ms^m
HD2R
.M414
ALFRED
P.
WORKING PAPER
SLOAN SCHOOL OF MANAGEMENT
Technological Progress and the duration of
contribution spans
Michael A. Rappa
Massachusetts Institute of
Technology
December 1991
Raghu Garud
Ko«nraacl Debackere
Rijksuniversiteit
Gent
New
Sloan
York University
WP # 3393-92
MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE. MASSACHUSETTS 02139
Massachusetts Institute of Technology
Technological Progress and the duration of
contribution spans
Michael A. Rappa
Massachusftts Institute of
Technology
December 1991
Forthcoming
New
Sloan
in Technological Forecasting
1991
Raghu Garud
Koenraad Debackere
Gfnt
Rijksuntversiteit
and Social Change,
York University
WP # 3393-92
Vol. 42,
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
Alfred P. Sloan School of
Management
Massachusetts Institute of Technology
50 Memorial Drive, E52-538
Cambridge,
02139-4307
MA
No. 4 (199*)
;M
9 1992
Technological Progress and the Duration
OF Contribution Spans
Michael A. Rappa, Koenraad Debackcre and Raghu Garud'
December 1991
ABSTRACT
This study uses the scientific
and patent
literature as
a source of data
analyze the relationship between author!inventor contribution spans
the rate
to
and
of technological progress in two chemical fields. Using survival
examine the probability that an individual
analysis statistics, the authors
of time and the probability
a specified period of
contribute in the future. The authors aLo test the
will contribute to the field for a specified length
that
an individual, having contributed
will cease to
time,
significance
of several
to the field for
covariates in predicting the length
of contribution
spans.
introduction
Predicting the rate of technological progress within a given field
problem
for those individuals
Ideally, if
who
are responsible for the allocation of scarce resources.
managers and government policy-makers had
them
indicators to enable
resource allocation
to predict the rate
would be
of technological progress in a
is
no small
an array of
field,
optimal
and policymakers
to
direction of technological advancement will largely
determine their firm's or nation's competitive performance
century. This
at their disposal
assured. Indeed, the ability of managers
comprehend the pace and the
an enduring
is
task. Historical
in
world markets into the next
accounts of industrial evolution, such as with the
development of semiconductors, videocassette recorders, and personal computers, show the
immense
some
difficulties
firms encounter
when
conft-onted by
new
technologies
[1
,
2, 3].
'Michael Rappa is an assisunc professor with the Massachusetts Institute of Technology, Sloan School of
Management. Koenraad Debackere is a Fulbright post -doctoral research fellow at MIT, and a research associate with the
Vierick School, Rijkuniversiteit Gent. Raghu Garud is an assisunt professor with the Stem School, New York
University. The authors thank Hans van Cool, Jef Vincent and George Evens for helpful comments and assistance. This
study was funded,
from
ICM
in part,
with a grant from the
(Brussels). Preliminary results
International
Meeting held
in Brussels,
of
DSM
Research. Dr. Debackere 's research was supported by a fellowship
this research
were presented
Belgium, June 24-26, 1991.
at the
Decision Sciences Institute First
Undoubtedly, there
new
is
technologies emerge.
an obvious need to enhance our understanding of the way in which
To
end, different methodologies, ranging from qualtitative
this
case studies to sophisticated quantitative forecasting models, have been developed [4, 5, 6,
7].
These developments have established technological forecasting
discipline in
own
its
right.
However, the many
and the often limited usefulness of
forecasts
characteristic of technological
pitfalls
outcomes have been noted with striking
their
As a consequence, technological
regularity [8, 9, 10].
forecasters face a
one hand, indicators of technological progress have often been
intense global competition in industry
technology have
One
made
approach
to
the search for such indicators
forward.
problems
The
On
illusive.
all
the
more
who
solve the
for science
urgent.
and reassessment of a particular
rapid, researchers will be
more
likely to stick
reap the rewards of their work. Conversely,
may
The
its
to
formulate their
own
field,
opinions
field.
Based
researchers will decide (or
Simply
stated, if progress
is
with their research agenda long enough to
if
progress
is
more
slow, researchers will be
out more promising research areas where they can apply their energies. As a
be possible to ascertain the relative rate of progress in a
the duration of the contributions
shaping
is
basic rationale for this perspective follows that in the process of solving
their assessment
it
field
problems that enable the technology
influence the decision) where to best apply their energies. 2
result,
and
are responsible for creating progress;
at the technological frontier, researchers continually
likely to seek
the
the other hand,
regarding the rate of progress and the probability of success within a particular
upon
On
understanding the rate of technological progress within a given
namely, the scientists and engineers
move
dilemma.
and constrained government budgets
on the aaions of those individuals who
to focus
an academic
as
made by
those individuals
who
field
by examining
are actively involved in
technological progress.
following study focuses on
how
long researchers persist in a
the relative rate of technological progress. Scientific
and patent
field in
order to assess
literature are
used
as a
source of data to measure the length of time that individual authors and inventors
^This basic rationale has been a recurring theme
scientists as "investors
of credibility.' That
is,
among many
scientists
sociologists of the sciences,
who
have
come
to view
are likely to invest their credibility in those specialties
they hypothesize the probability of reaping rewards to be the highest [11].
where
contribute to the field and to determine statistically the survival and hazard rates as well
some of
the factors associated with
panicipation of individuals
a field: that
is,
is
longevity.
examined through an
the time span between their
and
first
3
In panicular, the duration of the
analysis of their "contribution spans" in
last
an analysis of the contribution spans, estimates are
paper or patent contribution.
made
author/inventor's contribution span will extend a given
probability that, having contributed a given
to contribute in the future.
ofi (1)
as
years,
and
number of years, an author/inventor
(2) the
will cease
Furthermore, the relevance of a number of covariates
two stereoregular polymers
— EPDM
and polypropylene
(a plastic) as
on the independent evaluation of individuals
and polypropylene
5
The
—were
choice of
selected
EPDM
(a
was based
solely
chemical industry that these
fields
comparative
in the
in
examined.'*
is
the technical fields for comparative analysis in this study.
synthetic rubber)
From
the probability that an
number of
prediaing the duration of author/inventors' contribution spans
Catalysts for
as
test cases
experienced markedly different rates of technical progress over the past several decades.
Indeed, the historical record reveals that since the discovery of the Ziegler-Natta catalyst
process in the 1950s, ft-om which both processes are based, the rate of catalyst development
in each field diverges significantly. In the case
of polypropylene, four major breakthroughs
have lead to the development of three well-defined generations of
EPDM
catalysts
have evolved
less rapidly.
the past fifteen years no radically
EPDM
new
are generally considered to be
No
major breakthroughs have occurred, and
processes have been developed.
first
some
may
insight
researchers
is
^For lack of a better term, we
and polypropylene
of the rate of technological
will refer to these individuals as
author/inventors in order to erKomp>ass individuals
scientific publications or patents.
5Tlie historical development of
must be noted
used in
discussed in [12].
whose contributions may include
*It
catalysts
be gained into the question of whether or not the duration of
their participation in the field can serve as an accurate indicator
^Tlic methodology
The
in
generation technology.^
By comparing the contribution spans of EPDM
researchers,
catalysts. In contrast,
EPDM
and polypropylene
is
described in [13, 14, 15].
that although the fields realized different rates of technical progress in catalyst development, this
does not imply that one
field
is
necessarily unsuccessful in a
successes, although polypropylene has
commercial sense. Both
become much more widely
used.
fields arc
considered commercial
progress.
It is
expected that
EPDM
researchers
would be more
likely to leave their field
sooner than their counterparts in polypropylene.
DATA COLLECTION AND METHODS
Commercial
electronic databases were used to identify patents
related to the fields of
EPDM
and polypropylene
were searched on-Une using
a set of
lexicon of author/inventors
and might be
The
of a document.
catalyst
key terms that are
either in the
development. The databases
known
title,
to be
commonly used
searches resulted in the retrieval of 1,383 polypropylene-
of documents retrieved were patents
in the
abstraa or classification terms
EPDM-related patents and publications between 1955 and 1989. In each
EPDM
and publications
and 613
case, the majority
(60% of the polypropylene documents and 78% of the
documents).
The documents were
retrieved
and were temporarily placed
electronically
in
a
bibliographic relational database operating on a personal computer. This allowed for a
careful
inspeaion of each document
in
order to ensure the accuracy and integrity of the
search procedure. Since multiple source databases were used,
it
was necessary to remove
duplicate documents. In addition, while inspecting the database, an effort was
remove
misclassified
documents that did not pertain
to
EPDM
made
to
or polypropylene catalyst
development. In the process of inspeaing the documents, any that seemed inappropriate
were flagged, so that an individual active
to
its
in these areas
could make the
final
judgment
as
due
to
relevance. Furthermore, to avoid problems of incompleteness in the later years
patent lags, as well as in the earlier years, the databases were constrained to the period
fi-om
1960 to 1986.
The
data
collection
to
described
above
ultimately
resulted
in
the
613
EPDM
and 1,314 polypropylene patent applications and publications
1986.
The
contribution span data subsequently used in this study were
identification of
from 1960
procedure
derived from these documents. However, before the documents could be used as a source of
data, they required extensive editing in order to create consistency
among
author/inventor
names and
the
their affiliation names. It
name of an author
frequently the case with commercial databases that
is
or an affiliation
is
not standardized across documents. Sometimes
the inconsistencies arise because of misspellings, but mostly they are the result of variations
in the use
of abbreviations, middle
initials, capitalizations
and hyphenations. Although such
a 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
author/inventor contribution spans. Therefore,
name of each author and
could be eliminated.
literature databases
is
A
it
was
in
determining the duration of
essential to meticulously inspect the
affiliation in the relational
database so that
all
inconsistencies
particularly unfortunate complication specific to the patent
the frequent absence of inventor names from corporate patents. This
required the use of multiple databases and the cross-checking of patent
numbers
in order to
obtain the missing data.
Upon
completing the editing of the documents, the database was used to identify each
who
author/inventor
procedure yielded a
contributed to the
total
field
over the twenty-seven year period. This
of 3,280 individuals. At
was
this stage, a statistical database
created containing several covariates for each author/inventor that were derived from
information obtained from the published documents
provides a
list
as well as
number of years
variable for the analysis, the contribution span,
that have elapsed
from the
first to
the last
each author.7 Although calculating the contribution span
some methodological
issue
of concern
is
is
is
only
is
known
is
calculated as the
patent or publication for
relatively straightforward, there
issues that arise that require ftarther explanation.
that for those author/inventors
polypropylene during the
span
last
who
that the length of their contribution span
in
EPDM
and
year of the database, the ultimate length of their contribution
entry year to the present year).
implemented
The primary
are active in the fields of
indeterminate. In other words, since these individuals have not yet
known
1
of the variables and their definitions.
The dependent
arc
other sources. Table
To
account for
analyzing the data [16,
17,
is
some minimum value
this, survival analysis
18,
left
19].
the
field,
(that
statistics
Such techniques take
is,
it
the
were
into
^ For example, if a researcher first published in 1975 and last published in 1980, the researcher's contribution span
would be caJcuiated as six years. Furthermore, it is assumed that a researcher who publishes in only one year has a span of
one year. Note that the contribution span is unaffected by the frequency of publication within a given year.
consideration precisely this kind of problem in the calculations with a procedure that
adjusts for the biases that right-censored data create.
Having determined the distribution of contribution spans,
what
might affea
faCTors
literature, a
how
it
is
interesting to
long an author/inventor contributes to the
varying covariates (that
is,
as
Using the
field.
number of covariates were construaed. Although they could be
examine
treated a5 time-
having values that vary yearly in the course of an author's
contribution span), the present analysis does not implement such an approach to formulating
the data
Therefore, the value for each covariatc
set.
is
taken according to the
year of the
last
author's contribution span. In this manner, several covariates were created, including two
dummy
variables to control for factors that
population examined.
coded according
to
First,
might account
for heterogeneity within the
the kind of organization in which each author
whether they are employed
in
is
employed was
an industrial or non-industrial
academic or government) research laboratory. Second, the country
in
(i.e.,
which the author
is
located was coded, and a covariate was created to signify whether the individual's
affiliation
located in a Western industrial country or in an Eastbloc country.
is
Additional
covariates
were created which
reflect
individual,
organization
population anributes. At the individual-level, a covariate was constructed to
author's productivity in the field as measured by the cumulative
and patents
produaivity
the field
—
to their credit.
—
An
number of
reflect
or
an
publications
organizational-level covariate was created to reflect the
or what might be considered as an organization's cumulative investment in
terms of the cumulative number of patents and publications assigned to the
in
author/inventor's affiliation.
Three population-level covariates were created
each
field in
who
publish or patent in the field in a given year,
Figure
1.
A
each year. Population size
is
measured
to reflect the size
in
as
terms of the
and dispersion of
number of
individuals
described below and illustrated in
second-order covariate, the square of population
size,
was created
in order to
capture any quadratic association between population size and contribution spans.
covariate
that
is,
is
a
The
third
measure of dispersion of authors/inventors among different organizations:
the extent to which the population
is
concentrated in a few organizations or spread
many. For
across
this
purpose, a Hirfindahl concentration
sum of
calculating the
organization annually,
Lastly,
market
aggregate
is
determined by
to reflect the maturity of the field.
EPDM
The
produaion of Western Europe, Japan and the U.S.
is
the cumulative
first is
and polypropylene produced annually). Because
world produaion are not available over the entire twenty-seven
maturity covariate
each
used.
is
terms of kilotons of
statistics for total
which
the squared share of author/inventors affiliated with
two covariates were included
size (in
statistic,
used instead. 8
is
number of patents granted
years, the
The second
in each field.
RESULTS
Using data from the
scientific
and patent
literature
on
EPDM
and polypropylene
published between I960 and 1986, the contribution spans for 3,280 author/inventors and
several
explanatory variables associated with each were compiled into a
database:
3,280
2267 (69%)
cases,
individuals in polypropylene and 1013
739 (22.5%) were aaive the
classified as censored. Sixty percent
firms
and nearly twenty-five percent
The
historical
growth
last
of the
(31%)
in
statistical
EPDM. Of
the
three years of the database and were therefore
total
population are employed in industrial
are located in Eastbloc countries.
in participation in each field
number of author/inventors contributing
researchers in each field in a given year
can be seen clearly in terms of the
to the literature (see Figure
is
1).
The number of
calculated to be the cumulative
number of
individuals entering the field (as evidenced by an initial publication or patent application)
subtracted by the cumulative
by
number of
individuals
who
have
left
the field (as evidenced
their failure to continue to publish or patent in a future year).
The
data were analyzed using the LIFETEST and LIFEREG procedures of SAS (v5.18).
Using the LIFETEST, the
first
step in the analysis was to
the survival and hazard functions for the data.
The
make non-parametric
lifetable
estimates of
approach was chosen. The
^The market data were provided by the marketing research department of a major chemical firm. The data
were checked for accuracy with dau from Kline &C Co.. an organization that publishes statistics on the chemical industry.
results
of
this
procedure are illustrated
monotone decreasing function and
in
Figure
2.
TTie survival funaion for each field
are nearly identical. Tests of
is
a
homogeneity of the
survival curves stratified by field can not reject the null hypothesis that the strata have
identical survival distributions. The probability of a author/inventor's contributions span
lasting
two
years or longer
is
about 0.3. After two years, the survival rate continues to
diminish, eventually leveling-off at about 0.07 for contribution spans of 15 years or more.
likely
he or she will be to leave
with the
it,
points. Indeed, the risk of leaving the field
first
is
and
sixth years being particularly critical
highest within the
first
year.
>
E
3
C/5
13
10
Duration of Contribution Span
FIGURE
The
15
15
10
15
Duration of Contribution Span
(years)
(years)
Non-Parametric Estimates of Survival and Hazard Functions far Author!Inventor Contribution
Spans in EPDM and Polypropylene
2:
next step in the analysis wa5 to determine the parametric model that best
fits
the
distribution of contribution spans. Although non-parametric analysis permits certain
assumptions that can be made about the shape of the survival distribution
it
is
non-monotonic), nonetheless we decided
distributions for goodness of
fit.
The
basic
to
examine
model adopted
(for instance, that
statistically several different
for the analysis
is:
Y = Xp + ae
where
Y
is
unknown
the log of the contribution span,
regression parameters,
a
assumed distribution. This model
is
is
a scale
X
is
the matrix of covariates, P
parameter and £
is
a vector
is
a
veaor of
of errors from an
referred to as an accelerated failure time
model
10
because the effect of the explanatory variables
is
to scale a baseline distribution
times. In order to determine the underlying distribution that best
are provided in
Using LIFEREG, the
Table
2.
The parameters
Newton-Raphson algorithm. The
likelihood funaion.
the data, four
of
fit
procedure for the entire sample
this
by
are estimated
overall
Minus two times
results
maximum
of each model
is
likelihood using a
represented by the log-
the log-likelihood value has a chi-square distribution
with appropriate degrees of freedom. Using the baseline model, the goodness of
each distribution
is
failure
were evaluated: the exponential, Weibull, gamma, and log-
different types of distributions
logistic distributions.
fits
of
fit
for
evaluated in term of minimizing the absolute value of the log-
likelihood score. As a result, the log-logistic distribution (with a log-likelihood score of
-1287) was chosen and became the
basis for estimating the regression coefficients
explanatory variables in the model. This
findings, suggests a
stratify
which
consistent with the non-parametric
is
non-monotonic hazard funaion.
Table 2 also shows the
to
result,
of the
on the two
results
fields
being investigated.
distribution indicate no field effect.
of contribution spans for the two
non-parametric analysis. This
In the next step, the
dummy
of the inclusion of a
There
is
is
estimates for the log-logistic
no evidence to suggest that the distribution
fields are divergent,
result
The
variable (Technical Field)
which confirms the findings from the
consistent across
all
distributions.
model was estimated with LIFEREG
in a
sequence of steps by
adding each covariate into the equation using the log-logistic distribution. Since
desirable to investigate
whether the significance of covariates
differs
among
the two
it
is
fields,
the modelling results are presented separately (see Tables 3a and b). In the case of each
field,
the addition of each covariate has the effect of generally improving the log-
likelihood score.
Model 9 was chosen
as the baseline for
the comparison to understand the
effea of the covariates.
The
estimation results of
Model
9 indicate a
number of
differences in terms of the
significance of the covariates examined. First, several covariates that are significant in the
case of polypropylene are not significant for
industrial (-), patents (+),
and market
EPDM: namely,
organization productivity
size (+) are all significant for
(+),
polypropylene but not
11
for
EPDM. Second, although
the population size variables are significant in both cases, the
signs of the coefficients indicate a different relationship in each case.
between the two
fields are
The only
similarities
the significance of author/inventor produaivity (+), and the lack
of significance for the Eastbloc and concentration covariates.
In the case of the population variables, the
significant right
from
their initial inclusion in
the negative coefficient for population size
second-order term
implies
first-
and second-order population terms
both
fields.
combined with the
after
it
is
small,
size
its
is
population, at which
contribution spans.
this size (refer
population
its
may
size
1
is
critical
positive coefficient for the
data indicate that
1).
mass for the
become
sufficiently large to
In contrast to the
EPDM
EPDM
and the length of contribution spans
EPDM
1
00
-2
-3
size
and
when
the
-
1
1
author/inventor
significant in increasing
population never grew beyond
case, the relationship
in the case
negatively sloped, and increasingly so, as the population grows larger.
3
EPDM,
50 individuals the slope of the curve turns positive. This
be a point of
size
back to Figure
The
3).
interesting to note that the
It is
of
negatively related to the length of contribution spans; but
reaches the size of about
result suggests that there
in the case
U-shaped relationship between population
a
author/inventor contribution spans (see Figure
population
However,
are
between
of polypropylene
is
12
DISCUSSION
Using the
analysis
scientific
and patent
literature as a source
of data,
of the contribution spans of author/inventors
paper provides an
this
EPDM
the field of
in
and
polypropylene catalyst development. Non-parametric estimates of the survival rate and
hazard rate are made, and
most
it
is
found that the distribution of contribution spans follows
closely a log-logistic function. In addition, a statistical
between the hazard
The
rate
and
of covariates
a set
is
model of the relationship
examined.
findings of this analysis indicate that the sample survival
3,280 author/inventors
in
EPDM
and polypropylene
and hazard functions
are generally similar.
percent of the author/inventors have a contribution span of more than two years.
an author/inventor ceasing to contribute to the
contribution span.
remains
year,
for
fairly
when
The hazard
constant.
The
the hazard for
field
is
greatest in the
rate declines sharply after the first year
between the two
critical difference
EPDM
researchers
The
fact that
understandable, since at that point not
field.
much
The
risk
of
year of their
and subsequendy
comes
in
the second
suggestive of the basic hypothesis that
is
individuals confronted with slow progress in a field will be
alternative area of research.
About 30-
more than one-and-a-half times higher than
is
polypropylene researchers. This result
fields
first
for
it
more
likely to seek an
the early years that are most critical
is
of a researcher's career has been invested
Needless to say, the longer one stays in the
field,
the
less likely
is
in the
they are to leave
it,
regardless of the rate of progress.
An
examination of relationship between several covariates and the length of
contribution spans indicates that
noteworthy
spans.
It
is
EPDM
and polypropylene are quite
the difference in the relationship between population size
appears that in the rapidly progressing
field
of polypropylene
different.
Most
and contribution
catalysts, the larger
the field became, the shorter and shorter the duration of contribution spans became. This
perhaps the result of the competitive pressures that arise as more people work in a
contrast,
EPDM
shows
a
much
is
field. In
different relationship, suggesting that a lack of individuals in
the field had a detrimental effea on the length of contribution spans.
13
Another important difference
market
size
is
lies in
significant
is
author/inventor contribution spans. This
of technology
The
based studies and
set that
more
may have
is
and has a positive
In the case of
an imporunt "market pull" component.
some of which
to
is
on the length of
effect
are
more generic
understand progress
some of which
in nature.
in a field,
it
are peculiar to literature-
Given that the primary
will be necessary to
dynamic phenomena that
scrutiny of the
affect
interest
construa a data
implements a time-varying covariate data struaure. Such an approach
careful
EPDM,
suggestive of the notion that rapidly progressing
present analysis has certain limitations,
in this research
size.
not significant in influencing contribution spans. However, in the case of
polypropylene, market size
fields
the efFea of market
will
permit a
an author/inventor's
contribution spans. Furthermore, this approach will allow for the examination of whether or
not changes in the hazard rate of a
of the
field. It is also
findings from the
community can
serve as an indicator of future
momentum
necessary to determine the extent to which the present and future
EPDM
science and technology
and polypropylene
and
to
can be generalized to other
field
fields
examine the importance of other explanatory variables
of
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
currently being constructed, with fields varying in terms of their size and disciplinary
composition, the national and sectoral distribution of their author/inventors, their
commercial impact, and the degree
established,
institutionalized
to
research
which they have succeeded
this
becoming
communities. Third, funher studies
supplemented with other data, derived both from the
We
in
believe the approach towards the scientific
literature
and from other
and technological
well-
will
be
sources.
literature outlined in
paper offers new perspectives to the application of bibliometric methods to
technological forecasting. Instead of predicting the growth and the decline of particular
fields
by looking
at
publication or patent volumes, our research points to the usefijlness of
14
publication and patent information in determining the contribution spans of researchers.
Analyzing contribution spans
may
eventually serve
government policy-makers who
are responsible for
technologies globally. In essence,
we
monitoring the progress of emerging
are proposing a technique that allows
worldwide pulse of technological progress by measuring the
commitment
our aim
is
to a field.
By
managers and
as a useful tool for
rate
one
of change
to
gauge the
in researchers'
focusing on the determinants of researcher contribution spans,
to shift attention
away from prediaing the technological future and towards
understanding the underlying fijndamentals of researcher behavior. Improvements
understanding of survival and hazard
rates for researchers in a field
may
in
our
ultimately lead to
the identification of critical factors and events that can inform our policy decisions
regarding emerging technologies. In this manner,
researchers' persistence using contribution spans
rate
of technological progress
may
in a particular field.
we
suggest that measurements of
serve as an indicator of
Perhaps
we can most
change
in the
clearly envision
our technological future by understanding in a comprehensive and systematic manner the
sustained
commitment of
researchers to the ideas they are pursuing today.
15
NOTES
[I]
Braun, E. and Macdonald,
UK
Cambridge,
[2]
S.,
Revolution in Miniature,
Cambridge University
Press,
(1978).
Rosenbloom, R.S. and Cusumano, M., "Technological Pioneering and Competitive
Advantage: The Birth of the VCR Industry," California Management Review, 29, 4
(1987).
[3]
D.K. Smith and R.C. Alexander, Fumbling the Future:
William Morrow,
Ignored, the First Personal Computer,
[4]
[5]
How Xerox Invented,
New York (1988).
then
Rowe, C, Wright, G. and Bolger, F. "Delphi: A Reevaluation of Research and
Theory, " Technological Forecasting and Social Change 39,235-51 (1991)
Webler, T., Levine, D., Rakel H., and Renn, O., "A Novel Approach to Reducing
The Group Delphi," Technological Forecasting and Social Change 39, 253-
Uncertainty:
63(1991).
[6]
Martino,
[7]
Girifalco, L.,
Technological Forecasting
J.,
and Planning, North Holland, New York (1983)
The Dynamics of Technological Change, Van Nostrand Reinhold,
New
York (1991).
[8]
Einhom
Illusion
[9]
and Hogarth, R,M. "Confidence
in
Judgement: Persistence of the
of Validity," Psychological Review, 85,
5,
395-476 (1978).
H.J.,
Hogarth, R.M., and Makridakis, S. "Forecasting and Planning:
Science, 27, 2, 1 15-38 (1981).
An
Evaluation,"
Management
[10] Schnaars, S.P.,
Press,
New
[II] Latour B.,
and Edge,
Megamistakes and the Myth of Rapid Technological Change,
The
Free
York (1989).
and Woolgar,
eds.,
MIT
S.
Press,
"The Cycle of Credibility,"
in Science in Context,
Barnes
Cambridge, Mass. (1982).
[12]
Rappa, M.A., and Garud, R,, "Using the Literature in the Study of Emerging Fields
of Science and Technology," MIT International Center for Research in the
Management ofTechnolo^, Working Paper #59-92 (1991).
[13]
McMillan, F.M., The Chain
[14] Morris, P.J.T.,
Straighteners,
MacMillan
Press,
Lx)ndon (1979).
The American Synthetic Rubber Industry, University of Pennsylvania
Press, Philadelphia (1989).
16
[15] Sicilia, D.B.,
"A Most Invented Invention," Invention
& Technology
Spring/Summer,
45-50(1990).
[16] Elandt-Johnson, R.C.,
Wiley
&
Sons,
New
[17] KaJbfleisch J.D.,
Wiley
&
Sons,
[18] Allison, P.D.,
New
Analysis, ]o\\vi
Prentice, R.L.,
The
Statistical Analysis
ofFailure Time Data, John
York, (1980).
Event History Analysis, Sage Publications, Newbury Park (1984).
The Econometric Analysis of Transition Data, Cambridge University
Cambridge, (1990).
[19] Lancaster, T.,
Press,
and
and Johnson, N.L., Survival Models and Data
York, (1980).
TABLE
1
Variables used in the analysis and their definitions
CATEGORY
TABLE
ML estimation
2
of contribution spans using different distributions
c-
o^
:
d^
^
o
O
o
o
o
9
o
"
§s
MIT
3
\
TDflD
LIBRARIES DUPL
007Shfll7 D
-^iys :-
i*
'"
MIT LIBRARIES
3
TOaO 0D75bfll7
i
J f'^
D
i