HD28
.M414
-11
ALFRED
P.
WORKING PAPER
SLOAN SCHOOL OF MANAGEMENT
Using the Literature in the Study of emerging
FIELDS OF Science and Technology
Michael A. Rappi
Massachusetts Institute of
Raghu Garud
New
York University
Technology
July 1991
Sloan
WP# 3392-92
MASSACHUSETTS
TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
INSTITUTE OF
Massachusetts Institute of Technology
Using the literature in the Study of Emerging
Fields of Science and Technology
Raghu Garud
Michael A. Rappa
New
Massachusftts Institute of
Technology
July 1991
©
York University
Sloan
1991
WP # 3392-92
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
Alfred P. Sloan School of
Management
Massachusetts Institute of Technology
50 Memorial Drive, E52-538
Cambridge,
02139-4307
MA
Using the Literature in the Study of emerging
Fields of Science and Technology:
AN Examination of Researcher Contribution-Spans
Michael A.
RAPPA
Raghu GARUD'''
New
Massachusetts Institute of
York University
TechnoloQ
ABSTRACT
This study describes
how
the scientific
and technical literature can be used as a
source of data to analyze the emergence of new fields. In particular, the
literature is used to determine the length of contribution spans of individual
researchers in afield
analysis, the authors
to the field for
Through the use ofstatistical technicjues, such as survival
examine the probability that a researcher will contribute
a specified length of time
having contributed
and
the probability that a researcher,
for a specified period of time, will cease to
The authors aLo test the significance of several
to the field
contribute in the future.
variables in explaining researchers' contribution spans.
introduction
New
of science and technology typically emerge
fields
individuals
who
common
develop a
set
in the context
of ideas and techniques and
of a community of
who communicate
with
each other about the nature of their work. Frequently, a portion of the communication
between these individuals takes the form of conference presentations and
which then become documented
fullest
in
and
in
in
the scientific and technical literature.
conjunaion with other kinds of data,
this
used to
its
information can be extremely helpful
improving our understanding of the dynamics of a research community.
The
following study, which serves as one illustration of this approach, uses
literature to
measure the length of time researchers remain
statistically their survival
longevity. In particular,
it
and hazard
rates as well as
some of the faaors
associated with their
examines the duration of the participation of individual researchers
their first
and
is,
contribution to the published literature in a given
of the contribution-spans of researchers
of: (I)
in the field
the time span between
field.
From an
Michael Rappa
is
assistant professor
management with
analysis
of cochlear implants, estimates are made
the probability that a researcher's contribution-span will extend a given
assisunt professor of
the
thereby determining
through an analysis of their "contribution-spans": that
last
dau from
in a field,
in a field
'
scientific papers,
When
number of
of management with the Massachusetts Institute of Technology. Raghu Garud
New York
University.
The
is
authors are indebted to Susan Bjarner for her assistance
in conducting the literature searches and Gary Quick for his assisunce in creating the daubase. TTie authors also thank Joel
Baum, Hayagreeva Rao and Robert Yaffee for helpful comments and guidance, and Edward Roberts for his support and
encouragement. This study was funded, in part, with a grant from the Council on Library Resources.
years,
and
having contributed a given number of years, a researcher
(2) the probability that,
will cease to
number of
contribute in the future. Furthermore, this paper examines the relevance of a
individual-, organizational-,
and community-related variables
in explaining the
duration of researchers' contribution-spans.
Before elaborating
and
results
of
assessing the
upon
the concept of contribution-spans
this study, a brief
growth of new
fields
and discussing the procedures
review of earlier studies that have used the literature in
is
in order.
Prior Litrrature-based Studies ofErmrging Fields
The
use of the literature to understand the growth of a field has a long and well-
established tradition.
When
comparative anatomy
in
Cole and Eales published
1917, they were
among
the
literature in order to quantify the progress in a field.'
truly a painstaking effort
to compile their data.
by the two
The
research in comparative
result
scientists,
development of
to seriously utilize the published
Given the technology of the day,
who examined
was a detailed
anatomy over three
their study of the
first
nearly 6,500 books
was
account of the ebb and flow of
statistical
Cole and Eales
centuries.
it
and papers
clearly illustrated the
prevalence and magnitude of cyclical changes in the level of publication activity that occur in
a field over time. This finding led
attributed, in part, to the
The pioneering
who
them
to suggest that such
movement of individuals between
effort
waves of
activity can be
different research fields.
of Cole and Bales was joined subsequendy by Wilson and Fred,
published a study in 1935 of the growth and development of research on the subject of
nitrogen fixation by plants.^
It is
clear that
value of the literature as something
among
scientists.
"Study of the
Wilson and Fred had an
more than
literature as
a
an
intuitive grasp of the
means of communicating research
entity...,"
results
they claim, "...should provide
valuable information for the interpretation of past production and might afford
for prediction of future trends" within a given field. ^ Like their predecessors,
Fred showed empirical evidence of the cyclical nature of research within a
some
basis
Wilson and
field. Similarly,
they suggest that the fluctuating level of publication aaivity in nitrogen fixation research was
F.J.
Cole and N.B. Eales, "The History of Comparative Anatomy, Part
(1917): 578-96.
I.
—A
Sutistical Analysis of the Literature," Science
PmgrenU
^P.W. Wilson and LB.
41 (1935): 240-50.
3lbid., p. 240.
Fred,
"The Growth Curve of a
Scientific Literature:
Nitrogen Fixation by Plants,"
Scientific
Monthly
movement of
a result of the
would
rise
researchers between fields.
with an important breakthrough
and afterward
to further experiments;
problems too complicated to
It
it
as
it
They claim
anraaed new
would subside
1956 used the physics
led
found the remaining
solve.
The one most
literature
would once again
who
responsible for this resurgence was Price,
document the "exponential growth of
literature to
work brought legitimacy
and
scientists to the field
ai scientists
was not for another two decades that the study of the
receive serious attention.
that the level of aaivity
to the scholarly
examination of
in
science.""* Price's
literature, partly
because his
conclusions coincided with, and indeed reinforced, the belief that the proliferation of science
seemingly knew no bounds.
More
science and technology have
become
diverse.
However,
fairly
in general, there have
model the growth of a
different in that
recendy, literature-based studies of emerging fields of
it
field
common. The
emerged two
nature of this research
basic types of studies.
is
first
quite
seeks to
by measuring annual publication volume. ^ The second
is
quite
uses the citation as a unit of analysis.^ Citation-based studies seek to
understand the interlocking nature of citation panerns
as a
development of clusters of researchers who may ultimately come
field.
The
Both types of literature
although attention to the
studies cover a
latter has
wide range of topics
means
to
for observing the
form the
in science
basis for a
new
and technology,
recendy led to a more intensive examination of the patent
literature.^
The
contributions of Cole and Eales, Wilson and Fred, and Price,
have established a tradition of research that uses the literature to
direction of a field.
literature, their
'*D.J. Price,
However,
work
in
among many
assess the
others,
growth and
having led the way in revealing the promise of using the
has exposed
"The Exponential Curve of Science,"
some of
its
limitations, as well.
Perhaps the most
Durat/fry 17 (1956): 240-43.
'For example, see A. Granberg, 'A Bibliomctric Survey of Fiber-Optics Research
Institute, University of Lund, Sweden, 1985.
in
Sweden, West Germany, and Japan,"
Monograph, Research Policy
°TTic notion of "bibliographic coupling" was proposed by Kessler in 1962. Since then, numerous "co-citation" studies have
been published, including the pioneering work of Small and his colleagues. For a recent example, see H. Small and E.
A variant of co-ciution analysis is co-word
Greenlee, 'Collagen Research in the 1970s," Scienumutrics 10 (1986): 95-1 17.
combination with co-citation) sometimes referred to as "science mapping." See P. Healy, H. Rothman,
and P.K. Hoch, "An Experiment in Science Mapping for Research Planning," Research Policy 15 (1986): 233-51; R.R.
Brahm, H.F. Moed, and A.F.J, van Raan, "Mapping of Science: Critical Elaboration and New Approaches." Informetha
(1987/88): 15-28; and M. Gallon, J. Law and A. Rip (eds.) Mapping the Dynamics of Science and Technology (London:
analysis (aJone or in
Macmillian
Press, 1986).
'For an overview,
see B.L. Basberg, "Patents
Research Policy 16 (1987): 131-41.
An
and the Measurement of Technological Change: A Survey of the
is provided by W.D. Reekie, "Patent Dau as
early study of patents
Industrial Aaivity,' Research Policy 2 (1973): 246-64; for a
more
recent study, see
Literature,"
a Guide to
R,M. Wilson, "Patent Analysis using
Online Databases: Technological Trend Analysis," World Patent Information 9 (1987): 18-26.
troublesome criticism of prior studies
is
that the literature
While
indicator of the level of research aaivity in a field.
offers a useful
patent
measure of
alone
statistics
Although the
scientific
criticisms are valid,
it
would be unfortunate
of the literature for the study of new
literature
a rich source of information.
Many
fields
exploit the full potential of the literature.
it
beyond measures of publication
and sophistication, enabling
citations to take full advantage of the information
in
and of
itself,
investigators to go
from treating the
changing. Recent
is
possible to use the literature with
volume and
shifted
of these studies to
inability
this situation
greater ease
is
it
contains.
literature as
something
of molecular biology,
literature to
in
which he
this vein, a small
fields.
One
pioneering effort
a different approach,
literature can be analyzed to
major advances
in clinical
is
Mullins' 1972 study
takes advantage of the incidence of co-authorship in the
understand the communication network that forms
Taking
among
Comroe and Dripps show how
researchers in the
the content of the
understand the contribution of long-term basic research
medicine. ^ Yet another approach
is
uses the literature to identify individual researchers in order to
manpower
measured
in research
scholars have sought to use the literature as a source of data about research
communities and the emergence of new
field. ^
to be
of data about what goes on
to using the literature as a source
communities, many new avenues of research become readily apparent. In
number of
interest
of science and technology. The
from the
However,
advances in computer hardware and software are making
the perspeaive
dampen
they were to
if
field.
of the limitations of previous studies are not
necessarily inherent in the data, but rather they arise
Once
be true that the literature
about the process of the emergence of a new
in the use
is
a singular, coarse-grained
is
may
and technological output, the focus on publication and
us linle else
tells
it
in different countries."^
Notice that
in
that of Spiegel-Rosing,
compare the
level
to
who
of scientific
each of these instances, the investigators
went beyond numbers of publications and instead probed the
literature for specific
N.C. Mullins, 'TTie Development of a Scientific Specialty: The Phage Group and the Origins of Molecular Biology,"
Minerva 10 (1972): 51-82. Since Mullins, others have sought to use co-authorship data to investigate collaboration among
researchers, including constructing a map of the inter-organizational linkages within a community. For example, see D. deB.
Beaver and R. Rosen, "Studies in Scientific Collaboration," Scientorrutrus 1 (1978): 63-84; M.S. Sridhar, "A Study of Coauthorship and Collaborative Research Among Indian Space Scientists," R&D Mamgemmt 15 (1985): 243-49; and M.A.
Rappa, "Assessing the Emergence of New Technologies: The Case of Compound Semiconductors,"
Management of Innovation, A. Van dc Ven et al. (Cambridge, Mass.: Ballingcr, 1989).
-^J.H.
Comroe,
Jr.
and R.D. Dripps,
Two
on the
Support of Biomedical ScKntx," Science 192 (1976): 105-1
Authors as an Indicator of Scientific Manpower:
Germanics and Europe," Science Studial (1972): 337-59.
'^I.S. Spiegel-Rasing, 'Journal
the
"Scientific Basis for the
in Research
A
1
1.
Methodological Study Using Data for
information contained therein: that
other, or the nature of their
These studies have only
is,
about the people involved, their interaction with each
work.
begun to
just
within the literature. Journal
articles,
up
the abundance of information that
conference papers, and patents in a given
is
contained
field represent
a detailed, self-reported archival record of the effort generated by researchers to solve the
and technical problems confronting them. Furthermore, the
scientific
appealing source of
quality
dau
literature
is
in several respeas: the conventions of publication ensure a level
an
of
and authenticity; the data can be collected unobtrusively; the findings can be
replicated
and tested
for reliability;
inexpensive to collea. Clearly,
and longitudinal nature of the
and the data are publicly available and
would be very
it
difficult to
relatively
match the comprehensive scope
literature using other data colleaion techniques.
When
taken
together, the literature can be viewed as a unique chronology of the efforts of researchers to
establish a
new
where they
are
when
field,
and can provide information with respect to the individuals involved,
employed,
who
they were aaive in the
they collaborate with, what problems they are pursuing, and
field.
This paper
illustrates
how
such data can be used to obtain
a better understanding of the longevity of researchers' contributions during the
a
new
emergence of
field.
Researcher Contribution-spans
The
challenge to understanding the emergence of
in part, a
fields
of science and technology
is,
problem of understanding why researchers choose the topic they do, and why they
remain committed to that topic or leave
is
new
that fields
which
those that are
Furthermore,
attract
and
less attractive
it is
it
for
something
else.
The
fiindamental assumption
retain researchers are likely to progress
more quickly than
and
retain researchers.
and
are therefore less able to recruit
reasonable to assume that researchers' decisions to remain with a field are
influenced by their assessment of the rate of progress being made. Given this behavior,
be interesting to use the literature to understand in a
researchers' panicipation in a given field
statistical
and the faaors that may influence
Determining the number of years a researcher spends working
task. First,
it
requires
manner
some degree of historical
it
may
the duration of
it.
in a field
is
not a simple
investigation: if we simply ask everyone
who
currently works in a particular field
how
long they have panicipated,
experience of numerous individuals
who
have already come and gone over the years. Second,
the
number of
individuals
who
have participated in a
widely dispersed geographically, thereby making
it
field
we would
overlook the
can be quite large and can be
very hard to ascertain the extent of their
participation.
Confronted with these constraints,
literature as a source
"contribution-span"
who
of data about
—
publications in a field
that
the
is
becomes appealing
it
number of years spanning an
—
can serve
how
panicipates in a field and for
author's
first
to look to the
Thus, the
long.
and
last
known
unique and useful measure of the duration of their
as a
participation in a field.
Using the
statistical
techniques of survival analysis, data on researcher contribution-spans
derived from the literature can be used to estimate the probability that a researcher will
remain in the community a given number of years (the survival
rate).
Furthermore, the data
can also be used to estimate the conditional probability of a researcher leaving the
contributing a given
number of years
survivor and hazard functions
understood,
is
it
field after
However, once the distribution of the
(the hazard rate).
becomes pertinent
to ask
what faaors might
influence researcher contribution-spans. Here again, the literature can offer
some
insights in
terms of providing additional data regarding each researcher. For example, the length of an
might be affeaed by:
individual's contribution-span
cumulative produaivity in the
size
of their organization's research effort in the
they are employed and
the
among
dispersion
the
the
field; (6)
field.
factors
its
geographic location;
number of other
and
researchers
different organizations;
Using data from the
(1) their
own
or their organization's
the size of their collaborative social network; (3) the
field; (2)
and
field; (4)
(5)
the kind of organization in which
the accumulated
working
amount of knowledge
in the field
evolutionary stage of development of
(7) the
literature, the statistical relationship
researchers' contribution-spans will be
in
and the extent of their
between each of these
examined.
The Cochlear Implants Field
The
but
it is
field
of cochlear implants was chosen
just
one of several
fields that the
program on the assessment of emerging
of the
electrical stimulation
as
means
for enabling individuals
result
illustrative case for the present analysis,
areas
of science and technology.'' Although studies
of the ear have a long history,
researchers began the systematic investigation of
One
an
authors are studying as part of a larger research
how
it
was not
until the
elearical stimulation
1950s that
might provide
a
with sensorineural deafness to gain some sense of hearing.
of this research was the advent of the cochlear implant, a device that uses elearical
stimulation of the cochlear to provide a sense of sound for profoundly deaf individuals.'-^
For
a full
account of the historical development of cochlear implanti, see R. Garud and A.H.
Innovation and Industry Emergence: Tlie Case of Cochlear Implants," in
Van de Ven,
ct al.,
Van de Ven, "Technological
Management of
Research on the
Innovation (Cambridge, Mass.: Ballingcr, 1989).
A
cochlear implant
a receiver that
is
is
composed of a microphone,
surgically implanted
behind the ear
signal processor,
just
and transmitter worn on the outside of the body, and
under the skin cither into the cochlear or placed around
it.
For
a
One
of the
first trials
by William House, a
of a rudimentary cochlear implant device was performed
clinical physician,
who
center for cochlear implant research. At
later
first,
founded the House Ear
device to lack a scientific basis and believed that
experimentation.
It
implants emerged
was not
it
Many
A
as a viable research field.
field:
Otological Society meeting
(a
workshop held
were held
as part
group that was adamantly opposed
As a
result
of these meetings, a limited
and cochlear
in
1973 that
of the American
to the topic in previous
major international conference on the subject and a workshop
California, San Francisco.
major
in cochlear
was too premature for human
flurry of meetings
including, a
1961
considered the
until the early 1970s that the controversy subsided
galvanized the burgeoning
years), a
Institute, a
few researchers took an interest
implants, largely due to the intense controversy the field encountered.
in
at
the University of
number of
clinical trials
began. '3
One
study of particular importance
is
Bilger's
work, initiated in 1975, on the
comparative performance of cochlear implants.''* Funded by the U.S. National Institutes of
Health and released
in
1977, the "Bilger Report" showed that although the early claims
regarding performance were exaggerated, cochlear implants did indeed have merit. In
conclusions, the report lent
the door for a greater
some
sorely needed legitimacy to the field
number of researchers
be seen clearly in terms of the
number of
to participate. This
growth
its
and thereby opened
in participation
can
researchers contributing to the literature (See
Figure l).^^
It is
estimated that by 1990 there were about four-hundred individuals active in the
who were employed
in nearly
one-hundred different organizations worldwide. About
percent of the cochlear implant
progress has been
made over
community
is
field,
forty-
located in the U.S. Although tremendous
the years and about three-thousand patients have received the
detailed description of cochlear implants see: Gerald E. Loeb, 'TTie Functional
Replacement of the Ear"
Scientific
American,
225(2):104-n.
'Proceedings of the First International Conference on the Electrical Stimulation of the Acoustic Nerve as a Treatment for
Profound Sensorineural Deafness in Man, San Francisco, California, 1974 (edited by M.M. Merzenich. R_A. Schindler and
F. Sooy); Report on a Workshop on Cochlear Implants held at the University of California at San Francisco, October 2325, 1974 (edited by M.M. Merzenich and F. Sooy).
^R.C.
Bilger, et
al.,
'Evaluation of Subjects Presently Fitted With Implanted Auditory Prostheses," Ann. Otol. RhinoL
LaryngoL, (Suppl. 38) 86 (1977): 3-10.
-'The
number of
entering the field
researchers in the
(as
evidenced by an
community
initial
in a given year
is
calculated to be the cumulative
publication) subtracted by the cumulative
the field (as evidenced by their failure to continue to publish in a future year).
number of
number researchers
who have left
individuals
r
cochlear devices, the cochlear implant remains largely an experimental procedure with
many
challenging problems to be overcome.'^
400
—— — — — — — ——— — — —— — —
'
I
I
I
I
r~i
I
I
I
I
'
I
I
I
I
I
90
Yea
FIGURE
1:
Growth of Cochlear Implant Research Community, 1973-89
DATA COLLECTION AND METHODS
Five commercial electronic databases covering the medical science and engineering
literature
were used
to identify publications related to the field
daubases were searched on-line using a
in the lexicon
classification
set
of key terms that are
of cochlear implants.''' The
known
of cochlear implant researchers and might be either
terms of a document.
The
by the National Library of Medicine
to be
commonly used
in the title, abstract or
search strategy was derived fi'om an earlier one used
to create a research bibliography
on cochlear implants.'^
Several researchers have provide historicaJ accounts of the field of cochlear implants and its technical evolution. For
example, see W.F. House and K.I. Berliner, "Cochlear Implants: From Idea to Clinical Practice," in H. Cooper, ed..
Practical Aspects of Cochlear Implants (London: Taylor and Francis) in press; and F.B. Simmons, 'History of Cochlear
Implants
in
the United States," in R-A. Schindlcr and
M.M.
Mcrzenich, eds. Cochlear Implants
(New
York: Raven Press)
1985.
'
'The databases used
are MedJine,
Compendcx,
was reduced to
K. Patrias
1
Medica and INSPEC. A total of 3064 documents were
implementing an on-line duplication removal process, the database
Biosis, Exerpta
originally identified as related to cochlear implants. After
884 documents.
and R.F. Naunton, National Library of Medicine, Current Bibliographies in Medicine: Cochlear Implants,
DC: U.S. Department of Health and Human Services, National Public Health
January 1983-March 1988 (Washington,
Service, National Institute of Health, 1988).
The
cochlear implant documents retrieved electronically from the search were
temporarily placed in a bibliographic relational database operating on a personal computer.
This allowed for a careful inspection of each document
order to ensure the accuracy and
of the search procedure. Since multiple source databases were used,
integrity
to
in
it
was necessary
remove duplicate documents. In addition, while inspecting the database, an
made
remove
to
misclassified
documents
effort
was
that did not pertain to cochlear implants as well as
those that did not have a research orientation, such as editorials or journalist documents. In
the process of inspeaing the documents, any that seemed inappropriate were flagged, so that
an individual active
in cochlear
implant research could make the
final
judgment
as to its
relevance.
The
data collection procedure described above ultimately resulted in the identification of
1,329 unique documents related to cochlear implants published between 1973 and 1989.'^
The
data subsequently used in this study were derived from these documents. However,
before the documents could be used as a source of data, they required extensive editing in
order to create a consistency
case that the
especially
among author names and
name of an author
when using
misspellings, but
different
mosdy
this
is
or an affiliation
is
affiliation
names.
It is
frequently the
not standardized across documents,
commercial databases. Sometimes
this arises
because of
the result of variations in the use of abbreviations, middle
hyphenations, and other sources of inconsistency.^^ Although such a
initials, capitalizations,
lack of standardization might not be a problem for the typical user of an electronic literature
database,
it
would be
a major source of error in determining the duration of researcher
contribution-spans. Therefore,
author and
affiliation
in
it
was
essential to meticulously
the relational database so that
all
inspea the name of each
inconsistencies could be
eliminated.
Upon
completing the editing of the documents, the database was used
individual author
who
procedure yielded a
total
contributed to the
field
of 1,257 authors. At
to identify each
over the seventeen-year period. This
this stage, a statistical
database was created
containing several individual-, organizational-, and community-level variables for each
'-T^otc that the electronic databases do not provide any indication of the wori< in cochlear implants prior to 1973- This
due
It
in part to the fact that
also
is
due
many eienronic
services. Unfortunately, this
^he
literature
daubaie
to the lack of publication of the pre-seventies
prevalence of author
is
services did not begin until the late sixties
work
in
one limitation of the electronic databases
name
inconsistencies
is
examined by
Research," National On-Line Meeting (10th) Proceedingi,
May
M.L
9-11,
and
is
early seventies.
cochlear implants in journals abstracted by the database
1
that
is
extremely difficult and cosdy to circumvent.
Pao, "Importance of Quality Data for Bibliomctric
989.
10
author that were derived from information obtained from the published documents. Table
provides a
of the variables and their definitions.
list
The dependent
number of
variable for the analysis, the contribution-span,
years that have elapsed from the
some methodological
is
is
these individuals have not yet
account for
publication for each
issues that arise that require further explanation. TTie
that for those researchers
is
calculated as the
is
relatively straightforward, there are
who
aaive
are
left
the field,
some minimum
it
is
value (that
this, survival analysis statistics
is,
indeterminate: that
is
known
only
primary
issue
of cochlear implants
in the field
present time, the ultimate length of their contribution-span
contribution-span
known
to the last
first
author. 21 Although calculating the contribution-span
concern
1
at the
is,
since
that the length of their
the entry year to the present year).
were implemented
in
of
To
analysing the data.^^ Such
techniques take into consideration precisely this kind of problem in the calculations with a
procedure that adjusts for the biases that right-censored data
Determining whether or not
certain cases.
The
reason for this
a researcher
is
is
still
create.
aaive
that typically researchers
in the field
can be difficult in
do not publish every
and
year,
indeed, an author's contribution-span in a field can be characterized by "gaps" of several
which there
years in duration in
no publications
are
discontinuities in publication records raise the issue of
to their credit.
how
publish in order to be considered an active contributor to the
field.
nature and prevalence of gaps in a researcher's contribution-span
the proper censoring scheme to use in the analysis.
someone
answer to
ceases to publish
this
question
is
is
it
The
The
existence of
frequently a researcher must
is
Thus, understanding the
important
question
arises:
in
determining
How
long after
reasonable to assume they are no longer in the field?
necessary in order to determine
who
has exited the field and
The
who
continues to be a panicipant.
Normally, the time between publications
can be quite long. Therefore, a rule
is
after the last publication in order for
exited the field.
An
may
be a year or two, but in some instances
required to determine
it
how many
to be reasonable to classify a researcher as having
analysis of the frequency of gaps
between publication
in researcher
contribution-spans provides the evidence on which to base this decision (see Figure
For example,
be caJculated
Note
if a
researcher
as six years.
first
is
2).
The
in 1975 and last published in 1980, the researcher's contribution span would
assumed that a researcher who publishes in only one year has a span of one year.
published
Furthermore,
that the contribution span
it
years should transpire
it
is
unaffected by the frequency of publication within a given year.
^^Sce R-C. Elandt-Johnson and N.L. Johnson, Survival Modeb and Data Analysis (New York: John Wiley
and J.D. Kalbfleisch and R.L. Prentice, The Statistical Analysis of Failure Time Data (New York: John Wiley
& Sons, 980)
& Sons. 1980).
1
11
cochlear implant data indicate that 242 (19%) researchers have a gap in their contributionspans.
Among
those
who
have a gap
have a gap of three years or
researchers
who
less in
in their contribution-span, four
duration. Based
upon
out of
this evidence,
published within the past three years of the
last
it
five researchers
was decided that
year of the data set (1989)
would be censored.
2
1
3
4
3
Duration of Gaps
FIGURE
Only
2:
a small
than three years
6
in
7
9
8
13
12
Contribution Span (Years)
Frequency of Gaps Between Publication in Researcher Contribution-spans
number (3.7%) of all
in duration,
and
still
researchers have a gap between publications of longer
fewer have an exceptionally large gap, such as ten years
or more. Although rare, these "sparse" contribution-spans
who
n
10
may
does not in fact continuously contribute to the
be indicative of an individual
field;
and thus, although
contribution-span might be quite long, their aaual participation in the
field
is
far less
their
than
the span implies. Because of their relative rarity in the present data, researchers with sparse
contribution-spans were not treated
advisable to address this issue
more
as special cases
in the analysis.
Having determined the distribution of contribution-spans,
what faaors might affea how long a researcher contributes
a
number of explanatory
could be treated
as
Nevertheless,
it
is
closely in future studies.
variables were constructed.
time-varying covariates (that
is,
as
it is
interesting to
to the field.
Using the
Although the explanatory
examine
literature,
variables
having values that vary yearly in the
course of an author's contribution-span), the present analysis implements a "single-spell"
12
approach to formulating the data
taken according to the
last
Therefore, the value for each explanatory variable
set.-^
year in the author's contribution-span. In this manner, several
variables were created to control for certain factors that
among
is
coded according to whether they reside
in
might account
for heterogeneity
which each author
researchers. First, the kind of organization in
employed was
is
an academic, industrial, government, or
independent research laboratory.^'* Second, the country
in
which the author
located was
is
coded. Since nearly half of the authors were based in the United States, to simplify the
analysis, a location variable
was coded
as to
whether the author resided
Third, a variable was created to determine whether the author
in the
U.S. or not.
the field during the
left
"pioneering" phase (considered to be prior to the publication of the Bilger Repon), or during
the "rapid growth" phase that occurred from 1978 onwards.
measure of the accumulated knowledge
in the field, as
The
control variable
last
is
a
determined by the cumulative number
of publications.
Additional explanatory variables were created, which are grouped according to their level
of analysis: individual-, organizational-, and community-level. At the individual-level, a
variable
was constructed
to reflect an author's
cumulative productivity
in the field as
measured by the cumulative number of publications
to their credit. In addition, a variable
was created
embedded
to reflect the extent to
collaborators. This
network
is
which an author
with which an author has been associated with
Two
number of
The
size
as a
implant publications by individuals
Three community-level
of an author's organization
is
measured
affiliated
and produaiviry of an
variables
individual authors
who
analysis that
as
who
was created
were created to
in
measured
in
terms of
also publish in the field.
the cumulative
is
reflect the size
measured
publish in the field in a given year.
size,
is
number of
cochlear
with an author's organization.
cochlear implant field in each year. Population size
square of population
network of
co-author on publications.
individuals affiliated with that organization
Cumulative organizational productivity
^An
in a larger social
organizational-level variables were created to reflect the size
author's institutional affiliation.
the
is
measured by the cumulative number of unique individuals
A
in
and dispersion of the
terms of the number of
second-order variable, the
order to capture any quadratic association between
implements time-varying covariatcs would require
a muitiple-spcli data structure.
This wor^l
is
currently
being performed and the results will be described by the authors in a future report.
24 It
is
important to note that one unfortunate limiution of electronic literature databases
affiliation
data for the lead author only. Therefore, in
organization
may be
misclassified.
some
instances secondary authors
who
is
the tendency of providing
are affiliated with a different
13
population
size
among
authors
and contribution-span. The third variable
different organizations: that
is
a
measure of dispersion of
the extent to which the
is,
community
is
concentrated in a few organizations or spread across many. For this purpose, a Hirfindahl
determined by calculating the sum of the squared share of researchers
statistic,
which
affiliated
with each organization,
is
in used.
RESULTS
Using data from the
scientific
and technical
literature
on cochlear implants published
between 1973 and 1989, the contribution-spans for 1,257 researchers and several
explanatory variables associated with each were compiled into a
database.
statistical
The
were analyzed using the LIFETEST and LIFEREG procedures of SAS (version 5.18).
1,257 cases, 700 (55.7%) were active within three years of the
last
year of the data,
data
Of
the
and were
therefore classified as censored.
Using the LIFETEST procedure, the
first
step in the analysis
estimates of the survival and hazard functions for the data.
chosen.
The
results
of
this
procedure are illustrated
negatively-sloped and non-monotonic.
sample leave the
field
within
The median
five years
on
in
Figure
was
3.
make non-parametric
to
The
lifetable
The
approach was
survival function
survival time in 5 years: that
their first publication.
researcher's contributions-span lasting longer than
two
continues to diminish rapidly between two and
six years,
years
is
The
is,
half the
probability of a
about 0.6. The survival
and then
is
rate
levels-off eventually
reaching a value of about 0.33 for contribution-spans often years or more.
The
hazard rate
is
also a negatively-sloped
decreases very rapidly for researchers
is,
is
who
non-monotonic function. The hazard
have contribution-spans of
at least
two
the probabilit)' of a researcher ceasing to contribute after having contributed for two years
only about 0.08 compared to about 0.25 for a researcher in the
field
only one year.
hazard rate subilizes for researchers whose contribution-spans are between two and
and than diminishes rapidly between
hazard ftinaion
is
rate
years: that
to leave
it,
is
is
and twelve
years.
The
basic implication of the
that the longer a researcher contributes to the field, the less likely he or she
with the
leaving the field
six
The
six years,
first
and
sixth years being particularly critical points. Indeed, the risk of
highest within the
first year.
14
1
1
1
r
-
0.2
_i
i_
I
4
2
6
8
10
12
14
16
18
Year
FIGURE
Non-parametric estimation of survival and hazard functions for researcher
3:
contribution-spans in cochlear implants.
The
next step in the analysis was to determine the parametric model that best
fits
the
distribution of contribution-spans (or, in the terminology of survival analysis, the failure
time).
Although non-parametric
analysis permits cenain assumptions that
the shape of the failure distribution (for instance, that
was decided to
basic
statistically
model adopted
examine
for the analysis
Y
is
variables,
several different distributions for
=
Xp
P
a
veaor of unknown
regression parameters,
a
X
is
vector of errors from an assumed distribution. This model
accelerated failure time
goodness of
fit.
it
The
ae
+
the log of the contribution-span (the failure time),
is
can be made about
non-monotonic), nonetheless
is:
Y
where
it is
model because the
effect
is
the matrix of explanatory
a scale parameter
is
and e
is
a
often referred to as an
of the explanatory variables
to scale a
is
baseline distribution of failure times. Specifically, four different types of distributions were
evaluated: the exponential, Weibull,
procedure are provided
using a
in
Table
2.
gamma, and
TTie parameters are estimated by
Newton-Raphson algorithm. The
likelihood fiinaion.
log-logistic distributions.
overall
Minus two times the
fit
of each model
is
The
results
maximum
As
is
likelihood
represented by the log-
log-likelihood value has a y} distribution with
appropriate degrees of freedom. Using the baseline model, the goodness of
distribution
of this
fit
for each
evaluated in term of minimizing the absolute value of the log-likelihood score.
a result, the log-logistic distribution
was chosen and became the
regression coefficients of the explanatory variables in the model.
basis for estimating the
15
The model was
estimated with the LIFEREG procedure in a sequence of
adding explanatory variables into the equation according to the
control variables are added
variables,
first,
and the individual
(see
Table
3):
followed by the population variables, the organization
variables.
Model
1,
of -1299.
variables, has a log-likelihood score
variables has the effea
of analysis
level
six steps by-
which did not include any explanatory
The
addition of each set of explanatory
of improving the log-likelihood score, such that Model
score of -295, was chosen as the baseline for
comparing other models
6,
which has a
understand the
to
efFea of each explanatory variable.
The
estimation results of Model 6 are provided in Table
3.
Among
the
dummy variables
that control for organization type, only the variable that distinguishes research institutes from
other types of organizations
researchers
employed
is
significant (p<.05 level). Its positive coefficient suggests that
in research institutes
have longer contribution-spans
researchers with other types of affiliations.
researchers represent only a small
However,
as
compared
to
should be noted that industrial
it
segment of the sample (3%); the
largest
segment being
academic (58%).
In addition, the
is
non
significant,
dummy variable
that distinguishes
between
US
implying that the contribution-spans of
than their counterparts in other countries.
The remaining
who were
The
who were aaive
significant right
from
The
their initial inclusion in
accumulated number
Model
and second-order population terms
3.
(The third
small (<
250
it
The
(see Figure 4).
researchers), population size
spans and increasingly so until
variable, dispersion,
combined with the
is
are
not
positive
term implies a U-shaped relationship between population
and researcher contribution-spans
is
first-
negative coefficient for population size
coefficient for the second-order
community
growth phase. ^5
in the field, the longer the contribution-spans.
In the case of the population variables, the
significant.)
negative sign of the
in the
coefficient of cumulative publication suggests that the greater the
of publications
size
level).
active prior to the field's legitimacy (before
1978) have shorter contribution-spans than those
The
and non-US researchers
researchers are no different
control variables, the phase and
cumulative publication, are both highly significant (p<.001
phase variable implies that researchers
US
reaches a size of about
is
data indicate that
when
the
negatively related to contribution-
250
individuals; at
slope of the curve turns positive. This result suggests that there
may
which point the
be a point of
critical
•^'Upon careful rcflcaion, the true effect of this variable is difficult to asccruin given the structure of the present analysis.
Understanding the issue of phases of growth in the community might be more appropriately addressed with the use of timevarying covariaces.
16
ma5S
for the
communiry,
at
which
The
increasing contribution-spans.
and has
a positive coefficient.
its
size
is
become
sufficiently large to
poplation dispersion variable
significant in
also significant (p<.001)
is
Thus, the more disperse researchers are among organizations,
the shoner are researcher's contribution-spans.
200
100
Population
FIGURE
4:
in
Model
5,
variables,
and are both non
400
size
Relationship between population size
The second group of explanatory
added
300
and researcher
contribution-spans
which examine organizational-level
significant at
first.
Model
In
factors are
6, organizational size
is
weakly significant (p<.05) and has a negative coefficient. This suggests that the larger the
number of
researchers
working
researcher's contributions.
significant.
The
in cochlear
The cumulative
implants in one's organization, the shoner are a
productivity of a researcher's organization
of significance of organization variables
relative lack
continued significance of the population variables:
opposed to a
researcher's
own
it is
is
is
not
noteworthy given the
the size of the entire community, as
organizational effort, which
is
the contributing factor to the
length of contribution-spans.
The
last
two explanatory
variables, the
cumulative number of co-authors and an author's
cumulative produaivity, are highly significant (p<.001
expected, that researchers
who accumulate
contribution-spans. Perhaps
more
individual co-authors a researcher
The
data
show
a greater
interesting,
is
is
level).
The
data indicate,
number of publications
will
as
might be
have longer
the significance of the cumulative
number of
associated with in the course of their contribution-span.
that researchers with a larger network of co-authors will have a longer
contribution-span. Since the co-author network can be viewed as a measure of the extent to
which a researcher
is
socially
embedded
into the research
community
as a
whole,
this result
17
suggests that the degree to
their peers in the
which researchers
community
is
a contributing
are
conneaed
in collaborative aaivities
faaor to their longevity
with
in the field.
CONCLUSION
Using the
of data,
this
paper provides an analysis of the
contribution-spans of researchers in the emerging
field
of cochlear implants. Non-parametric
literature as a source
estimates of the survival rate
and hazard
rate are
made, and
it is
found that the distribution of
contribution-spans follows most closely a log-logistic function. In addition, a
model of the
The
relationship between the hazard
and
a set of explanatory variables
statistical
examined.
is
findings of this analysis indicate that the sample survival and hazard ftinaions for
1,257 cochlear implant researchers are negatively-sloped and non-monotonic.
About 60-
percent of the researchers have a contribution-span of more than two years, and the
estimated median contribution-span
contribute to the field
is
it
The
five years.
risk
first
The
hazard rate
year and remains fairly constant until the sixth year, after
continues to decline rapidly. This result might have an explanation in a behavioral
characteristic of the research process. For example, a researcher
new
of a researcher ceasing to
greatest in the first year of their contribution-span.
declines sharply after the
which
is
might be drawn
to enter a
by the prospect of making an important contribution and by the possibility of
field
attraaing the resources to underwrite the cost of his or her work. Barring success at either,
the researcher might exit the field shortly thereafter for
researcher can initially "survive" in the field, a research
in place
it
would require
fertile territory.
several years of data colleaion
and
analysis.
By
But
if
established;
the
once
the sixth year, a
reached, at which time the success of the program leads the researcher to
critical
point
remain
in the field
is
more
program might be
with linle probability of ever leaving, or lacking success (or intellectual
interest), the researcher
might decide
to
move
in
an alternative research direction.
After controlling for organizational type, geographic location, and the field's phase of
growth,
it
is
found that the
size
of the community, the community's organizational
dispersion, the researcher's productivity,
statistically significant variables in
and the
researcher's collaborative
TTie size of a researcher's organization (in terms of the
same
field)
is
found
produaivity in the
to be only
field
is
network are
explaining the length of a researcher's contribution-span.
number of researchers working
in the
weakly significant, and the organization's cumulative
found to be not
significant.
18
These findings provide support to the often-cited importance of the existence of critical
mass within a
field (that
the field
is,
rate of progress to be sustained
is
membership
sufficiently large in
minimum
community', the data indicate a
perhaps most interesting,
is
to enable a healthy
by researchers). In the case of the cochlear implants
about 250 researchers.
effective size of
that critical mass pertains to the
same
within the same
number of
organization
important than the number of researchers working in the
The
is
less
working
significance of a researcher's co-author network
is
the
in
is
as
organization: the
researchers
What
community
field
opposed
to the
whole.
field as a
also interesting in this regard, since
it
underscores the importance of the researcher being connected to the larger research
community
The
in collaborative activities.
present analysis has certain limitations,
based studies and
some of which
in this research
to understand the process of a
is
are
more generic
some of which
in nature.
new
field's
are peculiar to literature-
Given that the primary
emergence,
it
will
interest
be necessary to
construct a data set that implements a time-varying covariate data structure. Such an
approach
will
permit a more careful scrutiny of the dynamic phenomena that affect a
researcher's contribution-spans. Furthermore, this
whether or not changes
future
momentum
in
of the
the hazard rate of a
field. It
is
approach
will
community can
also necessary to
allow for the examination of
serve as an indicator of the
determine the extent to which the
present and future findings from the cochlear implants field can be generalized to other fields
of science and technology and to examine the importance of other explanatory variables
in
understanding contribution-spans.
Work
is
currently
underway
to address these issues. First, a preliminary investigation
suggests that data from the literature
is
structured in such a
manner
that time-varying
covariates should be feasible to create. Second, data sets for ten additional fields are currendy
being construaed, with
national
and
fields
varying in terms of their size and disciplinary composition, the
sectoral distribution of their researchers, their
commercial impact, and the
degree to which they have succeeded in becoming well-established, institutionalized research
communities. Third, further studies
will
be supplemented with other data, derived both ft-om
the literature and from other sources. For example, data from the cochlear implant literature
is
currendy being gathered with respect to the nature of the work being conduaed by each
researcher (such as basic research, applied research, development, or clinical research), and
the particular "technological trajectory" each researcher
versus multi-channel device).
Other kinds of
is
pursuing (such
as a single-channel
data, such as annual funding levels
extent of market commercialization, are being sought as well.
and the
19
TABLE
1
Variables used in the analysis
CATEGORY
and their definitions
20
TABLE
ML
2
ESTIMATION OF CONTRJBUTION-SPANS USING
DIFFERENT DISTRIBUTIONS
21
TABLE
ML
3
ESTIMATION OF CONTRJBUTION-SPANS:
LOG-LOGISTIC DISTRIBUTION
bb
204
MIT LIBRARIES DUPL
3
TOaO OQVSbTie
T
Date Duef-s-'iz.
JULO6«08
Lib-26-67
MIT LIBRARIES
3
IDflO
DD75bT12
T