access to the people who controlthe objects, a fact that illustrates how dependence onobjects can breed task interdependence. Thus, in the caseof simulation, although digitization permits separation ofpeople from the referents of representations, virtualitymay still demand proximity. Managers, therefore, shouldstructure simulation work such that people have readyaccess to the physical entities and processes they model.

Tasks and RolesResearchers rarely report that virtual teams alter eitherwork roles or the division of labor. Most likely,researchers are silent on the matter because such teamsare often formed to take advantage of distant expertswho play their existing work roles in the same skilledmanner regardless of where their teammates are located(Boh et al. 2007, Sole and Edmondson 2002). Bycontrast, researchers often report that remote controlthreatens significant change in the division of labor. Inpaper mills, for example, digital controls place consid-erable information in the hands of operators, potentiallyenabling them to make operational decisions tradition-ally reserved for managers. Although some mills haveembraced such fundamental changes in the division oflabor, most have not, because managers are hesitantto cede authority to operators and operators are hesi-tant to shoulder the responsibility of making decisions(Zuboff 1988).

Like remote control, simulation appears to spawnchanges in the division of labor. At IAC, these changeswere closely associated with the deployment of increas-ingly sophisticated, iconic simulations. When FEA mod-els replaced lumped-parameter models in Period 3,mathematical performance analysis shifted from R&Dto engineering, thereby creating the new role of a sim-ulation engineer. Unlike R&D modelers, whose modelswere symbolic, not iconic, simulation engineers requiredaccess to vehicle parts and were highly dependent onother engineering occupations for accomplishing theirwork. The move to engineering signified the rise in theimportance of the modeling role within product develop-ment; managers now expected modelers to inform designdecisions.

The story, however, does not end there. When man-agers became convinced that iconic representations ofperformance were so accurate that simulation engineersdid not need access to physical referents, they alteredthe division of labor once again. By offshoring modelbuilding to India, managers split the simulation engi-neer’s role into two new occupations: simulation mod-elers (in India) and simulation analysts (in the United


States). This change in tasks and roles reflects an attemptat work rationalization that was absent in earlier changesat IAC: engineers in the United States sent the lessinteresting and routine tasks to the Indian modelers,keeping highly analytic tasks for themselves. They alsothought the tasks they sent to India were ones that couldbe divorced from physical referents. This presumptionturned out to be untrue, as problems in the modelsquickly attested.

These changes in tasks and roles at IAC resonatewith arguments that digital modeling technologies areapt to bring profound changes to work roles withinand across organizations. The advent of technologiesthat permit the simulation of fire, smoke, and humanevacuation of buildings prompted the creation of a newengineering profession: fire engineering (Dodgson et al.2007). As new actors working amidst insurers, regu-lators, architects, building authorities, and engineers ofvarious hues (structural, mechanical, electrical, environ-mental, and so on), fire engineers carved out a rolefor themselves in the design process for safe buildings.Boland et al. (2007) noted how architect Frank Gehry’sadoption of digital 3D technologies disrupted establishedpractices among architects, engineers, and contractors onhis projects, allowing them to forge new relationshipsamong actors, and, in so doing, to innovate considerably.A European automaker defied its traditional functionalorganization to create a new team consisting of sev-eral designers, a simulation engineer, and a physical testengineer. By using the rapid feedback that simulationprovided, the team could quickly iterate through designsolutions, ultimately improving crash safety by 30% ata fraction of the cost of the traditional developmentprocess (Thomke 2003). Subsequently, the automakerbegan dramatic changes in work processes and orga-nization to extend the benefits that this small teamdemonstrated. These examples demonstrate the kinds ofsweeping changes in tasks and roles that are likely toemerge as the use of simulation spreads across occupa-tions and industries. Our study further points to potentialpitfalls that may arise if these changes fail to take intoaccount each role’s relationship to representations.

Related Problems of TrustOne common theme in Table 2’s list of problems asso-ciated with each type of virtual work is trust. Studentsof virtual teams report that members have difficultytrusting distant coworkers, a situation that is trouble-some because a lack of trust undermines team per-formance (Iacono and Weisband 1997, Jarvenpaa andLeidner 1999). Issues related to trusting others also arisein remote control, but here, they are primarily betweenmanagers and workers. In this sense, remote controlreopens the rift between conception and execution thathas long underwritten issues of power, authority, and

trust in industrial settings (Noble 1984, Shaiken 1984,Vallas and Beck 1996).

More interestingly, studies of remote control fore-ground an entirely different type of trust and distrust: theadvisability of trusting representations. In paper mills,operators were overly suspicious of the representations’accuracy because they were accustomed to gatheringinformation by seeing, hearing, touching, and smellingequipment and paper (Zuboff 1988). In fact, some oper-ators distrusted the symbols on their monitors so muchthat they insisted on walking by the mill’s vats andpresses to make sure, as one worker explained, that“the machine [control system] ain’t lyin’ to you” (Val-las and Beck 1996, p. 347). As Perrow (1999) andHirschhorn (1984) noted in their discussion of remotecontrol in nuclear power plants, potentially even moreserious problems occur when operators trust representa-tions too much.

Placing too much trust in representations was pre-cisely the problem that accompanied the developmentof FEA simulations at IAC. Lured by the robustnessof FEA’s iconic representations, managers made deci-sions based on their trust in simulation models, evenwhen engineers desired physical tests as confirmation.This trust is what prompted managers to order the off-shoring of model building to India against the advice ofthe American engineers, who understood the importanceof validating simulations against their referents.

Such findings resonate with studies showing that usersof models are often less skeptical of their results thanare the models’ creators (Shackley and Wynne 1996,MacKenzie 1990). For example, Boland et al. (2007)noted that construction managers encouraged windowcontractors to forgo taking field measurements of open-ings in concrete walls and to rely instead on simulationdata to determine window size. Similarly, managers inthe crash division at the European automaker encour-aged engineers to use data from simulation models asopposed to physical test data to redesign the front sec-tion of a sport utility vehicle (Thomke 2003).

The Indian engineers similarly placed too much faithin their model’s results. In the absence of knowledgegained from direct contact with physical objects and pro-cesses, the Indian engineers acted in a manner consistentwith Turkle’s (2009) engineering students: they treatedtheir representations as if they were “real.” In short, likemanagers and engineering students, the Indian engineerssuccumbed to the lure of the virtual.

Building Theory, Beginning with RepresentationsTable 1 constitutes a taxonomy of virtual work in orga-nization studies. Taxonomies are useful for illuminatingdescriptive differences and for identifying meaningfulconstructs. For example, the taxonomy in Table 1 helpedus conceptually differentiate digitization from virtual-ity. Digitization affords spatial separation, which virtu-ality may or may not be wise to exploit. Digitization


gives rise to the divide between the physical and virtualthrough the creation of digital representations; virtualityspecifies what the interaction between the physical andvirtual will be. Representations forge the link betweendigitization and virtuality. Students of digitization andwork ignore virtuality and representations at their perilbecause it is the combination of digitization, virtual-ity, and representations that permits distinctions amongtypes of virtual work.

Generally speaking, the type of representations thatvirtual workers use varies by the type of virtual work.Virtual teams often use indexical representations ofmembers; they are also likely to use symbolic repre-sentations that lack physical referents (e.g., budgets,reports). In remote control, symbolic and low verisimil-itude iconic representations populate the digital inter-faces of computer control systems. High verisimilitudeiconic representations typify simulation. Associated withdifferences in types of representation are differences inworkers’ relationship to the representations. Membersof virtual teams primarily operate with or on represen-tations, people in remote control operate through rep-resentations, and creators of simulations operate withinrepresentations. As our study demonstrates in the caseof simulation, these differences in types of represen-tations and relationships to representations are centralto explaining dependencies on objects and people atwork. As such, they also set the parameters for accessto objects and people. Beyond having implications forthe design of work, as we have discussed, differences inthe nature of representations point the way for buildingtheories of virtual work.

For example, our case study suggests that, in simu-lation, the fit between dependence on the referents ofrepresentations and access to those referents is predic-tive of performance. That is to say, when simulation atIAC was wholly situated in the United States, simula-tion engineers’ dependence on vehicle parts and otherpeople was paired with ready access to those parts andpeople. When IAC offshored simulation to India, how-ever, simulation engineers’ high dependence on vehicleparts and other people was suddenly paired with lowaccess to those parts and people. Problems in the qual-ity of simulation models quickly arose, underscoring thecriticality of access for performance when dependenceis high. These problems may not be universal to virtualwork. They cannot arise, for example, when individualswork on representations that have no referent, and forthis reason, they may be less prevalent among virtualteams than in simulation.

Most discussions of dependence in organization stud-ies focus on dependence among people, or what iscalled task interdependence (Wageman 1995, Thompson1967).4 Workers’ dependence on physical objects rarelyattracts scholars’ attention because the objects arealways proximate to the workers, making access simple.

When objects are digitized and then made virtual, readyaccess to them may no longer be available to work-ers. At that point, scholars can no longer take physicalobjects for granted; rather, objects must jump to the fore-front of our analyses. In short, when considering virtualwork, organizational theorists and work systems design-ers alike ought to pay attention to representations as afirst step, and then to workers’ dependence on and accessto the physical referents of representations as a second.

Confusing One Type of Virtual Work for AnotherWe conclude by noting that the developments at IACillustrate that the changes in work associated with virtualwork become more complicated when managers confuseone type of virtual work for another. IAC’s managersfailed to appreciate how operating within representa-tions differed from operating with or on representations.They correctly understood that having access to phys-ical objects was relatively unimportant for tasks thatinvolved operating on representations that have no refer-ents. They also understood that such work could be doneby virtual teams, although it might be subject to com-munication and coordination problems that arise whenpeople attempt to operate with indexical representationsof team members. What they failed to grasp was howimportant it was for model builders to have access tothe physical objects and processes they were simulating.Consequently, they confused operating within represen-tations with operating with or on representations. In fact,they chose Bangalore as the site for the India technicalcenter precisely because it was known as an offshoringhub, the perfect location for finding people experiencedin working with or on representations. Had they recog-nized that building adequate simulations requires valida-tion against physical referents, they may not have sentthe work to India.

IAC’s managers eventually realized that their dream ofa completely virtual automotive design process still layin the realm of science fiction. When it became clear thatproblems with offshored FEA models were not declin-ing, they recognized that simulation engineers neededaccess to the physical referents of iconic representations.Accordingly, IAC installed a teardown room in the Indiacenter where engineers could handle and inspect vehicleparts. The firm also joined other automotive companiesin India to build a proving ground for physical testing.Thus, IAC used a strategy similar to that which scholarsadvise firms to use to mitigate the problems of virtualteams: they removed the distance associated with virtu-ality (Maznevski and Chudoba 2000, Hinds and Bailey2003). In the case of virtual teams, eliminating distanceis typically temporary and involves face-to-face meet-ings or sending distant members to visit the team’s pri-mary location.5 In the case of simulation, eliminatingdistance between simulators and objects is a permanentproposition.


Ours has been a cautionary tale about the lure of thevirtual. Organizations usually turn to the virtual in allof its forms in the hopes of reducing costs by replacinghumans and objects with representations or by employ-ing less expensive labor. The lure of the virtual appearsdifficult for firms to resist. For example, despite consid-erable evidence that virtual teams are less effective andmore troublesome than colocated teams, organizationscontinue to employ them. Thus, when managers chooseto make objects, rather than people, virtual, it is perhapsfortunate that they will encounter problems that theycannot skirt by instructing people to exert more effort orto simply make do. Faulty remote control systems andsimulations can have catastrophic consequences, suchas plant disasters and product recalls. When faced withproblems of such magnitude, organizations must learnto resist the lure of virtual work. Our admonition is thatthey should learn to do so with foresight rather thanhindsight.

AcknowledgmentsThe authors gratefully acknowledge Vishal Arya, Will Barley,Jan Benson, John Cafeo, Daisy Chung, Aamir Farroq, AlexGurevich, Mike Johnson, Bill Jordan, Hallie Kintner, MarkNeale, Susan Owen, Dan Reaume, R. Jean Ruth, and RandyUrbance, who assisted in data collection. They thank the engi-neers for graciously allowing them to observe them at workand for sharing their thoughts with them. They also thankthe three anonymous reviewers and the editors of this spe-cial issue for their insightful comments on earlier versions ofthis manuscript. This research was made possible by NationalScience Foundation Grants ITR-0427173 and SES-0938934.

Endnotes1In the setting we studied, no one used remote control; henceno one operated through representations. Although remotecontrol technology can be found in some of IAC’s settings,particularly manufacturing, we did not encounter it in thegroups we studied.2A lumped-parameter model is so named because it describesthe relationships among system components without explain-ing the particular functioning of each parameter. Instead, theparameters are “lumped together” to render an acceptableoutput.3A plenum is a ventilation duct designed to allow air circula-tion for heating and air conditioning systems. By definition, noparts should penetrate the plenum to avoid obstructing air flow.4For an exception, see the Bailey et al. (2010) study of depen-dence among work technologies.5Some evidence suggests that even this mitigation strategy isinsufficient in the case of virtual teams. Metiu (2006) reportedthat when distant Indian members of a software developmentteam temporarily relocated to California to be with the rest ofthe team, they still remained socially isolated from, and hadlimited interaction with, their U.S. counterparts.

ReferencesBailey, D. E., P. M. Leonardi, J. Chong. 2010. Minding the gaps:

Understanding technology interdependence and coordination inknowledge work. Organ. Sci. 21(3) 713–730.

Barley, S. R. 1996. Technicians in the workplace: Ethnographic evi-dence for bringing work into organization studies. Admin. Sci.Quart. 41(3) 404–441.

Black, M. 1962. Models and Metaphors: Studies in Language andPhilosophy. Cornell University Press, Ithaca, NY.

Boh, W. F., Y. Ren, S. Kiesler, R. Bussjaeger. 2007. Expertise and col-laboration in the geographically dispersed organization. Organ.Sci. 18(4) 595–612.

Boland, R. J., Jr., K. Lyytinen, Y. Yoo. 2007. Wakes of innovationin project networks: The case of digital 3-D representations inarchitecture, engineering, and construction. Organ. Sci. 18(4)631–647.

Chaminade, C., J. Vang. 2008. Globalisation of knowledge productionand regional innovation policy: Supporting specialized hubs inthe Bangalore software industry. Res. Policy 37(10) 1684–1696.

Cooley, M. 1980. Architect or Bee? The Human/Technology Relation-ship. South End Press, Boston.

Cramton, C. D. 2001. The mutual knowledge problem and its conse-quences for dispersed collaboration. Organ. Sci. 12(3) 346–371.

Csordas, T. J. 2001. Computerized cadavers: Shades of representationand being in virtual reality. P. P. Brodwin, ed. Biotechnologyand Culture: Bodies, Anxieties, Ethics. Indiana University Press,Bloomington, 173–192.

Cunningham, J. 2007. Simulation makes big savings. ProfessionalEngrg. 20(21) 39–40.

Dodgson, M., D. M. Gann, A. Salter. 2007. “In case of fire, pleaseuse the elevator”: Simulation technology and organization in fireengineering. Organ. Sci. 18(5) 849–864.

Drew, C. 2009. Drones are weapons of choice in fighting Qaeda. NewYork Times (March 17) A1.

Eco, U. 1979. A Theory of Semiotics. University of Indiana Press,Bloomington.

Faraj, S., L. Sproull. 2000. Coordinating expertise in software devel-opment teams. Management Sci. 46(12) 1554–1568.

Frege, G. 1892/1980. On sense and reference. P. Geach,M. Black, eds. Translations from the Philosophical Writings ofGottlob Frege. Blackwell, Oxford, UK, 25–50.

Gibson, C. B., J. L. Gibbs. 2006. Unpacking the concept of virtual-ity: The effects of geographic dispersion, electronic dependence,dynamic structure, and national diversity on team innovation.Admin. Sci. Quart. 51(3) 451–495.

Griffith, T. L., J. E. Sawyer, M. A. Neale. 2003. Virtualness andknowledge in teams: Managing the love triangle of organi-zations, individuals, and information technology. MIS Quart.27(2) 265–287.

Handy, C. 1995. Trust and the virtual organization. Harvard Bus. Rev.73(3) 40–50.

Hinds, P. J., D. E. Bailey. 2003. Out of sight, out of sync: Understand-ing conflict in distributed teams. Organ. Sci. 14(6) 615–632.

Hirschhorn, L. 1984. Beyond Mechanization. MIT Press, Cam-bridge, MA.

Iacono, S., S. Weisband. 1997. Developing trust in virtual teams. Proc.30th Annual Hawaii Internat. Conf. System Sci., IEEE ComputerSociety Press, Los Alamitos, CA, 412–420.

Jarvenpaa, S. L., D. E. Leidner. 1999. Communication and trust inglobal virtual teams. Organ. Sci. 10(6) 791–815.

Kanawattanachai, P., Y. Yoo. 2002. Dynamic nature of trust in virtualteams. J. Strategic Inform. Systems 11(3–4) 187–213.


Kincaid, J. P., J. Donovan, B. Pettitt. 2003. Simulation techniques fortraining emergency response. Internat. J. Emergency Manage-ment 1(3) 238–246.

Kurowski, P. M. 2008. Expecting too much from FEA. MachineDesign (July 10) 89–93.

Levina, N., E. Vaast. 2008. Innovating or doing as told? Status dif-ferences and overlapping boundaries in offshore collaboration.MIS Quart. 32(2) 307–332.

Longridge, T., J. Bürki-Cohen, T. H. Go, A. J. Kendra. 2001. Simula-tor fidelity considerations for training and evaluation of today’sairline pilots. Proc. 11th Internat. Sympos. Aviation Psych., OhioState University, Columbus.

Macedonia, M. 2002. Games soldiers play. IEEE Spectrum 39(3)32–37.

MacKenzie, D. 1990. Inventing Accuracy: A Historical Sociology ofNuclear Missile Guidance. MIT Press, Cambridge, MA.

Marescaux, J., F. Rubino. 2004. Robot-assisted remote surgery:Technological advances, potential complications, and solutions.Surgery Tech. Internat. 12 23–26.

Maznevski, M. L., K. M. Chudoba. 2000. Bridging space over time:Global virtual team dynamics and effectiveness. Organ. Sci.11(5) 473–492.

Metiu, A. 2006. Owning the code: Status closure in distributedgroups. Organ. Sci. 17(4) 418–435.

Miles, M. B. 1979. Qualitative data as an attractive nuisance: Theproblem of analysis. Admin. Sci. Quart. 24(4) 590–601.

Miles, M. B., A. M. Huberman. 1994. Qualitative Data Analysis: AnExpanded Sourcebook, 2nd ed. Sage, Beverly Hills, CA.

Noble, D. E. 1984. Forces of Production: A Social History of Indus-trial Automation. Oxford University Press, Oxford, UK.

Peirce, C. S. 1932. Elements of Logic: Collected Papers of CharlesSanders Peirce, Vol. 2. Edited by C. Hartshorne, P. Weiss. Har-vard University Press, Cambridge, MA.

Perrow, C. 1983. The organizational context of human factors engi-neering. Admin. Sci. Quart. 28(4) 521–541.

Perrow, C. 1999. Normal Accidents: Living with High-Risk Technolo-gies, revised ed. Princeton University Press, Princeton, NJ.

Prentice, R. 2005. The anatomy of a surgical simulation: The mutualarticulation of bodies in and through the machine. Soc. Stud.Sci. 35(6) 837–866.

Quine, W. V. O. 1961. From a Logical Point of View: Nine Logico-Philosophical Essays. MIT Press, Cambridge, MA.

Satava, R. M. 1995. Virtual reality and telepresence for militarymedicine. Comput. Biol. Medicine 25(2) 229–236.

Saussure, F. d. 1916/1966. Course in General Linguistics. Edited byC. Bally, A. Sechehaye. Translated by R. Harris. McGraw-Hill,New York.

Shackley, S., B. Wynne. 1996. Representing uncertainty in global cli-mate change science and policy: Boundary-ordering devices andauthority. Sci. Tech. Human Values 21(3) 275–302.

Shaiken, H. 1984. Work Transformed: Automation and Labor in theComputer Age. Holt, Rinehart, and Winston, New York.

Sole, D., A. Edmondson. 2002. Situated knowledge and learning indispersed teams. British J. Management 13(S2) S17–S34.

Sproull, L., S. Kiesler. 1991. Connections: New Ways of Working inthe Networked Organization. MIT Press, Cambridge, MA.

Susman, G. I., R. B. Chase. 1986. A sociotechnical analysis of theintegrated factory. J. Appl. Behav. Sci. 22(3) 257–270.

Thomke, S. H. 2003. Experimentation Matters: Unlocking the Poten-tial of New Technologies for Innovation. Harvard BusinessSchool Press, Boston.

Thompson, J. D. 1967. Organizations in Action: Social Science Basesof Administrative Theory. McGraw-Hill, New York.

Thornton, J. 2010. The question of credibility. Mech. Engrg. 132(5)40–45.

Tschang, F. T. 2007. Balancing the tensions between rationalizationand creativity in the video games industry. Organ. Sci. 18(6)989–1005.

Turkle, S. 2009. Simulation and Its Discontents. MIT Press,Cambridge, MA.

Vallas, S. P., J. P. Beck. 1996. The transformation of work revisited:The limits of flexibility in American manufacturing. Soc. Prob-lems 43(3) 339–361.

Wageman, R. 1995. Interdependence and group effectiveness. Admin.Sci. Quart. 40(1) 145–180.

Yin, R. K. 1994. Case Study Research: Design and Methods, 2nd ed.Sage, Beverly Hills, CA.

Yoo, Y., R. J. Boland Jr., K. Lyytinen. 2006. From organization designto organization designing. Organ. Sci. 17(2) 215–229.

Zuboff, S. 1988. In the Age of the Smart Machine: The Future ofWork and Power. Basic Books, New York.

Diane E. Bailey is an assistant professor in the School ofInformation at the University of Texas at Austin. She studiestechnology and work in information and technical occupations.

Paul M. Leonardi is an assistant professor in the Depart-ments of Communication Studies and Industrial Engineer-ing and Management Science at Northwestern University,where he holds the Allen K. and Johnnie Cordell Breed JuniorChair in Design. He received his Ph.D. from Stanford Uni-versity. His current research explores how organizations canemploy advanced information technologies to more effectivelycreate and share knowledge.

Stephen R. Barley is the Richard Weiland Professor ofManagement Science and Engineering and the codirector ofthe Center for Work, Technology and Organization at Stan-ford’s School of Engineering. He received his Ph.D. in orga-nization studies from the Massachusetts Institute of Tech-nology. His research interests include work and occupations,technological change, and corporate power in representativedemocracies.

The Lure of the Virtual

Documents

Transcript of The Lure of the Virtual