proactive or reactive: an analysis of the effect of agent …zlin/isafm(1993).pdf · 2006-11-16 ·...

Proactive or Reactive: AnAnalysis of the Effect ofAgent Style onOrganizational Decision-making PerformanceZhiang Lin

I H. John Heniz III School of Public Policy and Management, Carnegie-MellonUniversity, Pittsburgh, PA 15213. USA

Kathleen CarleyDepartment of Social and Decision Sciences, Carnegie-Mellon University, Pittsburgh,PA 15213, USA

ABSTRACT This paper addresses the issue of agent style—proactive and reactive—from atheoretical perspective. The results show that agent style, though often consideredkey in decision making, only affects the organization's performance when theorganization is under moderate time pressure. Further, the effect of agent styledepends on the type of training given to organizational agents and the internalcondition under which the organization operates. This research suggests that,when resources are scarce, organizations should spend these resources onorganizational design and on increasing the accuracy of incoming informationrather than on altering the agents' style.

INTRODUCTION

Organizations vary in the rate at which theyrespond to organizational problems, even whenthey <»re in similar task environments or arefacing similar stress conditions (Scott, 1987).This rate of response and (presumably) itsaccuracy is due, in part, to the way agentsapproach problems (Kets de Vries and Miller,1986; LaPorte and Consolini, 1991; Roberts,1989, 1990). An agent's problem-solving style,which we refer to as agent style, varies onmany dimensions, such as proactiveness orcreativeness. A proactive agent engages indecision making and information gatheringwhenever possible. A reactive agent waits untilbeing asked or until it is absolutely necessaryto gather infonnation or make a decision

CCC 1055-615X/93/04027M7© 1993 by John Wiley & Sons, Ltd,

(Larson et al., 1986). Once they begin theirinformation search, proactive and reactiveagents spend the same amount of time on asingle search, and, once they begin the decision-making process, spend the same amount oftime making and communicating a decision.The difference between these agents lies intheir co-ordination. Proactive agents begintheir information search, decision making andcommunication without being told to do so bytheir manager. Reactive agents essentially donot do anything unless asked. We analyzehow the proactiveness or reactiveness of anorganization's agents affects organizational per-formance.

Organizations composed of proactive agentsshould respond to organizational problemsfaster than those with reactive agents. A conse-

Received March 1993Revised September 1993

INTELLIGENT SYSTEMS IN ACCOUNTING, FINANCE AND MANAGEMENT VOL. 2: 271-287 (1993)

quence is that organizations of proactive agentsshould outperform those of reactive agents(LaPorte and Consolini, 1991; Pauchant et ai,1990). The conventional evaluation of agentstyle typically regards the proactive style asbeing better for the organization than thereactive. The wisdom is that proactive organiza-tions are more active (Smiar, 1992), more co-operative (Rice, 1977), more prepared (Das,1986; Newman, 1989), and thus are supposedto be better performers Qauch and Kraft, 1986;Smiar, 1992). This argument can be summarizedby a famous Chinese proverb: One can alwaysget credit for sweat if not for achievement.

Given that we are interested in proactiveversus reactive behavior, another factor ofparamount consideration is timing. That is, formany tasks, it is important that the organizationmakes both accurate and timely decisions.Numerous studies show that under time press-ure, organizations tend to rearrange theirdecision-making process (Means et al, 1992;Rothstein, 1986) and perhaps co-ordinationprocesses. In many cases, time pressure causeserrors due to loss of information because itstresses the limit of human cognition(Magazannik, 1982). When time is short aproactive agent has an advantage because heor she is prepared and is always ready to makea decision (Pauchant et al, 1990). In contrast,the organization filled with a set of reactiveagents is at a disadvantage as precious timespent in problem solving and informationgathering is in telling the agents to startgathering information and making decisions.However, when timing is not critical, reactiveagents may be an economical solution as theymay need less training, and may have lowerinformation-processing costs (Pauchant et al,1990; Sussman, 1984).

The arguments that agent style matters arelargely drawn from micro-level studies oforganizational behavior and case studies oforganizations facing crises. While proactiveagents are often claimed to effect high organiza-tional performance, the relative benefit to theorganization of having proactive versus reactiveagents has not been systematically studied.Nevertheless, we know that organizations differin whether their agents are proactive or reactive.In the Chernobyl case (Silver, 1987), organiza-tional agents were proactive and tended to

make decisions on their own, some of whichwere incorrect. While in the Vincennes case(Rochlin, 1991), organizational agents werereactive and acted when ordered. Detailed casestudies of organizations under stress point toagent style as well as many other factors (suchas the structure of the organization and thenature of the task environment) when delineat-ing the determinants of organizational perform-ance. Theoretical and empirical studies haveshown that organizational structure (Carley,1991, 1992; Carley and Lin, 1992; Mackenzie,1978; Mintzberg, 1983; Padget, 1980), taskdecomposition scheme (Carley and Lin, 1992;Cohen et al, 1972; Mackenzie, 1978; Thompson,1967), training (Carley, 1992; Carley and Lin,1992; Hammond, 1973; Perrow, 1984), and stress(Carley and Lin, 1992; Krackhardt and Stem,1988; Staw et al, 1981) all affect performance.All of which suggests that a systematic analysisof organizations under stress should look atagent style, but should also control for theseother factors.

In contrast to the literature just discussed,there is a scientific tradition that treats agentstyle as irrelevant. Network studies (e.g.Mayhew, 1980; Wellman, 1988) argue thatperformance is a function of the structure ofrelations connecting agents in the organizationand in the roles that individuals hold. Contin-gency theorists (e.g. Burton and Obel, 1984,1990; Lupton, 1976; Woodward, 1965) arguethat performance is a function of the fit betweenorganizational structure and task environment.Both the network perspective and the contin-gency perspective implicitly suggest that agentstyle will not determine organizational perform-ance when these other factors are controlled.

In this paper, we will examine, from a meso-perspective, whether the agent's proactive orreactive problem-solving style affects organiza-tional performance. We are particularly con-cerned with the effect of time pressure onthe performance of organizations composed ofeither proactive or reactive agents. This studygoes beyond many studies of organizationaldesign as it not only considers both timing ofthe decision and the proactiveness/reactivenessstyle of the agents but it also controls forfactors such as organizational structure, task-decomposition scheme, training, nature of the

272 Z. LIN AND K. CARLEY

task environment, and stress. This study iscarried out using simulation.

We examine agent style using a dynamicand interactive computational model that inte-grates a set of factors that may influenceorganizational performance. There are severalreasons to examine the effect of agent styleusing simulation. First,, in the real worldthere is little consensus on what constitutesorganizational performance. Research hasshown that it is impossible to obtain the bestor sufficient indicator, and that whether anorganization is said to perform well dependson 'the purposes and constraints' placed on theorganizational performance measure (Cameron,1986). Performance has been viewed from avariety of perspectives, such as productivity(Argote and Epple, 1990), profitability(Lawrence and Lorsch, 1967), reliability(Roberts, 1989), etc. While such measures maytell you what the organization is doing, theydo not necessary indicate how well it is doingit. Because of the lack of a good measure oforganizational performance, empirical compari-son across real organizations such as thosewith different designs are somewhat suspect.

Second, assuming that we find a reasonableindicator of organizational performance that letsus gauge action against an objective indicator, itis virtually impossible to obtain sufficient datafor comparing organizations with a rangeof designs under both normal and stressfulsituations. The difficulty in getting such dataincludes time limits on data acquisition, confi-dentiality, insufficient infonnation on specificoperating conditions within organizations, andlack of comparability across industries. Suchdifficulties tend to result in an unbalanceddesign from an analysis standpoint and so callinto question the generalizability of the results.A related difficulty with field studies of per-formance is that they typically focus on success-ful firms (Child, 1974; Drazin and Van de Ven,1985; Lawrence and Lorsch, 1967; Long, 1980).As such, they provide little insight from adesign perepective as to whether 'failed' firmsdiffer in design from 'successful' ones.

In contrast, simulation has multiple advan-tages:

(1) We can conduct balanced simulation exper-

iments, and control certain factors to exam-ine the effect of other factors.

(2) We can consider both successful and failedfirms. Thus results will not be biased bylooking only at successes.

(3) Simulated organizations have been shownto resemble the real-world organizations inan idealized way (Lin and Cariey, 1993). Theperformance characteristics of simulatedorganizations are, under certain conditions,comparable to the performance character-istics observed in the real world.

(4) Researchers have also shown that organiza-tional performance is affected by factorssuch as organizational design (Houskissonand Galbraith, 1985; Lawrence and Lorsch,1967), task environment (Drazin and Vande Ven, 1985), and stress (Anderson, 1977).

Only a systematic examination of thesefactors on organizations can address theissue of what really constitutes organiza-tional performance. By using simulation,we can get an insight into these importantfactors with less cost than conductinghuman experiments or field studies. Oncethe dominant factors are examined, humanexperiments or field studies can be done totest the theoretical results.

To guide our analysis, we will consider howthe results of our model fit with, contradict orelaborate on propositions put forward in theliterature with respect to agent style. We willbriefly describe the model, analyze the resultsfrom the model and, finally, we will discussthe results and present conclusions.

MODEL DESCRIPTION

Task

The organizations we simulate are faced witha limited choice task in which organizationsmake decision choices regarding the stateof moving aircraft under stress from limitedalternatives according to infonnation they havethrough organizational communication pro-cesses. The simulation test-bed used in thispaper is an adaptation of that in Cariey andLin (1992) that takes into account time andagent style. Choice tasks are very common in

EFFECT OF AGENT STYLE ON ORGANIZATIONAL DECISION-MAKING PERFORMANCE 273

the real world. Such choice situations includelaw making, price setting, planning, and a hostof other similar things (Allison, 1971; Marchand Olsen, 1976; Shull et al, 1970).

We have operationalized this choice task asa stylized radar task. There is a single aircraftin the airspace. This aircraft is moving. Theorganization has three choices—decidingwhether the aircraft in the airspace is friendly,neutral or hostile. Each aircraft is characterizedby nine parameters (see Carley and Lin, 1992,for details).

We measure the organization's performanceas the percentage of these problems that theorganization has made the correct decision (thetrue states of all the aircraft are predefined bythe task environment and are unknown to theorganization), regarding whether the aircraft isfriendly, neutral or hostile. We treat type Ierror (making the decision that the aircraft ishostile, but in fact the aircraft is friendly) andtype II error (making the decision that theaircraft is friendly, but in fact the aircraft ishostile) with equal weight. In organizationalliterature, organizational performance has alsobeen measured by organizational effectiveness(Mackenzie, 1978; Pfeffer and Salancik, 1978),and organizational efficiency (Mackenzie, 1978;Scott, 1987). In many cases, accuracy, effective-ness and efficiency are indistinguishable andcan convert into each other.

Task environment

Task environments vary on a large number ofdimensions. Two such dimensions that havereceived some attention in the literature aredecomposability and biasness. Task environ-ment decomposability measures the inter-relationships among task components. A taskenvironment is decomposable if there are nocomplex interactions among components thatneed to be understood in order to solve aproblem, and non-decomposable if otherwise.In a decomposable task environment, a linearcombination of the features is sufficient toproduce the decision. In a non-decomposabletask environment there are interaction effects.As LaPorte and Consolini (1988, 1991) andRoberts (1989, 1990) note, interdependence oftask components (i.e. non-decomposability) isa factor that should be considered in designing

the organization. Task environment biasnessmeasures the distribution of all possible out-comes. A task environment is unbiased if allthe possible outcomes are equally likely tooccur, and biased if not. As Aldrich (1979)notes, a biased environment is one in whichthe problems faced by the organization areconcentrated and so highly similar. Based onthese two manipulations, we examine fourdifferent 'realities' or environmental situations.They are: (1) biased decomposable task environ-ment; (2) unbiased decomposable task environ-ment; (3) biased non-decomposable taskenvironment, and (4) unbiased non-decompos-able task environment.

Stress

Organizations are often affected by stress(Perrow, 1984; Shrivastava, 1987). In examiningorganizational performance it is important toconsider multiple sources of stress: internalsuboptimal operating conditions or murphies(caused by suboptimal operating conditionwithin organizational design), external hostileconditions or maydays (caused by hostile taskenvironment), and time pressure.

A murphy refers to a small disruption in thegathering or communicating of informationwithin the organization (Carley and Lin, 1992).Murphies are things that move the organizationfrom an optimal to a suboptimal operatingcondition and thereby create internal stress.We examine five types of murphies:

(1) Missing information—a piece of theincoming information for a particular prob-lem is not available;

(2) Incorrect information—a piece of incominginformation is erroneous;

(3) Agent unavailability—an analyst is notavailable to help the organization solve theproblem and so does not report his or herdecision to his or her manager;

(4) Communication channel breakdown—ananalyst is unable to report to a superiorbecause the communication channel isunavailable;

(5) Agent turnover—an analyst leaves theorganization and is replaced by a new one.

For each type of murphy, the number ofsimultaneous murphies ranges from 0 to 3.

274 Z. LIN AND K, CARLEY

A mayday refers to an external threat to theorganization (Carley and Lin, 1992). This canbe a problem such that the wrong decision mayhave disastrous consequences. For example, ahostile aircraft would be considered a mayday.We define as maydays those problems thatrepresent hostile aircraft.

In this paper, time pressure occurs when theaircraft is very fast, or when the time requiredto make the decision is very short. Timepressure is the number of time units beforethe organization must make a decision. Twofactors affect when the organization must makea decision: a decision is demanded (1) by someoutside or superior sources (e.g. Congress) and(2) because the aircraft reaches the danger point(the point at which if the organization doesnot respond and the aircraft is hostile, theaircraft can destroy the organization). In thismodel, this danger point occurs when theaircraft has a range of 1 mile and/or a rangeof 5000 feet. Both numbers were chosen basedon characteristics of radar systems.

In this model for each problem (i.e. an aircraftappearing in the airspace), we randomly assignthe number of time units prior to a demandfor a decision to be between 1 and 60. Thecharacteristics of the aircraft are also randomlychosen for each problem. Each aircraft can alsovary with different characteristics such asspeed, range and altitude. Consequently, thetime pressure varies randomly across all prob-lems.

We study three levels of time pressure: low(41-60 time units), medium (21-40 time units)and high {1-20 time units). Thus, time pressurevaries linearly across the three conditions. Aparticular organization will be faced with asequence of problems and each problem has aparticular time pressure associated with it. Thetime pressure defines how many time unitsthe organization has to determine the state ofthe aircraft for this problem, varying randomlyacross the problems. Because this is a dynamicenvironment, agents' interaction with oneanother in terms of whether to communicateand how to communicate, and the utilizationof decision-making procedures in terms ofwhich procedure to choose, all depend on theconstrained resource—time. How agents reactto time pressure is discussed later.

Organizational Design

While most existing definitions of organiza-tional design focus only on either structure ortask decomposition scheme or both, in thispaper organizational design is viewed as acombination of organizational structure, task-decomposition scheme, training and agentstyle. They virtually ignore procedures to pro-cess information learned through training.Through an examination of multiple designs,expected relations between design and perform-ance can be computationally deduced.

Organizational structures are defined in termsof the network of relations among agents. Westudy four stylized organizational structures:

(1) A team with voting—a totally decentralizedstructure in which organizational decisionis through majority voting of each membersof the organization;

(2) A team with a manager—basically a flathierarchy such that while each analystexamines information and makes a rec-ommendation, the ultimate organizationaldecision is made by the manager (or teamleader);

(3) A hierarchy-a multi-leveled communicationstructure in which each baseline agentexamines information and makes a rec-ommendation to his or her immediatesupervisor who in tum makes a recommen-dation to the top-level manager who makesthe ultimate organizational decision;

(4) A matrix—like the hierarchy, a multi-leveledcommunication structure, except that eachbaseline agent has two communication linkswith two middle managers across divisions.

The task-decomposition scheme defines therelations between agents and resources and/or information. We study four stylized taskdecomposition schemes:

(1) Segregated—each baseline agent has accessto one task component;

(2) Overlapped—each baseline agent has accessto two task components, with one taskcomponent being overlapped with anotherbaseline agent;

(3) Blocked—each baseline agent has accessthree task components, but usually eachthree agents within the same division havethe same three task components; and


(4) Distributed—each baseline agent has accessto three task components usually acrossdifferent divisions.

The third and final aspect of orgaiuzationaldesign, with which we will be concemed, isthe type of training. We study three trainingscenarios: (1) untrained^—each agent in theorganization makes decisions by basicallyguessing; (2) experientially trained—each agentin the organization makes decisions by refer-ring to historical experience; and (3) oper-ationally trained—each agent in the organiza-tion makes decisions by using standardoperating procedure. The number of timeunits required for different types of decisionprocedure is different given the same amountof information. Due to the different complexityof decision-making processes, experientialdecision-making procedure usually takes moretime than operational decision-making pro-cedure, and they both take more time thanuntrained decision-making procedure. In thispaper all agents are considered to be fullytrained. We examine performance as it variesacross the type of training received and notthe level of training.

Agent Style

Regardless of agent style, each agent can takethe following actions—ask for information,read information, make a decision, pass up adecision, and wait. The amount of time thatthese actions take depends on the amount ofinformation the agent must process and thetype of decision-making procedure they use(e.g. experiential) and not on the agent's style.In Table 1 the number of time units each actiontakes is displayed. Agent style affects the orderpreferred by the agent for engaging in theseactions and in when the agent initiates them.Time pressure does not affect how many unitsit takes to perform an action but how manyactions can be taken and how agents choosethe preferred order in which to take them.

A proactive agent (Figure 1) asks for infor-mation, reads information, then if there isinformation, makes a decision based on theavailable infonnation, then passes on thedecision. This cyclic process repeats until timeexpires, i.e. the aircraft goes out of range or

Table 1 Agent's possible actions

Name ofaction

Ask forinformationReadinformationMake adecision

Pass up adecisionWait

Time units neededfor action

11'Number of pieces ofinformation to be processed•Untrained or randomdecision—1•Operationally traineddecision—1'number ofpieces of information to beprocessed•experientialiy traineddecision—2'number ofpieces of information to beprocessed1

1

Prerequisiteaction

Ask forInformationReadInformation

Make aDecision

Note: Each time unit equals 2 seconds of real time. This isaccording to initial lab experiments using human subjectson decision making by Kathleen Cariey and Michael Prietulaof Carnegie-Mellon University. In the experiment each subjectprocesses 120 problems in about 40 minutes, which isatMut 20 seconds tor each problem. For every problem, asubject reads three pieces of information (3 units), makesexperiential decision (6 units), and passes the decision {1unit). Thus if we let x as the seconds in each time unit, wehave 3x + 6x + X = 20, or X = 2. Also, for analysts, thereis no prerequisite action for the action Read Information.

the top manager decides to stop the process.Each agent (except the top manager) can beinterrupted by a request from an upper managerfor decision. Manager interrupts can disruptthis cycle at the points indicated by a box andarrow in Figure 1. When the request from themanager arrives the agent responds to therequest during the next time unit. Thus, onlyactions that make more than one time unit(e.g. make a decision) can be disrupted. Whenan agent receives a request for informationhe or she will respond to the request bycommunicating a decision, if there is oneavailable. After responding to the request theagent returns to the action in which they wereengaged when the request arrived. Clearly,the procedures followed by the top manager,middle managers and baseline analysts differsomewhat on the basis of their organizationalposition. The top manager cannot be interrup-ted (since there is no superior) and a baselineanalyst cannot ask for information (since thereis no subordinate) while a middle manager can

276 2. LIN AND K. CARLEY

IF TIME NOT ENOUGHFOR DECISION

If previously mide decisioncxtsis. pus \ip the ikcif ion.t\ic if top maniger ind noprevioui decision, make arandofn decision u fmaJdecuion.

IF TIME ENOUGH FORDECISION

Sun or coniinuc.

KKQtKSTFROM UPPERMANAGER

Coniinuc

'RESPONSE TO IIEQUEST>

RFQl EST FROM UPPERMANAGER

If pfcviously made decisionexists, pasi up the decision,elie oontinue current action. J

Read Inrormation

Number of limc uiuu needed u btied on thenumber of piec£* of mftxmation to be proccased

Ask forInrormation

One lime unit Tweded forihis aciion.Analyiu ikip ihii action

Pass up DecisionOne time unit needed forthis action. Make a Decision

Number of time units needed abted pn the number of piecci ofinformition to be procesied andthe type of decision lo be made.

REQlfcSTUPPER MANAGER

^RESPONSE TO REQUEST^

1/ pnviouity made decuioncxiiu. pati up the dociaion.eUe ooniinue currtni actian.

Figure 1 Preference function of proactive agents (Note: in this and Figure 2 if there is no request from anupper manager, the agent continues current action. Also, for the top managef, there is no interruption.)

be interrupted as well as ask for information.Further, the top manager has the power todecide the final or organizational decisionaccording to certain standards (for a detaileddescription of the styles at three levels, seeTable Al in the Appendix).

A reactive agent {Figure 2) reads informationrepeatedly until time expires or until the agentis told by the manager to stop. Each agent(except the top manager) can be interruptedby a request from an upper manager fordecision as indicated in Figure 2. When therequest from the manager arrives the agentresponds to the request during the next time

unit. Thus, only actions that take more thanone time unit (e.g. make a decision) can bedisrupted. The agent will then stop readinginformation (if not finished yet), and begina cycle of decision-making process: ask forinformation, read information, make a decisionand pass up a decision. This cycle may also beinterrupted by a new request from an uppermanager. When an agent receives a request forinformation during the cycle, he or she willrespond to the request by communicating adecision, if there is one available, or simplycontinue current action. After responding tothe request the agent returns to the action in


REQUEST FROM UPPERMANAGER

ead Inrormahon

RESPONSE TO REQUESTThe top manager uks for iiuonnsuon Tin t buttill there ii inTormaiion being ptsicd up. then readsthe infopnnilion. mikci • decisiim, uid passes up (orconcludes) the deciiion.

If prcvioutty made docuton. pus up Ihc deciiion.

CIK no dccuior ti pasMd iqi. j

TIME NOTENOUGH FORDEas iON

TIMF ENOUGHFOR DECISION REQUEST FROM UPPER

MANAGER

RESPONSE TO REQUEST^Read Inforination

•f pnviousty made docuuncxbts. pass up the deciiion.eUe no deciiion ii paued up

Number of lime units needed ubased on the number of pieces ofnformaiion lo be procecsed.

Ask forInformatio

One lime unit isneeded for this iclion

REQLEST FROM UPPERMANAGER

Pass up DecisionOne time unit itreeded for this

Make a Decision REQUEST FROM UPPERMANAGER

Number of Ume umu rweded isbased on the number of piecesof bn/orniition to be proccisedand the type of deciiion to be

RKQl F.STFKO.M UPPERMA^^fiER

rrwiLinueIf pnviously made deciikmexists, pass up the decision.

. eke no decision is pusod up- .

Figure 2 Preference function of reactive agents.

which they were engaged when the requestarrived. After finishing the decision-makingcycle, the agent returns to the default action—read information—again. Unlike the proactiveagent, a reactive agent will not make a decisionand pass it on unless being requested by hisor her manager. Again, there are processingdifferences among agents due to their organiza-tional positions (top, middle and baseline).The top manager cannot be interrupted, and

initiates the decision-making process by askingfor information. A middle manager will ask forinformation only when a request from the topmanager is received for decision and when thereis no decision already being communicated bybaseline analysts. A baseline analyst alwaysreads infonnation and never asks for infor-mation (for a detailed description of styles atthree levels, see Table A2 in the Appendix).

The proactive and reactive agent take the

278 Z LIN AND K CARLEY

same amount of time for each action. However,as can be seen in Figures 1 and 2, the orderin which actions are taken by the agents isdifferent. Further, at any point in time, the'position' of the proactive and reactive agentsin the decision-making cycle may be different.Consequently, proactive agents typically takeless time than reactive ones because they arealready further along on the cycle when arequest for a decision arrives.

METHODOLOGY

The performance of the organizations is gener-ated using a computer simulation test-bed builtin C within a VAX/VMS system. Using thistest-bed, we are able to systematically altertask environment (4), organizational structure(4), task-decomposition scheme (4), trainingscenario (3), agent style (2) which are all builtwithin the test-bed. Thus, 384 organizationaltypes were examined. By varying type (5) anddegree (4) of internal stress, we examine 20internal conditions, with degree 0 as the optimaloperating condition. In this experiment, weexamine a total of 7680 cases.

Each case represents an organization facedwith a specific internal condition and taskenvironment. The performance of each case/organization is evaluated by examining per-formance on 1000 problems. Each problemrepresents a specific moving aircraft. This studyis carried out using Monte Carlo analysis. Allcharacteristics of each problem are randomlygenerated and include: the starting position,speed and nature of the aircraft, the locationof the murphies affecting the organizationinternal, and the maximum time units allowedthe organization to make a decision regardingthe aircraft.

The 1000 problems can be classified asbeing friendly, neutral or hostile or under low,medium or high time pressure. We can measureoverall performance (percentage of problemsfor which the organization made the correctdecisions) or performance for a subset of the1000 problems (e.g. percentage of problemsgiven the true state is hostile for which theorganization made the correct decisions). Weexamine only post-training performance. Thatis, all agents in all organizations examined are

trained on a large set of aircraft and thenperformance is measured.

Prior to running these 1000 problems (onwhich performance is measured) all agents inall organizations are trained. Agents are trainedin an environment in which there are nomurphies and no time pressure. In this study,we examine the behavior of these organizationscomposed of fully trained agents on lCKX)problems. Learning does not continue whileworking on these problems. We use the 1000problems so that we can get an estimate ofthe expected performance of the organizationcomposed of fully trained agents. Learning perse does not drive the results obtained from thismodel. However, the type of training doesaffect the organization's behavior.

This simulation test-bed offers us the chanceto examine systematically the interactionsamong, and effect of, the various organizationalfactors on organizational performance that haveconcerned researchers. Nevertheless, in thispaper our focus is on agent style.

RESULT AND ANALYSIS

Across All Different Levels of TimePressure

According to the general discussions of thenature of proactive and reactive agent styles inthe literature, one might expect the followingpropositions to hold:

Proposition 1: Agent style does not matter(Mayhew, 1980; Wellman, 1988).

What this says is that when the structure ofthe organization and the fit between the taskenvironment and organizational structure areconsidered, organizations composed exclus-ively of proactive agents (proactiveorganizations) and those composed exclusivelyof reactive agents (reactive organizations)should exhibit equivalent performance.

Proposition 2: Proactive organizations performbetter than reactive organizations (Smiar, 1982).

canProactive agents are better prepared and Lanrespond faster. Thus, organizations filled withproactive agents should perform better than


those with reactive agents, controlling for allother factors.

Our analysis supports Proposition 1. Whetherwe consider just experientially trained organis-ations or operationally trained ones, or both,there is no difference between the performanceof organizations with proactive and reactiveagent styles. This is true regardless of externalor internal conditions (Table 2). Using a one-tailed f-test for each pair of proactive andreactive organizations with the same character-istics, there is no difference at the 0.01 signifi-cance level. Degrees of freedom are reportedin Table 2. Similarly, for different organizationalstructures, for different task-decompositionschemes and for different task environments,there is no difference in the performance ofproactive and reactive organizations.

Agent style appears simply not to matter.This result supports Propjosition 1 but notProposition 2. It shows that proactive organiza-tions do not necessarily perform better overall.This result can be attributed to the fact thatorganizations composed of intelligent agentsact as though they have persistent beliefs. Newdecisions do not replace old ones unless theorganization is both highly confident in thenew decision and doubtful about the oldone. The unwillingness on the part of theorganization to change its decision regardlessof whether it is proactive or reactive leads toan entrenchment of ideas. In the reactiveorganization, this entrenchment has little effectsince they are so slow to gather new informationand change a new decision. However, in theproactive organization, entrenchment mitigatesthe value of proactive behavior. Despite con-tinually gathering new information and makingnew decisions, due to entrenchment, the likeli-hood of the new decision replacing the old oneis relatively low.

Under High Time Pressure

Under high time pressure, organizations areoften overloaded with information. They giveup normal 'time-spending procedures' or nor-mal decision-making procedures and mainlymake decisions based on hunches (Rosenthalet al., 1989). Thus, agent style should not matterwhen time is at a premium, controlling for

Table 2 Performance by agent style across all levelsot time pressure

External Internalcondition condition Proactive

With either experiential or operationalOverall

Across all

Murphies

Overall

Maydays

Murphies

47.00(7680,0.20)

48,64(1920,0,39)

46.44(5760,0,20)

57.14(7680,0.26)

59,01(1920,0,57)

56,53(5760,0,30)

With experiential trainingOverall

Across all ^

Murphies

Overall

Maydays

Murphies

46,11(3840,0,29)

47.28(960,0.57)

45,72(2880,0,31)

59,47(3840,0,32)

61.30(960.0 90)

58,85(2880.0,49)

With operational trainingOverall

Across all ^

Murphies

Overall

»«». ^ , ,^ OptimalMayaays

Murphies

47,86(3840.0,24)

49,99(960,0,53)

47,16(2880.0.24)

54,83(3840.0,33)

56,71(960,070)

54,20(2880.0.36)

Reactive

training46.94

(7680,0.20)48.69

(1920,0.39)46.36

(5760,0,20)

57.04(7680,0,26)

59.01(1920,0,57)

56,39(5760,0,30)

46.08(3840.0,29)

47.32(960,0,57)

45,66(2880,0,31)

59,39(3840.0,32)

61.35(960,0,89)

58.74(2880.0,49)

47,81(3840.0,24)

50,06(960,0.53)

47,06(2880.0,24)

54.70(3840.0,33)

56.68(960.0,69)

54,04(2880,0,36)

Note: Periormance is in percentages. Number o( cases andstandard errors are in parentheses below performance

all other factors. This suggests the followingproposition:

Proposition 3; Proactive and reactive organiza-tions have the equivalent performance underhigh time pressure (Rosenthal et ai, 1989).

In our analysis, when there is high timepressure on average there is no difference


between the performance of proactive andreactive organizations, regardless of externalconditions. Using a one-tailed t-test for eachpair of proactive and reactive organizationswith the same characteristics, there is nodifference at the 0,01 significance level. Degreesof freedom are reported in Table 3. However,if we consider only optimal internal conditions(no murphies), organizations filled with proac-tive agents perform worse than those filledwith reactive agents. This is true across all

external conditions (one-tailed Ntest, p < 0.001,DF shown in Table 3), and when just hostile(mayday) conditions are considered (one-tailed(-test, p < 0.005, DF shown in Table 3). It isimportant to note, however, that all organiza-tions under high time pressure are making theright decision approximately 33.33% of thetime. If the agents were simply guessing, thenorganizational performance would in fact be33.33%. While the proactive organization isslightly worse than the reactive one neither is

Table 3 Performance by agent style and level of time pressure

Trainingtype

AcrossbothTraining

Experientialtraining

Operationaltraining

Externalcondition

\JVKS' a l 1

Maydays

Ovprall

Maydays

rH/PrAllKJV^^i all

Maydays

Internalcondition

Overall(2560)

Optimal(640)

Murphies(1920)Overall(2560)

Optimal(640)

Murphies(1920)Overall(1280)

Optimal(320)

Murphies(960)

Overall(1280)

Optimal(320)

Murphies(960)

Overall(1280)

Optimal(320)

Murphies(960)

Overall(1280)

Optimal(320)

Murphies(960)

High (imeAgent

Proactive

33.47(0.05)

33.52(0.10)33.46(0.06)

33.43(0.08)

33.52(0.15)

33.41(0.09)

33.27(0.07)33.27(0-13)33.27(0.08)

33.18(0.10)

33.27(0.21)

33.15(0.12)

33.67(0.08)

33.77(0.16)

33.64(0.08)

33.69(0.11)

33.77(0.22)33.66(0.12)

Note. Pertormance Is in percentages Number of casesconditions Standard errors

pressurestyleReactive

33.47(0.05)

33.66(0.10)33.41(0.06)

33.41(0.07)

33.63(0.15)

33.34(0.09)

33.24(0.07)33.31(0.14)

33.22(0.08)

33.25(0.10)

33.41(0.20)

33.19(0.12)33.71(0.07)

34.01(0.15)

33.61(0.08)

33.57(0.11)

33.85(0.21)

33.48(0.12)

for each row1 are in parentheses twiow performance.

Level of time pressureMedium time pressure

AgentProactive

48.70(0.23)

50,41(0.52)

48.13(0.25)59.73(0.32)

61.67(0.68)

59,09(0.36)

43.69(0.29)44.38(0.60)

43,45(0.33)

55.38(0.45)

56.39(0.93)

55.05(0.51)

53.71(0.29)

56.44(0.71)52,80(0.31)

64.09(0.42)

66.96(0.91)63.13(0.47)

IS listed in the

styleReactive

48,55(0.23)50.39(0.52)

47,94(0.25)

59.51(0.31)

61.61(0,67)

58.81(0,35)43,54(0 29)44.39(0.59)43.26(0.33)

55.14(0.43)

56.42(0.90)54,71(0.49)

53.56(0.29)

56.40(0.70)5261(0.31)

63.88(0.42)

66.79(0,89)

62.92(047)

parenthesis in

Low timeAgent

Proactive

58,79(0.31)

61,98(0.68)

57,73(0.35)78,27(0.39)

81.82(0,78)77.09(0.44)

61,38(0.51)64,18(V02)

60.44(0,59)

89,84(0,42)

94.25(0,65)

88,37(0,51)

56.21(0.35)

59.77(0.86)55.02(0.37)

66.71(0,46)

69.40(101)

65,81(0.51)

the column i

pressurestyleReactive

58.81(0.31)

62.02(0.67)57.74(0,35)78.21(0.39)

81.80(0,78)77.02(0.44)

61.45(0.51)

64,27(102)

60.51(0.58)

89.79(0.42)

94.21(067)88.31(0.50)

56.17(0.36)

59,77(0,86)54.97(037)

66.64(0,46)

69.40(1,01)65,72(0.52)

[itied Internal


noticeably different from an organization thatis simply guessing. We must be cautious ininterpreting this result, as the fact that proactiveorganizations appear worse may be an artifactthat would be eliminated were we to examineorganizational performance over more prob-lems. Note also (Table 3) that, regardless ofwhether we consider only experientially trainedorganizations or operationally trained ones,performance always hovers around 33.33%.

In summary, under extremely high timepressure, organizational performance is essen-tially reduced to be equivalent to the bestthat can be expected when agents guess.Consequently, there are no consistent effectsthat can be attributed to agent style. Thisgenerates mixed support for Proposition 3.

Under Medium Time Pressure

Under moderate time pressure, the organiza-tional decision-making process should betightly constrained by time, though not com-pletely disrupted. Thus agent style shouldcritically affect performance, controlling for allother factors.

Proposition 4: Proactive organizations performbetter than reactive ones under medium timepressure (Jauch and Kraft, 1986; Pauchant etal, 1990; Smiar, 1982);

Our analysis suggests that, on average, thereis no difference between the performance ofproactive and reactive organizations. Using aone-tailed Ntest for each pair of proactive andreactive organizations with the same character-istics, there is no difference at the 0.01 signifi-cance level. Degrees of freedom are reportedin Table 3. This is true regardless of externalor internal conditions (Table 3). However, whenwe consider the type of training that the agentshave received, we see significant performancedifferences due to agent style under suboptimaloperating conditions (murphies).

In experientially trained organizations, if weaverage across all operating conditions, orconsider only optimal internal operating ones,there is no difference between the performanceof proactive and reactive organizations, regard-less of external conditions. Using a one-tailedt-test for each pair of proactive and reactiveorganizations with the same characteristics.

there is no difference at the 0.01 significancelevel. Degrees of freedom are reported inTable 3. Under suboptimal internal operatingconditions, organizations filled with experien-tially trained proactive agents outperform thosefilled with experientially trained reactive agents.This is true both across all external conditions(one-tailed f-test, p < 0.001, DF shown in Table3) and under maydays (one-tailed Ntest, p <0.001, shown in Table 3). The performancecomparisons between proactive and reactiveorganizations are also illustrated in Figure 3. Asimilar pattern occurs for operationally trainedorganizations.

In summary, under medium time pressure,agent style affects organizational performanceonly under suboptimal internal operating con-ditions. In this case, organizations filled withproactive agents outperform those filled withreactive agents. Thus, Proposition 4 is sup-ported only, under suboptimal conditions. Pro-active agents benefit the organization, but thatbenefit is most apparent when time is somewhatconstrained and things are going wrong.

Under Low Time Pressure

Under low time pressure, organizations haveenough time to make decisions regardless ofexternal or internal conditions. It is usuallyefficient to use reactive style as it offers acomparable performance with proactive style,but with lower operating costs (Sussman, 1984).Thus, we should expect no difference in theperformance of organizations with either proac-tive or reactive style when time is not critical,controlling for all other factors.

Proposition 5: Proactive and reactive organiza-tions have the equivalent performance underlow time pressure (Sussman, 1984).

In our analysis, when considering either experi-entially trained organizations or operationallytrained ones, or both, there is no differencebetween the performance of organizations withproactive and reactive organizations with thesame characteristics, there is no difference atthe 0.01 significance level. Degrees of freedomare reported in Table 3. This is true regardlessof external or internal conditions (Table 3).Thus, under low tim e pressure there is virtuallyno effect of agent style on organizational

282 Z. LIN AND K. CARLEY

PtRFORMANCE

PROACTIVE(bOTHj

DUttnmc*Sigaiflaat

4 8 *

4 6 -

4 4 -

4?

1 nI K

1

DifTcrcBCCSinincaBt

1 1. 1REACnVE PROACTIVE REACTIVE PROACTIVE REACTIVE

(BOTH) (EXPERIENTIAL) (EXPERIENTIAL) (OPERATIONAL) (OPERAtWNAL)

AGENT STYLE(TYPE OF TRAIN ING)

All Internal Conditions and All External ConditionsOptimal Internal Condition and All External ConditionsSuboptimal Internal Condition and All External Conditions

All Internal Conditions and Under MaydaysOptimal Internal Conditions and Under MaydaysSuboptimal Internal Conditions and Under Msydayt

Figure 3 Performance by agent style under medium time pressure.


performance. This result supports Proposition5.

DISCUSSION AND CONCLUSION

This analysis is mainly based on simulationtechniques. Those simulated organizationshave enabled us to examine the effect ofagent style across numerous factors that haveinterested researchers in organizations withina reasonably short period of time. In fact,to examine these same factors using humansubjects would have taken at least ten yearsand cost at least one million dollars. Computersimulation is a powerful extension of humancognition. As pointed out by Ostrom (1988),computer simulation offers a third symbolsystem in studying social science, besidesnatural language and mathematics, because'computer simulation offers a substantialadvantage to social psychologists attemptingto develop formal theories of complex andinterdependent social phenomena'. Fararo(1989) also regards computational process asone of the three processes (the other two aretheoretical and empirical processes) necessaryto the development of any discipline.

In this research, our focus is on the effectof one aspect of agent style (proactiveness/reactiveness) on organizational performance,not on the origin of agent style. We believefurther understanding of the sources of agentstyle is necessary for future research into theeffect of agent style. More aspects of agentstyle such as emotion and creativity may alsobe included in future research. In addition, wewould also like to examine more real-worldorganizations, thus providing new insight intothis theoretical study.

We categorized time pressure into threelevels according to the time units assigned tothe organization or those spent when theorganization has to make a decision before theaircraft reaches a certain point. It is difficult todisentangle time pressure from organizationaldesign and task environment, because a hightime pressure to some organizations may notbe as stressful as to others. In the case of ateam with a manager versus a matrix organiza-tion given the same task-decompositionscheme, the organization will need more time

in the matrix organization than in the team witha manager organization, thus time pressure maynot have the same effect on them.

Also, the effect of time pressure may also bedifferent on differently trained organizations.An operationally trained organization usuallyrequires less time to proceed a normal decisionthan an experientially trained one. In thispaper, this is represented by the time unitsneeded for each operation, with an operationaldecision-making procedure needing less timeunits than an experiential one given the sameamount of information.

The fact that organizational form (structureand task-decomposition scheme) and taskenvironment do not change the pattern of effectof agent style shows that this aspect of agentstyle is a relatively weak factor in organizationaldecision-making performance, compared withfactors such as organizational structure, task-decomposition scheme and task environment.

In this paper we have studied the effectof one aspect of agent style—proactiveness/reactiveness—on organizational performance.Our results show that, on average, there isvirtually no effect of agent style on organiza-tional performance (support for Proposition 1but not for Proposition 2). However, if weconsider the effect of time pressure we findthat, in general, only under medium timepressure does agent style play a large role.Further, under medium time pressure, organi-zations filled with proactive agents outperformthose filled with reactive agents, but only ifthe organization is facing suboptimal operatingconditions. Further, there is basically no inter-action effect between agent style (proactiveness/reactiveness) and structure, task-decompositionscheme, training scenario, or task environment,on performance.

There is, however, an interaction effectbetween agent style and stress. The resultssuggest that under optimal operating conditions(without internal stress) agent style does notmatter. Under maydays, again, agent style doesnot matter. However, if murphies are presentwhether or not the organization is in a maydaysituation, agent style matters.

On average, then, we find support for theimplication of network theory and contingencytheory that agent style does not matter. Acloser look, however, where we control for the

284 Z. LIN ANO K. CARLEY

effects of time pressure, reveals that behavior-ists and crisis analysts are indeed right thatagent style can matter. Our results suggest thatthis apparent contradiction is due to agentstyle being largely irrelevant except in highlyspecific situations and in those situations, agentstyle plays a critical role.

The situation where agent style has the mostconsistent and noticeable effect is when thereis moderate time pressure and the organizationsis facing various obstacles that limit its infor-mation-processing capabilities (erroneousinformation, incomplete information, etc.). Onecould argue that this situation is fairly common(March and Olsen, 1976).

The result that agent style in most situationsis irrelevant to organizational prformanceshows that the amount and frequency ofinformation used in decision making may notcharacterize the quality of decisions. The mytharound proactive agent style can probably beattributed to the fact that organizations oftencollect information for 'signal and symbol'purposes rather than for decision making(Feldman and March, 1981).

To conclude, we find support for the impli-cation of network theory and contingencytheory that agent style often does not matter.Organizations should spend more resources ondesigning better organizations and ensuringthe quality of information than on the amountand frequency of information in decision mak-ing.

APPENDIX

Table Al Preference function for proactive agent

Top managerrepeat the following actions until time expires or thedecision process is stopped:if remaining time is enough for another round ofdecision,check decision buffers for passed up decisions:

if there are decisions passed up, read decisionsfrom decision buffers, make a decision,

if the current decision is better according tocertain standards,update the organizational decision;else no update, ask for more information;

else ask for information, or wait if already asked;else if there is a previously made decision,

pass up (or conclude) the decision, and stop theorganizational decisionprocess;

else make a random decision, pass up (or conclude)the decision, and stop the organizational decisionprocess.

Middle managerif there is no request from upper manager for decision,repeat the following actions;

if remaining time is enough for making anotherround of decision.check decision buffers for passed up decisions;

if there are decisions passed up, read decisionsfrom decision buffers, make adecision, and pass up the decision;else ask for information, or wait if already asked;

else if there is a previously made decision,interrupt the current action, pass up the decision,and continue the current action;

else continue the current action;else If there is a previously made decision,

interrupt the current action, pass up the last madedecision, continue the interrupted action;else continue the current action.

Baseline analystif there is no request from upper manager for decision,repeat the following actions;

if remaining time is enough for making anotherround of decision,read radar equipment, make a decision, and passup the decision;else if there is a previously made decision, interruptthe current action, pass up the last made decision,continue the interrupted action;

else continue current action;else if there is a previously made decision, interruptthe current action, pass up the last made decision,

and continue the interrupted action;else continue the current action.

Table A2 Preference function for reactive agents

Top managerrepeat the following actions until time expires or thedecision process is stopped:if remaining time is enough for making another roundof decision,check the decision buffers tor passed up decisions:

if there are decisions passed up, read decisionsfrom decision buffers, make a decision, pans upthe decision, and stop the organizational decisionprocess;else if the action of asking for information was nottaken before,

ask for information;else wait;

else make a random decision, pass up (or conclude)the decision, and stop the organizational decisionprocess.

Middle managerif there is no request from upper manager for decision,repeat the action: wait;


else if there is a previously made decision,pass up the decision;else no decision is passed up:if remaining time is enough for making anotherround of decision,check the decision buffers for passed up decisions;

If there are decisions passed up, interrupt thecurrent action, read decisions from decisionbuffers,make a decision, pass up the decision, andcontinue the interrupted action;else if the action of asking for information wasnot taken before,

ask for information, and continue the interruptedaction;

else continue the interrupted action.

Baseline analystif there is no request from upper manager for decision,repeat the action: read radar equipment:else if there is a previously made decision,

pass up the last made decision;else no decision is passed up;if remaining time is enough for making anotherround of decision,read radar equipment (if not finished yet), make adecision, pass up the decision, and continue theinterrupted action;else continue the interrupted action.

Acknowledgements

This work was supported, in part, by the Officeof Naval Research, United States Navy, undergrant number N00014-90-J-1664.

References

Aldrich, H.E., Organizations and Environment, Pren-tice-HaU, Englewood Cliffs, NJ, 1979.

Allison, G.T., Essence of Decision: Explaining theCuban Missile Crisis, Scott, Foresman, Glenview,1L,1971.

Anderson, C.R., 'Locus of control coping behavior,and performance in a stress setting: a longitudinalstudy'. Journal of Applied Psychology, 62, No. 4,1977, 446-451.

Argote, L. and Epple, D., 'Learning curves inmanufacturing'. Science, 247, 1990, 920-924.

Burton, R.M. and Obel, B,, Designing Efficient Organi-zations: Modeling and Experimentation, ElsevierScience, New York, 1984.

Burton, R.M. and Obel, B., 'Devising expert systemsin organization theory: the organizational consult-ant', in M. Masuch (ed.). Organization, Manage-ment, and Expert Systems, Waller De Gniyter,Berlin, 1990.

Cameron, K.S., 'Effectiveness as paradox: consensusand conflict in conceptions of organizational effec-tiveness'. Management Science, 32, No. 5, 1986,539-553.

Carley, K., 'Designing organizational structures tocope with communication breakdowns: a simul-ation model'. Industrial Crisis Quarterly, Vol. 5,1991, 19-57.

Carley, K., 'Organizational learning and personnelturnover'. Organization Science, 3, No. 1, 1992,2-46.

Carley, K. and Lin, Z., 'Maydays and murphies: astudy of the effect of organizational design, task,and stress on organizational performance', paperpresented at the 87th Annual Meeting of theAmerican Sociological Association, Pittsburgh,Pennsylvania, 1992.

Child, J., 'What determines organizational perform-ance? The universal vs. the it-all-depends'. Organi-zational Dynamics, 3, No. 1, 1974, 2-18.

Cohen, M.D., March, J.B. and Olsen, J.P., 'A garbagecan model of organizational choice'. AdministrativeScience Quarterly, 17, No. 1, 1972, 1-25.

Das, H., 'Time: a missing variable in organizationalresearch', ASCI Journal of Management, 16, No. 1,1986, 60-75.

Drazin, R. and Van de Ven, A.H., 'Alternative formsof fit in contingency theory'. Administrative ScienceQuarterly, 30, 1985, 514-539.

Fararo, T., The Meaning of General Theoretical Soci-ology, Cambridge University Press, New York,1989.

Feldman, M.S. and March, J.G., 'Information inorganizations as signal and symbol'. AdministrativeScience Quarterly, 26, 1981, 171-186.

Hammond, K.R., 'Negative effects of outcome-feed-back in multiple-cue probability learning'. Organi-zational Behavior and Human Resources, 9, 1973,30-34.

Houskisson, R.E. and Galbraith, C.S., "The effect ofquantum versus incremental M-form reorganiza-tion on performance: a time series exploration ofintervention dynamics'. Journal of Management,11, 1985, 55-70.

Jauch, L.R. and Kraft, K.L., 'Strategic managementof uncertainty'. Academy of Management Review,11, No. 4, 1986, 777-790.

Kets de Vries, M, F, and Miller, D., 'Personality,culture, and organization'. Academy of ManagementReview, 11, No. 2, 1986, 266-279.

Krackhardt, D., and Stem, R.N,, 'Informal networksand organizational crises: an experimental simula-tion'. Social Psychology Quarterly, 51, No. 2, 1988,123-140.

LaPorte, T.R. and Consolini, P.M., 'Theoretical andoperational challenges of "high-reliability organi-zations": air traffic control and aircraft carriers'.Working paper, 1988.

LaPorte, T.R. and Consolini, P.M., 'Working inpractice but not in theory: theoretical challengers of"high-reliability organizations'", journal of Public


Administrative Research and Theory, 1, No. 1, 1991,19-^7.

Larson, L.L., Bussom, R.S., Vicars, W. and Jauch, L.,'PROACTIVE versus REACTIVE manager: is thedichotomy realistic?' journal of Management Studies,23, No. 4, 1986, 38S-400.

Lawrence, P.R. and Lorsch, J.W., Organization andEnvironment: Managing Differentiation and Inte-gration, Graduate School of Business Adminis-tration, Harvard University, Boston, 1967.

Lin, Z. and Cariey, K., 'Organizational performanceresponding to crises—theory and practice'. Papersubmitted, 1993.

Lincoln,].R., Hanada, M. and McBride, K., 'Organiza-tional structures in Japanese and U.S. manufactur-ing'. Administrative Science Quarterly, 31, 1986,338-364.

Long, R.J., 'Job attitude and organizational perform-ance under employee ownership'. Academy ofManagement Journal, 23, 1980, 726-727.

Lupton, T., '"Best fit" in the design of organizations',in E.J. Miller (ed.). Task and Organization, JohnWiley, New York, 1976.

Mackenzie, K.D., Organizational Structures, AHMPublishing Corporation, Arlington Heights, IL,1978.

Magazannik, V.D., 'Memory limits and the organiza-tion of memory', Voprosy Psikhohgi, No. 1, 1982,104-111.

March, J.G. and Olsen, J.P., Ambiguity and Choice inOrganizations, Universitetsforlaget, Bergen, 1976.

Mayhew, B., 'Structralism vs. individualism: Part 1,Shadowboxing in the dark'. Social Forces, 59,No. 3, 1980, 335-375.

Means, B., SaJas, E., Crandall, B. and Jacobs, T.O.,Training decision makers for the real world', inG. Klein, J. Orasanu and R. Calderwood (eds).Decision Making in Action: Models and Methods,Ablex Publishing, Norwood, NJ, 1992.

Mintzberg, H., Structures in Five: Designing EffectiveOrganizations, Prentice Hall, Englewood Cliffs, NJ,1983.

Newman, L.N., 'Exxon's lesson for other managers'.Business and Economic Review, 35, No. 4, 1989,8-10.

Ostrom, T.M., 'Computer simulation: the thirdsymbol system'. Journal of Experimental SocialPsychology, 24, 1988, 381-392.

Padgett, J.F., 'Managing garbage can hierarchies'.Administrative Science Quarterly, 25, No. 4, 1980,583-604.

Pauchant, T.C., Mitroff, I.I., Weldon, D.N. andVentolo, G.F., 'The ever-expanding scope of indus-trial crises: a systematic study of the Hinsdateteiecommunications outage'. Industrial Crisis Quar-terly, 4, 1990, 243-261.

Perrow, C, Normal Accidents: Living with High RiskTechnologies, Basic Books, New York, 1984.

Pfeffer, J. and Salancik, G.R., The External Control ofOrganizations. Harper and Row, New York, 1978.

Rice, C.A., 'Observations on the unexpected andsimultaneous termination of leader and group'.Croup, 1, No. 2, 1977, 100-117.

Roberts, K., 'New challenges to organizationalresearch: high reliability organizations'. IndustrialCrisis Quarterly, 3, No. 3, 1989, 111-125.

Roberts, K., 'Some characteristics of one type of highreliability organizations'. Organization Science. 1,No. 2, 1990, 160-176.

Rochlin, G.L, 'Iran Air Flight 655 and the USSVincennes: complex, large-scale military systemsand the failure of control'. In T.R. LaPorte (ed.).Social Responses to Large Technical Systems: Controlor Anticipation, Kluwer Academic Publishers,Amsterdam, 1991.

Rosenthal, U., Charles, M.T. and Hart, P.T. (eds).Coping with Crises: the Management of Disasters,Riots and Terrorism, Charles C. Thomas, Springfield,IL, 1989.

Rothstein, H.G., 'The effects of lime pressure onjudgment in multiple cue probability learning'.Organizational behavior and Human Decision Pro-cesses, 37, No. 1, 1986, 83-92.

Scott, W.R., Organizations: Rational, Natural, andOpen Systems, Prentice-Hall, Englewood Cliffs, NJ,1987.

Shrivastava, P.B., Anatomy of a Crisis, Balliager,Cambridge, MA, 1987.

ShuU, F.A. Jr, Delbecq, S.L. and Cummings, L.L.,Organizational Decision Making, McGraw-Hill, NewYork, 1970.

Silver, L., Fallout from Chernobyl, Deneau Publishers,1987.

Smiar, N.P., 'Cool heads: crisis management foradministrators'. Child Welfare. 71, No. 2, 1982,147-156.

Staw, B.M., Sandelands, L.H. and Dutton, J.E.,'Threat-rigidity effects in organizational behavior:a multilevel analysis'. Administrative Science Quar-terly, 26, 1981, 501-524.

Sussman, L., 'Grounding executives' future shock'.Training and Development Journal, 38, No. 6, 1984,24-27.

Thompson, J.D., Organizations in Action, McGraw-Hill, New York, 1967.

Wellman, B., 'Structural analysis: from method andmetaphor to theory and substance', in B. Wellmanand S. Berkowitz (eds). Social Structures: A NetworkApproach, Cambridge University Press, New York,1988.

Woodward, J., Industrial Organization: Theory andPractice, Oxford University Press, London, 1965.


proactive or reactive: an analysis of the effect of agent …zlin/isafm(1993).pdf · 2006-11-16 ·...

Documents