From: AAAI Technical Report WS-93-03. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved.
I)()ES
AGENT STYI,E MATTER
()R(~ANIZATI()NAL
DECISION MAKING
I
PERF()RMANCE?
Zhiang Lin
H. John HeinzIII Schoolof Public Policy ,and
Mimagement
(412)268-8892
zh I h@andrew.cmu.edu
KathleenC~u’ley
l)cpartmcnt ofSocild and DecisionSciences
(412)268-3225
carley@cenlro.solu’.cs.cm
u .edu
C~u’negieMellonUniversity
Pittsburgh, Pit 15213
April 9, 1993
INTR()DUCTi()N
Organizations vltry in the way they respond to
orglmizational problems, even whenthey are in similar
task environmentsor are facing simihu stress conditions
(Scott 1987). This way of response is due, in part,
agent style (Kets de Vries and Miller 1986; LaPorle and
Consolini1991), whichis also related to the normsin, or
culture of, thc organization (Parsons 1956; Blumberg
1987)2. By agent style: we mean the general way in
which the agent approaches the problem. However, the
link betweenorganizatiomdculture (e.g., as seen in agent
style) and perfonnancehits received insufficient attention
(Saffold 1988). This is due, in p~u’t, to the complexand
somclimescontroversial natures of culture, agent style,
lind pcrll~nn~utcc.
Agent style varies on manydimensions. Weare
here concernedwith pn~activeness/reactiveness,as there are
strong links between this aspect of agent style and
perform;race. Wcdefine an agent to be proactive if the
agent engages in decision making and information
gathering wheneverpossible. In contrast, a reactive agent
is one that waits until being asked or until absolutely
necessary to gather information or make a decision
(L,’trson el al. 1986). The link betweenagent style and
perfonnance,however,hits receivedinsufficient attention.
The conventional evaluation of agent style
typically regards proactive style as being better for the
organization than reactive style. Organizationscomposed
of proactive agents are supposed to respond to
organizational problems faster than organizations
composed of reactive agents. A consequence is that
organizations of proactive agents should outperform
org~mizationsof reactive agents (Pauchantet al. 1990).
The wisdomis that proactive organizations are more
IThis workwassupported,in part, by the Office of Naval
Research, United States Navy,under grant numberN000149t)-J- 1664.
2Agentstyle mayalso he due to personality (Ban-on1982;
Kets de Vries and Miller 1986), although this point is
controversial.
53
active (Smiar 1992), morecooperative (Rice 1977),
prepared (Das 1986; Newman1989), and thus ,arc
supposedto be better performers (Jauch and Kraft 1986;
Smiar 1992). This argument c,’m be summarized by
famous Chinese proverb: One can always get credit for
sweatif not for achievement.
Given that we are interested in proactive versus
reactive behavior, another factor of paramount
considerationis timing. This is, for manytasks, it is not
important that the organization respond accurately, but
,also that it makesthe decision in a timely fashion. One
expects proactiveness to be particularly important when
time matters. Moreover, numerous studies show that
undertime pressure, org~mizationstend to rearrange their
decision makingprocess (Rothstein 1986; Meanset ~d.
1992). In manycases, time pressure causes errors due to
loss of informationbecauseit stresses the limit of hutmm
cognition (Magazannik 1982). Whentime is short
pn~active agent mayhave ~madvantage because he or she
is prepared(Pauch~mt
et al. 19911).In contrast, the reactive
agent, though less prepar~ whentime is short, maybe a
more economical solution when time is less critic,’d,
because they mayneed less trltining, and mayhave lower
information processing costs (Sussman1984; Pauchant el
¯ ,d. 1990).
The ,argumentsthat agent style mauersarc htrgely
drawnfrom micro-level studies of org~mizatiomdbehavior
and case studies of org~mizations fitcing crises. While
proactive agents are often lauled as effecting high
organizational performance, the relative benefit to the
org~mizalion
of havingproactive versus reactive agents hit,;
not been systematically studied. Detailed case studies of
orglmizationsunder stress point to agent style as well its
many other factors (such as the structure of the
org~mization~mdthe nature of the task environment)when
delineating
the determinants of org:mizational
perfortmmce.Theoreticld and empiricld studies have shown
thai organization’,d structure (Mackenzie1978; Lin and
Carley 1992), task decomposition scheme (Thompson
1967; Lin and Carley 1992), training (Hammond
1973;
Carley 1992), task environment (Dri~in and Van de Ven
1985), and stress (Slaw et al. 1981; Lin and Carley 1992)
all ,affect perform,’mce.A study of the impact of agent
style onperform,’mce
shouldcontrolfi~r the other fitctors.
Manystudies of organizations, however, suggest
thai agent style is simply irrelevant. Networkstudies
(e.g.,
Mayhew 1980; Welhnan 1988) argue that
performanceis a function of the structure of rehttions
connecting agents in the organization. Contingency
theorists (e.g., Woodward
1965; Burlon and Obel 1984)
argue that performanceis a function of the fit between
organizational structure and task environment. Both the
network perspective and the contingency perspective
implicitly suggest that agent style will not detennine
org~mizatiomdperformancewhenthese other fitctors are
controlled.
In this paper, we will examine, from a mezoperspective,
how agent style (proactiveness
or
reactiveness) affects organizational performance.Weare
particularly concernedwith the effect of time pressure on
the performance of organizations composedof either
proactive or reactive agents. Many studies of
org;mizaliona]designdo not v~u-ythe agentslyle (Cohen,
March,andOlsen1972;P;tdget 1980;Carley]992; ]..,in
=tndC’;u’lcy 1992).Thisstudygoesbeyond
thosestudiesof
organizationaldesignas it not only considersboth timing
ol the decisionandthe proactiveness/reactiveness
style of
the agcnls, hut also controls for factors such as
organizational structure, task decompositionscheme,
training, natureof the task environment,
andstress. This
sludyis c;u’ried out usingsimulation.
Wcexamineagent style using a dynamic and
imcracliv¢ computationalmodelthat integrates a set of
I;tclors Iha! mayinfluence organizationalperformance.
There;u’c severedre;t,~onsto examine
the effect of agent
style usingsimulation.Firsl, in the real worldthereis
lilllc consensuson whal constitutes organizational
pcrfortnancc and whetheran organization is said to
pcrfornt well depends
on "the purposesandconstraints"
pl;tccd on the organizational performancemeasure
(Cameron
1986). Second,it is virtually impossible
ohlain sufficienl data for ¢omp;uing
organizationswith a
r=mge
of designsunderbothnorm,’dandstressful situations
due to either time constraints or confidentiality. In
conlrasl, simulation has multiple advantages, in
particular, (]) wecan control for various factors and
conducl systematic examination; (2) Simulated
org;mi~ationshavebeenshownto resemblethe real world
organi/:ttions in an ide;diz~ way(Lin and Carley1993);
(3) Wec;machieveour go;dof rese;u’checonomic;dly,
and
quickly.
To guide our analysis, wewill considerhowthe
results of our modelfit with, contradict, or elaborateon
propositions forwardedin the literature with respect to
agcn! slylc. Wewill briefly describe the model, analyze
the results from the model, and finally, we will discuss
the results ,’rod concludethe paper.
MODEL DESCRIPTION
Task
The organizations we simulate ,are faced with a
limited choice utsk in whichorganizations makedecision
choices regardingthe state of moving,aircraft understress
from limited ,’alternatives accordingto information they
3.
have thnmghorganization;d communicationprocesses
Choice tasks are very commonin the real world. Such
choice situations include law-making, price-setting,
planning, and a host of other similar things.
Wehave operationalized this choice task as a
stylized radar Utsk. There is a single aircraft in the
airsp.’tce. This aircraft is moving.The organization has
three choices ~ deciding whether the aircraft in the
airspace is friendly, neutral, or hostile. Eachaircraft is
chantcteriz~ by 9 parameters(see Lin and Carley 1992for
debtils).
Wemeasurethe organization’s performanceas the
percentage of these problemsthat the organization hits
madethe correct decision4 regardingwhetherthe aircraft is
friendly, neutral, or hostile. Wetreat type i error (making
the decision that the ,aircraft is hostile, but in fact the
aircraft is friendly) and type II error (makingthe decision
that the aircraft is friendly, but in fact the aircraft is
hostile) with equal weight. In org:mizational literature,
organization,’d performance has also been measured by
organizational effectiveness (Mackenzie1978; Pfeffer and
Salancik 1978), and organization;d efficiency (Mackenzie
1978;Scott 1987). In manycases, accuracy, effectiveness,
and efficiency arc indistinguishable and can convert into
each other.
Task Environment
Task environments vary on a large number of
dimensions. Twosuch dimensions that have received
someattention in the literature are decompo.~tbilityand
biasness. Task environmentdecomposabilityme,’~suresthe
interrelationships
among task components. A task
environment is decomposable if there are no complex
interactions mnongcomponent~
th~tt need to he underst~xl
in order to solve a problem, and non-decomposableif
otherwise. In a non-decompo,~tblc
lle;k enviromnentthere
are interaction effects. As LaPorle and Consolini (1991)
and Roberts (1990) note interdependence of task
components(i.e., non-decomposability)is It factor that
should be consideredin designingthe organiz,ation.
Task environment bi;tsness
measures the
distribution of all l~ssible outcomes.A task environment
is unbiasedif ,’dl the IX)ssible outcomesare equallylikely
to occur, and biased if not. As Aldrich (1979) notes,
biased environmentis one in whichthe problemsfaced by
the organization are concentrated ~mdso highly similar.
Based on these two mzmipulations, we examine four
different "r~dities" or environmenud
situations. Theyare:
(1) biased decomposableutsk environment; (2) unbiased
decomposable task environment; (3) biased nondecomposabletask environment, and (4) unbiased nondecomposabletask environment.
Stress
Organizationsare often affected by stress (Perrow
1984; Shrivastava 1987). In ex;unining organizational
performanceit is imlx~nantto consider multiple sources
of stress: external hostile conditions or maydays(caused
by hostile task environment), internal suboptimal
operating conditions or murphies(caused by sub-optimal
operating condition w.ithin organizational design), and
time pressure.
Weexamine five types of internal stress or
murphies (Lin and Carley 1992). They are: (l) missing
information -- a piece of the incominginfonnation for a
particular problem is not available, (2) incorrect
information -- a piece incoming information is
erroneous, (3) agent unavailability -- an analyst is not
avaihtble to help the organization solve the problemand
so does not report his or her decision to his or
her manager, (4) communicationchannel breakdown
an analyst is unable to report to a superior because the
communicationchannel is unavaihtble, and (5) agent
turnover -- an analyst leaves the organization and is
replaced by a new analyst. For each type of murphy,the
numberof simulumeousmurphies ranges from 0 to 3.
3The simulation test-bed used in this paper is an adaptation
of thai in Lin andCarley(1992)that takes into accounttime
and agent style.
4The true states of all the aircraft are predefined by the task
environment and are unknown to the organization.
54
In Ihi~, paper,timepressure
occurswhenthe aircraft
is vt.ry l asl, ¢)r whenthe lime required to makethe
decision is very short. Timepressure is the numberof
time units before the organization must makea decision.
Twof,’ictors affect whenthe organization must makea
decision: (it a decision is demandedby someoutside or
superior sources (e.g., congress) and (2) a decision
dem~mded
becausethe aircndt reaches the d~mgerpoint (the
point at whichif the organizationdoes not respondand the
aircraft is hostile, the aircraft can destroy the
organization)5. in this modelfor each problem(i.e., an
aircraft appearing,in the airspace), werandomlyassign the
numberof time units prior to a demandfor a decision to
be betweenI and 60. Thecharacteristics of the aircraft ,are
also randomly chosen for each problem6. Consequently,
the lime pressure vltries randomlyacross all problems.We
study three levels of time pressure: low (41-60 time
units), medium(21-40 time units), and high (1-20
units). Becausethis is a dyn,’unic environment,agents’
interaction with one another in terms of whether to
communicatelind howIt communicate,and the utilization
of decision makingprocedures in lerms of which decision
makingpn)cedureto ch(~)se, all dependon the consmtined
rcsottrce -- time. Howagents react It time pressure is
discus.~dlater.
()rganizational Design
in this p,’tper, organizationaldesignis viewedas a
combination of organizational
structure,
task
decompositionscheme,trltining, and agent style. Through
an examin,’ttion of inulliple designs, expected relations
between design and perflmn~mcecan be compuultionally
tkxluctxl.
Org:mizalion~d
structures are defined in terms of the
networkof relations lunongagents. Westudy h)ur stylized
organizational structures. Theyare: (1) team with voting
-- a toudly deccntrldizedstructure in whichorganizational
decision is through majority voting of each membersof
the organization, (2) teaun with a manager-- basically
flal hierarchy such that while each analyst examines
information and makes a recommendation,the ultimate
organizational decision is madeby the manager(or telun
leader). (3) hierarchy -- ,’t multi-leveled communication
structure in which each baseline agent examines
inl’onnation and makes a recommen&ttionto his or her
immediate supervisor
who in turn makes a
recommendationIt the top-level manager whomakes the
uhimateorganizational decision. (4) matrix -- like the
hierarchy, is it multi-leveled communicationstructure,
except that each baseline agent has two communication
links with Iwouniddle m~magers
across divisions.
The task decomposition scheme defines the
rel,’ttions betweenagents and resourcesand/or information.
Westudy hmrstylized utsk decompositionschemes. They
iu’c: ( 1 ) scgregat~-- eachbaselineagenth~usaccess It one
task component,(3) overlapped~ each baseline agent has
access to two task components,with one task component
being overlappedwith another baseline agent, (3) blocked
-- each baseline agent has access three task components,
but usually each three agents within the samedivision
havethe samethree task components,,and (4) distributed
each baseline agent hits access to three task
components
usually across different divisions.
Thethird and finld aspect of organizational design.
with which we will be concerned, is trltining. Westudy
three training scenarios. They are: (I) untrained ~ each
agent in the organization makes decisions by basically
guessing, (2) experientially trained -- each agent in the
organizational m’,tkesdecisions by referring to historicld
experience, and (3) operationally trained -- each agent
the organization makes decisions by using standard
operating procedure7.
Agent Style
Withina proactive organization, each agent asks for
information, reads information if there is information,
makesa decision based on the available infi,rmation, then
passes on the decision. This process repeats until time
expires, i.e., the aircraft goes out of range, or the top
nmmagerdecides It stop the process. Each agent (except
the lop manager)can be interrupted by a request from an
uppermanagerfor decision. The agent will respondto the
request by communicatinga decision, if there is one
available. Clearly, the procedures fol.lowed by the top
manager, middle managers, and baseline analysts differ
somewhaton the basis of their org~mizational position.
The top managercannot be interrupted (since there is no
superior), and a baseline analyst cannot ask for
information(since there is no suN3rdinate),while a middlc
tnanagercan be interrupted as well It,~ ask for information.
Further, the top managerhas the powerto decide the linal
or organizationlddecisionaccordingto certain stan&trtts.
Withina reactive orglmization, each agent simply
reads information again and again until time expires or
until the agent is told by the mlmagerto stop. Eachagent
(except the top manager)can be interrupted by a request
from an upper managerfor decision. The agent will then
stop reading information (if not finished yet), make
decision, and pass on the decision. Unlike the proactive
agent, a reactive agent will not makea decision and pass
on the decision unless being requested by his or her
manager. Again, there are processing differences among
agents due to their orglmization~dpositions (top. middle,
and baseline). The top managerclmnotbe interrupted, and
initiates the decision making process by asking for
information. A middle managerwill ask for infonnation
only whena request from thc lop tnanager is received h)r
decision and when there is no decision already being
comununicatedby baseline analysis. A baseline analysl
7Numberof time units required h~r different types of
decision procedureis different given the sameamountof
information. Dueto the different complexityof decision
makingprocesses, experiential decision makingprocedure
usually takes moretime than operational decision making
procedure, and they both take more time than untrained
decision makingprocedure.
51nthis model,this dangerpoint occurswhenthe aircraft
has a range of l mile and/or a range of 5,000 feet. Both
numberswere chosen based on characteristics of radar
systems.
6Aircratt can vary with different characteristics such as
speed,range,andaltitude.
55
alwaysre;ids inl0rm:ttion and never asks for information
(please refer Io Appendixii)r moredetailed algorithms
i~ro;iclivc midreactiveagentstyles).
METH()I)()LI)I;Y
The performanceof the org~mizations is generated
using, compulersimulation lest-bed built in C within a
VAX/VMS
system. Using this test-bed, we are able to
syslcmalically aller task environment(4), org~mizational
structure (4), utsk decompositionscheme(4), training
scenm-io(3), agent style (2) whichare all built within
lest-bed. Thus, 384 organizational types were examined.
By varying typc (5) and degree (4) of internal stress,
exmninc 20 inlcrnal condilions, with degree 0 as the
~q)tim;ll operating condition. In this experiment, we
cxamiuc
a I()l;tl ill’ 7,680
clt,~es.
Eachca.,~e represents an organization faced with a
spccilic inlcrnal condilion lind task environment. The
performance of each casc/organizalion is evaluated by
exantining performanceon I,(XX) problems. Eachproblem
represents a specific movingaircntfl. This study is carried
using Monle-Carh)analysis. All characteristics of each
problem are randomly generated. These include: the
sl;trling posilion, speed, and nature of the aircraft, the
location of Ihc murphies affecting the organization
internal, and the maximumtime units allowed the
orgmli/alionIo In;lke a decisionrcgm’dingthe aircraft.
The 1,0()1) problems can he classified as being
friendly, neutral, or hostile or as being under low,
medium, or high lime pressure.
We can measure
pcrli)rtnmtcc ovcntll (percentageof problemsfor whichthe
t)rg:mi/~tlion madethe correct decisions) or wecan mea,sure
performance for a subset of the 1,000 problems (e.g.,
pcrcenlageel pn)blemsgiven the true suite is hostile for
whichIhc organizalion madethe correct decisions). This
simulation lest-bed offers us the chanceto systematically
examinethe inlemctions mnong,and effect of, the various
organizational I~tclors on organizational perfonnancethat
have concernedresearchers. Nevertheless, in this paper,
our focus is on agent style.
TableI: I)erfonn,-mce
by AgentStyle acrossall Levelsof
TimePressure
Acn,ssExperienti;dl~¢Trained()r~anizations
~xlcrn;dInternal
AszentStyle
Proactive
Reactive
Overall I 46.11(3840,0.29)I 46.08(3840,0.29)
Overall ()primal I 47.28(960,0.57) ] 47.32(960,0.57)
Murphy[ 45.72(2730,0.31) 1 45.66(2730,0.31
Overall [ 59147(384010132)
1 59.39(38J1010.32)
Mayd.’ty
()ptimal[
61.30(960.0.90) I 58.74(2730,0.49)
61.35(960.0.89)
Murphy,58.85/2730,0.49)
I
Across()perationall~,Trained()rLzanizations
Internal
At, ent Style
Proaclive
Reactive
Overall [ 47.86(3840,0.24) 47.81(3840,0.24)
Overall Optimal 49.99(960,0.53)
50.06(960,0.53)
Murphy47.16(2730,0.24) 47~06(27:30,0.24)
()verall [ 54.83(3840,0.33) 54.70(3840,0.33)
Mayday ()ptimal 56.71(96(I,0.70) 56.68(960,0.69)
Murphy54.20(~2730,0.36~) 54,04~2730,0.36~
Note: l)erformance is in percentage. Numberof cases and
standarderrors are in parenlheses.
56
RESULT AND ANALYSIS
AcrossAll Levelsof Time Pressure
Accordingto the general discussions of the nature
of proactive and reactive agent styles in the literature, one
might expect the folh)wingpropositions to hold:
Propositign i; Agent style does not matter
(Mayhew1980; Wellman1988).
This says that when the structure
of the
organization and the fit between task environment and
organizational structure are considered, organizations
compose exclusively of proactive agents (proactivc
orglmizadons) and organizations composeexclusively of
reactive agents (reactive organizalions) should exhibit
equivalent perlormance.
Proposi00n 2; Proactive organizations perform
better than reactive organizations (Kraft 1986; Smiaz
1992).
Proactive agents are better prepm’edand c;mres[xmd
faster. Thus, organizations filled with proactive agents
should perform better than organizations filled with
reactive agents, controllingfor all other htctors.
Our ;malysis supports Proposilion I. Whetherwe
consider just experientially trained organization, or
operalionally trained organizations, or both, there is no
difference betweenthe perh)nnanceof organizations with
proactiveand reactive agent styles (oc=0.()1). This is
regardless of extermdor internal conditions(Table 1).
Similarly,for different organizationalstructures, for
different task decompositionschemes,;rod different task
environments,there is no difference in the pcrhmn;mce
of
proactive,and reactive organizations.
Under High Time Pressure
Underhigh time pressure, organizations are often
overloaded with information. They give up normal "time
spending procedures", or normal decision making
procedures and mainly makedecisions based on hunches
(Rosenthai el ~d. 1989). Thus, agent style should not
matter when time is at a premium,controlling for all
other hictors. This suggests the followingproposilion.
Pro r~sition 3: Proactive;rod reactive organizations
have the equivalent perform;race under high time pressure
(Rosenthalet al. 1989).
In our analysis, whenthere is high time pressure
on avenige there is no difference betweenthe performance
of proactive ;rod reactive organizations tot = 0.01),
regardless of external conditions. However,if we consider
only optimal internal conditions (no murphies),
organizations filled with proactive agents perfonn worse
than org~mizationsfilled with reactive agents. This is true
across all exlernal conditions (p < ().(X)I), and when
hostile (may&ty)conditions are considered (p < 0.005)
(Table2).
It is important to note, however, that all
organizations under high time pressure are makingthe
right decision approximately33.33~ of the time. If the
agents were simply guessing, then organizational
performancewouldin facl be 33.33~.. Whilethe proactive
organization is slightly worse than the reactive
organization neither is noticeably different than lit)
organizationthat is simplyguessing. Theslight difference
in performancemaywell be the artifact of randomness.
in summary,under extremely high time pressure,
organizational performanceis essentially reduced to be
equivalent to the best that can be expected whenagents
guess. Consequently,there are no consistent effects that
can be attributed to agent style. This generates mixed
support for proposifon 3.
In summary,under mediumtime pressure, agent
style affects organizational performance only under
suboptimal internal operating conditions. In this ease,
proactive organizations outperformreactive organizations.
Thus, Proposition 4 is supported only under suboptimal
conditions, Proactive agents benefit the organization, but
that benefit is most apparent when time is somewhat
constrained and things are going wrong.
Under Medium Time Pressure
Undermoderate time pressure, the organizational
decision makingprocess should be tightly constrained by
time, though not completely disrupted. Thus agent style
should critically affect performance,controlling for all
other factors.
Proposition 4: Proactive organizations perform
better than reactive organizations under mediumtime
pressure (Jauch and Kraft 1986; Pauchant el al. 1990;
Smi;tr 1992):
Our an,’dysis suggests that on average there is no
difference between the performance of proactive and
reactive organizations(cx = 0.01). This is true regardless
of external or internal conditions (Table 2). However,
whenwe consider the type of training that the agents have
received, we~e significant performancedifferences due to
agent style under suboptimal operating conditions
(murphies).
In experientially trained organizations, if we
averageacross ,all operating conditions, or consider just
optimal internal operating conditions, there is no
difference between the performance of proactive and
reactive organizations (oc = 0.01), regardless of external
conditions. Under suboptimal internal operating
conditions, organizationsfilled with experienti,’dly trained
proactive agents outperform organizations filled with
experientially trained reactive agents. This is true both
across all extern:d conditions (p < 0.001), and under
maydays (p < 0.001). A similar pattern occurs for
operation;dlytrained organizations.
Under Low Time Pressure
Underlow or no rime pressure, organizations have
enoughtime to makedecisions regardless of external or
internal conditions. It is usually efficient to use reactive
style as it offers a comparableperformancewith proactive
style, but with lower operating costs (Sussman1984).
Thus, we should expect no difference in the performance
of organizations with either proactive or reactive style
whentime is not critical, controllingfor all other factors.
proposition 5: Proactive and reactive organizations
have the equivalent performanceunder low rime pressure
(Sussman1984).
In our analysis,
when considering either
experientially trained organizations, or operationally
trained organizations, or both, there is no difference
betweenthe performanceof organizations with proactive
andreactive agent styles tot = 0.01), Thisis true regardless
of external or internal conditions (Table 2). Thus, under
low or no time pressure, there is no effect of agent style
on organizational performance. This result supports
Proposition 5.
DISCUSSION
AND CONCLUSION
In this paper, we have studied the effect of one
aspect of agent style -- proactiveness/reactiveness -- on
organizational performance.
Our results showthat on average, there is virtually
no effect of agent style on organizational performance
(support for Proposition 1 but not for Proposition 2).
However,if we considerthe effect of time pressure we find
Table2: Performance
by AgentStyle and Levelof TimePressure
lnternal
Hieh
Medium
Low
Condition
TimePressure
TimePressure
TimePres~ur~
Agent Style
Agent Style
Agent Style
Proactive
Reactive
Proactive
Reactive
Proactive
Reactive
Overall(1280) 33.27(0.07) 33.24(0.07) 43.69(0.29) 43.54(0.29) 61.38(0.51) 60.5..)..(9...5..S)
61.45(0.51)
Overall
Optimal(320) 33.27(0.13) 33.31(0.14) 44.38(0.60) 44.39(0.59) 64.18(1.02) ’64.27(1.02)
Expe
............................ M._urp_h.Y_!9.60.)..
33.27~-.~.~8)~.~.~3~3~2~2(~8~.~3~-~45(~:33.)-~-4~3.26(~:33)~.~6~A4(~59
)
Training
()verall(1280) 33.18(0.10) 33.25(0.10) 55.38(0.45) 55.14(0.43) 89.84(0.42) 89.79(0.42)
Mayday
Optimal(320) 33.27(0.21) 33.41(0.20) 56.39(0.93) 56.42(0,90) 94.25(0.65) 94.21(0.67)
Murph),(960) 33.15(0.12~ 33.19(0.12) 55.05(~0.51) 54.71(0.49) 88.37(0.51) 88.3,1(0.50)
Overall(1280) 33.67(0.08) 33.71(0.07) 53.71(0.29) 53.56(0.29) 56.21(0.35) 56.17(0.36)
Optimal(320) 33.77(0.16) 34.01(0.15) 56.44(0.71) 56.40(0,70) 59.77(0.86) 59.77(0.86)
Overall
()per
Murphy(960) 33.64.~0.08) ,,33.61(0.08
52.80~0.31)52.61(0..31), 55.02/0.37
54.9"1(0.37
)
)
)
Training
Overall(1280) 33.69(0.1!) 33.57(0.11) 64.09(0.42) 63.88(0.42) 66.71(0.46) 66.64(0.46)
Mayday
Optimal(320) 33.77(0.22) 33.85(0.21) 66.96(0.91) 66.79(0.89) 69.40(1.01) 69.40(1.01)
Murphy/960
33.66/0.12
33.4810.12}, 63.13(047~ 62.9210,47
65.8110.51
65.7210.52
,
~
)
)
~
~
Note:Performance
ss in percentage.Number
of cases for each roware listed in the columnInternal Condition.Standarderrors ate in
parentheses. Expe~ Experiential. Oper~ Operational.
Traimng External
Type
Condition
57
that in general, only under medium
time pressure, does
agent.~tylt" play a large role. Undermedium
timepressure,
organizations filled of proactive agents outperform
organizationsfilled with reactiveagents,but only if the
organizationis facing suboptimaloperatingconditionsin
whichv~u-iousobstacleslimit its informationprocessing
capabilities (erroneous information, incomplete
inl~)rmation,andso forth).
Furlher, there is basically no interaction effect
helweenagent style (proacliveness/reactiveness)and
slructure, utsk decompositionscheme,training scenario,
or utsk environment,tin perhirmance.
The resull that agent style in most situations is
irrclcvanl Io organizational perfl)rmance showsthai the
a.m~unt and frequency of information used in decision
makingmaynot characterize the qualily of decisions. The
myth around proaclive agent slyle can probably be
allrihulcd to the facl that organizations often collect
inlornmlion flit "signal and symbol"purpose rather than
li~J decision makingpurpose(Feidman,’rod March1981).
In this rese,arch, our focus is on the effect of one
aspect of agent style (proactiveness/reactiveness)
organizational performance, not on the origin of agent
slyle. Webelieve further understandingof the sources of
agent style is necessaryfor future research into the effect
of agent style. More aspects of agent style such as
ein(ition and creativity mayalso be included in future
restatrch on agent style. In addition, future rese~uchshould
examinemore real world organizations, thus providing
newinsight into this theoretical study.
Wccategorized time pressure into three levels
accordingto the time units assigned to the organizationor
the time units spent whenthe organization has to makea
decision before the aircraft reachesa certain point. It is
difficult to disentangle time pressure from organizational
design and task environment,because a high time pressure
Io someorganizationsmaynot be its stressful as to other
organi~’ations.
This analysis is mainly based on simulation
techniques. Thosesimulatedorganizations have enabled us
to ex~unine the effect of agent style across numerous
factors that have interested researchers in organizations
within a reasonably short period of time. In fact, to
examinethese s~une factors using humansubjects would
havetaken at leas! ten years andcost at le,’t~l twomillion
dollars. Computersimulation is a powerful extension of
human cognition. As pointed out by Ostrom (1988),
computer si,nulation offers a third symbol system in
studying social science, besides natural h’mguageand
mathematics, because "computer simulation offers a
substantial advantageto soci,al psychologists attempting
to develop formal theories of complexand interdependent
social phenomena". Fararo (1989) also regards
computationalprocess as one of the three processes (the
other two are theoretical and empirical processes)
neces~try to the developmentof any discipline. However,
we should still view the results from this paper with
c,’luti(in.
Toconclude, wefind suplx~rt for the implication of
networktheory ~mdcontingency theory that agent style
often does not matter. Organizations should care more
about the quality of infor*nation than about the amount
and frequencyof information in decision making.
REFERENCES
Aldrich, Howard E. (1979), Organizations and
Environment.EnglewtxxtCliffs, NJ: Prentice Hall.
Barron, Frank. (1982), "Personology focuses on the
Examinationof MentalLife, Including Creativity,"
Personality and Social PsychologyBulletin, 8(4):
611-614.
Blumberg, Rhoda Lois. (1987), Organizations in
ContemporarySociety. Prentice-Hall, Englewood
Cliffs, NewJersey.
Cameron, KimS. (1986), "Effectiveness as Paradox:
Consensus and Conflict in Conceptions of
Organizational Effectiveness,"
Management
Science. 32(5): 539-553.
Carley, K.M. (1992), "Organizational Learning and
PersonnelTurnover."Organi’ationS(’iem’e. 3( 1 ):2-46.
Cohen, M. D., March, J. B and J. P. Olscn. (1972),
Garbage Can Mt~lel of Organizational Choice,"
AdministrativeScienceQuarterly.! 7( 1 ): 1-25.
Das, Hart. (1986), "Time: A Missing Variable
organizational
Research," ASCI Journal of
Management,16(1): 60-75.
Drazin, Robert, Andrew H. Van de Ven. (1985),
"Alternative Formsof Fit in ContingencyTheory,"
Administrative Science Quarterly, 30:514-539.
Fararo, Thomas (1989), The Meaning of (;eneral
Theoretical Sociology. Cambridge University
Press.
Feldman, Martha S. and James G. March. (1981),
"Information in Organizations as Signal and
Symbol," Administrative Science Quarterly, 26,
171-186.
Hammond,Kenneth R. (1973), "Negative Effects
Outcome-Feedbackin Multiple-Cue Probability
Lelu’ning," Organizational Behavior and Human
Resources. 9, 30-34.
Jauch, Lawrence R. ,and Kenneth L. Kraft. (1986),
"Strategic managementof Uncerutinty," Academy
of management
Review, 11(4): 777-790.
Kels de Vries, Manfred F. and Danny Miller. (1986),
"Personality, Culture, and Organization," Academy
of Management
Review, ! !(2): 266-279.
LaPorte, Ttxld R., Paula M. Consolini. (1991), "Working
in Practice But Not in Theory: Theoretical
Challengers of ’High-Reliability Organizations’,"
Journal of Public Administrative Research and
Theory. 1(1991):1:19-47.
Larson, Lars L., Roberts S. Bussom,Willi,’un Vic~u’s, and
Lawrence Jauch. (1986), "PROACTIVE
versus
REACTIVE Manager: Is the Dichotomy
Realistic?" Journal of Management
Studies, 23(4):
385-400.
Lin, Zhiang ,and Kathleen C~urley. (1992), "Maydaysand
Murphies:A Study of the Effect of Organizational
Design, Task, and Stress on Organizational
Performance." Working Paper / Carnegie Mellon
University, Presented at the 1992 AnnualMeeting
of the AmericanSociological Association.
Lin, Zhiangand KathleenCarley. (! 993), "Organizational
58
PerformanceResponding
to Crises -- Theoryand
Praclicc",
Working Paper/ Carnegie Mellon
[/niversity,
Lincoh’~, J;unes R., MitsuyoHanada,and Kerry McBride.
(1986), "Organizational Structures in Japanese and
U.S. Manufacturing," Administrative Science
Quarterly, 31:338-364.
Mackenzie,KennethD. (! 978), OrganizationalStructures.
AHM
Publishing Corporation, Arlington Heights,
Illinois.
Magazannik, V. D. (1982), "MemoryLimits and the
Org~ufizalion of Memory,"VoprosyPsikhologii. I:
1(14-11
I.
Mayhew,
Bruce. (1980), "Structural[sin vs. Individu,’dism:
Pan I, Shadowboxing
in the Dark," Social Forces.
pp 335-375.
Me,ms,B~trhara, EduardoSalas, Beth Crandall, and T.
OwcnJacobs. (1992), "Training Decision Makers
for Ihc Real World," in G. Klein, J. Orasanu, and
R. (’alderwood(Eds.), Decision Makingin Action:
Models attd Methods. Norwood, N J: Ablex
Publishing Corp.
Newman,Lloyd N. (1989), "Exxon’s Lesson for Other
M~magers,"Business and EconomicReview, 35(4):
8-10.
Ostrom, Tholna.,~ M. (1988). "ComputerSimuhttion: The
Third Symbol System." Journal of Experimental
Sm’ial Psychology. 24: 381-392.
Padgett, John F. (1981)). "Managing Garbage
Hierarchies." Administrative Science Quarterly.
pp:583-604.
Parsons. Talcott. (1956), "Suggestionsfor a Sociologic,’d
Approach to the Theory of Organizations,"
Administrative Science QuarterlyI: 63-85.
Pauchanl, Thicrry C., hmI. Mitroff, DouglasN. Weldon,
and Gerald F. Ventolo. (199(I). "The Everexpanding Scope of Indus[rind Crises: a Systemic
Study of the Hinsdale Telecommunications
Outage,"Industrial Crisis Quarterly.4: 243-261.
Perrow, C. (1984), NormalAccidents: Living with High
Risk Technoh)gies.Basic Bt’~)ks, Inc.
Pfeffer, Jeffrey, and Gerald R. Sahmcik. (1978), The
L:rternal Control of Organizations. Harper and
Row, NewYork.
Rice, Cecil A. (1977), "Observations on the Unexpected
and Simultaneous Termination of Leader and
Group,"Group,i(2): 1110-117.
Roberts, K. (1990), "SomeCharaclerislics of One Type
High Reliability Organizations," Organization
Science.1 (2): 160-! 76.
Rosen[hal, U., Michael, T. Charles, and Paul ’T Hart
(eds.),
(1989), Coping with Crises: the
Management
of Disasters. Riots attd Terrorism.
Charles C Thom;Ls,Publisher, Springfield, Illinois,
U.S.A.
Rothslein, HowardG. (1986), "The Effects of Time
Pressure on Judgmentin Multiple CueProbability
Learning," Organizational Behavior and Human
Decision Processes. 37(1): 83-92.
Saffold, GuyS., !II. "Culture Traits, Strength, and
Organizational Performance: Moving Beyond
’Strong’ Culture," Academy of Management
Review, 13(4): 546-558.
Scott, W. Rich,’u’d. (1987), Organizations: Rational.
Natural, and OpenSystems. Prentice H,’dl, Inc.,
EnglewoodCliffs, NewJersey.
Shrivastava, Paul. (1987), Bhopal: Anatomyof a Crisis.
Cambridge,Mass. Ballinger Pub. Co.
Smiar, Nicholas P. (1992), "Cool Heads: Crisis
Managementfor Administrators," Child Welfare,
71(2): 147-156.
and Jane E. Dutton.
Staw, B,’u’ry M., LanceE. Sandehmds,
(1981), "Threat-Rigidity Effects in Organization~d
Behavior: A Multilevel An;dysis," Administrative
ScienceQuarterly., 26:5[)1-524.
Sussman, Lyle. (1984), "Grounding Executives’ Future
Shock," Training and DevelopmentJournal. 38(6):
24-27.
Thompson,James D. (1967), Organizations in Action.
NewYork: McGrow-Hill.
Welhnan, Barry. (1988). "Structural Analysis: From
Methodand Metaphorto Theory~mdSubstance," in
Welhnanand Berkowitz(Eds.) Social Structures: A
NetworkApproach,CambridgeUniversity Press.
Woodw~u’d,J. (1965), Industrial Organization:Theol7 and
Practice, OxfordUniversity Press, London.
APPENDIX
TablcA 1: PreferenceFunctionsfi)r ProactiveAgent
Note:
TIME_REM
~ Time remaning for decision making
TIME_REQ
-- Timerequired fi~r a round of decision
SIGN_PASS
~ Signal Ior passed up decision
SIGN_REQ
~ Signal for request of decision infi~nnation
SIGN_EDEC
~ Signal for earlier made decision
ASK_INFO
-- Ask h)r decision information
READ_INFO
-- Re~d decision infonnation
MAKE_DEC
-- Makea nonnal decision
MAKE_RAN
-- Make a r~mdomdecision
UPD_DEC
~ Update decision
PASS_DEC
~ Pass up decision
WAIT-- W~dt
INT_ACT
~ Interrupt current action
CON_ACT
~ Continue currenl action
RETURN
-- Return final organizatiomd decision
Top Manager
while (TIME_REM>0){
if (TIME_REM>TIME_REQ)
if (SIGN_PASS){
READ_INFO;
MAKE_DEC;
UPD_DEC;}
el~ {
ASK INFO;
WAIT;}
if (SIGN_EDEC){
UPD_DEC;
RETURN:}
eL~{
59
MAKE_RAN;
RETURN;}
READ_INFO;
MAKE_DEC;
PASS_DEC;
RETURN;]
eLsel
ASK_INFO;
WAIT;}
else[
MAKE_RAN:
RETURN;}
MidtfleMunat,er
while (TIME_REM_>f)){
if not (SIGN_REQ)
if (TIMEREM_>TIME_REQ)
if (SIGN_PASS)[
READ_iNFO:
MAKE_DEC;
PASS_DEC;)
else{
ASK_INFO;
WAIT;}
Middle Manaeer
while (TIME_REM_>0)[
if not (SIGN_REQ)
CON_ACTor WAIT;
else
if (SIGN_EDEC)
PASS_DEC;
WAIT;]
else
if (TIME_REM_>TIME_REQ)
if (SIGN_PASS)[
INT_ACT;
READ_INFO;
MAKE_DEC;
PASS_DEC;
CONACT; I
eLse{
INT_ACT;
ASK 1NFO;
CON_ACT;I
if (SIGN_EDEC){
INT_ACT;
PASS_DEC;
CON_ACT;}
else
CON_ACT;
if (SIGN_EDEC){
1NT_ACT;
PASS_DEC:
CONACT: ]
else
CON_ACT;
Baseline Analyst
while (TI ME_REM_>0){
if (SIGN_REQ)
if (TIME_REM>TIME_REQ)[
READ_INFO;
MAKE_DEC;
PASS_DEC; }
else
if (SIGN_EDEC)[
INT_ACT;
PASS_DEC;
CONACT; }
else
CON_ACT;
else
if (S1GN_EDEC)[
INT_ACT:
PASSDEC:
CON_ACT;}
else
CON_ACT;
BaselineA nal)’st
while (TIME_REM_>O){
if (SIGN_REQ)
if (SIGN_EDEC)
PASS_DEC;
if (TIME_REIvI~TIME_REQ)
I
READ_INFO;
MAKEDEC;
PASS_DEC; }
else
if (TI ME_REM_>TIME_REQ)
INT_ACT;
READ_INFO;
MAKE_DEC;
PASS_DEC;
CON ACT; ]
else
CON_ACT;
I
READ_INFO;
Table A2: Preference Functions for Reactive Agents
Too Manager:
while (TIMEREM_>0)I
if (TIME_REM>TIMEREQ)
if (SIGN PASS)[
60