Changing Patterns or Patterns of Change: The Effects of a Change in Technology on Social

Network Structure and Power

Author(s): Marlene E. Burkhardt and Daniel J. Brass
Source: Administrative Science Quarterly, Vol. 35, No. 1, Special Issue: Technology, Organizations
, and Innovation (Mar., 1990), pp. 104-127
Published by: Sage Publications, Inc. on behalf of the Johnson Graduate School of Management,
Cornell University
Stable URL: http://www.jstor.org/stable/2393552
Accessed: 18-08-2015 14:43 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/2393552?seq=1&cid=pdf-reference#references_tab_contents

You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Sage Publications, Inc. and Johnson Graduate School of Management, Cornell University are collaborating with JSTOR
to digitize, preserve and extend access to Administrative Science Quarterly.

http://www.jstor.org

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
ChangingPatternsor
Patterns-of Change:
The Effects of a Change
in Technology on Social
Network Structure
and Power
Marlene E. Burkhardt
Daniel J. Brass
The Pennsylvania State
University
?
1990 by CornellUniversity.
04/$1.00.
0001-8392/90/3501-01
0 arrangementsact as the conduits of technologicalchange
The authorsare indebtedto Mark
Sharfmanfor his helpfulsuggestions
duringthe developmentof this research.
We would also liketo thankDavidKrack-
hardt,MichaelTushman,and anonymous
ASQ reviewersfor theirinsightfulcom-
ments and suggestions. Thisstudywas
fundedby the Centerfor the Management
and Technological
of Organizational
Changeand The PennsylvaniaState Uni-
versityCenterfor Research.Copiescan
be obtainedfromthe authorsat Depart-
ment of Managementand Organization,
410 Beam B.A.B.,The PennsylvaniaState
University,UniversityPark,PA 16802.
This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
The effects of a change in technology on organizational
structure and power were investigated in a longitudinal
study of the introduction and diffusion of a computerized
information system. Employees increased their power and
network centrality following the change in technology. In
particular,early adopters of the new technology increased
their power and centrality to a greater degree than later
adopters. Results of cross-lagged correlation analyses
suggest that centrality precedes power. While the diffu-
sion process occurred via the network structure, it also
imposed changes in the structure. Adoption patterns were
found to be more closely related to network structure
after the change than priorto the change.-
Does technology drive structure?Or does technology adapt
to existing structure,reinforcingestablished, stable patterns?
The relationshipbetween technology and structurehas been
the topic of much writingand research (Woodward,1965;
Perrow,1967; Thompson, 1967; Hickson,Pugh, and
Pheysey, 1969; Mohr,1971; Hage and Aiken, 1969; Barley,
1986). Althoughthe accumulationof researchstudies has
modifiedthe concept of technologicalimperative,technology
is still considered an importantvariablein relationto organiza-
tionalstructure(Rousseau, 1979). Yet, after decades of re-
search relatingorganizationaltechnology to organizational
structure,"the evidence for technology's influenceon struc-
ture, is at best, confusing and contradictory"(Barley,1986:
78). The same may be said for the multitudeof conceptions
and methodologies employed in such studies (Rousseau,
1979). While technology may be generallydefined as the
transformationof organizationalinputs into organizational
outputs (Perrow,1967; Rousseau, 1979), numerousdefini-
tions and operationalizationsat varyinglevels of analysisand
contexts demonstrate the diversityof technology research
(Comstockand Scott, 1977; Rousseau, 1979; Fry,1982).
Despite this diversity,littleattentionhas been paidto the ef-
fects of technology over time. Cross-sectionalresearch has
typicallyfocused on existing technologies and corresponding
formalorganizationalstructures.The majorityof these cross-
sectional studies treat technology as the independentvari-
able, based on an assumptionthat organizationaltechnology
is inflexibleand, correspondingly,that there is a need for
structureto adaptto the requirementsof technology. These
assumptions are questionable.Technologycan be a flexible
organizationalstrategy that can be modifiedby an organiza-
tion's structure,in particular,the informalstructure.Structural
and, as such, may influenceorganizationaltechnology as well
as be influencedby it. Investigationof the effects of a change
in technology may illuminatethe process by which structure
affects technology, or vice versa.
Few studies relatingtechnology to structurehave considered
the relationshipbetween organizationalstructureand power.
Structuralposition is an importantsource of power in that it
providesaccess to people, information,and other resources.
As Pfeffer (1981) noted, power is first and foremost a struc-
turalphenomenon. Likewise,power strengthens existing
structuralconfigurations.Those in power seek to maintain
104/AdministrativeScience Quarterly,35 (1990): 104-127
 Patterns of Change

power by reinforcingthe existing organizationalstructure

(Pfeffer, 1981). Thus, a change in structuremay necessitate
a change in the distributionof power, and vice versa.
Althoughminor,incrementalchanges in power and structure
may occur graduallyover long periodsof time, the likelihood
of a majorrestructuringmay only occur when the organization
encounters an "exogenous shock" (Barley,1986: 80) such as
the implementationof a new technology. Such a shock might
be conceptualizedas a sudden, dramaticincrease in uncer-
tainty(Tushmanand Anderson, 1986). Attempts to reduce
uncertaintymay foster changes in interactionpatterns,with
those able to cope with uncertaintyadjustingtheir social lo-
cation and increasingtheir power (Salancikand Pfeffer, 1977;
Tushmanand Romanelli,1983). Thus, it is possible that a
change in technology may producechanges in structure,
power, or both.
However, as Pfeffer (1981) noted, stability,not change, is
typicalof the distributionof power and influencein most or-
ganizations.Those in power seek to perpetuatetheir power
advantage.Such processes as commitmentto previousdeci-
sions, institutionalization
of beliefs and practices,and the
abilityof those in power to generate additionalpower con-
tributeto stability(Pfeffer, 1981). Likewise,structuralpatterns
of interactionbecome institutionalizedover time and con-
tributeto organizationalstability.Thus, while a technological
change may providethe opportunityfor a redistributionof
power and organizationalstructure,it does not guaranteeit.
The currentstudy is a longitudinal,cross-level investigationof
a change in technology withinan organization.The change in-
volved the introductionand development of a computer
system with distributedprocessing capabilitiesavailableto all
employees. We adopted a social networkperspective on
structureand includedpower as a key variable.Althoughthe
technology change occurs at the organizationallevel, the ef-
fects of this change are evidenced at the individuallevel. In-
dividualsadopt or reject the new technology and maintainor
change their interactionpatternsand influence relationships.
Interactionand influenceare relational;a change by indi-
vidualsresults in a change in the entire system. Social net-
work measures reflect the cross-level natureof this research.
Ourstudy exploredpossible changes in social networkstruc-
ture and individualinfluencebroughtabout by the introduction
and diffusionof new technology in an organization.The focus
was on stabilityversus change. Does the diffusionof new
technology follow established networkpatterns,with those in
power reinforcingtheir positions? Or does the introductionof
uncertaintyresult in changingpatterns of interactionand in-
fluence?
To investigate the process of change, we used a contingency
model whereby technology providesthe occasion for struc-
turing.Whether stabilityor change occurs is a functionof the
power and social network position of those who are first to
adopt the new technology. When central,powerfulem-
ployees are earlyadopters, existing patternsare reinforced
and stabilityis maintained.Conversely,if less powerful,pe-
ripheralpersons are first to adopt, changes in structureand
the distributionof power will result.
105/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 TECHNOLOGICAL CHANGE:THEINTRODUCTION
OF UNCERTAINTY
A change in an organization'stechnology entails adjustingthe
tools, devices, knowledge, or techniques that mediate be-
tween inputs and outputs and/orcreate new productsor ser-
vices (Rosenberg, 1972; Tushmanand Anderson,1986). In
their industry-levelstudy, Tushmanand Anderson(1986) de-
scribed technologicalchange as an incremental,cumulative
process, punctuatedby majordiscontinuitiesthat represent
majorbreakthroughsin process or product.Technological
changes can be classified as competence-enhancingor com-
petence-destroying.Competence-enhancingadjustments,
which buildon existing know-howwithinthe organization,
tend to consolidate industryleadership:"the richget richer"
(Tushmanand Anderson, 1986: 460). In contrast,the intro-
ductionof fundamentallydifferenttechnologies or compe-
tence-destroyingdiscontinuitiesis associated with major
changes in the distributionof power and control(Chandler,
1977; Barley,1986; Tushmanand Anderson, 1986). Compe-
tence-destroyingdiscontinuitiesdisruptindustrystructure
(Mensch, 1979).
Regardlessof the extent to which a technologicaldisconti-
nuityis competence-enhancingor competence-destroying,
the change in technology increases uncertaintyas attempts
are made to master new tools, devices, or techniques
(Tushmanand Anderson, 1986). Uncertaintycan be generally
defined as an inabilityto predictfutureoutcomes. More spe-
cifically,uncertaintyhas been defined as "the difference be-
tween the amount of informationrequiredto complete a task
and the amount of informationalreadypossessed" (Galbraith,
1977). Both types of discontinuitiescreate technologicalun-
certaintyas individualsstruggle to understandnew and in-
completely specified processes or products.This introduction
of uncertaintyis the theoreticalkey to hypothesizedchange,
or stability,in both structureand power.
Effects of Uncertainty on Structure
Intheir review of the literature,James and Jones (1976: 76)
defined structure"as the enduringcharacteristicsof an orga-
nizationreflected by the distributionof units and positions
withinan organizationand their systematic relationshipsto
each other." Inthis paperwe have adopted a social network
perspective on organizationalstructureconsistent with this
definition:structureis viewed as patterned,repeated interac-
tion among social actors (Weick,1969; Mintzberg,1979). Al-
though this approachdiffers from the traditionalviews of
structure,social networkresearchershave providedexamples
of the successful applicationof this approachto organizations
(Weiss and Jacobson, 1955; Rogers and Rogers, 1976; Tichy
and Fombrun,1979; Tushman,1979; Robertsand O'Reilly,
1979; Tichy, 1981; Brass, 1981, 1984, 1985; Krackhardt and
Porter,1985, 1986).
As technologicaluncertaintyis introduced,changes in inter-
action patterns may occur. As Galbraith(1977) proposed and
researchfindingssubstantiate,increased complexityand un-
certaintyresult in increased communication(Vande Ven, Del-
becq, and Koenig,1976; Katzand Tushman,1979). Individuals
are uncomfortablewith uncertaintyand will work to structure,
106/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

organize,and interpretthe world they experience (Katz,

1980). This structuringand organizingwill take the form of in-
creased communicationwithinthe organizationto interpret
the change in organizationaltechnology and reduce uncer-
tainty.The result may be a change in organizationalstructure.
Althoughchanges in communicationpatternsmay occur,
many of the earlysociotechnicalstudies have pointedto the
stabilityof established social ties (Rice, 1958; Susman, 1976).
Taylorand Utterback(1975) found that althoughintensityof
communicationincreased, communicationroles (such as
gatekeeper) remainedintactdespite changes in projectas-
signments and groupingsin a researchand development lab-
oratory.More recently, Robey (1981) reportedon eight
organizationsthat had introducednew technology in the form
of computersystems. Using post-hoc interviewsand focusing
on formalstructure,he concluded that existing structurewas
reinforcedby technologicalchange. Otherstudies focusing on
the decision-makingstructurehave reportedboth centralizing
(Whisler,1970) and decentralizingtendencies (Brunsand Wa-
terhouse, 1975; Blauet al., 1976).
Effects of Uncertainty on Power Distribution
The introductionof technologicaluncertaintymay affect the
distributionof power just as it may affect structure."Those
who get the upper hand in the game are those who control
most of the crucialuncertainties"(Crozierand Friedberg,
1980: 8). Because a new technology introducescrucialun-
certainties,it represents an opportunityfor employees to gain
influence.Those who are able to reduce uncertaintyfor
themselves and others can increase their power (Hicksonet
al., 1971; Pfeffer, 1981; Tushmanand Romanelli,1983). The
resultmay be a redistributionof power withinthe organization.
However, as previouslynoted, the distributionof power
withinan organizationis relativelystable (Pfeffer, 1981).
Those in power are unlikelyto relinquishtheir power. As in
the case of competence-enhancingdiscontinuities,powerful
individualsas well as industry-leadingorganizationsmay in-
crease their dominance (Tushmanand Anderson, 1986). Influ-
ential individualsmay buildon their existing power bases to
become more powerful.
Early adopters. In investigatinga change in technology, we
used a model that differentiatestwo types of technology
adopters, those who arefirst to adopt the new technology
and those who adopt later(Mohr,1987). Earlyadopters are
those who become early,frequent,and effective users of the
new technology.They are the first in the organizationto cope
with the uncertaintycreated by the change. They are likelyto
be identifiedas experts and be sought out by others within
the organization.Thus, earlyadopterswho have the abilityto
reduce technologicaluncertaintyfor others withinthe organi-
zationhave what is tantamountto a recipe for increased net-
work centralityand power. As an example, when a new
computer system is introduced,technologicaluncertaintymay
become highfor those individualswithinthe organizationwho
had previouslyreliedon the manualmanipulationof informa-
tion to meet workf\low requirements.These individualsmay
seek out and come to depend on those capable of reducing
1O1/ASQ,March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 this technologicaluncertainty.The lattermay become more
centralin the networkof interactions.Thus, we hypothesize:
Hypothesis 1: Earlyadopters will increase their networkcentrality
followinga change in technology.
Earlyadopters may also increase theirpower by being able to
reduce technologicaluncertaintyfor others. This, however, is
contingent on the extent to which there are few other substi-
tutes for their abilityto reduce uncertainty(Hicksonet al.,
1971). As technology is first introduced,theirabilityto reduce
others' uncertaintyis expected to be highlynonsubstitutable,
since only a few individuals(the earlyadopters)will be adept
at workingwith the new system. Thus, we hypothesize:
Hypothesis 2: Earlyadopterswill increase their power followinga
technologicalchange.
While earlyadoptionof a technology providesthe opportunity
for some individualsto increase their centralityand power,
there is also a riskthat the new technology will not be suc-
cessful. Tushmanand Anderson(1986) noted several ex-
amples at the industrylevel of innovativetechnologies that
did not become the dominantdesign in the industry.As they
noted, technologicaldiscontinuitiesand dominantdesigns are
only known as such in retrospect,since competence-
destroyingdiscontinuitiescreate a periodof technological
competitionuntila dominantdesign emerges. Forindividuals
within an organization,some of the riskand uncertaintyof
early adoptionis absorbed by the organization,because the
organizationadopts and approves the new technology. When
an organizationis a late adopter(adopts an alreadyestab-
lished technology),there is littleriskfor the individualadopter,
althoughthe firm may lose its competitiveadvantagewithin
the industryif it waits too long to adopt a technology.
Even withinearlyadoptingfirms the riskof earlyadoptionfor
individualsmay be less than for the organization.At the in-
traorganizational level of analysis,the dominanttechnology
has been established once the decision to change technolo-
gies has been made. There may be a great deal of uncertainty
and competitionamong the decision makers in the organiza-
tion, but, once the decision is made, the design is established
and change is mandated.
Thus, we predictearly-adoptingindividualswill increase their
centralityand power within the-organizationregardlessof the
success of the technology industrywide.When the organiza-
tion's decision is correct, in that it adopts the industry's
eventual dominantdesign, individualearlyadopterswill main-
tainand even strengthentheirpower and position.When the
organization'sdecision to change technology is wrong (an al-
ternativedominantdesign emerges in the industry),individual
increases in power and centralitymay be temporaryand last
only untilthe organizationmakes the decision to abandonand
replace the unsuccessful technology (or, in the extreme case,
the firmfails).At this point, a new technology will be man-
dated, and the process of technologicalchange begins again.
Uncertaintyis again created by the change, and the early
adopters of the new technology increase their power and
centrality.
108/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

Stability versus Change

Just as the previousstudies of technology and structurehave
found mixed effects, we have thus far suggested the possi-
bilityof both stabilityand change in organizationalstructure
and the distributionof power. In Barley's(1986) longitudinal
study, he found differenteffects from the same technology.
Similarly,Tushmanand Anderson(1986) found differentef-
fects in industrystabilitydepending on the type of disconti-
nuityintroduced.Likewise,we have adopted the notionthat
a change in technology providesthe occasion for structuring.
We predictthat stabilityor change is contingent on the social
networkpositionand power of earlyadopters.
Whether or not changes in structurewill occur is contingent
on the previouscentralityof those who are able to reduce
uncertaintyfor others. If centralityprovidesaccess to infor-
mationallowingfor uncertaintyreduction,these central
people become the earlyadopters and it is unlikelythat a
change in structurewill occur. This propositionis consistent
with Tushmanand Anderson's(1986) findingthat littlechange
occurredin industrystructurewhen the technologicaldiscon-
tinuitywas competence-enhancing.Additionalsupportis
found in the adoptionof innovationliterature.When the inno-
vationwas normative(consistent with the social system's
norms),earlyadopterswere centraland well integratedin the
system, often referredto as opinionleaders (Rogers, 1971).
However, earlyadoptionis not relatedto centralitywhen the
innovationruns counter to system norms (Rogers, 1971).
If uncertaintyis absorbed by individualswho were previously
less central,theirgain in centralitymay adjustthe overall
structureof the organization.Interactionpatternswill change
as those who were previouslyperipheralare sought out by
others. Thus, we generate the followingcontingency hypoth-
eses:
Hypothesis 3a: Ifearlyadoptersare more centralthan late adopters
priorto a technologicalchange, the existing structurewill be rein-
forced.
Hypothesis 3b: If earlyadoptersare less centralthan late adopters
priorto a technologicalchange, a structuralchange will occur.
Similarly,those who become earlyadopters may increase
their level of influence relativeto late adopters, thereby redis-
tributingpower throughoutthe organization.Or those cur-
rentlyin power may remainin power by adoptingearly,thus
reinforcingthe existing-power distribution.These two pos-
sible scenarios, politicalstabilityversus the redistributionof
power, are thus contingenton the previouspower of early
adopters. The followingcontingencyhypotheses are thus
generated:
Hypothesis 4a: If earlyadoptersare more powerfulthan late
of power is
adopters priorto a technologicalchange, a redistribution
unlikely.
Hypothesis 4b: If earlyadoptersare less powerfulthan late
adopters priorto a technologicalchange, a redistributionof power
will occur.
A redistributionof power is most likelyto occurwhen the less
powerfulbecome the earlyadopters and the more powerful
are late adopters. This is more likelyto happen if earlyadop-
109/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 tion is not relatedto centralityor power. Tushmanand Ro-
manelli(1983) found that the greaterthe task uncertainty,the
greaterthe influenceof boundary-spanning individualsrela-
tive to those internallycentralto the organization.At the in-
dustry level, Tushmanand Anderson(1986) found that
industrydominancechanged when technologicaldisconti-
nuities were competence-destroying.Withinan organization,
the most likelyscenario would includesome early adopters
among both those in central,powerfulpositions and those
less centraland powerful.Thus, it is likelythat a change in
technology would not result in a total upheavalbut, rather,in
a moderate redistributionof power.
A total redistributionof power becomes even less likelywhen
the connection between power and networkcentralityis con-
sidered. As earliernoted, those persons occupyingcentral
positions in the organization'snetworkare likelyto be per-
ceived as powerful(Tushmanand Romanelli,1983; Brass,
1984). Because this study is longitudinal,we can explore
whether power leads to centrality,or vice versa. Are powerful
individualssought out by others, thus increasingtheir cen-
tralityin the network?Or does being in a centralpositiongive
one access to people and informationsuch that one be-
comes powerful?
Regardlessof whether these centralpositions are the source
of power or the result of power, they will be instrumentalin
the diffusionof technology. They representthe key nodes
throughwhich informationflows and is dispersed throughout
the organization.Therefore,change becomes particularly diffi-
cult when persons in centralpositions are resistant.Thus,
any change in the power distributionmay necessitate a cor-
respondingchange in the informalstructure.Overtime,
changes in the informalstructuremay necessitate changes in
the formalorganizationalstructure.Whilethese changes are
thoughtof as difficultones, they are more likelyto occur with
the introductionof new uncertaintiescreated by technological
discontinuity.

Predicting EarlyAdoption
Identifyingthe attributesof earlyadopters of a change pro-
cess may aid in predictingthose who are able to reduce or-
ganizationaluncertaintiesand thus in predictingpossible
changes in power and structure.Adoptionof innovation
studies have found that attitudeand education level are re-
lated to earlyadoption;results-concerningage have been
mixed (Rogers, 1971). Inour study, we predictthat individual
characteristicswill be relatedto earlyadoptionof the new
computersystem. In particular,we hypothesize:
Hypothesis 5: The followingcharacteristicswill be relatedto early
adoption:(1) age, (2) educationlevel, (3) previouscomputertraining,
(4) attitudestowardcomputers, and (5) feelings of efficacy re-
gardingcomputeruse.
Individualcharacteristicshypothesizedto be relatedto early
adoptionmay also be relatedto power and centrality.Roberts
and O'Reilly(1979) found that individualcharacteristicswere
relatedto variousroles in communicationnetworks.This pos-
sibilitysuggests stability.Only if individualcharacteristicsare
relatedto earlyadoptionand not positivelyrelatedto power
11OASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

and centrality would we expect changes in power and struc-

ture.
Rogers (1971) noted that when organizational change is first
introduced, the relevance of individualcharacteristics is
heightened. As the diffusion process continues, individualat-
tributes are overshadowed by structural characteristics. The
spread of ideas and practices becomes contingent on the way
in which social structure brings people together (Burt, 1987a).

The Diffusion Process

Researchers in sociology and related disciplines have made
extensive use of network analysis in investigating the diffu-
sion of innovations (see Rogers, 1962, 1971; Burt, 1982).
These studies trace the communication of new ideas and
adoption of innovations over time through channels of
communication in a social system. Although the types of in-
novations and the social systems studied have varied tre-
mendously, some consistent results concerning
communication channels have been found. Awareness of in-
novations is often accomplished via mass media input from
outside the social system; the evaluation and decision to
adopt an innovation is primarilythe result of interpersonal
communication within the system (Rogers, 1971). Although
initialadopters tend to be more cosmopolitan, the diffusion of
innovations to later adopters tends to follow social network
patterns of interaction (Rogers, 1971).
Late adopters are expected to adopt as a result of a socializa-
tion process referred to as contagion. In particular,an indi-
vidual is likely to adopt an innovation based on contact,
communication, or competition with an individualwho has al-
ready adopted (Rogers, 1971; Burt, 1987a). This likelihood is
based on an analysis of the social structural circumstances of
the individualwho has not yet adopted. The contagion model
focuses on the spread of innovation attributable to communi-
cation between the individualwho has not adopted, or ego,
with an individualwho has already adopted, the alter. The
model also accounts for the adoption of innovation to the ex-
tent that ego and alter are in a similar position in the social
structure, i.e., the extent to which they occupy the same
roles and talk to the same other people within a social unit.
Thus, it is hypothesized:
Hypothesis 6: The diffusionof a technologicalchange will occur
throughstructuralpatternsof interaction.
Thus, whether or not people adopt a new innovation is a
function of the social context in which they act and speak
(Burt, 1987a). It is also likely, however, that the social context
may be changed by the introduction of the innovation itself.
Diffusion may occur through existing structural patterns, or
changes in these patterns may occur as a result of a techno-
logical change. As suggested previously, individuals may ad-
just whom they communicate with in order to reduce
technological uncertainty. Thus,
Hypothesis 7: If existing patternschange, adoptionof a new tech-
nologyviacontagionwillbe morecloselyrelatedto networkstruc-
tureafterthe changethanpriorto the change.
March 1990
111/BASQ,

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 The Change
Ourstudy involveda four-partlongitudinalanalysistracking
the introductionand diffusionof a computersystem in a fed-
eralagency responsiblefor the analysisand disseminationof
a nationaldata base of nutrientdata. The computersystem
offered distributedprocessing capabilities,includingfile
editing, data-base management, statisticalanalysis, spread-
sheet analysis,and word processing to all employees. Priorto
the introductionof distributiveprocessing capabilities,an ex-
ternalcomputingfacilitywas accessed for computer analysis
of research data for all employees. The prohibitivecost of this
service was the primarymotivatorfor the purchaseof the
computersystem. The head of the agency's survey statistics
branchproposed the implementationof the computersystem
to the agency's directoras a feasible undertakingthat would
providesubstantialcost savings to the agency.
While the system was still in the planningstages, employees
were queried as to their computer needs. However, lack of
computerexperience left the majorityof employees uncertain
as to what their needs were. Althoughtrainingwas originally
scheduled to occur shortlyafter system implementation,
problemswith scheduling and the decision to use trainers
from outside the agency delayed it considerably.Employee
interviews conducted throughoutthe study suggested that
the delayed trainingincreased the employees' uncertaintyand
aggravation.
Most employees had not had directwork experience with
computerapplicationspriorto the introductionof this system.
The computerfunctions substantiallychanged their method
of analyzingnutritiondata and preparingdocuments for publi-
cation. Forexample, ratherthan submittinga request for data
analysis, nutritionanalysts began to programand runtheir
own statisticalanalyses. Thus, by almost any of the varied
definitionsof technology, the introductionof computercapa-
bilitiescan be regardedas a majorchange in the organiza-
tion's technology.
METHOD
Data Collection
All questionnairedata were collected on site at four points in
time. The first questionnaireadministration(T1)was approxi-
mately three months priorto the introductionof the computer
system; the second (T2)occurredthree months after the
system configurationwas in place and approximatelysix
months followingT1. The thirdquestionnaireadministration
(T3)was three months followingthe second and immediately
precedinga formalthree-daytrainingperiod.The last data
collection (T4)was three months after training,approximately
one year after system implementationand 15 months fol-
lowing T1. Interviewswith variousinformantswere con-
ducted by the researchers before, during,and followingthe
questionnaireadministrations.Participation was voluntaryand
respondents were assured that their individualresponses
were confidentialand would be used for research purposes
only.
Ninety-fourfull-timeemployees were employed by the
agency at T1. Of these, thirteen left the agency duringthe
112ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

time of the study, resultingin a sample size of 81 over the

four time periods for measures of power and centrality.Be-
cause centralityand power scores were obtainedfrom
sources independentof the focal person, it was possible to
receive 81 scores for these variablesat all four questionnaire
administrations.Thus, for analyses of changes in power and
centralityover time, our sample was 81. Forty-nineof these
81 employees completed questionnairesat all four times.
The total numberof persons completingquestionnairesat
each of the four time periodswas 75, 84, 74, and 66, respec-
tively. Thus, for analysis of correlatesof earlyadoption,in-
volvingonly T1 data, the sample includedthe 75 employees
who completed questionnairesat T1.

Measures
Network analysis. Respondents were providedwith a list of
all agency employees and were asked to circlethe names of
people with whom they communicatedas partof theirjob
duringa typicalweek. Priorto each questionnaireadministra-
tion the roster of names was updated.Names of employees
who quit were droppedand names of new employees were
added. This data was entered as a binarymatrixand analyzed
to determine the followingnetworkmeasures. Two opera-
tionalizationsof networkcentralitywere calculated,closeness
and in-degree.
Closeness. The closeness measure of centralityaccounts for
both direct and indirectlinksand conceptuallyrepresents
ease of access to others. Forexample, in additionto em-
ployees who are directlyconnected to the focal person, there
are typicallymany other employees who are indirectlycon-
nected to him or her. These others are indirectlyconnected to
the focal person by being directlyconnected to a person with
whom the focal person is directlyconnected. Forexample,
focal person A talkswith B, B talkswith C, but A does not talk
with C. A has a one-link,directconnectionto B and a two-link,
indirectconnection to C. Forthis closeness measure of cen-
trality,we ignoredthe directionof the linksand treated them
all as reciprocated(Knokeand Burt,1983).
The closeness measure of centralitywas calculatedfor each
of the 81 individualsin the sample by addingthe minimum
numberof linksbetween the focal individualand all others
withinthe organization(Freeman,1979; Knokeand Burt,
1983). This sum was then dividedby n - 1, where n equals
the numberof persons in the organization.The closeness
centralitymeans were transformedby the formula1 - [(d -
1)/dmax](Lincolnand Miller,1979; Brass, 1984), where d
equals the path distance and dmax equals the largest ob-
served value of d. This transformationnormalizescloseness
scores to a range of zero to one and results in higherscores,
reflectinghighercloseness centrality.This transformation
does not change the magnitudeof the relationshipsof other
variableswith closeness, but it reverses the sign of the rela-
tionships.
In-degree.A second measure of centralityfocuses simplyon
the numberof employees with whom an individualis directly
connected, referredto as degree (Freeman,1979). Degree
centralitytypicallyincludes direct connections in which the
l1l3/ASQ,March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 focal person is either the source or object of the connection.
In-degreetakes into account the directionof the link,in-
cludingonly those linksin which the focal person is the object
of the connection. In-degreecentralitywas operationalizedas
the numberof times an individualwas chosen by coworkers
on the communicationroster, dividedby the numberof
persons completinga particularquestionnaire.
An example will illustratethe practicaldifference between the
two centralitymeasures. A focal employee can increase his or
her closeness centralityby seeking out a highlycentralother.
The other's direct linksbecome the focal person's indirect
links,thereby substantiallyincreasingthe focal person's
closeness centrality.Ifwe ignorethe directionof the link,the
other's closeness centralityis also increased. However, by
consideringthe directionof the link,only the other's in-degree
centralityis increased-the other is chosen by the focal
person.
If diffusionfollows established communicationpatterns,nei-
ther closeness nor in-degreecentralitymeasures should
change. If communicationpatternschange, both earlyand
late adopters may increase theircloseness centrality,but only
early adopters are expected to increase their in-degreecen-
trality.
Power. Individualpower was assessed by asking each re-
spondent in contact with the focal individualto rate that indi-
vidualon a 5-pointLikert-typescale (1 = very littleinfluence,
5 = very much influence).The individualratingsobtainedfor
each focal person were averagedto obtainan overallpower
score for that person. Thus, all 81 persons in the sample re-
ceived power scores for all four time periods.The average
numberof ratingsper focal person was 18.19.
Individualcharacteristics.Five individualcharacteristicswere
measured on the questionnaire.Respondents were asked to
indicatetheir age in years, their educationlevel (1 = high
school, 2 = some college, 3 = bachelor's degree, 4 =
master's degree, 5 = Ph.D.),and hoursof previouscomputer
training.To measure computerattitude, respondentswere
asked to consider eight pairsof adjectives, each pairan-
choringthe ends of a 7-point Likert-typescale (Shaft, 1986).
Forexample, polaradjectives such as helpful/harmful; easy to
use/difficultto use; threatening/nonthreatening; boring/in-
triguing;and enjoyableto use/frustratingto use were in-
cluded. The average score on the eight items was used as a
measure of computerattitude(alpha= .84).
The finalindividualattributemeasured on the questionnaire
was computerefficacy. Respondents were asked the extent
to which they agreed or disagreed on a 7-pointLikert-type
scale (1 = disagree strongly,7 = agree strongly)with three
statements about their feeling of efficacy regardingcom-
puters. One example was "I have the capabilityto effectively
use computers in my job." Scores on the three items were
averagedto form an index of computerefficacy (alpha= .92).
These five individualcharacteristics(age, education level,
computertraining,computerattitude,and computerefficacy)
were chosen because of their hypothesizedrelationshipto
adoptingthe new technology. Withthe exception of age, we
l1l4/ASQ,March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

hypothesizedthat allwould be positivelyrelatedto computer

adoption.
Earlyadoption.Earlyadopters were identifiedthroughanal-
ysis of data collected after system implementation(T2ques-
tionnaireadministration)but priorto the formaltraining
providedby the agency. It was duringthis time that early
adopters were expected to be highlydifferentiatedfrom co-
workers. Individualswere categorizedas earlyor late
adopters depending on their response to three different
questionnaireitems at T2. We asked respondents to indicate
the date on which they started using the new computer
system. Date was coded in terms of months followingT1 (X
= 10.5, S.D. = 8.2). We asked them to indicate how many
hours per week they were currentlyusing the new computer
system (T2 X = 8.0, S.D. = 9.7), and we asked them to re-
spond to the following item: "I am effectively using the new
computer system" (1 = strongly disagree, 7 = strongly
agree; T2 X = 3.58, S.D. = 1.89).
To be categorizedas an earlyadopter,an employee must
have been using the computerat least 10 hours per week at
the T2 survey distribution,list his or her date of adoptionas
priorto T2, and agree that his or her computer use was ef-
fective by indicatinga 5, 6, or 7 on the computereffective-
ness item. Interviewsconducted by the researchersfollowing
the T2 administrationwere used to confirmmany of the early
adopters. Others were checked against the agency's roster of
assigned computer IDs for confirmation.Seventeen em-
ployees who met all the above criteriawere identifiedas early
adopters. All other employees were considered late adopters.

RESULTS
Correlates of EarlyAdoption
To test hypotheses regardingcharacteristicsrelatedto early
adoptionand as our first test of the stabilityversus change
hypotheses, we correlatedindividualattributesand centrality
and power priorto the change (Ti) with earlyadoption.To the
extent that power and centralityare predictiveof earlyadop-
tion, patternsof organizationalstructureand power are ex-
pected to remainthe same. If individualattitudes are
predictiveof earlyadoptionand not positivelyrelatedto
power and centrality,changes in organizationalpatternsare
expected. Table 1 presents the intercorrelationsof power,
centrality,individualcharacteristics,and earlyadoption.Be-
cause hours of trainingwas highlyskewed, correlationswere
calculatedwith the logarithmof the values. With the excep-
tion of age and education,the individualattributeswere posi-
tively and significantlyrelatedto earlyadoption,consistent
with hypothesis 5.
As expected, power and centralitywere highlycorrelatedat
T1; however, neitherwas positivelyrelatedto earlyadoption.
Infact, both measures of centralityat T1 were negativelyas-
sociated with earlyadoption.Results of hierarchicalregres-
sion analyses were consistent with the zero-order
correlations.Individualattributes,when entered following
centralityand power, added significantlyto the varianceex-
plainedin earlyadoption.When individualattributeswere en-
l1l5/ASQ,March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
Table 1
Correlates of EarlyAdoption (N = 75)
Variable Mean S.D. 1 2 3 4 5 6 7 8
1. T1 Power 2.67 .67
T1 Centrality
2. Closeness .71 .08 .62"
3. In-degree .27 .11 .55" .72"
characteristics
Individual
4. Age 41.80 11.07 .24- .21- .17
5. Education 3.38 1.16 .43" .10 .19 .08
6. Hoursof training* 34.58 54.05 -.17 - .01 -.02 -.17 .11
7. Computerefficacy 5.21 1.54 -.13 .08 -.02 -.23- .11 .24-
8. Computerattitude 5.32 .85 .07 .23- .05 - .13 .13 .23- .53"
Earlyadoption -.01 -.13 -.11 -.13 .19 .21- .27" .26"
p < .05; Up < .01.
* Logarithmof hoursof trainingused to calculatecorrelations.

tered first, centralityand power did not add significantlyto the

regression equation.
Table2 presents the results of multivariateanalysis of vari-
ance assessing changes in power over time for earlyversus
late adopters.The results show a significantinteractioneffect
between earlyadoptionand time. The means across time in-
dicate that earlyadopters gained more in power over time
than late adopters, lendingsupportfor hypotheses 2 and 4b.
Furthermore,there was a significantmaineffect for time. In
general, power increased followingthe technologicalchange.

Table2
MANOVA:Results for Power over Time of Earlyvs. Late Adopters
Effect d.f. F pa
Earlyvs. late 1, 79 1.33 .252
Time 3, 77 18.33 .000
Earlyvs. late x time 3, 77 4.14 .009
Means
T1 T2 T3 T4
Entiresample 2.63 2.79 2.81 2.90
Earlyadopters 2.65 3.05 2.96 3.13
Lateadopters 2.62 2.74 2.77 2.85

MANOVAanalysis of both operationalizations of centrality,

closeness and in-degree(Table3), showed similarresults. An
interactioneffect was significantat the .05 level for the in-
degree measure of centralitybut not for the closeness mea-
sure. A main effect for time was evidenced for both close-
ness and in-degree.These results indicatethat
communicationpatternschanged, with all employees in-
creasing their closeness centrality.Employees increasedtheir
interactionwith earlyadopters more than with late adopters,
as indicatedby the significantinteractioneffect of earlyadop-
116/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

Table 3

MANOVA: Results for Network Centrality over Time of Early vs.

Late Adopters

Closeness

Effect d.f. F p<

Earlyvs. late 1, 79 .03 .855

Time 3, 77 9.22 .000
Earlyvs. late x time 3, 77 1.71 .172

Means
T1 T2 T3 T4

Entire sample .692 .721 .715 .702

Early adopters .671 .721 .712 .715
Late adopters .696 .721 .715 .700

In-degree

Effect d.f. F p

Earlyvs. late 1, 79 .01 .911

Time 3, 77 3.90 .012
Earlyvs. late x time 3, 77 2.79 .046

Means
T1 T2 T3 T4

Entire sample .263 .239 .246 .256

Earlyadopters .239 .236 .266 .274
Late adopters .268 .240 .242 .253

tion by time for the dependent variable in-degree. This

finding lends support for hypotheses 1 and 3b.
An overall measure of network density (MacEvoy and
Freeman, 1986) was used to assess the degree to which all
actors were interconnected. Density measures the extent to
which actors in a system are connected, on average, to one
another (Burt, 1982). The network densities for each time pe-
riod were as follows: T1 = .262; T2 = .1 16; T3 = .286; and
T4 = .287. These findings lend additional support to predic-
tions of structural change.
Network Structure and Power
As expected, centrality was significantly related to power at
all four time periods (closeness: T1, r = .62; T2, r = .43; T3,
r = .32; and T4, r = .33; in-degree: T1, r = .55; T2, r = .51;
T3, r = .39; T4, r = .41). In order to investigate a temporal
ordering of increases in centrality and power, cross-lagged
correlation analysis (Campbell and Stanley, 1963) was per-
formed. If power precedes centrality, the correlation between
T1 power and T2 centrality should be greater than the corre-
I
lation between T1 centrality and T2 power. If centrality leads
1k = (r12- r24rl4)(r34- r24r23)
+ (r13- r12r23)
to power, the reverse should be true, i.e., T1 centrality with
(r24- r12r14)
+ (r12- r13r23)
(r34- r13r14) T2 power greater than T1 power with T2 centrality.
+ (r13- r14r34)
(r24-r34r23)
where 1, 2, 3, and 4 are variablesand N Table 4 presents the results of this analysis for both mea-
is sample size. The followingthen has an sures of centrality, closeness and in-degree, for all four time
approximately standardnormaldistribu-
tion: periods. To test for statistically significant differences be-
tween the cross-lagged correlations, we used a test that
(N)"2(r1- r23)
allows the correlations to be correlated (Kenny, 1979).1 Al-
[(1 - 1224)2 + (1 1-223)2 - k]"2 though few of the differences are significant,the\overall
117/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Table4
Cross-lagged CorrelationAnalyses of Power and Network
CentralityMeasures
Power and Closeness
P1C2 P2C1 P1C3 P3C1 P1C4 P4C1
Early .42 > .30 .19 < .20 .19 < .40
Late .57 > .50 .41 < .51 .37 < .47
Entiresample .55 > .44 .39 < .44 .35 < .43
P2C3 P3C2 P2C4 P4C2
Early .26 < .54 .08 < .53
Late .30 < .50- .25 < .43
Entiresample .29 < .49w .24 < .43-
P3C4 P4C3
Early .01 < .41
Late .34 > .30
Entiresample .31 > .30
Power and In-degree
P1N2 P2N1 P1N3 P3N1 P1N4 P4N1
Early .42 < .59 .38 < .61 .23 < .69
Late .61 > .44- .41 < .45 .46 > .45
Entiresample .58 > .43- .41 < .45 .43 < .45
P2N3 P3N2 P2N4 P4N2
Early .57 < .65 .47 < .57
Late .32 < .54" .37 < .50
Entiresample .36 < .55w .39 < .50
P3N4 P4N3
Early .49 < .66
Late .43 > .31
Entiresample .44 > .36
*p < .05; "p < .01.

trends are clear. Correlations with closeness preceding power

were stronger than correlations with power preceding close-
ness. There are, however, two anomalies to this trend, P1C2
versus P2C1 and P3C4 versus P4C3. The same trends exist
when using in-degree measures of centrality. However, sub-
group analysis of early adopters shows no anomalies for in-
degree and power cross-correlations and only one anomaly
for cross-lagged analysis with closeness scores. That is, for
early adoption, centrality precedes power in almost all com-
parisons.
Diffusion Processes
For analysis of the diffusion process (hypothesis 6), we
adopted a structural equivalence model (Burt, 1982). Diffusion
by structural equivalence reflects the extent to which indi-
viduals adopt a technology based on the adoption of those
with whom they have similar patterns of interaction. This
model does not require that two individuals communicate
with each other directly, only that they have similar positions
in the social structure. Similar or structurally equivalent posi-
tions are defined by the extent to which individuals talk to the
same other individuals in the social system. A special form of
118/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

structuralequivalence in which individualsdirectlyinteract

with each other is referredto as cohesion (Burt,1982).
Althoughthe structuralequivalence model of contagionin-
cludes the possibilityof direct interaction,theoreticaland
methodologicaldifferences exist between this model and one
based solely on cohesion. Burt(1982) provideda detailed re-
view of the differences between the positional(structural
equivalence)and relational(cohesion) approaches.Although
there is some debate about the merits of the two ap-
proaches, the structuralequivalencemodel includesa broader
range of possible types of interactionand has been shown to
be a useful method for analyzingcontagion(Burt,1987a).
Measures of structuralequivalencewere used to weight the
date of adoptionof individualswithinthe networkto deter-
mine contagion effects throughthese social processes. The
normativeor predictedresponses were correlatedwith the
observed or actualdates of adoptionto determinethe extent
to which adoptiontiming is a functionof structuralequiva-
lence. A high correlationindicatesthat the diffusionof the in-
novationfollows the interactionpatternsas depicted by
structuralequivalence.The observed dates of adoptionare
the same for T1 and T4 analyses. Patternsof interaction,
however, may be different.A differentcorrelationbetween
observed and normativeresponses for T1 and T4 would indi-
cate that patternsof interactionchanged. Whetherthe T1 or
T4 autocorrelationis higherreflects the extent to which the
diffusionprocess more closely followed pre-existing(if the T1
autocorrelationis higher)or new (if the T4 autocorrelationis
higher)interactionpatterns.A detaileddiscussion of the anal-
yses is contained in the Appendix.
Jackknifeestimates of FisherZ-transformedcorrelationswere
obtainedto determinethe contagioneffect's significance(see
Appendix).The jackknifeestimate of the FisherZ-transformed
correlationfor T1 was .278 with a standarddeviationof .206.
A t-test on the jackknifeanalysiswas 1.352 (n.s.) for a corre-
lationof .271. The jackknifeestimate of the FisherZ-trans-
formed correlationfor T4 was .524 with a standarddeviation
of .150, yieldinga t-test of 3.491 (p < .001). These results
supporthypothesis 6: diffusionoccurredthroughstructural
patternsof interaction.Furthermore,they indicatethat conta-
gion more closely followed the T4 interactionpatternsthan
the T1 interactionpatterns.This findinglends supportto hy-
pothesis 7: patternsof communicationchanged to enable the
diffusionprocess to occur.

DISCUSSIONAND CONCLUSIONS
Despite the forces supportingstability,considerablechange
in both structureand power occurredfollowinga technolog-
icalchange withinan organization.Being centraland powerful
priorto the introductionof a new technology was not related
to earlyadoption.Rather,earlyadoptionwas a functionof in-
dividualcharacteristicsrelevantto the change process. Thus,
in accordancewith theoreticalpredictions,the ingredientsfor
structuralchange were in place.
Earlyadopterswere able to reduce uncertaintyfor others, and
this uncertaintyreductionabilityenabled them to gain power
and centrality.Because they were not central,powerful
119/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 playerspriorto the technologicalchange, the possibilityfor
structuraland power redistributionon the organizationallevel
was likely.Earlyadopters gained more in-degreecentrality
and power than lateradopters. At T2, late adopters actually
decreased their in-degreecentralityto a great extent. This
drop is reflected in an overalldecrease in networkdensity at
T2. Thus, while structureis difficultto change, a technological
change providedopportunityfor restructuringand consequent
changes in the organization'spower configuration.
Results also lend supportto the view that diffusionitself oc-
curredas a result of the restructuringprocess. Individualsad-
justed their patternsof interactionin orderto learnfrom
those who were alreadyadept at using the new technology.
Hence, contagion is a process that occurrednot as a resultof
priorstructuralconfigurationsbut, rather,structuralconfigura-
tions changed to enact contagion.
Prior influence. Althoughresults indicatedchanges in net-
work centralityand power those employees who were pow-
erful,centralfigures in the organizationpriorto the change
(Ti) were not totallydisplaced by earlyadopters.Although
earlyadopters gained substantiallymore influence,those with
priorpower maintainedmuch of their power. The zero-order
correlationbetween T1 power and T4 power was .84. One
possible explanationis that those in power at T1 derivedtheir
influencefrom sources that were not affected by the change.
They may also have taken advantageof their organization
centralityto be the first of the lateradopters to contact and
learnfrom the earlyadopters. Thatis, persons in centralposi-
tions may have a better understandingof the network(Krack-
hardt,1989) and be able to use it to theiradvantageto adapt
to a change. Ourinterviewsat T2 providedsome evidence of
the latter.As one earlyadoptersaid, "Allof a sudden, the
bigwigs are coming to me, asking my advice."
The furtherpossibilityexists that those in power priorto the
change were responsiblefor makingthe decision to change
the technology. These individualsmay not have been the first
to adopt but may have maintainedtheir power by being re-
sponsible for and receivingcreditfor the decision. The
decision to implementthe computersystem would not nec-
essarily requirethe specific expertise needed to operate the
system.
Greater total influence. Overall-,
the total amount of indi-
vidualinfluence in the organizationincreased as individual
centralityincreased and the networkbecame more intercon-
nected. One possible explanationis that employees felt less
dependent on externalsources of data processing. The new
computersystem gave them more controlover theirwork
outcomes. This possibilitywas indicatedduringour interviews
and might have been more evident had we obtainedself-
ratingsof power. Anotherexplanationfor the increased influ-
ence was also suggested in interviewsand is consistent with
our use of externalsources for ratingindividualinfluence. In
establishingnew communicationlinkswith earlyadopters,
late adopters became aware of the expertise of the early
adopters and ratedthe latterhigheron influence.This expla-
nationof contact precedingawareness of influence is consis-
120/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

tent with the results of our cross-lagged correlationanalysis

suggesting that centralitypreceded power.
Training. A formaltrainingprogramconducted by represen-
tatives of the manufacturerof the new system was available
to all employees directlyfollowingthe T3 questionnairead-
ministration.Althoughoriginallyscheduled to occur sooner,
the trainingwas delayed, and in interviewspriorto T3, em-
ployees indicatedthat the lack of trainingwas their major
complaint.We felt that this delay might have emphasized the
nonsubstitutabilityof the earlyadopters.Althoughcloseness
centralityand power means showed littlechange between T2
and T3, in-degreecentralityincreased significantlyfor early
adopters. Interviewsindicatedthat some of the lateradopters
were simplywaitingfor trainingbefore attemptingto learn
the system.
Althoughwe had expected that trainingmight decrease the
nonsubstitutabilityof the earlyadopters, it appears that
traininghad the opposite effect. Mean changes between T3
and T4 suggest that earlyadopters again increased closeness
centralityand power as they had between T1 and T2. Those
who had been waitingfor trainingwere attemptingto learn
the system followingT3. Interviewsfollowingtrainingindi-
cated that lateradopters were seeking out earlyadopters for
their expert help on the new system.
Limitations. One limitationof the currentstudy involves the
lackof an appropriatecontrolgroup. Unfortunately,all the
employees in the agency were subject to the change in tech-
nology.Thus,we could not identifyand study a groupof em-
ployees who did not experience the change.
We were also limitedby the lackof additional,nonreputa-
tional measures of power. Althoughthe study was based on
perceptions of influence,we feel that the multiple-rater
methodologyprovidescrediblereliabilityand validity.Power is
a social phenomenon, dependent on the attributionsof
others. If behavioris consistent with attributions,then those
perceived as powerfulare powerful.
Generalization of findings. The particulartype of technolog-
ical change may limitthe generalizabilityof these results to
other changes in technology or to organizationalchange in
general. Forexample, there is a long historyof the effects of
automationon the routinizationof work and loss of controlby
employees (cf. Kipnis,1984). While our results suggest that
computersystems do not decrease employee influence,
other types of technologicalinnovationsmay producedif-
ferent results. Takenin combinationwith previousstudies,
our results indicatethe importanceof accountingfor the type
of technology introduced.
It is also importantto consider individualcharacteristicsin re-
lationto power and centrality.Inthe event that centralorga-
nizationalmembers possess characteristicsassociated with
earlyadoption,a redistributionof power and centralityis un-
likelyto take place. In our study, only age and educationlevel
were significantlyrelatedto T1 power, and neitherwas
significantlyrelatedto early adoption.However, it is not un-
reasonableto imagine a situationin which individualcharac-
teristics such as internallocus of control(Rotter,1966) or
121/ASQ, March1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 general self-efficacy (Bandura,1986) are relatedto power,
centrality,and earlyadoption,with all contributingto stability
in power and communicationpatterns.
This study also differs from many previousadoption-of-
innovationstudies (Rogers, 1971; Burt,1987a; Fennelland
Warnecke, 1988) in that adoptionwas not voluntarybut man-
datory.Althoughemployees could choose not to use the
system and continue relyingon previouswork methods, there
were clear expectations that the new system be used. This
may explainwhy we found very little evidence of organized
resistance to the change. OurT2 interviewssuggested the
possibilitythat one departmentmight resist. However,T3 and
T4 data indicatedits late but eventual adoption.
When adoptionis voluntaryratherthan mandatory,it is likely
that complete diffusionwill not occur, as socializationis never
completely successful (Bergerand Luckmann,1967). Re-
sisters may try to influencecoworkersagainst using the new
technology, sharingtheir resistance with other workers in the
network. If resistance is particularly
strong, these "late"
adopters may never adopt the change.

Implications for Theory and Research

One might be tempted to includethe results of this study
with those of previous studies that supporttechnology's in-
fluence on structure.However, the importanceof this re-
search is not whether it adds, one way or the other, to the
debate concerningthe existence of a technologicalimpera-
tive. One more set of results will not conclusivelydecide the
argumenteither way. Rather,its significancelies in the at-
tempt to understandthe process by which technology may
affect structure,or vice versa. As a longitudinalstudy, it is
best considered in relationto Barley's(1986) study. Although
it employs considerablydifferenttheories and methodologies,
the results are not inconsistent. Changes in technology pro-
vide the occasion for structuring.
The results of this study are also consistent with those found
by Tushmanand Anderson(1986) in their industry-levelstudy
of technologicalchange. In retrospect,the installationof the
computer system can best be classified as a competence-
destroyingprocess discontinuity.Whilethe service remained
essentially unchanged,the process by which it was rendered
was fundamentallychanged. The skills and knowledge base
requiredto transformthe inputs into outputs shifted dramati-
cally.When combined with Tushmanand Anderson's
findings,our results suggest a multilevelperspective on tech-
nologicalchange. Individualswho are the first to recognize
and exploit technologicalopportunities(earlyadopters)in-
crease their power and centralitywithinthe organization,just
as innovativeorganizationsincrease their competitive advan-
tage within an industry.
Othersimilaritiesacross levels and studies can be noted. For
example, the process of change appearsto be the same.
Technologicalchange produces uncertainty.Just as indi-
vidualsattempt to cope with uncertainty,so do organizations.
Althoughthe present study did not measure uncertainty,our
interviewswith employees, coupled with Tushmanand An-
derson's (1986) measures of uncertainty,providesupportfor
122/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 Patterns of Change

this underlyingassumption. However, the types of uncer-

tainty may not be the same across levels. At the interorgani-
zationallevel of analysis, environmentaluncertaintyoccurs as
organizationsand technologies compete for the dominantin-
dustrydesign. At the intraorganizational level, the design is
mandated;the uncertaintyfor individualsinvolves learningthe
new technology.
Furtherresearch is needed to extrapolatethe industry-level
findingsto the intraorganizational level, and vice versa. We
attempted to predictearlyadopters at the individuallevel
based on individualcharacteristics.Similarly,we might spec-
ulate that particularorganizationalcharacteristicssuch as age,
culture,or strategy might predictearlyadopters within indus-
tries. Forexample, new organizationswith "prospector"
strategies and innovativeculturesare likelyto be the first to
take advantageof competence-destroyingdiscontinuities
(Milesand Snow, 1986; Tushmanand Anderson, 1986).
Networkanalysis can also be appliedat the organizational
level of analysis in orderto study the structureof an industry.
Forexample, Boje and Whetten (1981) found that centrality
withina networkof manpoweragencies was associated with
influence. Miles and Snow (1986) have suggested the con-
cept of dynamicnetworkswithin industries.Withinindustries,
certainorganizationsmay take the early-adopterrole. Miles
and Snow suggest that "prospectors"may possess the dis-
tinctive researchand development competence to generate
technologicalinnovationsthat push the industryforward.The
study of an industrynetworkmay also provideinformationon
the diffusionof innovationamong organizations.The struc-
turalequivalence model of contagion is particularly suited for
analysis of competing organizations(Burt,1982), which need
not be directlylinkedin orderto be considered equivalent.
FollowingTushmanand Anderson's suggestions, future re-
search might combine intra-and interorganizational levels of
analysis. Forexample, do adoptionand diffusionpatterns
among individualsin early-adoptionorganizationsdifferfrom
those in later-adoptionfirms? Inour study, the federal
agency's adoptionof the computersystem could only be
classified as a late adoptionwithinthe industry.
Late adoptionby an organizationmay indicatethat the tech-
nologicalchange runs counter to prevailingorganizational
norms.Althoughwe did not measure the agency's norms,we
found that individualattitudes toward computertechnology
were positivelyassociated with earlyadoptionand were not
relatedto T1 power. Our results were consistent with the ex-
tensive diffusion-of-innovation literature.When innovation
was counter to system norms, persons not well integrated
into the system tended to be earlyadopters (Rogers, 1971).
Althoughit would seem likelythat industry-levelcompe-
tence-destroyingtechnologicaldiscontinuitieswould violate
intraorganizationalnorms, it is possible that organizational
norms differentiateearly-and late-adoptingorganizations.
Tushmanand Anderson (1986) suggested that if compe-
tences will be destroyed, an organizationis unlikelyto be an
earlyadopter. Mitchell's(1989) findingsindicatethat the
greater the competitive threat, the less likelyan industryin-
cumbent is to enter a new technicalsubfield but the earlierit
123/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
 will do so if it does planto enter. However, a competence-
destroyingchange may not be inconsistentwith the organiza-
tional norms of an early-adoptingorganization.Forexample,
Mitchellsuggests that an organizationmay possess industry-
specializedsupportingassets. In such a situation,the techno-
logicalchange may result in a majorshift in competitive
advantagewithinthe industrybut no majorshift in power and
structurewithinthe early-adoptingorganization.Conversely,a
competence-enhancingindustrydiscontinuitymay result in
majorchanges in power and structurewithina late-adopting
organization.Inthis case, the industrystructurewould not
change, but the organizationalstructureof the late-adopting
organizationwould change.
Thus, it is possible that organizationalnorms ratherthan tech-
nologicaldiscontinuitiesmay be more predictiveof intraor-
ganizationalchange. Althoughspeculative,we cannot ruleout
this possibilityin that this study involvedboth a competence-
destroyingdiscontinuityand the late adoptionby the agency.
The possibilityis also consistent with Barley's(1986) finding
that the same technologicalchange had differenteffects on
differentorganizations.
We are suggesting that future studies adopt a multilevel
strategy. Industry-levelcompetence-enhancingtechnological
discontinuitieswill be consistent with the norms of a majority
of industry-leading organizations.The powerfulorganizations
will be the first to adopt and thereby maintainand increase
their industryposition. Likewise,the centralpowerfulindi-
vidualswithinthese leadingorganizationswill maintainand
increase their intraorganizational positions.
The opposite will occur when the industry-leveltechnological
discontinuityis competence-destroying.This change will run
counterto the norms of industry-leading organizations,re-
sulting in shifts in competitiveadvantagewithinthe industry
and shifts in power and structurewithin the previouslypow-
erful but late-adoptingorganizations.At the same time, this
competence-destroyingdiscontinuitywill be consistent with
the norms of a small minorityof new and/orinnovativeorga-
nizationswithinthe industry.While individualpower shifts
withinthese early-adoptingorganizationswill not occur, these
organizationswill acquirepower withinthe industry.
We have attempted to providea frameworkfor examiningthe
process of technologicalchange. Althoughcautionshould be
used in generalizingthe specific results, we believe that the
proposed change model is applicableto other types of tech-
nologicalchange and to organizationalchange in general. We
expect earlyadopters to increase or reinforcetheir influence
and centralityby virtueof their abilityto cope with the uncer-
tainties created by the change. Whetherthese increases re-
sult in majorchanges in the existing structureand power
distributionwill depend on the match between earlyadoption
and established power and communicationchannels.
Overall,this study highlightsthe importanceof investigating
an organizationover time. The relationshipbetween a tech-
nologicalchange and structureand power may have been
interpretedquite differentlyif one had investigateda cross-
sectional pictureof organizationalprocesses. Instead,this re-
search illuminatedhow technology is diffused and how this
124/ASQ, March 1990

This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions

REFERENCES
Bandura, Albert A.
1986 Social Foundations of Thought
and Action. Englewood Cliffs,
NJ: Prentice-Hall.
Barley, Stephen R.
1986 "Technology as an occasion
for structuring: Evidence from
observations of CT scanners
and the social order of radi-
ology departments." Adminis-
trative Science Quarterly, 21:
78-108.
Berger, Peter L., and Thomas
Luckmann
1967 The Social Construction of Re-
ality: A Treatise in the Soci-
ology of Knowledge. Garden
City, NY: Anchor Books.
Blau, Peter M., Cecilia McHugh
Falbe, William McKinley, and
Phelps K. Tracy
1976 "Technology and organization
structure in manufacturing."
Administrative Science Quar-
terly, 21: 20-40.
Boje, David M., and David A.
Whetten
1981 "Effects of organizational
strategies and constraints on
centrality and attributions of
influence in interorganizational
networks." Administrative
Science Quarterly, 26:
378-395.
Brass, Daniel J.
1981 "Structural relationships, job
characteristics, and worker
satisfaction and performance."
Administrative Science Quar-
terly, 26: 331-348.
1984 "Being in the right place: A
structural analysis of individual
influence in an organization."
Administrative Science Quar-
terly, 29: 518-539.
1985 "Technology and the struc-
turing of jobs: Employee satis-
faction, performance, and
influence." Organizational Be-
havior and Human Decision
Processes, 35: 216-240.
This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
Patterns of Change
diffusionprocess affects structureand power. Froma longi-
tudinalperspective, two areas remainfor furtherresearch.
First,an investigationof the decision-makingprocess priorto
the change is needed. As Tushmanand Anderson(1986)
suggested, those who controlthe decision-makingprocess
may controltheir own and their organization'sfuturepower.
Secondly, furtherresearchis needed on the process by which
changes in technology, structure,and power become institu-
tionalized.Willthe changed distributionof power and struc-
ture continue to reinforceitself, or will the former(priorto
change) relationshipsresurfaceto challenge the newly estab-
lished patterns?Thus, while evidence of change is apparent,
questions on the institutionalization of change are yet to be
determined.
Bruns, WilliamJ., and John H. Fennell, Mary L., and RichardB.
Waterhouse Warnecke
1975 "Budgetarycontroland organi- 1988 The Diffusionof MedicalInno-
zationalstructure."Journalof vations. New York:Plenum
AccountingResearch,13: Press.
177-203. Freeman, LintonC.
Burt, Ronald S. 1979 "Centralityin social networks:
1982 Towarda StructuralTheoryof Conceptualclarification."So-
Action.New York:Academic cial Networks,1: 215-239.
Press. Fry, Louis W.
1987a "Socialcontagionand innova- 1982 "Technology-structure
tion: Cohesionversus struc- re-
search: Threecriticalissues."
turalequivalence."American Academyof Management
Journalof Sociology,92: Journal,25: 532-552.
1287-1335.
1987b STRUCTURE. TechnicalRe- Galbraith,Jay R.
port#TR2, ResearchPrograrr 1977 OrganizationDesign. Reading,
in StructuralAnalysis,Center MA:Addison-Wesley.
for SocialSciences, Columbia Hage, Jerald, and Michael M.
University.
Aiken
Campbell, Donald T., and Julian C. 1969 "Routinetechnology,social
Stanley structureand organization
1963 Experimentaland Quasi-Exper- goals." AdministrativeScience
imentalDesigns for Research. Quarterly,14: 366-376.
Chicago:RandMcNally. Hickson, David J., ChristopherR.
Chandler,Alfred D., Jr. Hinings, Charles A. Lee, Rodney E.
1977 The VisibleHand:The Mana- Schneck, and Johannes M.
gerialRevolutionin American Pennings
Business. Cambridge,MA: 1971 "A strategiccontingencies
BelknapPress. theoryof intraorganizational
power."Administrative
Comstock, Donald E., and W. Science Quarterly,16:
RichardScott 216-229.
1977 "Technologyand the structure
of subunits: Distinguishingin- Hickson, David J., Derek S. Pugh,
dividualand workgroupef- and Diana C. Pheysey
fects." AdministrativeScience 1969 "Operations,technologyand
Quarterly,22: 177-202. organizationstructure:An em-
piricalreappraisal."
Adminis-
Crozier,Michel, and ErhardE. trativeScience Quarterly,14:
Friedberg
1980 Actorsand Systems: The Pol- 378-397.
itics of CollectiveAction.Chi- James, Lawrence R., and Allen P.
cago: Universityof Chicago Jones
Press. 1976 "Organizational structure:A
review of structuraldimen-
Dow, Malcolm, M., Michael L. sions and theirconceptualre-
Burton,and Douglas R. White lationshipswith individual
1982 "Networkautocorrelation:A attitudesand behavior."Orga-
simulationstudy of a founda- nizationalBehaviorand Human
tionalproblemin regression
Performance,16: 74-113.
and survey research."Social
Networks,4: 169-200.
125/ASQ, March 1990
Katz, Ralph
1980 "Time and work: Toward an
integrative perspective." In
Barry M. Staw and L. L. Cum-
mings (eds.), Research in Or-
ganizational Behavior, 2:
81-127. Greenwich, CT: JAI
Press.
Katz, Ralph, and Michael Tushman
1979 "Communication patterns,
project performance, and task
characteristics: An empirical
evaluation and integration in an
R&D setting." Organizational
Behavior and Human Perfor-
mance, 23: 139-162.
Kenny, David A.
1979 Correlation and Causality. New
York: Wiley.
Kipnis, David
1984 "Technology, power, and con-
trol." In Samuel B. Bacharach
(ed.), Research in the Soci-
ology of Organizations, 3:
125-156. Greenwich, CT: JAI
Press.
Knoke, Davis, and Ronald S. Burt
1983 "Prominence." In Ronald S.
Burt and Michael J. Minor
(eds.), Applied Network Anal-
ysis: 195-222. Beverly Hills,
CA: Sage.
Krackhardt, David
1989 "Structure, cognition and
power in organizations."
Working paper, Johnson Grad-
uate School of Management,
Cornell University.
Krackhardt, David, and Lyman W.
Porter
1985 "When friends leave: A struc-
tural analysis of the relation- Rice, A. Kenneth
ship between turnover and
stayers' attitudes." Adminis-
trative Science Quarterly, 30:
242-261.
1986 "The snowball effect: Turnover
embedded in communication
networks." Journal of Applied
Psychology, 71: 50-55.
Lincoln, James R., and Jon Miller
1979 "Work and friendship ties in
organizations: A comparative
analysis of relational net-
works." Administrative
Science Quarterly, 24:
181-199.
MacEvoy, Bruce, and Linton C.
Freeman
1986 UCINET:A Microcomputer
Package for Network Analysis.
Irvine, CA: University of Cali-
fornia.
Mensch, Gerhard
1979 Stalemate in Technology: In-
novations Overcome the De-
pression. Cambridge, MA:
Ballinger.
This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
Miles, Raymond E., and Charles C.
Snow
1986 "Organizations: New concepts
for new forms."California
ManagementReview, 28:
62-73.
Mintzberg,Henry
1979 The Structureof Organizations.
EnglewoodCliffs,NJ: Pren-
tice-Hall.
Mitchell,Will
1989 "Whetherand when? Proba-
bilityand timingof incum-
bents' entryinto emerging
industrialsubfields."Adminis-
trativeScience Quarterly,34:
208-230.
Mohr, Lawrence B.
technologyand
1971 "Organizational
organizational structures."Ad-
ministrativeScience Quarterly,
16: 444-459.
1987 "Innovationtheory:An as-
sessment fromthe vantage
pointof the new electronic
technologyin organizations."
InJ. M. Penningsand A. Bui-
tendam (eds.), New Tech-
nologyas Organizational
Innovation:The Development
and Diffusionof Micro-elec-
tronics: 13-31. Cambridge,
MA: Ballinger.
Perrow, Charles
1967 "Aframeworkfor the compar-
ative analysisof organiza-
tions." AmericanSociological
Review, 32: 194-208.
Pfeffer, Jeffrey
1981 Power in Organizations.
Marshfield,MA: Pitman.
1958 Productivity and SocialOrgani-
zation,The AhmedabadExper-
iment. London:Tavistock.
Roberts, KarleneH., and Charles A.
O'ReillyIll
1979 "Some correlatesof commu-
nicationroles in organizations."
Academyof Management
Journal,22: 42-57.
Daniel
Robey,
1981 "Computerinformation
systems and organization
structure."Communicationsof
the ACM,24: 679-687.
Rogers, Everett M.
1962 Diffusionof Innovations.New
York:Free Press.
1971 Communicationof Innovations.
New York:Free Press.
Rogers, Everett M., and Rekha
Agarwala Rogers
1976 Communicationin Organiza-
tions. New York:Free Press.
126/ASQ, March 1990
Rosenberg, Nathan
1972 Technologyand American
EconomicGrowth.Armonk,
NY:M. E. Sharpe.
Rotter,Julian R.
1966 "Generalizedexpectancies for
internalversus externalcontrol
of reinforcement."Psycholog-
ical Monographs,1: 80.
Rousseau, Denise M.
1979 "Assessment of technologyin
organizations:Closed versus
open systems approaches."
Academyof ManagementRe-
view, 4: 531-542.
Salancik, Gerald R., and Jeffrey
Pfeffer
1977 "Whogets power-and how
they holdonto it: A strategic
contingencymodel of power."
Organizational Dynamics,5:
3-21.
Shaft, Teresa
1986 "Computerattitudescale."
Unpublishedmanuscript,De-
partmentof Management
Science, The Pennsylvania
State University.
Susman, Gerald I.
1976 Autonomyat Work:A Socio-
technicalAnalysisof Participa-
tive Management.New York:
Praeger.
Taylor, Robert L., and James M.
Utterback
1975 "A longitudinalstudy of com-
municationand research:
Technicaland managerialinflu-
ences." IEEETransactionson
EngineeringManagement,
EM-22:80-87.
Tichy, Noel M.
1981 "Networksin organizations."
In PaulC. Nystromand William
H. Starbuck(eds.), Handbook
of Organizational Design, 2:
225-249. New York:Oxford
UniversityPress.
Tichy, Noel M., and Charles
Fombrun
1979 "Networkanalysisin organiza-
tionalsettings." HumanRela-
tions, 32(11): 923-965.
Thompson, James D.
1967 Organizationsin Action.New
York:McGraw-Hill.
Tushman, Michael L.
1979 "Workcharacteristicsand sub-
unitcommunicationstructure:
A contingencyanalysis."Ad-
ministrativeScience Quarterly,
24: 82-98.

Tushman, Michael L., and Philip
Anderson
1986 "Technologicaldiscontinuities
and organizationalenviron-
ments." Administrative
Science Quarterly,31:
439-465.
Tushman, Michael L., and Elaine
Romanelli
1983 "Uncertainty,social location
and influencein decision
making:A sociometricanal-
ysis." ManagementScience,
29: 12-23.
This content downloaded from 128.163.7.131 on Tue, 18 Aug 2015 14:43:55 UTC
All use subject to JSTOR Terms and Conditions
Patterns of Change
Van de Ven, Andrew, Andr6 Weiss, Robert S., and Eugene
Delbecq, and RichardJ. Koenig Jacobson
1976 "Determinantsof coordination 1955 "A method for the analysisof
modes withinorganizations." the structureof complex orga-
AmericanSociologicalReview, nizations."AmericanSociolog-
41: 322-338. ical Review, 20: 661-668.
Weick, KarlE. Whisler, Thomas L.
1969 The Social Psychologyof Or- 1970 The Impactof Computerson
ganizing.Reading,MA:Ad- Organizations.New York:
dison-Wesley. Praeger.
Woodward, Joan
1965 IndustrialOrganization:Theory
and Practice.New York:Ox-
ford UniversityPress.
APPENDIX: Contagion Analysis
Normative response data were generated using the following network auto-
correlation model (Burt, 1987b):
Xi* = Wj1X1 + Wj2X2 * * * + Wrn,
where x; is the date of adoption for person j, and wj, are network weights re-
flecting the extent to which some person i is structurally equivalent with j.
The normative response is thus a predicted date of adoption based on the
date of adoption of other workers whereby those individuals who are more
structurally equivalent with the focal individualare assumed to be more influ-
ential in terms of adoption timing. Thus, the normative response of j is
simply the weighted average of observed responses by people defining the
social context of j's responses (Burt, 1987b). The network weight, wp.,equals
(dmax,,- d,1)I[E
(dmax1- d,1)],i = j, where d,1is the distance between the
pattern of i's relations and the pattern of j's relations and dmaxi is the largest
distance between person j and anyone in the system (Burt, 1987b).
For T1, the correlation between observed and normative responses was
.126. The correlation between observed and normative responses was .313
for T4. However, these ordinary least squares estimates of contagion effect
are not maximum likelihood (Burt, 1987b), thus they cannot be assessed with
routine statistical tests (see Burt, 1987b; Dow, Burton, and White, 1982, for
further explanation of the statistical problem). Therefore, to determine the
contagion effect's significance, jackknife subsampling results were obtained
by using Burt's (1987b) jackknife analysis in his network program, STRUC-
TURE.
Jackknife analysis entailed using the observed data distribution to construct
a sampling distribution for the contagion effect in order to draw statistical
conclusions about its magnitude (Burt, 1987b). The autocorrelation, r, of ob-
served and normative data was recomputed N times from the data without
each actor j. Each generation of the network autocorrelation varies from the
overall estimate as a function of individual responses to contagion. The dif-
ference between the complete and subsample results is represented by the
following difference between the estimates: r*i = Nr - (N - 1)ri, which
provides an estimate of the contagion effect in the deleted observation (Burt,
1987b). The mean of these values is the jackknife estimate, r*, of the net-
work autocorrelation model.
A t-test with N-1 degrees of freedom is carried out for the null hypothesis of
no contagion effect, whereby t* = r*ls*. The Fisher's Z transformation is
also calculated, because such a transformation is more nearly normal than
the raw correlations (Burt, 1987b).
127/ASQ, March 1990