Q Academy of Management Review2015, Vol. 40, No. 3, 446–460.http://dx.doi.org/10.5465/amr.2013.0350

TASK COMPLEXITY: EXTENDING A CORE CONCEPT

THORVALD HÆREMBI Norwegian Business School

BRIAN T. PENTLANDKENT D. MILLER

Michigan State University

We reexamine the assumptions of current theory to update and extend the concept of taskcomplexity to tasks that include multiple actors at any level of analysis. Tasks can bemodeled as networks of required actions and information cues carried out or processed byparticular actors. Counting pathways in the task network provides an index of task com-plexity that incorporates insights from organization research but is more consistent withcontemporary complexity science than prior approaches and better reflects the exponentialnature of the phenomenon. The revised concept of task complexity can be readily used asan independent or dependent variable, and it can be used to compare observed and ide-alized descriptions of tasks. We discuss its implications for developing theory.

In this article we address a basic question:how can we conceptualize the complexity oftasks that involve multiple actors carrying outa set of interdependent actions to achieve acommon goal? Established definitions of taskcomplexity (Campbell, 1988; Wood, 1986) focusprimarily on tasks performed by individuals andare based on Hackman’s (1969) principle of sep-arating the task from the individual task doer(i.e., analyzing the task qua task). This concep-tualization has been useful in controlling for taskcomplexity in experimental research, but it is notadequate for addressing situations where workis distributed among multiple actors, or wherethe actors are organizational units. We find thatthe past concept of task complexity does nottranslate across levels of analysis into settingswhere tasks are collaborative. It does not takeinto account the material context of task execu-tion, and it is merely additive, not exponential. Ithas been useful as a control variable but not asa dependent variable. Established concepts andmeasures of task complexity predate contempo-rary ideas about complexity (e.g., Gell-Mann,1995), so it should not be surprising that we needto revisit the topic.

Following Alvesson and Sandberg (2011), werevisit assumptions in the existing literature

(Campbell, 1988; Wood, 1986) and offer a newframework for conceptualizing andmeasuring taskcomplexity that incorporates key insights from thephysical and biological sciences (Gell-Mann, 1995;Gell-Mann & Lloyd, 1996) and organization studies(Moldoveanu & Bauer, 2004; Rahmandad, 2008;Rivkin & Siggelkow, 2007; Zhou, 2013). The newframework contributes to the literature in severalways:

1. It extends the concept of task complexity totasks performed by multiple actors at dif-ferent levels of analysis.

2. It integrates the concept of task complex-ity with the material context of taskperformance.

3. It proposes a general way to represent tasksas networks of actors, actions, and in-formation and to quantify the associatedcomplexity.

4. By extending the concept of task complexityto include empirical observations as well asidealized descriptions, it provides the basisfor developing and testing theory aboutcomplexity as a dependent variable, not justa control variable.

We chose to retain the original label—“taskcomplexity”—because our revised concept retainsa close connection to the original individual-levelconcept. Drawing on Naylor, Pritchard, and Ilgen(1980), Wood defined tasks in terms of three ele-ments—products, acts, and information cues: “Thetwo types of input components (i.e., acts and in-formation cues) and products can be used to de-scribe any task and, therefore, represent the basis

This work was supported in part by the National ScienceFoundation (NSF SES-1026932). We thank Avinash VenkataAdavikolanu, Neal Ashkanasy, Peng Liu, Elaine Yakura, andthe anonymous AMR reviewers for their help.

446Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyrightholder’s express written permission. Users may print, download, or email articles for individual use only.

for developing a general theory of tasks” (1986: 65)We retain this definition and the basic elementsused by Wood (1986), Campbell (1988), and othersbecause any meaningful measure of complexitymust build on “identified regularities” in the do-main of interest (Gell-Mann & Lloyd, 1996: 49). Inthe case of tasks, the identified regularities arethe products, required actions, and informationcues. In Wood (1986), the product defines the pur-pose of the task (e.g., assembling a bicycle orlanding an airplane), while an idealized de-scription of the required actions and informationcues is used to compute the complexity of the task.We accept acts, information cues, and goals askey components of tasks, but we reconceptualizethe way those features contribute to the com-plexity of tasks involving multiple actors.

An example can serve to illustrate the nature oftasks extending beyond the individual level ofanalysis. Imagine that you have been assignedthe task of protecting oil rigs in the North Seaagainst a possible attack from terrorists, whomay be operating on a large, commercial fishingboat. There are many such fishing boats in theNorth Sea and many oil rigs, spread out overa vast area. You must detect potential threats,verify that they are foes, and intercept them be-fore they can attack their target. To carry out thistask, you must coordinate the action of multipleactors: surveillance planes, to scan the vast areaof the North Sea and detect vessels in the area;fast patrol boats, to get to the detected vesselsand investigate whether they have friendlyintentions or not; and perhaps large, well-armedfrigates, to intervene and stop the terrorists.

Although this example may seem esoteric, thestructure of the task is representative of a broadrange of collaborative tasks that involve multipleactors with interdependent roles. It illustratesaspects of classic organizational phenomena,such as differentiation and integration of func-tions (Lawrence & Lorsch, 1967), coordination andreciprocal interdependence (Thompson, 1967),problem solving and search (March & Simon,1958), and collective interpretation of environ-mental stimuli (Daft & Weick, 1984). Our goal inthis article is to extend the concept of task com-plexity so it can be incorporated into theoriesabout these kinds of phenomena. A key insight isthat the complexity of a collective task is notmerely the sum of the complexity of the constit-uent tasks, because interdependence betweenmultiple actors can have an exponential effect on

task complexity. The existing concept of taskcomplexity tends to mask this effect.Our focus on tasks sets an important limit on

the scope of our work. Over the years, organiza-tional scholars have been interested in the com-plexity of many different kinds of phenomena,and they have, in some cases, developed metricsfor the complexity of those phenomena: organi-zational complexity (Damanpour, 1996; Dooley,2002; Moldoveanu, 2004; Moldoveanu & Bauer,2004), product complexity (Hobday, 1998; Novak &Eppinger, 2001), job complexity (Fields, 2002),cognitive complexity (Bieri, 1955; Crockett, 1964;Mayer & Dale, 2010), behavioral complexity(Denison, Hooijberg, & Quinn, 1995), environ-mental complexity (Miller & Friesen, 1984;Mintzberg & Waters, 1985), and network com-plexity (Butts, 2001), to name a few. The physical,biological, and computational sciences providemany more examples. For each of these, a mean-ingful definition of complexity starts froma description of the regularities within the par-ticular empirical domain (Gell-Mann & Lloyd,1996). Here we limit our inquiry to the complexityof tasks.

BACKGROUND ON TASK COMPLEXITY

An Increasingly Relevant Concept

Task complexity became a variable of interestin the mid-1960s, when a young organizationalpsychologist named Karl Weick (1965) observedthat conflicting results in individual- and group-level research might be traced to uncontrolledvariation in the kinds of tasks being used inexperiments. In reaction to this insight, Hackman(1969), Wood (1986), and Campbell (1988) de-veloped and refined a framework for analyzingtask complexity that has remained the standardfor nearly thirty years, and may even be gainingin popularity. As of March 2014, Wood’s 1986 ar-ticle, “Task Complexity: Definition of the Con-struct,” had over 340 citations in the SocialSciences Citation Index, more than 90 of thoseoccurring since 2010, and Campbell’s 1988 article,“Task Complexity: A Review and Analysis,” hadover 350 citations, with more than 120 since 2010.Weick (1965) and Hackman (1969) criticized re-

search that confounded the task and the taskdoer, arguing that this was a key problem in task-related research. Hackman (1969) and sub-sequent theorists of task complexity identified

2015 447Hærem, Pentland, and Miller

the value of carefully separating the “task itself”from (a) the person or persons doing the task, (b)the context in which the task is performed, and (c)the behavioral pattern required to perform thetask. Following Hackman, Wood (1986) arguedthat “task as behavior description” would bea bad basis for studying task complexity becauseit would conflate the independent variable (taskcomplexity) with the dependent variable (taskperformance). As such, Wood preserved the sep-aration of task from doer (actor) and doing (actualbehaviors). Furthermore, he argued that taskcomplexity could be expressed as a linearcombination of three factors: component com-plexity, coordinative complexity, and dynamiccomplexity.

Component complexity. Wood (1986) arguedthat tasks that have more steps are more com-plex. Component complexity refers to the numberof distinct acts required to complete a task, where“distinct” means “nonredundant with other acts”(Wood, 1986: 67). If an act is repeated severaltimes in a task, it is only counted once.

Wood (1986) and Campbell (1988) both recog-nized the contribution of information cues to taskcomplexity—that is, information processingrequirements add to task complexity. Hence, thedistinct steps included within Wood’s summativemeasure of component complexity include bothrequired physical actions and information cuesto process cognitively. His equation for compo-nent complexity sums all information cues overall acts making up a task.

Coordinative complexity. Wood (1986) recog-nized that interdependence between steps alsoincreases complexity. Coordinative complexity isbased on the precedence relations among the re-quired actions only. It does not include precedencerelations among information cues and betweeninformation cues and action steps. By “coordinativecomplexity,”Wood was referring to the precedencerelationships of the required actions that converttask inputs into task products. Wood argued thatthe longer the sequence of such dependencies, themore complex the task.

Dynamic complexity. Wood (1986) recognizedthat change over time was also an important partof task complexity. He defined dynamic com-plexity in terms of changes in the other twodimensions of complexity. Wood’s approach todynamic complexity was one of comparativestatics, and his index includes the differences incomponent and coordinative complexity at two or

more distinct points in time. As such, dynamiccomplexity reflects changes in the requiredactions and information cues and in the means-ends hierarchy to which the task doer mustadapt. These changes are assumed to be exoge-nous to the process of task completion. Unlike the“dynamic complexity” of the decision-makingtasks studied by Sterman (1989, 1994), Wood’s(1986) concept does not incorporate feedbackloops, where task performances alter the state ofthe system where the task is being performed.

Extensions and Applications

Campbell (1988) adopted the basic frameworkdefined by Wood (1986) and suggested a numberof additional types of complexity (e.g., multipleoutcomes, uncertain means-ends relations, andmultiple pathways to task completion). Liu and Li(2012) have also identified several additionaldefinitions of task complexity. Our review of theliterature suggests that these dimensions are notoften operationalized in empirical research ontask complexity. Even Wood’s (1986) core dimen-sions of coordinative and dynamic complexityhave rarely been operationalized in empiricalresearch.Of the 705 studies in the Social Sciences Cita-

tion Index citing either Wood (1986) or Campbell(1988), we found only 39 in which scholarsattempted to operationalize task complexity inline with Wood’s and Campbell’s original defi-nitions. Our analysis of these studies is pre-sented in supplementary materials (see http://home.bi.no/fgl94010/taskcomplexity). Typically,the operationalization of task complexity dis-tinguishes, more or less arbitrarily, betweencomplex and simple tasks. Operationalizationis most often done experimentally by varyingthe number of information cues that subjectsconsider (e.g., Asare & McDaniel, 1996;Timmermans & Vlek, 1992). This reflects Wood’s(1986) component complexity. A few studies aremore specific in their operationalization, statingwhich dimensions are manipulated and how.Only one study (Banker, Davis, & Slaughter,1998) operationalizes all three dimensions ofWood’s (1986) framework. Across all of thesestudies, task complexity appears as an inde-pendent variable or a moderator, but never asa dependent variable.In very few empirical studies have researchers

sought to operationalize the complexity of a task

448 JulyAcademy of Management Review

carried out by a group or an organization (e.g.,Argote, Insko, Yovetich, & Romero, 1995; Espinosa,Slaughter, Kraut, & Herbsleb, 2007; Nahrganget al., 2013). As with individual-level research,this operationalization generally has been lim-ited to some dichotomous indicator of componentcomplexity used as an independent variable ormoderator.

REVISITING THE ASSUMPTIONS OFTASK COMPLEXITY

We adopt the strategy suggested by Alvessonand Sandberg (2011) of generating new theoryby revisiting the assumptions of current theory.The middle column of Table 1 shows a list ofassumptions in Wood’s (1986) version of taskcomplexity. The right column summarizes new,alternative assumptions.

In revisiting these assumptions, we are di-rectly challenging the key justification and the-oretical foundation for the earlier concept oftask complexity. Nobody would deny the ne-cessity of controlling for the task when con-ducting individual-level experimental research,and the guidance provided by Hackman, Wood,Campbell, and others has repeatedly demon-strated its value as a basis for doing so (e.g.,Stajkovic & Luthans, 1998; Utman, 1997; Wood,Mento, & Locke, 1987). Unfortunately, theseassumptions limit the applicability of task com-plexity in other areas, such as organizationallearning, control, routinization, and design, where

the antecedents and consequences of task com-plexity have not been examined.

Separability of Task from Behavior and Context

Old assumption: Task should be separatedfrom behavior. As mentioned above, Hackman(1969) and subsequent theorists assumed that itwas possible to separate the task itself from thepeople, context, and behavior used to perform thetask. In this view, a task consists of a set ofrequirements that constrain but do not fully de-termine the set of actions contributing to taskcompletion, and these requirements can be quiteconsistent across varying contexts. An idealizeddescription of the task may function well forcontrolling for variation in tasks, and the idealmodel has demonstrated its usefulness by show-ing that task complexity moderates the effectson task performance of individual-level moti-vation variables, including self-efficacy, goals,and performance versus mastery orientations(Stajkovic & Luthans, 1998; Utman, 1997; Woodet al., 1987).New assumption: Tasks are inseparable from

behavior. This revised assumption is consistentwith theories of practice (Bourdieu, 1990;Feldman & Orlikowski, 2011; Giddens, 1984). Asa result, it is more consistent with organizationtheories adopting a practice perspective, such astheories of organizational routines (Feldman &Pentland, 2003; Parmigiani & Howard-Grenville,2011) and organizational change (Farjoun, 2010;

TABLE 1Task Complexity Assumptions

Dimension Old Assumption New Assumption

Separability of task from behaviorand context

Tasks should be separated frombehavior.

Tasks are inseparable from behavior.

Tasks are separate from their materialcontext.

Tasks are inseparable from materialcontext.

Complexity is a property of anidealized task description.

Complexity is indexed by observablebehavior.

Complexity is observer independent. Complexity is observer dependent.Level of analysis Task complexity is an individual-level

construct.Task complexity applies for any

number of actors at any level ofanalysis.

Types of complexity There are a few predetermined “types”of complexity (component,coordinative, etc.).

There are many mechanisms that cancontribute to task complexity.

Functional form Complexity is a linear function of taskcomponents.

Complexity is an exponential functionof task components.

2015 449Hærem, Pentland, and Miller

Pentland, Hærem, & Hillison, 2010, 2011). Thesetheories emphasize duality, rather than separa-tion, of ostensive and performed tasks. In onto-logical terms, there is no task unless someone isdoing it, and doing always entails more than canbe scripted (Polanyi, 1962).

Even if one prefers not to accept the ontology ofcontemporary theory, the separability of task andenactment is problematic. In the framework ar-ticulated by Wood (1986) and Campbell (1988), in-formation cues are treated as objective. Ironically,by including information cues in the definition oftask complexity, the separation of the task fromthe task doer is blurred because information cuesare, by definition, a product of the environ-ment and the interpreter (Gibson, 1979). From acontemporary perspective, which acknowledgessensemaking and enactment (Weick, 1995), theassumption of observer-independent informationcues is difficult to defend.

Furthermore, Wood’s formula for componentcomplexity is based on counting requiredactions and information cues, but the number ofacts to perform and the number of informationcues to process depend, at least in part, on theexpertise and the individual differences of theperceiver/enactor (e.g., Hærem & Rau, 2007;Hunter, Schmidt, & Judiesch, 1990; Levin, Gaeth,& Schreiber, 2001). Even if we concede thata well-defined, individual-level task with veryclear cues and actions corresponds closely to itsidealized depiction, the correspondence breaksdown for collective tasks, where the interpretationof contextual stimuli occurs through a socialprocess and actions are interdependent (Daft &Weick, 1984; Weick & Roberts, 1993).

Old assumption: Tasks are separate from theirmaterial context. Hackman (1969) also arguedthat tasks should be analyzed as separate fromtheir physical setting. As a result, there was noneed to account for the way that the use (andpotential depletion or creation) of resourcesmight influence task performance and complex-ity over time. Raw materials, tools, equipment,and facilities are all missing from this portrayalof tasks.

New assumption: Tasks are inseparable fromtheir material context. Task performance is nec-essarily a situated activity, in which task per-formers make use of the resources at hand. Tasksites contain people, artifacts, organisms, andother things (Schatzki, 2005). Claiming the es-sential materiality of tasks is consistent with

recent calls for greater recognition of the mate-rial context in organizational research (Leonardi,2012; Orlikowski & Scott, 2008).Old assumption: Complexity is a property of an

idealized task description. A closely related as-sumption concerns the nature of the evidenceused to determine the complexity of a task. Be-cause the task is assumed to be separate fromthe behavior of the task doer(s), the complexityof the task must be inferred from an idealizeddescription of the task (the task qua task). Forexample, Wood (1986) analyzed an idealizedversion of the air traffic control task. Task com-plexity in this sense does not take into accountexperience, learning, the operating environment,or any other individual or organizational char-acteristics. In order to study consequences of in-dividual differences, at any level of analysis, weneed a concept of task complexity that capturesbehavioral consequences.New assumption: Complexity is indexed by

observable behavior. The mathematical scienceof complexity offers useful guidance here. Forexample, the Lempel-Ziv algorithm (Kaspar &Schuster, 1987; Lempel & Ziv, 1976) can be used tomeasure the complexity of a process by observ-ing the string of actions within the process.Old assumption: Task complexity is observer

independent. Another corollary to the traditionalassumption about separability concerns thepoint of view from which task complexity can beobserved. In separating the task from the taskdoer, Hackman, Wood, and Campbell assumedresearchers could adopt an objective, universalpoint of view not dependent on the identityof the task doer or the outside observer. Theresearcher privileges a single acontextualidealized perspective. Thirty years ago this as-sumption could be made without explicit ex-planation or defense. Today it not only seemsindefensible but forecloses a variety of valuableopportunities for theory development that in-volve comparing multiple perspectives on tasks,such as how and why actual behavior departsfrom idealized versions of the task.New assumption: Task complexity is observer

dependent. This assumption is consistent withthe contemporary idea that complexity is ob-server dependent (Gell-Mann, 1995). Consider-ation of observer dependence led Gell-Mann tonote that when we attempt to measure the com-plexity of anything in the real world, all mea-sures that we can define and compute “are to

450 JulyAcademy of Management Review

some extent context-dependent or even sub-jective” (1995: 1). This is why, for each domainwhere we wish to investigate complexity, weneed to establish a distinct set of “identifiableregularities” on which our complexity measurecan be based.

Level of Analysis

Old assumption: Individual level of analysis.Because tasks are considered to be individual,so is task complexity. In air traffic control, forexample, complexity generally has been treatedas an individual-level phenomenon (Mogford,Guttman, Morrow, & Kopardekar, 1995). AlthoughWood (1986) treated air traffic control as anindividual activity, the presence of multiplecontrollers and multiple planes suggests that airtraffic control is, at least in some respects, a col-laborative activity (Fairburn, Wright, & Fields,1999).

New assumption: Any level of analysis. Manytasks or processes involve multiple actors. This istrue of all organizational routines by definition(Feldman & Pentland, 2003), all kinds of serviceencounters (Leidner, 1993), and any task thatinvolves communication or interactions with otheractors. Further, in many cases the actors are notindividual humans; they may be machines, orga-nizational subunits, or entire organizations. In ourNorth Sea counterterrorism task, for example, theactors are boats and planes that have crews ofpeople. The inclusion of actors such as organiza-tional subunits or computer systems is an estab-lished part of models used in process engineering(Eppinger, Whitney, Smith, & Gebala, 1994; Smith& Eppinger, 1997), but it has beenmissing from ourconcept of task complexity. It should be possibleto measure the complexity of a task regardless ofwho or what is carrying it out.

Types of Task Complexity

Old assumption: There are many types of taskcomplexity. Because Wood (1986) and Campbell(1988) defined “types” or “dimensions” of taskcomplexity, pluralism seems to have entered ourfield as a necessary part of the concept (Liu & Li,2012). Wood (1986) identified three forms, andCampbell (1988) added more. Each type of com-plexity, along with its corresponding measure,has a distinct conceptual basis. As noted in ourreview of the literature, these distinctions generally

have been glossed over in favor of a binaryjudgment in empirical research (complexity iseither “high” or “low”). In spite of researchers’reluctance or inability to measure them, weassume that the various dimensions of taskcomplexity are important.New assumption: There are many antecedents

of task complexity. All of the mechanisms iden-tified by Wood (1986) and Campbell (1988) can beunderstood as causal antecedents that may (ormay not) contribute to the complexity of a giventask. In this view, task complexity can be seen asa single construct, and there is no need for mul-tiple types or dimensions of task complexity. Theextent to which particular mechanisms contrib-ute to task complexity is an empirically testablequestion.

Functional Form

Old assumption: Complexity is additive. Theequations that Wood (1986) used to compute taskcomplexity are all linear. For example, compo-nent complexity is computed by counting thenumber of information cues across actions. Ifa task has ten components, then adding onemorecomponent makes it 10 percent more complex.Total complexity is the sum of component, co-ordinative, and dynamic complexity.New assumption: Complexity is exponential.

Current mathematical models of complexity de-rive from either information theory or algorithmiccomplexity (Prokopenko, Boschetti, & Ryan, 2009).Moldoveanu and Bauer (2004) offer an extensivelist of examples applied to organizations. None ofthe basic models for complexity is linear. In-stead, task complexity increases exponentiallyin its components. If we want to be consistentwith current science, our concept of task com-plexity should capture nonlinear contributions ofactions, information cues, and actors. Task com-plexity depends not only on the number of com-ponents but on the relations among components(Flood, 1987).

EXTENDED CONCEPT OF TASK COMPLEXITY

These new assumptions require us to recon-ceptualize task complexity in two substantialways. First, the new assumptions concerning(in)separability, material context, and level ofanalysis require us to consider how a broader setof constituents contributes to task complexity. In

2015 451Hærem, Pentland, and Miller

particular, locating tasks in their social and ma-terial context requires the inclusion of acts andinformation cues as they become enacted by oneor more actors throughout the task resolution.Second, the new assumptions about dimensionsand functional form require us to redefine howtask complexity is computed. In the followingsections we introduce a concept of task com-plexity that meets the requirements imposed bythe new assumptions.

Modeling Tasks As Networks of Events

To incorporate these new assumptions intoa new concept of task complexity, wemodel tasksas networks of events, where an event is an ac-tion performed by some actor at some moment intime. Events generate information cues that may(or may not) be processed by other actors. Theactors may be people, machines, or organi-zational subunits (Latour, 2005). Pentland andFeldman (2007) proposed the use of this kind ofnetwork as a model for organizational routinesand referred to the nodes as functional events.

Figure 1 shows the simplest possible exampleof such a network. It could be part of a trans-action, a communication, or a simple service,such as answering a question. (See Figures 3and 4 for examples of larger networks in matrixformat.)

The network in Figure 1 contains two events:actor A asks a question and then actor B answers.The events are sequentially related, and if eitherevent is missing, the overall task will be in-complete. Each event in the network has an ac-tion and an actor (Pentland & Feldman, 2007). As inWood (1986), events are related by sequential pre-cedence, and paths are defined by the sequencesthat lead to the completion of the task. But inkeeping with our revised assumption about in-separability, actors are explicitly included in themodel of the task.

It should be evident that each of these eventsgenerates information cues that may (or may not)be processed by other actors or observers. In-formation cues, which provide a mechanismthrough which events are related, are created bythe actions of actors and by changes in the ma-terial context. Thus, actions by one actor can beinterpreted as information cues by others actors.This is especially important when the taskincludes reciprocal interdependence (Thompson,1967) or network arrangements (Van de Ven,

Delbecq, & Koenig, 1976). Information cues areinherently subjective; they require interpretationby an observer (Gibson, 1979; Weick, 1969).

A Unified Concept of Task Complexity

Once the task has been modeled as a network,task complexity can be indexed by identifying allpossible paths to each goal of the task and bysumming the number of ties making up thesepaths using the method described by Oeser andO’Brien (1967):

Task Complexity5+g+ptiesg;p

where paths (p) are routes to particular goals (g)representing attainment of an outcome. Follow-ing Campbell (1988), we allow tasks to havemultiple goals or completion events. The primaryunderstanding reflected in this measure is thattask complexity is indexed by the number of pathsin the network of events that lead to the attain-ment of task outcomes. (The code for implement-ing this algorithm is available at http://home.bi.no/fgl94010/taskcomplexity.)Comparison to the existing concept. This ap-

proach preserves basic insights fromWood (1986)concerning component complexity and coor-dinative complexity, but it incorporates them intoa single computation that produces a nonlinearindex of task complexity.Figure 2 shows how this measure of task com-

plexity varies as a function of the number of nodesand ties in the network of events. The horizontalaxis represents the number of ties beyond theminimum number required to complete the taskin a “straight line,” as an invariant sequence ofevents, without alternatives or repetitions. Whenx 5 0, there is only one possible path. Additionalties introduce branches and possible loops intothe network of events. For x . 0, x nonredundantties between events are added at random to thetask network. Each point in the graph representsthe average computed index of complexity for fiftysimulated tasks, each of which includes x ran-domly added ties. The graphed results providea useful way to compare the new concept of taskcomplexity with the old one.First, consider Wood’s (1986) component com-

plexity, which is represented by the number ofnodes in the network. Figure 1 shows that thenumber of nodes has little effect on complexityunless there is also a substantial number of ties.

452 JulyAcademy of Management Review

Although component complexity has been thedominant concept used in empirical research,focusing on nodes to the exclusion of ties pres-ents an incomplete picture of task complexity.

Second, consider Wood’s (1986) coordinativecomplexity, which relates to the number of ties.When the number of ties is small, adding a fewties has little effect on task complexity, just asadding a few nodes does not make much dif-ference. However, when the number of tiesexceeds about ten or twelve, the picture changescompletely, because task complexity growsexponentially. The effect is stronger for largernetworks, which represent tasks with moreevents (i.e., greater component complexity).Paths and ties interactively affect complexity, sotheir relation is not simply additive. Figure 2illustrates why simply counting all ties, as doneby Zhou (2013), is not sufficient. As the numberof ties increases linearly, computed task com-plexity increases exponentially. Our approach

draws in part on conceptual elements proposedby Wood (1986), but it produces results that arestrikingly different.Example. We use the North Sea counterterror-

ism (NSCT) example introduced earlier to furtherillustrate the difference between the old and newapproach. We begin by modeling each role in theNSCT organization individually: airplanes, pa-trol boats, and frigates. Figure 3 lists events as-sociated with each role, and the matrix structureshows the directional ties (from row events tocolumn events). Each role is quite simple, withtask complexity equal to 3, 18, and 14, re-spectively. It is important to note that thesemodels are based on idealized task descriptions.When the roles are combined into a single

network, the actions by one actor can be in-formation cues for the other actors. For example,if a patrol boat notices that another actor haschanged direction, it might adjust its directionaccordingly. These kinds of reactions represent

FIGURE 1A Simple Multiactor Task Model

FIGURE 2Task Complexity As a Function of Network Nodes and Ties

2015 453Hærem, Pentland, and Miller

what Thompson (1967) would call “mutual ad-justment,”where each actor can notice actions byother actors and interpret those as informationcues. The resulting task network, shown inFigure 4, has a computed task complexity of over44,000.

To help interpret this result, we compared theNSCT task to the air traffic control task as

depicted by Wood (1986). That task has twenty-three nodes (required act and information cues)and thirty-eight ties, and it is carried out bya single individual. When we compute the com-plexity of air traffic control using our algorithm,its value is 91. The comparison is interestingbecause, according to Wood’s model, air trafficcontrol has more required acts and information

FIGURE 3Event Network for Each NSCT Role

FIGURE 4Event Network for Multiactor NSCT task

454 JulyAcademy of Management Review

cues than any of the NSCT roles. In fact, it hasmore than all three NSCT roles combined. Thus,under the old assumptions, we might concludethat the complexity of air traffic control is “high”while the complexity of NSCT “low.” In themultiactor model, with interactions and mutualadjustment, the counterterrorism task is orders ofmagnitude more complex. Although the numberof nodes in the network for counterterrorism isrelatively small, the required acts are in-formation cues for the other actors, and thesepathways of mutual influence and adjustmentstrongly increase the task complexity.

Summary. When the event network is ap-propriately constructed, this concept of taskcomplexity conforms to the new assumptionsarticulated earlier in this article. First, by virtueof the way the events in the network are defined,it reintegrates the task, the task doer(s), and thematerial context. Second, it can be applied to anypoint of view on a task—idealized or enacted.Third, it can be applied at any level of analysis,because the events can be defined for any kind ofactor. Fourth, it is elegant; it does not have dif-ferent metrics for different types of task com-plexity. Fifth, it is an exponential function of taskproperties. And sixth andmost important, it opensup new opportunities for theory development, asdiscussed in the next section.

IMPLICATIONS FOR THEORY DEVELOPMENT

Because this revised concept of task complexitycan be applied across a wide range of organiza-tional phenomena, it creates a variety of oppor-tunities for developing new theory. It is easy toidentify specific examples (e.g., task complexityas a moderator of organizational learning). Herewe survey the broad landscape of theoreticalpossibilities.

Focus on Behavior in Context

In organizational behavior and organizationtheory, behaviors are important sources for the-oretical explanations. Our revised assumptionsencourage us to consider what people actuallydo, rather than an abstract, idealized version ofwhat they do that privileges the researcher’sperspective. There are several precedents for thiskind of reconceptualization in organizational re-search. For example, behavioral decisionmakingand behavioral economics have progressed

through careful study of what people actually do(Camerer, Loewenstein, & Rabin, 2011). Likewise,areas of organizational research where there hasbeen a “practice turn” (Feldman & Orlikowski,2011) have focused on actual, rather than idealized,behavior. Both within-unit analysis of changes inenactments and across-unit analysis of differencesin enactments provide a basis for understandingways of organizing and their consequences.At the same time, idealized descriptions will

always be useful and theoretically interesting,and our framework can be used equally well toanalyze idealized descriptions, as we have donein the NSCT example. We can characterize thecomplexity of best practices, shortest path, stan-dard operating procedures, alternative designs,or any other task description that is exemplary ordistinctive in some respect. Such task descrip-tions provide important points of reference andcomparison, but they need not be privileged as“the task.” Disparities between idealized taskdescriptions and what people actually do area subject of research interest (e.g., Brown &Duguid, 1991). Idealized and empirical descrip-tions can coexist, as long as they are clearlyunderstood for what they represent.The reconnection of a task and its material

context also holds important implications fortask complexity. Tasks make use of and alterresources. For example, when chasing terroristsin the North Sea, planes and ships can only beginfrom positions established by prior tasks and canonly maintain pursuit as long as their fuel lasts.Because tasks often change the material context,the possibilities for enacting a task change overtime. Research on tasks involving dynamic sys-tems shows that even relatively simple flows ofmaterials and changes in inventory introduce a lotof complexity (Sterman, 1989). These features areessentially absent from the traditional model oftask complexity, because idealized tasks are as-sumed to be separable from their material context.Our model integrates sociality and materialityinto the concept of task complexity (Latour, 2005;Leonardi, 2012; Orlikowski & Scott, 2008).

Tasks As Dynamic Processes

Because task complexity is a function of thenetwork of events involved in a task, it relatesclosely to process theories of organizing (e.g.,Czarniawska, 2004; Farjoun, 2010; Puranam,Alexy, & Reitzig, 2014; Tsoukas & Chia, 2002;

2015 455Hærem, Pentland, and Miller

Weick, 1969). These theories emphasize actionsover time, rather than stasis. Although organiz-ing has become foundational in theories of or-ganizational stability and change (Farjoun, 2010;Sydow, Schreyogg, & Koch, 2009; Tsoukas & Chia,2002) and routines (Feldman & Pentland, 2003), ithas not been integrated into other theoreticalareas, such as task complexity.

Tasks are not frozen in time. They are per-formed, and as they are performed, their com-plexity can change. Sterman (1989) identifiedchanges over time as an intrinsic property ofsystems that are, by their nature, dynamic. Aslong as task complexity is based on an idealizeddescription of the task, our ability to explain dy-namics will be limited, because the idealizedtask does not change. It cannot incorporate thepossibility that the task itself might change asa result of events within the task context.

Because the new model is more sensitive tothe temporal dynamics of processes, it createsmore opportunities for theoretical development.First, incorporating actions, actors, and in-formation cues into the task model makes itsensitive to the various ways tasks can changeas they are performed. When applied to actualbehavior, our measure of enacted complexitycan change quickly and dramatically. As newactors enter or leave the scene, the potentialnumber of information cues and ties betweenactions can increase or decrease, with a corre-sponding nonlinear effect on task complexity.The notion that additional actors could dynam-ically change the complexity of the task is out-side the traditional framing, but it is obvious inthe revised concept.

We expect that the addition or removal ofactors could have an especially strong effectin situations where there is reciprocal inter-dependence (and mutual adjustment) among theactors (Thompson, 1967). The actions by one ac-tor can become information cues for all the otheractors. The resulting spike in complexity maycontribute to the classic “mythical man-month”phenomenon in software development (Brooks,1995), where adding programmers tends to delayproject completion.

Antecedents and Consequences ofTask Complexity

When Hackman (1969) andWood (1986) rejectedthe concept of “task as behavioral description” as

a useful approach to task complexity, they as-sumed that theorists and managers were not in-terested in the determinants of how tasks aresolved. Similarly, they assumed a lack of interestin the consequences of the complexity of taskresolutions. Of course, theorists and managersare often interested in both.Task complexity as a dependent variable.

When applied to observed behavior, task com-plexity provides an indicator that summarizesone property of the network of actions involved incarrying out a task. When referring to task com-plexity as a dependent variable, we are hypoth-esizing that some set of factors causes the tasknetwork to have more (or fewer) nodes (actions byactors) and/or connections between these nodes.Based on the analysis presented in Figure 1, weknow that modest changes in these essentialelements can have a dramatic impact on taskcomplexity.Treating task complexity as a dependent vari-

able opens up some interesting research possi-bilities. The general question is what causes taskcomplexity? Zhou (2013) has demonstrated thatorganizational design can influence complexity,and familiar principles of organizational design(e.g., Galbraith, 1973) emphasize modularity asa method to reduce complexity. Research onprocess design (Eppinger et al., 1994) has estab-lished techniques for simplifying and stream-lining processes.Other research indicates that task complexity

may change independent of the intentions of or-ganizational designers. Anything that affects howa task is accomplished can potentially influencetask complexity. Although the traditional per-spective would treat these as confounds, they canalso be legitimate topics for inquiry. For example,at the individual level, we might examine the roleof expertise (Hærem & Rau, 2007). At the subunitlevel, we might examine the role of changingcommunication patterns and routines. For exam-ple, Sydow et al. (2009), on the one hand, arguethat path dependence may cause simplification(and inertia) as patterns of action converge toa “locked-in” routine. On the other hand, Garud,Kumaraswamy, and Karnøe (2010) suggest thatpatterns of action can change through a genera-tive process of path creation. At the organizationlevel, we might examine the role of hierarchy(Zhou, 2013). Many factors can potentially in-crease or decrease complexity, depending on thesituation, but that is an empirical question.

456 JulyAcademy of Management Review

Task complexity as an independent variable.As mentioned, task complexity is usuallytreated as a binary category—high or low. Thissimplification is unfortunate for complexity notonly as a dependent variable but also as anindependent variable, because complexity isa phenomenon that often varies by orders ofmagnitude (Gell-Mann, 1995). A “simple” taskmight have dozens or hundreds of paths. Incontrast, a “complex” task might have tens ofthousands of paths. Our framework providesa way to conceptualize complexity that capturesorders of magnitude of difference. Given theenormous range of the construct, it might be bestto report it using a logarithmic scale (like theRichter scale for earthquakes). Reporting thelogarithm rather than the raw complexity scorewould focus attention on meaningful variationsin complexity.

Clarifying task complexity as we have donehere opens the door for using the construct to ex-plain a broad range of phenomena. The generalquestion is where and how does task complexityinfluence organizational outcomes—directly andas a moderator? Taylorist managerial norms sug-gest that having a single path is best, but thismay not always be the case, especially in ser-vice encounters (Leidner, 1993) or in other sit-uations where external contingencies affecttask execution. The complexity of a task mayinfluence efficiency, effectiveness, and a host ofother outcomes. Task complexity may also bea moderator for organizational learning and or-ganizational control, since simple patterns ofaction are easier to learn, control, and changethan are complex ones.

Whether it appears as an independent ordependent variable, it is helpful to have a uni-fied conceptualization for this construct. Pro-liferation of concepts and measures of taskcomplexity presents an unclear path for re-searchers. For the concept of task complexity tosupport a stream of research, there must besome consensus that it is valid, general, andmeasurable. The approach presented here isintended to fill that need.

Units and Levels of Analysis

The idealized version of task complexity por-trays tasks as neatly separated from their socialand material context. Furthermore, since tasksare defined at the individual level, the nested,

hierarchical nature of tasks and subtasks seemsto correspond to the nested, hierarchical natureof individuals, groups, and organizations. How-ever, tasks like delivering a service or complet-ing a transaction require the involvement ofmultiple, potentially heterogeneous actors. Themultiactor model of tasks that we present herehighlights the way actors are entangled throughtasks. Events can be part of multiple paths withina task, and also may be part of multiple tasks.Tasks are recognized by their product (or goal),but their boundaries are not always apparent.Research on process management demonstratesthat task networks in real organizations may beextensively interconnected (Eppinger et al., 1994).The potential for theory development becomes

clear when we compare Wood’s (1986) analysis ofair traffic control to Weick and Roberts’ (1993)analysis of flight deck operations on an aircraftcarrier. Weick and Roberts considered all of theroles involved in landing airplanes, not just theair traffic controller. The high degree of inter-dependence between the actors and actions onthe flight deck led Weick and Roberts to introducethe concept of “heedful interrelating.” Withoutminimizing the value of individual-level research,the reality is that important tasks (like landingairplanes) are often carried out bymultiple actors.If we restrict our view to single actors, we missopportunities to theorize about the complexity oftasks that include many actors.In this article we are not proposing a theory of

how complexity at one level may influencecomplexity at another. Rather, we are offeringa concept that can be applied equally at anylevel of analysis and that can include multipleactors. The grouping of actors in different unitsand different levels of analysis will influence theresulting event networks. The consequences ofsuch groupings for the overall task complexitymay then be studied empirically.Furthermore, the framework does offer some in-

sight into what Kozlowski, Chao, Grand, Braun,and Kuljanin (2013) refer to as “top-down” versus“bottom-up” theories of influence across levels.When entities at a higher level (e.g., organizations)act, they shape the context and provide informationcues for lower-level entities. For example, if a shipstops or changes direction, everyone on boardwill know. The bottom-up effect does not neces-sarily hold: actions by lower-level actors (individu-als) may or may not be sufficient to influenceactions by higher-level actors (organizations).When

2015 457Hærem, Pentland, and Miller

represented as a network that includes actionsby heterogenous actors, top-down effects may bemore evident than bottom-up effects.

CONCLUSION

As surely as we know that checkers differs fromchess, we know that complexity is a significantfactor in a wide range of organizational phenom-ena. Indeed, organizations can be understood asmechanisms for managing complexity (Simon,1969). However, our theory and empirical studiesoften oversimplify this core construct or omit itentirely. By updating and extending task com-plexity to include multiple actors at any level ofanalysis, the concept proposed here offers a way tofill this important gap in management research.

REFERENCES

Alvesson, M., & Sandberg, J. 2011. Generating researchquestions through problematization. Academy of Man-agement Review, 36: 247–271.

Argote, L., Insko, C. A., Yovetich, N., & Romero, A. A. 1995.Group learning curves: The effects of turnover and taskcomplexity on group performance. Journal of AppliedSocial Psychology, 25: 512–529.

Asare, S. K., & McDaniel, L. S. 1996. The effects of familiaritywith the preparer and task complexity on the effective-ness of the audit review process. Accounting Review, 71:139–159.

Banker, R. D., Davis, G. B., & Slaughter, S. A. 1998. Softwaredevelopment practices, software complexity, and soft-ware maintenance performance: A field study. Manage-ment Science, 44: 433–450.

Bieri, J. 1955. Cognitive complexity-simplicity and predictivebehavior. Journal of Abnormal and Social Psychology, 51:263–268.

Bourdieu, P. 1990. The logic of practice. Stanford, CA: StanfordUniversity Press.

Brooks, F. P. 1995. The mythical man-month. Reading, MA:Addison-Wesley.

Brown, J. S., & Duguid, P. 1991. Organizational learning andcommunities-of-practice: Toward a unified view of working,learning, and innovation. Organization Science, 2: 40–57.

Butts, C. T. 2001. The complexity of social networks: Theoret-ical and empirical findings. Social Networks, 23(1): 31–72.

Camerer, C. F., Loewenstein, G., & Rabin, M. (Eds.). 2011.Advances in behavioral economics. Princeton, NJ: PrincetonUniversity Press.

Campbell, D. J. 1988. Task complexity: A review and analysis.Academy of Management Review, 13: 40–52.

Crockett, W. H. 1964. Cognitive complexity and impressionformation. Progress in Experimental Personality Research,2: 47–90.

Czarniawska, B. 2004. On time, space, and action nets.Organization, 11: 773–791.

Daft, R. L., & Weick, K. E. 1984. Toward a model of organi-zations as interpretation systems. Academy of Manage-ment Review, 9: 284–295.

Damanpour, F. 1996. Organizational complexity and in-novation: Developing and testing multiple contingencymodels. Management Science, 42: 693–716.

Denison, D. R., Hooijberg, R., & Quinn, R. E. 1995. Paradox andperformance: Toward a theory of behavioral complexityin managerial leadership. Organization Science, 6:524–540.

Dooley, K. 2002. Organizational complexity. International En-cyclopedia of Business and Management, 6: 5013–5022.

Eppinger, S. D., Whitney, D. E., Smith, R. P., & Gebala, D. A.1994. A model-based method for organizing tasks inproduct development. Research in Engineering Design, 6(1): 1–13.

Espinosa, J. A., Slaughter, S. A., Kraut, R. E., & Herbsleb, J. D.2007. Familiarity, complexity, and team performance ingeographically distributed software development. Orga-nization Science, 18: 613–630.

Fairburn, C., Wright, P., & Fields, R. 1999. Air traffic control asdistributed joint activity: Using Clark’s theory of lan-guage to understand collaborative working in ATC. Pro-ceedings of the 10th European Conference on CognitiveScience: http://citeseerx.ist.psu.edu/viewdoc/download?doi510.1.1.35.8402&rep5rep1&type5pdf.

Farjoun, M. 2010. Beyond dualism: Stability and change asa duality. Academy of Management Review, 35: 202–225.

Feldman, M. S., & Orlikowski, W. J. 2011. Theorizing practiceand practicing theory. Organization Science, 22:1240–1253.

Feldman, M. S., & Pentland, B. T. 2003. Reconceptualizingorganizational routines as a source of flexibility andchange. Administrative Science Quarterly, 48: 94–118.

Fields, D. L. 2002. Taking the measure of work: A guide tovalidated scales for organizational research and di-agnosis. Thousand Oak, CA: Sage.

Flood, R. L. 1987. Complexity: A definition by construction ofa conceptual framework. Systems Research, 4: 177–185.

Galbraith, J. R. 1973.Designing complex organizations. Boston:Addison-Wesley Longman.

Garud, R., Kumaraswamy, A., & Karnøe, P. 2010. Path de-pendence or path creation? Journal of ManagementStudies, 47: 760–774.

Gell-Mann, M. 1995. What is complexity? Remarks on sim-plicity and complexity by the Nobel Prize-winning authorof The quark and the jaguar. Complexity, 1(1): 16–19.

Gell-Mann, M., & Lloyd, S. 1996. Information measures, ef-fective complexity, and total information. Complexity,2(1): 44–52.

Gibson, J. J. 1979. The ecological approach to visual percep-tion. Hillsdale, NJ: Lawrence Erlbaum Associates.

Giddens, A. 1984. The constitution of society. Oxford: PolityPress.

458 JulyAcademy of Management Review

Hackman, J. R. 1969. Toward understanding the role of tasks inbehavioral research. Acta Psychologica, 31: 97–128.

Hærem, T., & Rau, D. 2007. The influence of degree of expertiseand objective task complexity on perceived task com-plexity and performance. Journal of Applied Psychology,92: 1320–1331.

Hobday, M. 1998. Product complexity, innovation and in-dustrial organisation. Research Policy, 26: 689–710.

Hunter, J. E., Schmidt, F. L., & Judiesch, M. K. 1990. Individualdifferences in output variability as a function of jobcomplexity. Journal of Applied Psychology, 75: 28–42.

Kaspar, F., & Schuster, H. G. 1987. Easily calculable measurefor the complexity of spatiotemporal patterns. PhysicalReview A, 36: 842.

Kozlowski, S. W., Chao, G. T., Grand, J. A., Braun, M. T., &Kuljanin, G. 2013. Advancing multilevel research designcapturing the dynamics of emergence. OrganizationalResearch Methods, 16: 581–615.

Latour, B. 2005. Reassembling the social—An introduction toactor-network-theory. Oxford: Oxford University Press.

Lawrence, P. R., & Lorsch, J. W. 1967. Organization and environ-ment: Managing differentiation and integration. Boston:Harvard University Press.

Leidner, R. 1993. Fast food, fast talk: Service work and theroutinization of everyday life. Berkeley: University ofCalifornia Press.

Lempel, A., & Ziv, J. 1976. On the complexity of finite sequences.IEEE Transactions on Information Theory, 22: 75–81.

Leonardi, P. M. 2012. Materiality, sociomateriality, and socio-technical systems: What do these terms mean? How arethey related? Do we need them? In P. M. Leonardi, B. A.Nardi, & J. Kallinikos (Eds.), Materiality and organizing:Social interaction in a technological world: 25–48. Oxford:Oxford University Press.

Levin, I. P., Gaeth, G. J., & Schreiber, J. 2001. A new look atframing effects: Distribution of effect sizes, individual dif-ferences and independence of types of effects. Organiza-tional Behavior and Human Decision Processes, 88: 411–429.

Liu, P., & Li, Z. 2012. Task complexity: A review and concep-tualization framework. International Journal of IndustrialErgonomics, 42: 553–568.

March, J. G., & Simon, H. A. 1958. Organizations. New York:Wiley.

Mayer, B. W., & Dale, K. 2010. The impact of group cognitivecomplexity on group satisfaction: A person-environmentfit perspective. Journal of Behavioral and Applied Man-agement, 12: 24–47.

Miller, D., & Friesen, P. H. 1984. Organizations: A quantumview. Englewood Cliffs, NJ: Prentice-Hall.

Mintzberg, H., & Waters, J. A. 1985. Of strategies, deliberateand emergent. Strategic Management Journal, 6: 257–272.

Mogford, R. H., Guttman, J. A., Morrow, S. L., & Kopardekar, P.1995. The complexity construct in air traffic control: A re-view and synthesis of the literature. McKee City, NJ: CTA.

Moldoveanu, M. 2004. An intersubjective measure of organi-zational complexity: A new approach to the study of

complexity in organizations. Emergence: Complexity &Organization, 6: 9–16.

Moldoveanu, M. C., & Bauer, R. M. 2004. On the relation-ship between organizational complexity and organi-zational structuration. Organization Science, 15:98–118.

Nahrgang, J. D., DeRue, D. S., Hollenbeck, J. R., Spitzmuller, M.,Jundt, D. K., & Ilgen, D. R. 2013. Goal setting in teams: Theimpact of learning and performance goals on process andperformance. Organizational Behavior and Human De-cision Processes, 122: 12–21.

Naylor, J. C., Pritchard, R. D., & Ilgen, D. R. 1980. A theory ofbehavior in organizations. New York: Academic Press.

Novak, S., & Eppinger, S. D. 2001. Sourcing by design: Productcomplexity and the supply chain. Management Science,47: 189–204.

Oeser, O. A., & O’Brien, G. 1967. A mathematical model forstructural role theory: iii. Human Relations, 20: 83–97.

Orlikowski, W. J., & Scott, S. V. 2008. Sociomateriality:Challenging the separation of technology, work andorganization. Academy of Management Annals, 2:433–474.

Parmigiani, A., & Howard-Grenville, J. 2011. Routines revis-ited: Exploring the capabilities and practice perspectives.Academy of Management Annals, 5: 413–453.

Pentland, B. T., & Feldman, M. S. 2007. Narrative networks:Patterns of technology and organization. OrganizationScience, 18: 781–795.

Pentland, B. T., Hærem, T., & Hillison, D. 2010. Comparingorganizational routines as recurrent patterns of action.Organization Studies, 31: 917–940.

Pentland, B. T., Hærem, T., & Hillison, D. 2011. The (n)ever-changing world: Stability and change in organizationalroutines. Organization Science, 22: 1369–1383.

Polanyi, M. 1962. Personal knowledge: Towards a post-critical philosophy. Chicago: University of ChicagoPress.

Prokopenko, M., Boschetti, F., & Ryan, A. J. 2009. An information-theoretic primer on complexity, self-organization, andemergence. Complexity, 15(1): 11–28.

Puranam, P., Alexy, O., & Reitzig, M. 2014. What’s “new” aboutnew forms of organizing. Academy of Management Re-view, 39: 162–180.

Rahmandad, H. 2008. Effect of delays on complexity oforganizational learning. Management Science, 54:1297–1312.

Rivkin, J. W., & Siggelkow, N. 2007. Patterned interactions incomplex systems: Implications for exploration. Manage-ment Science, 53: 1068–1085.

Schatzki, T. R. 2005. The sites of organizations. OrganizationStudies, 26: 465–484.

Simon, H. A. 1969. The sciences of the artificial. Cambridge,MA: MIT Press.

Smith, R. P., & Eppinger, S. D. 1997. Identifying controllingfeatures of engineering design iteration. ManagementScience, 43: 276–293.

2015 459Hærem, Pentland, and Miller

Stajkovic, A. D., & Luthans, F. 1998. Self-efficacy and work-related performance: A meta-analysis. PsychologicalBulletin, 124: 240–261.

Sterman, J. D. 1989. Modeling managerial behavior: Mis-perceptions of feedback in a dynamic decision makingexperiment. Management Science, 35: 321–339.

Sterman, J. D. 1994. Learning in and about complex systems.System Dynamics Review, 10: 291–330.

Sydow, J., Schreyogg, G., & Koch, J. 2009. Organizational pathdependence: Opening the black box. Academy of Man-agement Review, 34: 689–709.

Thompson, J. D. 1967. Organizations in action: Social sciencebases of administrative theory. New York: McGraw-Hill.

Timmermans, D., & Vlek, C. 1992. Multi-attribute decision supportand complexity: An evaluation and process analysis of aidedversusunaideddecisionmaking.ActaPsychologica, 80: 49–65.

Tsoukas, H., & Chia, R. 2002. On organizational becoming:Rethinking organizational change.Organization Science,13: 567–582.

Utman, C. H. 1997. Performance effects of motivational state: Ameta-analysis. Personality and Social Psychology Review,1: 170–182.

Van de Ven, A. H., Delbecq, L. A., & Koenig, J. R. 1976. Deter-minants of coordination modes within organizations.American Sociological Review, 41: 322–338.

Weick, K. E. 1965. In J. G. March (Ed.), Laboratory experimen-tation with organizations: Handbook of organizations:194–260. Chicago: Rand McNally.

Weick, K. E. 1969. The social psychology of organizing. Read-ing, MA: Addison-Wesley.

Weick, K. E. 1995. Sensemaking in organizations. ThousandOaks, CA: Sage.

Weick, K. E., & Roberts, K. H. 1993. Collective mind in organi-zations: Heedful interrelating on flight decks. Adminis-trative Science Quarterly, 38: 357–381.

Wood, R. E. 1986. Task complexity: Definition of the construct.Organizational Behavior and Human Decision Processes,37: 60–82.

Wood, R. E., Mento, A. J., & Locke, E. A. 1987. Task complexityas amoderator of goal effects: A meta-analysis. Journal ofApplied Psychology, 72: 416–425.

Zhou, Y. M. 2013. Designing for complexity: Using divisionsand hierarchy to manage complex tasks. OrganizationScience, 24: 339–355.

Thorvald Hærem (thorvald.hærem@bi.no) is an associate professor at the BI NorwegianBusiness School. He received his Ph.D. in management from Copenhagen BusinessSchool. His research interests include organizational and individual routines, decisionmaking, and coordination in organizations.

Brian T. Pentland (pentland@broad.msu.edu) is a professor in the Department of Ac-counting and Information Systems at Michigan State University. He received his Ph.D. inmanagement from the Massachusetts Institute of Technology. His research interestsinclude organizational routines and methods for describing, explaining, and predictingpatterns of action.

Kent D. Miller (millerk@broad.msu.edu) is a professor of management at the Eli BroadGraduate School of Management, Michigan State University. He received his Ph.D. inbusiness administration from the University of Minnesota. His current research exam-ines organizational learning and change, as well as methodological and philosophicalissues in management research.

460 JulyAcademy of Management Review