HD28
.M414
no.wi
ALFRED
P.
WORKING PAPER
SLOAN SCHOOL OF MANAGEMENT
Technological Communities and the
Diffusion of
Michael A. Rappa
Massachusetts Institute of
Technology
December 1991
Knowledge
Koenraad Debackere
Gent
Rijksuniversiteit
Sloan
WP# 3391-92
MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
Massachusetts Institute of Technology
Technological Communities and the
Diffusion of
Knowledge
Michael A. Rappa
KoenraaiJ Debackere
Massachusetts Institute of
Rijksuniversiteit
Gent
Technology
December 1991
Forthcoming
©
in
Sloan
R&D Management,
VOL.
22,
WP # 3391-92
NO.
3
1991 Massachusetts Institute of Technology
Sloan School of
Management
Massachusetts Instirjte of Technology
50 Memorial Drive, E$2-538
Cambridge,
02139-4307
MA
GULY
1992).
.IT LIBRARIES
^AR
3 1992
Technological Communities and the
Diffusion of
Knowledge
Michael A. Rappa and Kocnraad Dcbackere'
Massachusetts Institute of Technology
December 1991
ABSTRACT
The external acquisition of technological knowledge is a central theme in
technology management. In this paper it is argued that
research on
may provide a useful unit of analysis to study
information and knowledge exchange among scientists and engineers
working on a particular research agenda. Based on a worldwide survey of
technological communities
more than 2,000 individuals engaged
in the research
and development of
neural network technology, the dynamics within that particular community
are explored. The primary focus
and
is
to
compare the
characteristics
of academic
of their
industrial researchers, with special attention given to the timing
entry into the field.
1.
INTRODUCTION
Understanding the external acquisition of technological knowledge has become an
important theme for
of recent trends
in
many
R&D
students of the
warrant
this attention;
namely, the increasing
specialization of research; the internationalization of
organizational
research
links
A number
management of technology and R&D.
R&D;
the growing
number of
(Williams and Gibson, 1990). As a consequence, the location of research
increasingly "fluid," both in an operational
and engineering
same things and
and organizational
becoming
is
national
to
and
and
is
becoming
and to companies that make the needed investments
in science
1990). Generic technological knowledge
capabilities. Well-trained engineers
will
activities
sense, as research contacts
expand from localized research systems
international networks (Howells,
easily accessible to nations
inrer-
and ventures; and the potential impact of computer
communication networks and information technology on the conduct of research
administrative boundaries
and
scale, intensity
communicate with each other
and
scientists will
regardless of
know
where they
roughly the
are located
(Baumol, 1990; Nelson, 1990a).
'
Michael Rappa
is
doctoral fellow at
assistant professor
MIT
and
of management
at
the Sloan School,
MIT. Kocnraad Debackere
is
a Fulbright post-
with the Vlerick School voor Management, Rijksuniversitcit Gent. An
the 1991
Conference in Kiel, Germany G"ly 8-10). The authors are
a research associate
RADMA
of this paper was presented at
two anonymous referees for helpful comments and to Thomas Allen, Edward Roberts and Roland Van
Dierdonck for their support and encouragement. Kocnraad Debackere was supported, in part, by a doctoral fellowship
from ICM, Brussels.
earlier version
grateful to
An
R&D
industrial
laboratory faces a dual challenge.
internally, trying to build
knowledge. This
development
own
its
On
the one hand,
it
looks
competitive advantage based on proprietary technological
the well-entrenched notion of the "local" character of technology
is
On
(Allen, 1977).
the other hand, the industrial laboratory looks toward the
external world to monitor developments that might yield opportunities or threats to the
Monitoring the external technological environment requires an active
firm.
the firm.
The
on behalf of
role
participation of researchers in national and international networks, their
mobility and their information exchange behavior are key ingredients in this outward
perspeaive (Pavitt, 1991; Rothwell and Dodgson, 1991).
In the
find
it
wake of
these findings,
it
has also
become
most firms
increasingly apparent that
block information from flowing to their competitors:
difficult to
What may
be more surprising, it appears that in many cases firms do not try
block information flow, and in others actively support it by
encouraging employees to publish, to talk at technical society meetings,
to
(Nelson, 1990b).
etc.
The
question
Why
is:
firm can strengthen
knowledge from
to be
its
would firms behave
its
in
such a manner.' Contrary to the notion that a
competitive advantage by carefully protecting
environment, Nelson claims, "There are industry-wide efficiency gains
had by sharing technology. Everyone would be
1990b). But
it
technological
its
better off if everyone shared"(Nelson,
not because sharing increases industry gains that individual firms are
is
inclined to share. Rather,
it
is
much more
likely that firms share competitively relevant
technical information as a form of quid-pro-quo (von Hippel, 1988). Nonetheless, fostering
technical information flows
may
offer a
number of
additional advantages to the firm even
without requiring reciprocity.
Indeed,
if
one wants
information. This
is
to stake claims to
newly developed knowledge, one has
the intent of the patent system. At the same time, this disclosure
enables the firm to attract customers, to enhance
compete
for
capabilities,
competent
and
scientists
its
to attract the interest of investors.
(so that the
image
as a
technological leader, to
and engineers, to inform suppliers about
information through a publication or presentation
of knowledge
to release
work can be
is
technological
its
Moreover, although the
often viewed as an act of
easily replicated
release of
full
by other researchers),
disclosure
this
is
rarely
the case. Even with purely scientific work, seldom does an article disclose everything one
needs to
know
information
is
in
order to repeat the experiment (Collins, 1982). In some cases the
simply too complex or requires too
much
detail in order for
it
to be fully
Rappa and Debackere (December, 1991)
detailed in a journaJ article or presentation.
disclosure occurs because a researcher
is
However,
other cases, the lack of complete
in
not necessarily motivated to provide
full details, at
not until the claims to his ideas can be adequately proteaed.
least
It
three
is
interesting to note that academic
means
— patenting,
publishing,
and
industrial researchers aJike are
and presenting
— almost
employing
simultaneously
all
when
attempting to protect claims to their research findings. For example, the aaivity in high-
temperature superconductors and cold fusion reveals more than one incidence where a
researcher ajinounced his
applied for a patent
all
work
conference, submitted a paper for publication, and
at a
within days,
not hours, of one another. Moreover, firms that do
if
publish research typically have an internal review process for "scrubbing clean" a manuscript
before
it
is
sent to a journal editor or presented at a conference, in order to
does not contain any information that management might view
make
certain
as proprietary. In this
firms can receive the benefits that flow from divulging information without
it
way,
compromising
confidentiality.
It
is
obvious that
external technological
emphasis on assessing and monitoring the firm's
this increasing
environment
calls
for empirical research that will illuminate the
fijndamentaJ dynamics of the extra-organizational environment in fostering technological
development.
2.
The concept of the
technological
TECHNOLOGICAL COMMUNITIES
A
number of
AS
community
precisely attempts to
this.
A LEVEL OF ANALYSIS
scholars have recently pointed to the influence of communities of
researchers in shaping technological progress (Constant, 1980;
study, the technological
are
do
community
working on an interrelated
is
set
community
1988). In this
defined as the group of scientists and engineers,
of technological
organizationally and geographically dispersed but
other. In particular, the
Thomson,
level
who
who
problems and who may be
nevertheless
communicate with each
of analysis allows us to focus on the extra-
organizational environment in studying technology development, as well as the actors
shaping
this
development.
The community
First, it
level
of analysis complements current research on technological change.
turns attention away from the organizational or project levels of analysis, which have
dominated
in past studies
of the firm. Second,
it
of technological development, toward the external environment
focuses on the actors of technology development. Although
Rappa and Dehackere (December, 1991)
economists and management scholars have demonstrated an interest in the process of
technological innovation, their focus has mainly been on contextual factors influencing
innovative performance. As a consequence, the aaors involved in the process have received
much
less
community concept
anention. In the same vein, the
forms of interorganizational linkages, such
information exchange behavior
understanding
need
this
behavior
to investigate to
among
imperative
is
as joint-ventures
turns anention
and
alliances, in
technologists themselves.
if
one wants to
what extent communities
capitalize
on
We
away from
order to study
conjecture that
In other words,
it.
all
are relevant loci of technological
we
knowledge
and information.
The analogy with
the sociological writings on the functioning of scientific communities
(Hagstrom, 1965; Crane, 1972; Hull, 1988)
community
not the same
is
community can be
as a scientific
community. For instance, a technological
truly interdisciplinary in nature; whereas, a scientific
demarcated
by highly specific
communities
are loosely-coupled systems in
even thousands
be obvious. However, a technological
will
—of
disciplinary
boundaries.
community
Moreover,
(i.e.
is
technological
which the micro-motives of hundreds
individuals converge to one macro-objective
often
— perhaps
solving the problems
related to a particular technology). Expressed in an alternative way, just as "organizations are
a means of achieving
collective action in situations in
which the price system faib" (Arrow,
1974), technological communities could be hypothesized as a means of achieving collective
aaion
in situations in
which the organization
fails.
Focused attention and shared values are
two potential ingredients preventing the system from breaking down (Orton and Weick,
1990).
Given the
development,
scarcity of empirical data
this
study
community with respea
3.
sets
out to examine
to the diffusion of
how
role
of communities
in
technological
researchers function within a panicular
knowledge.
RESEARCH METHOD AND DATA
In this paper
we examine some of
the behavioral charaaeristics of researchers within
technological communities. In panicular,
in
on the
we
investigate the similarities
and the differences
information exchange behavior for two important subsets of a community: academic
researchers
and
industrial researchers.
Much
scholarly writing has focused
that exist between academic and industrial research.
We
existence of differences between both types of institutions.
on the
differences
do not deny the likelihood of the
However,
it
has
become
generally
accepted that both academic and industrial research have a prominent role in knowledge
Rappa and Debackere (December, 1991}
development, even
the realm of technological activity (Swann,
in
1988; Jaffe,
1991). Therefore, a detailed empirical analysis of both groups
Mansfield,
1989;
may
yield
valuable insights into the dynamics of this process.
Given the actor-oriented nature of the
community of
research, our first choice
researchers. Following sociologists' advise
research site" (Bijker et
community encompasses
community was
and engineers working on the
scientists
to find a relevant
on the selection of
1987), the neural network research
al.,
was
set
a "strategic
chosen. This
of scientific and
technological problems related to the development of a paradigm fijndamentally different
from
traditional
von Neumann computing. The
today controversies persist
as to the feasibility
field has
known
a turbulent history, and even
and ultimate usefulness of neural network
computing (Minsky and Papen, 1988; Papert, 1988).
For the two-year period from 1988 and 1989, over 3,000 neural network researchers
were identified worldwide through a
carefijl analysis
journal articles, conference proceedings, and books).
specify the
exaa address
for
of the materials they published
From
their decision
(b)
ultimately lead
their
them
demographic
large-scale survey,
to
start
(a)
their current neural
neural network research;
we made
Given the
able to
network
factors that
difficulty in closely following-up
might
and
(e)
on such
a
use of elearonic mail bulletin boards in order to reach neural
Finally, since
37 researchers had more than one address during the
period considered, a total of 2,074 questionnaires were sent out.
a total of
162 were returned undelivered.
among
37
the
(c)
to leave the field; (d) their information exchange behavior;
characteristics.
network researchers.
we were
2,037 researchers from 35 different countries. These researchers
were sent a twelve-page questionnaire inquiring about
activities;
this material
(i.e.
researchers with
None
more than one
Of the
2,074 questionnaires,
of the undelivered questionnaires were
address.
A
final
response rate of
38.4% (720 of
the 1875 questionnaires presumed delivered) was obtained.
A number
of comparisons were carried out to see whether serious differences existed
between the survey population and the respondent sample. The
are
reponed
in Tables
1
and
2. In a
results
of these comparisons
geographic comparison of survey respondents and the
survey population, there appears to be an adequate representation of the entire population.
Researchers based in the U.S. are very heavily represented in both the respondent sample and
the survey population. In an institutional comparison between the respondent sample and
population, researchers were classified into three categories: academic, industrial and other.
The
last
category
is
composed of
researchers in a disparate collection of organizations,
Rappa and Debackere (December, 1990
including a
number of government managed and/or funded
universities.
Given the wide range of
countries,
it is
institutional
difficult to
types;
institutions not based at
institutional variation across the nearly three
begin analyzing this group without a more
therefore,
we
will
difference appears between the respondent sample
Geographic Region
carefiil classification
limit the present analysis to a
respondents employed in academic and industrial institutions.
No
and the population.
dozen
of
comparison of
statistically significant
have a doaorate (21 out of 89, or 23.6%). For the majority of "engineers"
opposite
is
true (19 out of 25, or
Overall, respondents
out of 400, or 7%).
The
who
76%, do not hold
classify
themselves
among
charaaer of the new technology. (Note:
when considering
doctoral degree or
Highest
is
a doaorate).
as
engineers are an absolute minority (28
vast majority of respondents hold a doctoral degree (335 out of
400, or 83.8%). Thi^ pattern of education
scientific
in industry, the
the complete sample,
in the process
i.e.
of obtaining
researchers
this level
81.8% of
one).
is
certainly an indication of the
of education
is
not
much
different
the complete respondent set holds a
Some
final
demographic sector comparisons are given
of the respondent;
(b) the
number of years
neural network technology (EXPNN); and
(EXPJOB). This
experience
respondent's graduation.
technology and the
It is
last
Table
in
5.
These
are: (a) the
the respondent has been involved with developing
(c)
the respondent's
variable was defined
number of years of professional
as
the time elapsed since the
obvious that the number of years involved with neural network
respondent's
professional
experience
will
not
be
completely
independent. At the same time, though, they need not be strongly correlated.
respondents
may indeed have
graduating, while others
age
Some
entered the neural network field only several years after
may have
entered long before graduating. (In faa, rexpnn.expjob=0-42;
p<.001.)
Besides the faa that industrial researchers are, on average, younger than their academic
colleagues,
While
we
also find their
involvement with neural network technology to be more recent.
their professional experience
difference in
EXPNN
is
quite similar to that of academic researchers, the
interesting. Indeed, the majority
is
into the field only after
it
began to expand rapidly and, presumably, had become more
legitimate. Prior to the early 1980s neural networks
to find researchers
researchers
who
of industrial researchers entered
was judged harshly, and
it is
not unusual
continue to have serious doubts (Papert, 1988). Only a handful of
worldwide were willing to pursue a neural network research agenda during the
"wilderness years," as they are called in a recent report
industrial researchers have
been present throughout,
it
increasingly significant subset of the total neural network
(DARPA, 1988). Thus, although
has only been since 1984 that an
community
has started to form.
year as a critical point in the community's evolution and have used
respondent sample into early and
late entrants.
A
it
sensitivity analysis (using discriminant
analysis techniques) indicates that the choice of a cut-ofFyear to separate the
robust as long as
it
to separate the
two groups was
occurred somewhere between 1980 and 1985.
Further analyses of variance, using sector of employment and highest degree
independent variables, show that respondents
neural networks for a longer period of time
who hold
a
doaorate have been involved
(EXPNN comparison,
F(l, 389)=4.
1,
as
in
p<.05) and
have had more years of professional experience (EXPjOB comparison, F(l, 399)=4.9, p<.05).
No
significant interaaion effects
were found.
100.0
712
75.0
n
c
E
3
2
50.0
S
3
E
3
u
FIGURE
25.0
1:
Cumulative distribution of entries into neural network research hy
survey respondents
4.
INFORMATION AND KNOWLEDGE EXCHANGE AMONG
NEURAL NETWORK RESEARCHERS
In this section
we compare
the information exchange behavior of academic and
industrial researchers in the neural
between early and
late entrants
is
network community.
When
warranted, a distinction
also taken into account.
Rjippa
and Debackere (December, 1991)
10
4.
1
Communication behavior: general remarks
.
The
questionnaire investigated a
number of
issues related to the respondent's inclination
to share information with the rest of the neural
network community. The diffusion
mechanisms include attending conferences, publishing
as directly
seem odd
made
communicating with
publicly, patent applications
is
original (unadjusted) results for a
Indicator
However,
do provide a means
information. As outlined, the main focus
The
applying for patents,
researchers in other organizations.
to include patent applications in the analysis.
researchers.
articles,
for
At
first
glance,
as well
it
may
to the extent that they are
communicating technical
on comparing academic and
number of variables
are
shown
in
industrial
Table
6.
11
not attain
statistical significance at aJl.
Thus,
in
terms of information acquisition by means
of conference attendance and professional association membership, researchers
show
similar behavior.
However, when
it
comes
both sectors
to information diffusion via publishing,
presenting, and/or patenting the differences that could be
Indicator
in
expeaed
are
bom
out.
12
industry apply for
more patents (p<.001), while
more conference
publications (p<.001),
With rcspea
attendance (p=.019).
academic counterpans repon more
presentations (p=.025), and
more frequent conference
membership, no
to professional association
Thus, when controlling for education
significant differences are found.
with respea to seaor differences shown
4.2
their
in
Table 7 are
statistically
our findings
level,
clarified fijrther.
Amount and diversity of external communications
The number
of hours spent talking to researchers outside their organization does not
differ significantly for
(1.40 hours per week,
academic (1.61 hours per week, N=256) and industrial researchers
N = 102).
Controlling for the
of education does not significantly
level
DoCTorates in universities spend 1.63 hours per week talking with researchers
alter the results.
outside their institution, while doctorates in industry spend an average of 1.36 hours
Non-doaorates
in universities
spend an average of 1.72 hours per week, compared with 1.36
hours for their industrial counterparts
(n.s.).
Moreover, when taking the other independent
we
variables into account as well as the respondent's professional experience,
significant
exists
main
effects
within the
communication
(n.s.).
find neither
nor interaction effeas. Thus, a remarkable degree of homogeneity
community when comparing
the
and
across institutional sectors, early
amount of
extra-organizational
late entrants, doctorates
and non-
doaorates, and respondents with differing degrees of professional experience.
The
respondents were further asked to categorize the
organization with
results
whom
external
is,
outside their
they regularly confer about neural network related issues.
of a Mann-Whitney non-parametric
seaor difference; that
number of researchers
test,
shown
in Figure 2, reveals
no
The
significant
academic and industrial researchers report the same number of
communication partners (Mann- Whitney U=l4l4l, z= -1.54,
n.s.).
As shown
in
Figure 3, the entry period variable yields a statistically significant difference (nonparametric
Mann-Whitney U=10,500,
communications
(as
measured
in
Of
course,
it
if
the
amount of outside
hours per week spent talking to external researchers)
similar for both groups, the external
the early entrants (where the
z= -4.02, p<.001). Thus, even
communication partners seem
number of communication panners
could be argued that
this
effea
is
is
to be
more
anributable to the difference in the
The
professional career, the greater the opportunity one has to build a network of
— even
if
a researcher
order to fijrther investigate this
is
new
to the field.
An
diverse for
a proxy for this diversity).
length of professional experience between early and late entrants.
partners
is
analysis
longer one's
communication
of cohorts was performed
in
issue.
Rappa ami Debackere (December, 1991)
13
80.0
Academic (N= 271)
Industrial
(N= 110)
60.0
I
^
-o
S c
40.0
I
o -a
20.0
ri
0.0
U—i«A
I
I
P
11-15
6-10
1-5
Number
16-20
>20
of external
communication partners
FinURE
2:
External communication partners for academic
14,141; z = -1.54: n.s.)
r««r<:Am
and
industrial
(M-WU=
80.0
D
Early entrant (N=97)
Late entrant (N=284)
«
^
60.0
-
40.0
-
20.0
-
3
2 =
U "O '3
'^
.
.
8
6-10
Number
11-15
16-20
>20
of external
communicarion partners
FIGURE
3:
External communication partners for early
(M-WU=
10,500;
Z
= -4.02;
and
late entrants
p<.001)
fUppa and Debackere (December, 1991)
14
The 401 respondents were
Within each cohort,
non-parametric
early
and
The
tests.
partitioned into cohorts based
late entrants
of
results
professional experience of
variable.
were subsequently compared using Mann-Whitney
mixed. In the respondent cohorts with
this analysis are
more than
on the EXPJOB
the difference between early and late
five years,
more
entrants never anains a p-value below 0.05. As a consequence, for respondents with
professional experience, the difference between early
and
late entrants as to the diversity
their
communication panners
with
five years or less professional experience, the difference
entrants
eliminated. For the cohort containing those respondents
is
between early entrants and
who have been
highly significant (p<.001). Thus, researchers
is
—
enter the field as a student
colleagues
who
Along
are likely to have
late
involved in neural
networks a long time but have only recently obtained their highest degree
who
of
—
more communication
that
those
is,
partners than
have only just entered the community.
similar lines, the respondents were asked to
name
the neural network research
teams outside their organization with whose work they are well-acquainted. Out of 110
who answered
industrial researchers
least
this question,
48 (43.6%) were unable
one such team. For the 277 academic researchers who answered
(39.4%) were unable
to
do
so.
this question,
respondents in industry and academia
the
EXPJOB
who mentioned
i.e.
109
We were
any significant interaction effeas from
one research
was carried out similar to the
with the sector and entry dichotomy
variable as a co-variate.
For those
n.s.).
familiarity with at least
their organization, an analysis of co-variance
ones reported above,
at
As a consequence, the null-hypothesis of no association
between institutional type and familiarity could not be rejeaed (x^=0.435,
team outside
mention
to
as
independent variables and
not able to find any significant main effects nor
this analysis.
academic and industrial researchers report the same
Thus, both early and
level
late entrants
of familiarity (that
is,
and
an average
of about 3.4 research teams for each group).
Furthermore, for the 62 industrial researchers and for the 168 academic researchers
mention some
familiarity
with
at
least
one team, we examined
similarities
who
across
institutional types.
The
mentioned
one academic team; whereas only a minority of academic respondents
at least
vast majority of industrial respondents (56 out of 62, or
90.3%)
(53 out of 168, or 31.5%) mention at least one industrial team (x^=30.2, p<.001). Finally,
for each
respondent
who mention
at least
one team, we calculated the percentage of teams
mentioned with which the respondent was
exchange.
Once
again,
we were unable
actively involved in technical
information
to find statistically significant differences
between
seaors. Information exchange occurred with about one external team in two.
Rappa and Debackere (December, 1991)
15
4.3.
Formal collaboration with
other research teams
Collaborative projeas between researchers in different organizations were considered as
yet another
form of communal
interaction.
who completed
For 106 industrial respondents
the question, a total of 43 (40.6%) repon that they are involved in at least one collaboration.
For 277 academic respondents, 110 (39.7%) report having collaborations, resulting
X^=.001,
n.s.
The
average
number of
collaborations per respondent for the total sample
(excluding those reporting no collaborations at
found
significant differences could be
in a
all)
is
1.73.
Once
no
again,
statistically
of institution, the entry period, or their
for the type
interaaion.
For academic respondents, 78 (70.9%) do not have collaborations involving industrial
partners. For the industrial
respondents, only 16 (37.2%) report collaborations with
Funhermore, with respect
researchers in other firms (x^=13.4, p<.001).
collaborations,
we
find
that
the
of academic
majority
large
to the
number of
respondents
report
collaborations with other academic researchers. Collaborations between academic and
industrial researchers are a minority.
The
reverse
is
true for the projects
mentioned by
industrial respondents.
4.4.
Knowledge difjusion
Respondents were asked
might take
to rank order five possible actions they
making an important advance
in
neural networks.
They were
able to choose between
immediately publishing the result in a rapid publication journal, announcing
press conference, seeking patent protection, assessing
disseminating
it
its
after
it
publicly at a
potential commercial value, or
to other researchers in the field via telephone, fax or
computer network.
Again, four groups were considered in the analysis according to institutional type and
entry period.
do
exist
The
results are
shown
in
Table
8.
As
is
clear
from the
data,
differences
between academic and industrial respondents. In general, industrial respondents are
more inclined towards examining patent protection and commercial
academic
institutions, early
and
their counterparts in industry.
industry
some
is
late entrants
The commercial
particularly noteworthy.
early entrants, since
show
both are pioneers
We
value.
Within
similar behavior. This cannot be said for
orientation
on behalf of the
early entrants in
expected them to be more similar to academic
in the field,
however, the opposite
is
true.
Rappa and Debackere (December, 1991)
16
17
Third,
it
is
clear that the present data allow for
difFerent direaions, such
2S:
more
detailed analysis in several
government researchers,
the role of graduate students, the role of
behavior.
and international comparisons of communal
Work
is
now
proceeding along
all
of
these fronts.
Rappa and Debackere (December. 1991)
18
REFERENCES
Allen, T.J. (1988) "Distinguish Engineers from Scientists," in
Innovative Organizations, R. Katz
(ed.).
Managing
Professionals in
Cambridge, Mass.: Ballinger Publishing
Company.
Allen, T.J. (1977)
Managing
the
Flow of Technology. Cambridge, Mass.:
Arrow, K.J. (1974) The Limits of Organizations.
New
Yoric Norton
MIT
Press.
& Company.
Baumol, W.J. (1990) "Technology-sharing canels," Mimeograph, Princeton University.
Bijlcer,
W.E.; Hughes, T.P.; Pinch, T. (1987) The Social Construction of Technobgical Systems.
Cambridge, Mass.: MIT Press.
H.M. (1982) "Tacit Knowledge and Scientific Networks,"
Barnes and Edge (ed.), Cambridge, Mass.: MIT Press.
Collins,
in Science in Context,
Constant, E.W. (1980) The Origins of the Turbojet. Baltimore: Johns Hopkins University
Press.
Crane, D. (1972) Invisible Colleges: Diffiision of Knowledge in Scientific Communities.
Chicago: University of Chicago Press.
DARPA
(1988) Neural Network Study. Fairfax, Virginia;
Hagstrbm,
Howells,
J.
W.O.
(1965) The Scientific Community.
AFCEA
New York:
International Press.
Basic Books, Inc.
(1990) "The Location and Organisation of Research and Development," Research
Policy, 19:
133-46
Hull, D.L. (1988) Science as a Process. Chicago: University of Chicago Press.
(1989) "The Real Effects of Academic Research," American Economic Review,
December: 957-970.
Jaffe, A.
Mansfield, E. (1991) "Academic Research and Industrial Innovation," Research Policy 20: 112.
Minsky, M.L.; Papert, S.A. (1988) Perceptrons. Cambridge, Mass.:
Nelson,
R.R
did
it
(1990a) "U.S. Technological Leadership:
1 17-32.
Where
did
MIT
it
Press.
come from and where
go?" Research Policy 19:
Nelson, R.R. (1990b) "Capitalism
as
an Engine of Progress," Research Policy 19: 193-214.
Orton, J.D.; Weick, K.E. (1990) "Loosely Coupled Systems:
Academy of Management Review, 15, 2: 203-23.
A
Reconceptualization,"
Rappa and Dehaekere (December, 1991)
19
Papert,
SA. (1988) "One AI
Pavitt, K.
(1991)
109-19.
or
"What Makes
Rothwell, R.; Dodgson,
Many?"
M. (1991)
"External Linkages and Innovation in Small- and
R&D
Management
J. P. (1988) Acadrmic Scientists and
Hopkins University Press.
Thomson,
117: 1-14.
Basic Research Economically Useful?" Research Policy 20:
Medium-sired Enterprises,"
Swann,
DaeMus
2\
,
2:
125-37.
the Pharmaceutical Industry. Baltimore:
Mechanized Shoe Production in the United
of North Carolina Press.
R, (1988) The Path to
Hill: University
von Hippel,
E. (1988)
The Sources of Innovation.
F.; Gibson, D.V. (1990) Technology
London: Sage Publications.
Williams,
Wortmann, M. (1990) "Multinationals and
New
Johns
States.
Chapel
York: Oxford University Press.
Transfer:
A Communication
the Internationalization of
Perspective.
R&D,"
Research
Policy 19: 175-83.
Rappa and Dehackrre (Decrmber, 1991)
MIT LIBRARIES DUPL
111
3
TDflD
DDTSbflll
3
Date Due ^-^-y^^
APR.
1
5
r^s,
-'4./
JUN13'93
ici^j-
MAY 2 91995
,JUN.
»
5
199''
Lib-26-67
MIT
3
TQfiO
LIBRARIES
0D75bail
3