DOWNS & MOHR - Conceptual Issues in The Study of Innovation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Conceptual Issues in the Study of Innovation

Author(s): George W. Downs, Jr. and Lawrence B. Mohr


Source: Administrative Science Quarterly, Vol. 21, No. 4 (Dec., 1976), pp. 700-714
Published by: Sage Publications, Inc. on behalf of the Johnson Graduate School of Management, Cornell
University
Stable URL: http://www.jstor.org/stable/2391725 .
Accessed: 22/11/2013 02:20

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 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
ConceptualIssues in the This paper discusses a number of complex and currently
unresolved conceptual issues arising in research on inno-
Study of Innovation vation in complex organizations. In an effort to linkthe
issues they are approached through an exploration of the
George W. Downs, Jr. factors responsible for instability in empirical findings.
and Four separate sources of instability are defined and their
Lawrence B. Mohr theoretical and methodological implications are treated at
some length.
The analysis of the relative adoptability of innovations
and the innovativeness of organizations are found to be
related by mirror-image theoretical symmetry. The four
sources of instability are seen to have the same implica-
tions for the development of a theory of adoptability.
Seven prescriptions for research on innovation are
suggested. Although this study suggests how one might
arrive at a general theory of innovation, it does not actu-
ally construct any specific theory.
Innovationhas emerged over the last decade as possibly the
most fashionableof social science areas; a frequentlyshort-
livedand perhapsdubiousdistinctionheld in recent memory
by small group research. Likethe latter,the study of innova-
tion has not been confinedto any single disciplinebut is being
explored in fields as diverse as anthropologyand economics.
This popularityis not surprising.The investigationsby innova-
tion researchof the salient behaviorof individuals,organiza-
tions, and polities can have significantsocial consequences.
The latterimbue even the most obscure piece of research
with generalizabilitythat has become rareas social science
becomes increasinglyspecialized.The act of innovatingis still
heavilyladenwith positive value. Innovativeness,likeeffi-
ciency, is a characteristicwe want social organismsto pos-
sess (cf. Nelson, 1972: 39). Unlikethe ideas of progress and
growth, which have long since been casualties of a new
consciousness, innovation,especiallywhen seen as more
than purelytechnologicalchange, is still associated with im-
provement.
Unfortunately,the theoreticalvalue of the researchthat has
been done is problematic.Perhapsthe most alarmingcharac-
teristic of the body of empiricalstudy of innovationis the
extreme varianceamong its findings,what we call instability.
Factorsfound to be importantfor innovationin one study are
found to be considerablyless important,not importantat all,
or even inverselyimportantin anotherstudy. This phenome-
non occurs with relentless regularity.1One should certainly
expect some variationof results in social science research,
but the recordin the field of innovationis beyond interpreta-
tion. Inspite of the largeamount of energy expended, the
results have not been cumulative.This is not to say that the
body of existing researchis useless; when organizedproperly,
when we know just how to examine it, it may well consti-
tute a powerfulsource of evidence bearingon important
theoreticalelements. But progress towardwhat has been
called integrativetheory (see Rogers and Shoemaker, 1971:
Of 38 propositionsbearingdirectlyon the
act of innovationcited by Rogersand
346) will have to be made before much of this data will be
Shoemaker(1971:350-376),34 were useful.
supportedin some studies andfound to
receive no supportin others.The4 prop- This troublinginstabilityresults from a lackof clarityon sev-
ositionswith a consistent recordwere eral conceptual issues. We will therefore elucidate certainkey
treatedin veryfew studies.
December1976, volume21 700/AdministrativeScience Quarterly

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

conceptual difficultiesin research and theory, showing their


necessary effect on the stabilityof research results from
study to study, and suggest what must be done to make
researcheffort more cumulative.Ourstudy focuses on inno-
vation in complex organizations.We define four primary
sources of instability:(1)variationamong primaryattributes,
(2) interaction,(3) ecological inferences, (4) varying
operationalizationsof innovation;and we offer seven
methodologicalprescriptionstowardthe development of an
integrativetheory. We will be employingthe ratherbroad,
conventionaldefinitionof innovationas the adoptionof means
or ends that are new to the adoptingunit.Althoughwe will
mainlydiscuss the question of the relativeinnovativenessof
organizationsratherthan the relativeadoptabilityof different
innovations,we will suggest that the two are closely related
methodologicallyand that this relationshipdescribes an im-
portantsymmetry.
We would liketo emphasize that we are not constructinga
specific theory of innovation,but are describinghow one
might be arrivedat throughresearchand the roughform a
general theory of innovationmight take.

TYPOLOGIESOF INNOVATIONS
The purpose of investigatingthe reasons some organizations
are more likelythan others to adopt a given innovationhas
been to develop a generaltheory of innovation.Because
previous researchindicatesthat determinantsimportantfor
one innovationare not necessarily importantfor others,
these efforts have not yielded such a theory. In industry,
for example, the large,successful firms have been found to
lead in adoptingsome innovations(Mansfield,1963), while
the relativelyunsuccessful firms were clearlythe leaders with
respect to others (Adamsand Dirlam,1966). Inthe public
sector, Becker (1970: 275) found that the publichealth offi-
cials leadingin the adoptionof measles immunizationwere
young, urban,liberal,and cosmopolitan,while the pioneers in
the adoptionof diabetes screening were old, rural,conserva-
tive, and parochial.The instabilityof the determinantsfrom
case to case frustratestheory-buildingefforts.
Perhapsthe most straightforwardway of accountingfor this
empiricalinstabilityand theoreticalconfusion is to rejectthe
notionthat a unitarytheory of innovationexists and postulate
the existence of distincttypes of innovationswhose adoption
can best be explainedby a numberof correspondinglydistinct
theories (cf. Rowe and Boise, 1974: 289-290). These theories
may includedifferentvariables,or they may containthe same
explanatoryvariableswhile positingdifferentinterrelation-
ships among them and differenteffects upon the dependent
variable.The existence of empiricallydistinguishable
categories of innovationsand theirassociated models would
help to explainwhy studies employingroughlythe same
predictorsachieve widely varyingR2s and why the explana-
tory power of individualvariablesis unstableacross them.
Eventhe suggestion thata single theoryand set of determi-
nants are applicableto the entireset of newly implemented
techniques, programs,rules, and norms that are lumped
underthe genericheadinginnovations shouldbe considered
701/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
suspect. While most students of.innovationmight agree with
this statement in the abstract,its implicationshave rarely
been considered in either researchor theory building.Most
emphasis has been placed on uncoveringthe effect of cen-
tralization,executive ideology,or professionalization.Sum-
maries are presented of those attributesof organizationsor
individualsassociated with innovativeness but are unaccom-
panied by any hintthat their impacts may depend on the
innovationstudied. When findingssurroundingthe impactof
a variableon innovationare contradictory,the common reac-
tion has been to view the matteras yet undecidedand call for
furtherstudy, ratherthan to explore the divergentstudies in
search of linkagesbetween the kindsof results obtainedand
the kindsof innovationsconsidered.
Examplesof potentiallyfruitfulcategorizationschemes or
typologies of innovationsare plentiful.Forinstance, the de-
terminantsof high-cost innovationswould seem to be mark-
edly differentfrom those of low-cost innovations.Wealthor
resources would clearlypredictthe formerdifferentlyfrom
the latter.Resource levels might explainmuch of the variation
in the adoptionof elaborate EDPsystems by corporationsbut
be relativelyunrelatedto the same corporations'adoptionof
new supervisorytechniques or job enrichmentprograms.
Similarly,variationsin the communicabilityof an innovation
would be expected to affect the impactof professionalismor
cosmopoliteness on adoption.It may therefore not be merely
coincidentalthat the role of resources and cosmopoliteness
has varieda great deal from one study to another(see Rogers
and Shoemaker,1971: 361, 369).
Primaryattributes of innovations. A typologydividinginno-
vations into groups or categories must itself be based on a
characteristicor attributeof the innovation.Forexample, it
might be based on the cost of the innovation(highor low), or
on its communicability(a simple innovationor a complicated
one). Such characteristicsmay be either primaryor secon-
dary,and this distinctionhas importantimplicationsfor how
typologies should be employed and what may be their rela-
tionships to theory building.James Jeans (1966: 196) con-
cisely explainsthe usage of these terms as follows:
Galileo,Descartes, Locke,and others dividedthe qualitiesof objects and
substances intotwo classes which Lockedesignatedas primaryand secon-
dary.Secondaryqualitiesare those which are perceivedby the senses, and
so may be differentlyestimated by differentpercipients;primaryqualitiesare
those which are essential to the objector substance and so are inherentin it
whether they are perceivedor not.
When a typology is based on a primaryattribute,an innova-
tion can be confidentlyclassified without reference to a
specified organization.Regardlessof its size, wealth, com-
plexity,decentralization,and so forth,each organizationwould
place the innovationin the same cell of the typology.Thus,
both richand poor and largeand small organizationswould
describe the innovationas type A. However, when a typology
is based on a secondaryattribute,the classificationof the
innovationdepends on the organizationthat is contemplating
its adoption.Largeorganizations,for example, might classify
it as type A while small ones might classify the same innova-
tion as type B.
Decidingwhether or not there are purelyintrinsic(primary)
characteristicsof innovationsthat are logicallyindependentof
702/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

the organizationswhich might adopt them is not essential for


our purpose. What matters instead is constancy withinan
individualstudy. We define an attribute(forinstance, cost) as
primaryin a given piece of researchwhen there would be
essentially no variationamong the organizationsstudied in
categorizingthe innovationon that attribute-for example, it
is a low-cost innovationto all.This lackof variationleads to a
restrictedabilityto generalizefrom a given study of innovation.
The existence of such attributesis less relevantto the design
of researchthan to its interpretation.The consequences for
the evaluationof researchand for theory buildingare great,
since at least some instabilitycan be explainedby differences
among innovationsin theirprimaryattributes.Forexample,
some findingsof researchinto the determinantsof high-cost2
innovationsare generalizableonly to other high-cost innova-
tions. Variationamong studies in the cost of the subject
innovationwill cause variationamong the coefficients of cer-
tain independentvariables,and there is littlethat can be done
about this instabilityother than to understandand accept it.
There would be no point in tryingto generalizeabout the
impactof a variablelikewealth on innovativeness because
wealth would not have a constant impact. Itis no doubt an
importantdeterminantof the adoptionof high-cost innova-
tions but may have no bearingat all upon the adoptionof
low-cost innovations.The challenge to the theorist is, by
comparingstudies of disparateinnovations,to determine the
coefficient on wealth as a functionof cost, so that, given the
cost of the innovation,theory is definite.
Inadditionto cost, there may be several other characteristics
of innovationsthat are candidatesfor the basis of a primary-
attributetypology.There is the danger in all scientific research
that such constants place unknownlimits upon generalizabil-
ity. Moreover,it is an obstacle to science that the constants
that matterare generallydifficultto divine,so that the limits
often go unrecognized.
Variationof a characteristicof an innovationbetween one
study and anotherbut not withinthe respective studies is
almost certainlyan importantsource of instabilityin innova-
tion research. If such a characteristicwere a true, Lockean
primaryattribute(thatis, actuallyintrinsicto the innovation),
then this instabilitycould not be resolved. We would need as
many theories of organizationalinnovativenessas there were
categories of such primaryattributes.However, if the charac-
teristicwere in trutha secondary attribute,differenttheories
would not be needed. Itwould be possible to manage the
resultinginstabilitywith a general theory by postulatingin-
teractionamong independentvariables.Fortunately,most if
not all characteristicsupon which one might consider basing a
typologyturnout to be secondary attributesof innovations.
Secondary attributes of innovations. Whilethere have been
few attempts to categorize innovationsby primarycharac-
2 teristics, several social scientists have suggested typologies
We use discretecategoriessuch as based on secondary attributesof innovations.The rationale
"low-cost"and "high-cost"as a matterof
terminologicalconvenience.Actually,cost
behindthese categorizationschemes is identical,in that the
is a continuumand its impactis such that determinantsof adoptionare differentfor differentcategories
there is a slightlydifferentcurvilinear
rela- of innovations.Some examples fromthe literatureare prob-
tionshipof wealth,for example,to innova-
tionat each pointon the continuum. lem versus slack, routineversus radical,variationversus
703/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
reorientation,majorversus minor,and means versus ends
(see Cyertand March,1963; Knight,1967; Normann,1971;
Wilson, 1966). Whatthese categorizationschemes have in
common is that the same innovationmay be classified in
differentcategories for differentorganizations,for example,
an innovationmight be seen as minoror routineby some
organizationsbut as majoror radicalby others. When we
thinkin terms of a continuousdimension, such as the extent
to which an innovationis compatiblewith the organization's
present mode of operation,the amount of varianceacross
organizationsis even more conspicuous. Winter(1968) has
made the pointforcefullythat an innovationis rarelythe same
thing to two organizations.Secondaryattributesare no doubt
more numerous by farthan primaryattributes.
Because an innovation'sclassificationin a secondary-attribute
typology can varyfrom organizationto organization,we face a
situationwhere two or more theories are presumablyapplica-
ble to th6 adoption of the same innovation. If researchers
ignorethis fact and proceed with analysis,the resultwith
respect to a single innovationwould often be similarto arbi-
trarilytakingan average of the findingsamong high-costand
low-cost innovations.Thatis, if differentorganizationsin a
study cause the same innovationto be classified differently,
the coefficient of an independentvariablethat possesses
differentialimpactwithintwo categories of the typology
would be a misleadingaverage of the values obtainingin each
category.The value that actuallyemerges from a regression
analysiswould therefore depend upon the proportionof or-
ganizationsin the sample that considered the innovation
major(ora reorientation,and the like)as opposed to those
that considered it minor.As to correlationcoefficients, which
are frequentlyemployed, the mixingor superimposingof one
causal functionupon anotherin this fashion almost always
acts to increase the varianceof the dependent variable(inno-
vation)relativeto the independentvariable(forinstance,
executive ideology),so that the correlationcoefficient will
almost always be smallerthan it would be in any single
category of the typology.This may possibly help to show why
the explanatorypower of some variablesthought to be
theoreticallyrelevanthas been low.
The existence of secondary-attributetypologies would appear
to have implicationsfor researchdesign. The crucialdiffer-
ence between secondaryand primaryattributes-that any
secondary attributeof an innovationmay varyfrom organiza-
tion to organization-is not necessarilya liability.It indicates
that we must buildthe idea of statisticalinteractioninto our
models of innovation.When we recognizethat differentor-
ganizationsclassify the same innovationinto different
categories, and also that determinantsvaryin existence or
strength depending upon the category into which the innova-
tion is classified, we are by these very facts recognizingthe
existence of interaction.Further,the interactionmay be built
into our designs and used to advantage.
Suppose that executive ideology is a more importantdeter-
minantof the adoptionof reorientationinnovationsthan it is
of variations.Ifwe are studyingan innovationthat would be a
reorientationfor some organizationsand a variationfor others,
all we need do is insert a variablewhich measures how
704/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

compatible3the innovationis to each organizationin the sam-


ple. This would provideus with even more informationthan
we would obtainfrom separate studies of variationsand
reorientations.Specifically,we can obtainthe followingfour
types of information.(1) By includingthe organization'scom-
patibilitywith the innovationas a separate independentvari-
able, we can determinethe extent to which this in itself
affects the adoptionof the innovation-note that to make
compatibilityan independentvariablewe must transformit
from a propertyof the innovationto a propertyof the organi-
zation,and that this can easily be done. (2) By employingone
or more interactionterms-for example, the productof com-
patibilityand executive ideology-we can determinethe dif-
ferentialimportanceof a variable,such as executive ideology,
for innovation,given any particularlevel of compatibility.(3)
We can also investigate which characteristicor set of charac-
teristics of the organizationdetermines the classificationof
the innovation.(4) If,for the sake of discussion, size is found
to play such a determiningrole,then we mightalso investi-
gate the interactionof size and executive ideology as predic-
tors of innovation.This may be an interestingand prescrip-
tivelyvaluablefindingthat could not be expected to emerge
from the separate study of variationsand reorientations.
There is no reason for developingseparate models of innova-
tion for differentcategories of secondaryattributes.Instead,
the secondary attributemay simplybe measured as an or-
ganizationalpropertyand then used in the ways described.
Shouldthere be more than one secondaryattributeresponsi-
ble for variationamong the determinantsof adoption-and
this may well be true, given the numberof typologies that
have been suggested-there is of course the requirement
that new interactionterms be added and that, therefore,
sample sizes be larger.The consequence of hypothesizing
interactionin this fashion should be an increased understand-
ing of the determinantsof innovation,greaterexplanatory
power, and less randomness in the appearanceof results.
Secondary attributes of organizations. Just as many
characteristicsturnout to be secondary attributesof innova-
tions, so also may a great manywell knownvariablesfunction
as secondary attributesof organizations.Centralization, for
instance, is a loose term for the degree to which authorityand
influence regardingdecisions are consolidated in highereche-
lons ratherthan being spread among lower ones. It is com-
monly considered to be a general propertyof an organization
and it used to be measured in global, nonspecificterminology
(see for example, Hall,1963). However, organizationtheorists
have long since recognizedthat centralizationvaries within
organizationsas well as between. It is still includedin analysis
as though it were an organization-wideproperty,but it is
measured by aggregatingthe responses to questions or ob-
3
servations about several differentkindsof decisions-
has been
The attribute,"compatibility," personnel, purchasing,and so forth(Pugh,et aL.,1968: 102-
used extensivelyin researchas a deter- 104; Child,1972; Negandhiand Prasad,1971). Thus, it is
minantof the rateof adoptionof innova- recognizedin measurement proceduresthat centralizationdif-
tions (Rogersand Shoemaker,1971:
145-154).Its meaningis actuallyquite fers depending upon what is being decided, and this observa-
close to Normann's(1971)"variations-
reorientations,"but perhapsnot precisely tion clearlyextends into the realmof innovation.Fora variety
the same. We use the two dimensions
interchangeably simplyforterminological
of reasons, in any given organizationsome innovationdeci-
convenience. sions mightbe made at a high level and others at a lower
705/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
one; some mightbe takenwith broadorganizationalinput
and some with the involvementof only one or two persons.
Centralization,in sum, is a secondary attribute.Its measured
value depends upon the innovationbeing considered. The
same is true of complexity,formalization,integration
(conflict-resolutionmechanisms), specialization,professionali-
zation,traditionalism,and no doubt many other factors that in
some sense help to determine innovation(see Hage and
Aiken, 1967; Zaltmanet al., 1973: 134-154).
THE INNOVATION-DECISION DESIGN
It is apparentthat much of the conceptualand methodological
complexitysurroundingthe issue of instabilityis brought
about by the existence of secondary attributes.Because this
category of variablesseems to be composed of a large por-
tion of the most importantpredictorsof innovation,the
strategy of increasingstabilityby simply eliminatingsecon-
daryattributesfrom our models would hardlyappeara viable
one.
One way of coming to grips with secondary attributesis to
thinkof them not as being composed wholly of characteris-
tics of the innovationor the organizationbut as characterizing
the relationshipbetween the two. The unitof analysis is no
longerthe organizationbut the organizationwith respect to a
particularinnovation,no longerthe innovation,but the innova-
tion with respect to a particularorganization.Neitherthe
organizationnorthe innovationwould be describedas com-
patible,for example, but ratherthe pairtaken in conjunction.
Fromthis perspective, secondary attributescan be viewed as
variablesthat characterizethe circumstancessurroundinga
particulardecision to innovate.
Inorderto integratethis perspective into the research on
innovation,it is helpfulto employ an innovation-decisionde-
sign, a considerationof the unit of analysis as an organization
in relationto an innovation.Ifwe were studyingthe adoption
of 10 innovationsby 100 organizations,we would be working
with a sample of 1,000. This design eliminates any confusion
that might stem fromvolatilesecondary attributes.Itwould
no longerseem inappropriate to assign a particularorganiza-
tion one set of scores with respect to its perceived profit,
level of centralization,formalization,and so forthin connec-
tion with innovationX and anotherset of scores for these
same dimensions in connectionwith innovationY. As for
primaryattributes,the organization'sscores would by defini-
tion remainconstant across the set of innovations.4
A benefit of employingthe innovation-decisiondesign is that
4 it serves to remindus of the dangers of thinkingin terms of
Forexample,the organization mightscore something called the organization,a reificationwith constant
largewith respectto all innovationsin the
study. Butgeneralizabilityneed not be propertieswhose probabilityof adoptionnever varies, regard-
completelyconstrainedeven here be- less of the kindof innovationit considers. Rather,it focuses
cause we mightalso have a smallorgani-
zationin the study.facingthe same or a our attentionon the shiftingincentives and constraintsthat
similarset of innovations.It is as though are relevantto the decision to innovate.
we conductedsome studies of largeor-
ganizationsand some of smallones and
then comparedthe results.The same The innovation-decisiondesign is an extension of the single-
logicwould obtainwith respectto primary innovationdesign. It has all of the advantages of the latter,
attributesof innovations.Providingthis
sort of varietywithina single study is one plus the greatergeneralizabilitythat results from considering
way in which the innovation-decision de- many innovationsratherthan one. Furthermore,we will show
sign is an improvementover the ordinary
single-innovation research. that it connects the study of innovationwith the study of
706/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

adoptabilityin a mannerthat is both illuminatingand efficient.


We therefore highlyrecommend its adoptionfor innovation
research.
MULTIPLE-INNOVATION RESEARCH
Fromthe standpointof theory buildingand conceptual issues,
what are the advantages and disadvantagesof treatingsev-
eral innovationsat once, not as in the innovation-decision
design, but as an aggregate?The strategy has been
employed in scores of studies(such as Mansfieldj1963;Hageand
Aiken, 1967; Mohr,1969; Walker,1969, 1971), the apparent
motive being to increase the reliabilityand generalizabilityof
findingsover a single-innovationstudy. But does it work?
Ina multiple-innovation study, we encounter dependent and
independentvariablesin a form such as "numberof
technologicalinnovationsadopted" and "averageperceived
profitability," respectively.The only clue indicatingthe extent
to which the differentialprofitability of a given innovationfor
variousorganizationsdetermines the orderin which they
adopt it is the relationshipof these two aggregate measures.
The clue mightor might not be a good one, depending upon
certainpropertiesof theory and data; if it is not good in a
certaincase, if the inference from the collective or ecological
level to the individuallevel is either biased or intolerably
variable,then the inference is often termed an "ecological
fallacy."Ininnovationresearch,this kindof cross-level infer-
ence is very likelyto be incorrect.There are two reasons for
this. One is that, given the interactionswe have discussed
and that almost certainlyexist, the impactof a variablelike
profitabilityupon individualinstances of innovationvaries sub-
stantiallyfrom one organizationto another.This connotes the
ecological difficultygenerallycalled aggregationbias (see
Theil, 1971: 556-566). The other is that the groupingby
organizationof the values of independentvariablessuch as
expected profitabilityis almost necessarily inefficient;it
would be most efficient if all of the highlyprofitableexpecta-
tions were concentratedin some organizationsand the unpro-
fitableexpectations in others, ratherthan a mixturein each
(see Praisand Aitchison,1954; Cramer,1964; and Hannan
and Burstein,1974). The bias and variabilityresultingfrom
these sources can be quite largeand may either aggrandizeor
understatethe true coefficient. We can be misled easily in
innovationresearch by aggregate statistics.
It should be noted that an ecological fallacycan occur only
when the independentvariableinvolvedis a secondaryattri-
bute of the organization,one whose value differsfrom one
innovationto anotherso that averagingthese values becomes
necessary. Profitabilityand similarmotivationalforces are ob-
viously in this category,as are variableslikeorganizational
compatibility,the organizationalcounterpartsof importantat-
tributesof innovations.We have seen that centralizationand
complexityare secondaryattributesas well, even though they
might not be recognizedas such in measurement procedures.
So are certainresources relevantto innovation,such as pro-
fessionalizationand expertise. Thus, correlationand regres-
sion coefficients involvingsuch variableswill be unstable in
multiple-innovationdesigns; they will1fluctuate mysteriously
aroundthe true microlevelvalues that they are supposed to
707/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
represent. But these variablesare extremely importantfor the
development of theories of innovation.We cannot relysolely
on primaryattributessuch as raw organizationalwealth and
the age and cosmopoliteness of the chief executive. Since
secondary attributesmust be prevalentin broadlyapplicable
models, interpretationbased on aggregate data will be
grossly misleading.
A question remains,however, whether this difficultymight
not be offset by the gains in generalizabilitythat are realized
when multiple-ratherthan single-innovationdesigns are
used.
It is not meaningfulto offset an evil of one kindwith a good
of another.The centraldisadvantageof the multiple-innova-
tion design is so serious as to be incapacitating.Further-
more, generalizability is not better attainedby the multiple-
innovationdesign. The dependent variable,aggregate adop-
tion of a mixtureof innovations,generallyrepresents a large
varietyof values on both primaryand secondary attributesof
the innovationsconsidered,all blended together so as to
obscure totallythe special implicationsof each. There is in
fact nothingto generalizeto. We learnnothingabout varia-
tions and reorientations,high and low cost, and so forth.The
innovation-decisiondesign, on the other hand, does ac-
complishwhat has been sought in multiple-innovation re-
search. It is a combination,in a sense, of the single- and
multiple-innovation designs. It uses many innovationsand so
has substantialgeneralizability,but it does not aggregate
them. Itconsiders them as discrete units, thereby preserving
the special theoreticalimplicationsof each.
Some might considerthat studyingthe aggregate adoptionor
nonadoptionof a numberof new ideas is a reasonablesubsti-
tute for studyingextensiveness of commitment to any given
new idea. Thatis clearlynot the case. Itis quite possible for
organizationsto adopt a great many innovationsat the most
superficiallevel. Infact, in at least one empiricalstudy (Mohr,
1969: 121), this propensitywas found to exist among the
largerorganizations.
It mightalso be considered that aggregate adoptionor
nonadoptionmay substitute for time-of-adoptionas a depen-
dent variableor that one can at least inferrelativeearliness to
adopt from relativenumberof innovationsadopted. This is no
more true of multiple-than of single-innovationstudies. If
organizationA has adoptedn innovationsand organizationB
has adoptedn+2, we know only that B has adopted two
innovationsearlierthan A,-but nothingabout the time at
which either adopted the othern. Ifwe were awardingpoints
for earliness of adoption,A could accumulate many more
points than B by havingadopted the n innovationsearlier.We
should not hastilyclassify as earlyadopters organizationsthat
have adopted many innovations.
We conclude that the multiple-innovation design is undesira-
ble for investigatingthe basic theoreticalquestion "Whatare
the determinantsof the adoptionof an innovation?"Despite
its drawbacks,however, it might be considered that the
multiple-innovation design is uniquelycapableof tappingan
entirelydifferentdimension of innovation.Considerthe
aggregated or macrolevelmeasure of innovativenessas a
708/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

dependent variablein its own right,ratherthan as a substitute


for some microlevelconcept. There does, after all, appearto
be something differentabout adoptinglots of innovationsas
opposed to adoptingone, and we would benefit from a model
that predictsthe formeras well as the latter.Again,however,
a multiple-innovation study is not the appropriateway to
accomplishthis end. Itpresumes that the adoptionof one
blended set of innovationsfollows the same laws as the
adoptionof another,regardlessof the kindsof innovations
includedin each. The assumption is not valid.Shouldwe
actuallywish to predictthe adoptionof a given set of innova-
tions, the correct method is to plug the appropriatevalues of
primaryand secondary attributesfor each innovationinto a
statement of the theory and then to aggregate the individual
equations.
Thus, we findthat the multiple-innovation design solves no
problems in the study of innovation,but only compounds
conceptualdifficulties.

THE MEASUREMENTOF INNOVATION


Anotherpotentialsource of instabilityin innovationresearch
involves the operationalization of the dependent variable.Just
as the results of studies of differentkindsof innovationsare
commonly considered together in general discussions of
what determines innovation,so are the results of studies in
which quite differentbehaviorsare being explained.There are
three principal,interrelatedoperationalizationsof innovation
found in the literature.The first, and by farthe most common,
is the assignment to each organizationof an innovationscore
based on its time of first adoptionor use. The second is a
simple, dichotomous adoptionor nonadoption.This may be
seen as merelya crude measure of the time of adoption,
allowingmanyties. The thirdoperationalization is determined
by the extent to which an organizationhas implementedan
innovation,or the degree to which an organizationis commit-
ted to it.
The prevalenceof the first operationalization over the third
can probablybe attributedto the relativeease with which
pertinentdata can be gathered and to the lightshed by its
analysis on the natureof the diffusionprocess. It is consider-
ablyeasier to learnapproximatelywhen a physicianor hospi-
tal first used an experimentaldrug,or when a legislature
passed affirmative-actionlegislation,than it is to discover the
proportionof eligiblepatients who received the drugor the
proportionof jobs filledin accordancewith the new regula-
tion. Inaddition,innovationresearch has traditionallyman-
ifested great interest in the mannerand rate by which an
innovationis spread througha population.A knowledge of the
orderin which members of a populationfirst employed an
innovationis a necessary prerequisiteto inferences about
theirsearch patterns,communicationprocesses, reference-
groupbehavior,and so forth.Extensiveness of use contrib-
utes littleto such discussion.
On the other hand, it is often the case that operationalizing
innovationby the extent of implementationcomes closer to
capturingthe variationsin behaviorthat we reallywant to
explain.Whileit is useful to knowwhat determinedthe se-
709/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
quence in which states first experimented with a new hybrid
strain of corn, it is much more desirable that the researcher
uncover what determines variation across states in the extent
to which the hybrid strain has replaced the traditional strain
(see Griliches, 1957). Of course, there are some innovations
in which virtually no variation in the extent to which they are
adopted is possible. An individual either purchases soft con-
tact lenses or does not, a city fluoridates its water or it
doesn't. In the cases where such variation is possible, it is
frequently this variation rather than time of first usage that we
would like to account for.

The two operationalizations are functionally related only if we


assume that all units first employ the innovation at the same
level and that each then increases its utilization at the same
rate. This may be true of some portion of innovations, but we
suspect that the assumption is generally tenuous. Few would
agree to the even weaker generalization that the first organi-
zation to adopt an innovation will be the one found to have
implemented it to the greatest extent at any point in the
diffusion process; most students of innovation are familiar
with a number of cases where early adopters do not have the
highest degrees of commitment to the innovation.

Reinforcing this belief that the two operationalizations are


tapping different aspects of innovation is the suspicion that
the determinants of the time of adoption are not the same as
the determinants of the depth of adoption. For example,
because a certain amount of prestige is known to be ex-
tended to organizations and individuals who are among the
first to adopt new innovations, we might hypothesize that the
desire for prestige will be a more powerful predictor of the
time of adoption than of the extent (cf. Mohr, 1969: 121-123).
We might also expect organizational slack to be a better
predictor of how quickly an organization adopts than of how
extensively the innovation is employed (Cyert and March,
1963: 278-279), something that may depend heavily on moti-
vational variables. This is the same sort of differential impact
of organizational variables that is responsible for instability in
connection with primary characteristics, and the effect is the
same here. Further, unless a clear and consistent relationship
can be established between the two operationalizations,
which is unlikely, the results of research on them are simply
irreconcilable. It would be wise to conceive of the two
operationalizations as two different behaviors to be explored,
not because they are invariablyindependent of one another,
but because of the need to avoid biased, muddled generaliza-
tions.

If one attempts to ignore the instability and generalize impres-


sionistically about the impact of a specific variable across a
number of studies that operationalize innovation differently,
the result will be an unenlightening average that would de-
pend on the proportion of studies having employed each
operationalization. Since the major proportion of studies have
used time of adoption as the operationalization of innovation,
that interpretation would be emphasized. This would no doubt
inflate the role of variables related to organizational slack and
a desire for prestige.
71 O/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

THE ADOPTABILITY OF INNOVATIONS


We have simplifiedour study by consideringone important
researchaim in detailwhile neglecting anotherone, that is,
we have postponed the question of the relativeadoptabilityof
innovations.Ourjustificationis that what holds for the first
should hold for the second by mirror-imagesymmetry;the
two theories are not independentbut are essentially two
sides of the same coin.
What makes one innovationmore likelythan anotherto be
adopted by any given organization?
Just as there may be typologies of innovations,so may there
be typologies of organizations,and these may also be based
either on primaryor secondary attributes.Primary-attribute
typologies of organizationswould functionhere just as
primary-attribute typologies of innovations.Just as organiza-
tionalwealth would undoubtedlybe a better predictorof the
adoptionof a high-cost than a low-cost innovation,so would
the cost of innovationsbe a better predictorof theiradoption
in a poor organizationthan a richone. Wealthwould therefore
appearto be a primaryattributeupon which organizations
should be classified for development of a general theory of
adoptability.How importantthis primaryattributeis and
whether there are others of importance(forinstance, public
versus privatesector) is a question that must be addressed in
future research.There has been littlework bearingupon the
question, and the complexitiesthat it raises are formidable
enough that it is difficulteven to volunteera speculative
response.
The study of innovationmeets the study of adoptabilityin the
concept of secondaryattributes.The secondary attributesof
innovationsare essentially the same phenomena as the sec-
ondaryattributesof organizations.Forexample, just as the
technicaland proceduralrequirementsof the innovationare
highly(or minimally)compatiblewith the structureand ac-
tivities of the organization,so is the organizationby that fact
precisely as compatiblewith the innovation.Inthe same
sense, what we have called an organization'sbenefits are
identicalto what is calledthe relativeadvantageof an innova-
tion (Rogersand Shoemaker, 1971: 138-145). By the nature
of secondary attributes,we always referto only one organiza-
tion and one innovationat a time, and once the pairis joined,
propertiessuch as expected benefit and relativeadvantage
are identical.
The possibilityof a varietyof operationalizationsof the de-
pendent variablehas identicalsignificancein researchon
adoptabilityas on innovation.We are studyingdifferent,
though overlapping,phenomena depending upon which de-
pendent variableis employed, and there is no a priorireason
to believe that the explanatorytheories of the three will be
congruent.
Shallwe use one organizationor many in our researchon
adoptability?This is a most interestingquestion, largelybe-
cause "many"has been the answer given by every inves-
tigatorwho has publishedon this subject, to the best of our
knowledge. This differs markedlyfrom the study of innova-
tion, where the single-innovationdesign has been used fre-
quently(hybridcorn, Gammanym,and so forth).Itwould be
711/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
difficultto say why this difference in practiceexists, perhaps
because there has frequentlybeen a strong practicalinterest
in the diffusionof a particularinnovation,such as hybridcorn,
whereas there has rarelybeen a-comparableinterest in the
innovativenessof a particularorganization.Whateverthe
reason, it would not appearadequate to justifythe common,
multiple-adopterstudy in which both independentand depen-
dent variablesare aggregate measures.
Allof the difficultieswith multiple-innovationstudies applyto
the multiple-organization design in adoptabilityresearch.
Once again, the problemsof secondary attributesand ecolog-
ical inferences are of particularconcern. We simply cannot be
certainfrom a strong correlationbetween "averagecomplex-
ity rating"and "proportionwho have adopted" that those
who did not find the innovationcomplex are the same as
those who adopted. These studies have frequentlyused cor-
relationdesigns, and the obtainedcorrelationsare in general
quite high (Fliegeland Kivlin,1966; Petrini,1966; Singh,
1966). But correlationsbased on aggregated measures are
notoriouslyinflatedover the true microlevelrelationships
(Cramer,1964). The less consistency there is among potential
adopters in their ratingsof a trait-such as the complexityof
an innovation-the less confidence we should have in the
strength of the correlationinvolvingthat trait.
The symmetryand close relationamong the theories of inno-
vationand adoptabilityare best demonstratedby considering
the innovation-decisiondesign, the design in which the sam-
ple size is equal to the numberof innovationstimes the
numberof organizations.Assume that we wish to study both
innovationand adoptability.Because the measured secondary
attributesof the innovationsare identicalto those of the
organizations,these need be measured only once. Givena
single data set based on the innovation-decisiondesign, we
can study adoptabilitysimply by eliminatingthe primaryattri-
butes of organizations;reinsertingthose primaryattributes,
we can study innovationby eliminatingthe primaryattributes
of innovations.
We suggest, then, that both single-organizationand
innovation-decisiondesigns should be affordedextensive trial
in the study of adoptability.

CONCLUSION
We conclude by offeringa set of prescriptionsfor researchon
innovationsuggested by our analysis.
1. Use studies of different innovations to expose the impact of primary-
attribute variation on models of innovation. This will involve observing and
reporting the primaryattributes of innovations and restricting generalizations
from a given study to innovations in the same category of a primary-attribute
typology rather than expecting all results to be identical.
2. Measure the secondary attributes of innovations (compatibility, relative
advantage, and so forth) with respect to each organization and consider them
as characteristics of adopters.
3. Use interactive models. In terms of the development of integrated theory,
this would probably be the single most important departure from current
practice.
4. Use the innovation-decision design as the basis for analysis. It is essen-
tially a single-innovation design, which explicitly recognizes that a great many

712/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Conceptual Issues

organizational characteristics are secondary attributes, that is, they can be


validly measured only in relation to a particularinnovation.
5. Do not conduct multiple-innovation studies in which the organization is
assigned an aggregate score for innovation.
6. Recognize that extent of adoption and time of adoption are distinct con-
ceptualizations of innovation. Do not generalize from one dependent variable
to the other. Do not use either as a comprehensive measure of innovative-
ness. There is not a single, unitary theory, but rather different theories to
explain different aspects of innovation.
7. Study the adoptability of innovations by using either many innovations in
relation to one single organization or by using the innovation-decision design.
George W. Downs, Jr., is an assistant professor in the
Department of Political Science, University of Californiaat
Davis. Lawrence B. Mohr is an associate professor in the
Department of Political Science and an associate research
scientist in the Institute of Public Policy Studies, Univer-
sity of Michigan.

REFERENCES
Adams, Walter, and Joel B. Dirlam Griliches, Zvi Menzel, Herbert,and ElihuKatz
1966 "Bigsteel, invention,and in- 1957 "Hybridcorn: an exploration in 1955 "Socialrelationsand innova-
novation."QuarterlyJournalof the economics of technologi- tion in the medicalprofession:
Economics,80: 167-89. cal change." Econometrica, the epidemiologyof a new
Becker, MarshallH. 25: 501-22. drug."PublicOpinionQuar-
1970 "Sociometriclocationand in- Hage, Jerald, and Michael Aiken terly, 19: 337-352.
novativeness: reformulation 1967 "Program change and organi-
and extension of the diffusion zational properties: a compara- Mohr, Lawrence B.
model."AmericanSociological tive analysis." American Jour- 1969 "Determinantsof innovationin
Review,35: 267-82. nal of Sociology, 72: 503-519. organizations."AmericanPolit-
Child,John icalScience Review,63:
1972 "Organizational structureand Hall, Richard H. 111-1 26.
strategies of control:a replica- 1962 "The concept of bureaucracy:
tion of the Aston study."Ad- an empirical assessment."
American Journal of Sociology, Negandhi, Anant R., and S. Ben-
ministrativeScience Quarterly, jamin Prasad
17: 163-77. 69: 32-40.
1971 ComparativeManagement.
Coleman, James, ElihuKatz,and Hannan, Michael T., and Leigh New York:Appleton-
HerbertMenzel Burstein Century-Crofts.
1957 "Thediffusionof an innovation 1974 "Estimation from grouped ob-
among physicians." servations." American
Sociometry,20: 253-70. Nelson, RichardR.
Sociological Review, 39: 1972 "Issues and suggestions for
Cramer,J. S. 374-392. the study of industrialorgani-
1964 "Efficientgrouping:regression zationin a regime of rapid
and correlationin engel curve Jeans, James H. technicalchange." In Policy
analysis."Journalof the 1966 Physics and Philosophy. Ann Issues and ResearchOppor-
AmericanStatisticalAssocia- Arbor: University of Michigan tunitiesin IndustrialOrganiza-
tion, 59: 233-250. Press. tion,V. Fuchs (ed.). National
Cyert, RichardM., and James G. Knight, K. Bureauof EconomicResearch.
March 1967 "A descriptive model of the New York:ColumbiaUniver-
1963 A BehavioralTheoryof the intra-firminnovation process." sity Press.
Firm.EnglewoodCliffs,N.J.: Journal of Business, 40:
Prentice-Hall. 478-496. Normann, R.
1971 "Organizationalinnovative-
Downs, George W., Jr. Mansfield, Edwin ness: productvariationand
1976 Bureaucracy,Innovation,and 1963 "The speed of response of reorientation."
Administrative
PublicPolicy.Lexington, firms to new techniques." Science Quarterly,16: 203-
Mass.: D. C. Heath. Quarterly Journal of 215.
Economics, 77: 290-309.
Fliegel, FrederickC., and Joseph E.
KivIin Menzel, Herbert Petrini,Frank
1966 "Attributesof innovationsas 1966 "Innovation, integration, and 1966 The Rateof Adoptionof
factors in diffusion."American marginality: a survey of physi- Selected Agricultural
Innova-
Journalof Sociology,72: cians." American Sociological tions. Reprint53. Uppsala:Ag-
235-248. Review, 25: 704-713. ricultural
Collegeof Sweden.
71 3/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions
Prais, S. J., and J. Aitchison Rowe, LloydA., and William B. Warner, Kenneth E.
1954 "The grouping of observations Boise 1974 "The need for some innova-
in regression analysis." Re- 1974 "Organizational innovation: tive concepts of innovation: an
view of the International currentresearchand evolving examination of research on
Statistical Institute, 22: 1-22. concepts." PublicAdministra- the diffusion of innovations."
tion Review,34: 284-392. Policy Sciences, 5: 433-451.
Pugh, D. S., D. J. Hickson, C. R. Singh, Ram N. Wilson, James Q.
Hinings, and C. Turner 1966 Characteristicsof FarmInno- 1966 "Innovation in organization:
1968 "Dimensions of organization vationsAssociated with the notes toward a theory." In
structure." Administrative Sci- Rateof Adoption.Reprint14. Approaches to Organizational
ence Quarterly, 13: 65-1 06. Guelph:OntarioAgricultural Design, James D. Thompson
ExtensionEducationReprints. (ed.). Pittsburgh: University of
Roessner, J. David Pittsburgh Press.
1974 Designing Public Organizations Theil, Henri
for Innovative Behavior. Paper 1971 Principlesof Econometrics. Winter, Sidney G., Jr.
delivered at the Thirty-fourth New York:John Wiley. 1968 Toward a Neo-Schumpeterian
Theory of the Firm. Publication
Annual Meeting of the Walker,Jack L. No. P-3802. Santa Monica:
Academy of Management, 1969 "Thediffusionof innovations Rand Corporation.
Seattle, Washington. amongthe Americanstates.
AmericanPoliticalScience Re- Zaltman, Gerald, Robert Duncan,
Rogers, Everett M., and F. Floyd view, 63: 880-899. and Jonny Holbek
Shoemaker 1971 "Innovationsin state politics." 1973 Innovations and Organizations.
1971 Communication of Innovation: InPoliticsin the American New York: John Wiley.
A Cross-Cultural Approach. States-A Comparative
New York: The Free Press. Analysis,2nd ed., Herbert
Jacobs and KennethVines
(eds.). Boston: Little,Brown
and Co.

714/ASQ

This content downloaded from 130.194.20.173 on Fri, 22 Nov 2013 02:20:55 AM


All use subject to JSTOR Terms and Conditions

You might also like