'r*H;».*
-
l!
ilfllr
}mim
IHlMlli!
'''i!'i"!|.j'''''i!?!''r;P^-^
iiiyikUi
"^o/
TBC'^^V
htOm
HD28
.M414
ALFRED
P.
WORKING PAPER
SLOAN SCHOOL OF MANAGEMENT
Scientists at Major and Minor Universities:
Mobility along the Prestige Continuum
Koenraad Debackere
Gent
Rijksuniversiteit
September 1992
Michael A. Rappa
Massachuietts Institute of
Technology
WP # 3479-92
MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
Massachusetts Institute of Technology
Scientists at Major and minor Universities:
Mobility along the Prestige Continuum
Michael A. Rappa
Koenraad Debackere
Rijksumversiteit Gent
Massachusetts Institute of
Technology
WP # 3479-92
September 1992
©
1992 MASSACHUSETTS INSTITUTE OF
Presently under review
at
TECHNOLOGY
Sociology of Education.
Alfred P. Sloan School of
Massachusetts Institute
oi-
Management
Technology
50 Memorial Drive, E52-538
Cambridge,
MA
02139-4307
M.I.T.
NOV
UBr,...,.Jl
2 1992
REGtivtLi
.
Scientists at
Major and minor
Universities:
Mobility ALONG the prestige Continuum
ABSTR.ACT
This paper investigates the relevance of institutional prestige on career patterns of
"major and minor
scientists at
involved in the early stages
prestigious universities? (2)
paths in terms ot
of-
Do
employment
working
to
in
is
development more
a fields
( 1 )
likely to
Are
scientists
sector?
and
(3)
Are graduates from prestigious universities who
at
prestigious
presented trom an international survey
ol"
more than 700
find
to
The
universities?
scientists
community.
is
found
in
the
initial
graduation. Beyond five years, the effect of graduate school prestige
This finding holds for
who
scientists
prestige of a scientist's graduate school
be an important indicator of the prestige of his or her academic appointment
significant.
entered the
are
graduate from more
employment
likely
the field of neural networks.
five years after
who
graduates from prestigious universities pursue different career
choose academic careers more
Empirical evidence
posing three questions:
universities,"
field
scientists
after
it
who
entered the field early, as well
gained widespread legitimacy
in
is
non-
as tor
those
the scientific
Scientists at
Major and Minor
Universities:
Mobility along the Prestige Continuum
Given
sociologists' interest in occupational
and career patterns (Giddens 1989),
surprising that the scientific profession has received close scrutiny. Indeed,
of the sociological dynamics ot a
more
scientist's career
closely than the determinants of career
research institutions
success
— the
numerous
aspects
have been examined, but perhaps none
advancement along the prestige continuum of
so-called major and
minor
universities.
How much
ot career
attributable to the intrinsic quality ot a scientist's research accomplishments as
is
reflected in published
words,
not
it is
work? Or, are other more "particularistic" factors
how important
is
the institutional prestige of
status of
ones family? The
back to the Research Program
in
origins of
many
work? In other
ones doctoral degree granting
university, the standing of one's thesis supervisor in the scientific
economic
at
community, and the
socio-
of these investigations can be traced
the Sociology of Science initiated at
Columbia
in
the late
been directed toward understanding the
relative
1960s.
A
considerable
amount of
effort has
influence of individual productivity and accomplishment versus particularistic criteria in
determining
who
institutions.
Although the
receives
academic appointments
results are
at
the most prestigious departments and
sometimes contradictory (Finkelstein 1984), the
empirical evidence usually confirms the importance of institutional prestige. For example,
studies by
Hargens and Hagstrom (196^) and Cole and Cole (1973), find that an
individual's
accomplishments are equally
as
important
as one's
academic background
securing a prestigious academic appointment. Crane (1965 and 1970) and
that the prestige factor
—
supervisor
is
— both
significantly
in
more
Long (1978)
in
find
terms of one's degree granting university and ones graduate
influential than one's research
a position. In reviewing the evidence
accomplishments
in
securing
from both streams of research, Finkelstein (1984)
is
led
CO
conclude chat
che cime of
"...ac
iniciaJ
appoincmenc,
ic
is
much more
che prescige of one's
terminal degree and one's graduace sponsor than one's scholarly produccivirv' which will lead
CO a
good academic appoincmenc."
Turning che quescion around, some
invescigacors suggesc that che prestige of an
academic deparcmenc may be an imporcanc concribucing faccor
co a sciencisc's produccivity
(Hargens and Hagscrom 1967; Allison and Long 1990). Alchough Hargens and Hagstrom
(1967) are unable to show chac inscicucional scanding influences productivity on che
individual level, they do provide evidence on che aggregate level. Furthermore, Cole and
Cole (1973) and Long
(
1978) find that institutional prestige
research performance, while
research performance in
and citation
decermming
within the scientific
pioneering a
scienciscs
the
be nearly as important
finds institutional prestige
amount
who
we
more imponant than
ot recognition (in terms oi rewards, honors,
invescigace (urther the significance ot inscicucional scratification
community
new
as
it
relates to che inclinacion ot scientists
field ot research.
are involved
from more prestigious
m
In particular,
we examine
che early scages of a fields
universities.' (2)
different career paths in terms of
Do
who
are involved
three questions: (1) Are
developmenc more
likely co
graduate
graduates trom prestigious universities pursue
employment
sector (academic, industry,
government)
within che sciencific community? and (3) Are graduates from prestigious universities
choose academic careers more
likely to find
L'nlike earlier sociological studies,
early, before
we mean
it
is
widely accepted by che
chose scienciscs
recognized
as
t-requenc\') that accrues co a sciencisc.
In che presenc paper
in
Crane (1965)
may
as significanc,
who
iniciate
we
employment
who
at prestigious universities?
focus specifically on scientists
resc of che sciencific
who
communicy. By
and continue working
in a field
enter a field
"early encrancs,"
before
ic
is
widely
or perhaps even legicimace by their peers. Empirical evidence
suggests that such scientists, scaciscically speaking, are relacively rare: alchough the probability
—
of a
It,
remaining with
scientist
a given field
the likelihood a scientist will persist
of research increases the longer he or she stays with
more than
1992). Despite their scarcity, the scientists
behind change
in science.
determination, they
we have
mean
to
may
By
who
few years
a
low (Rappa and Garud
fairly
enter a field early are essentially the catalysts
virtue of their unconventional
ultimately lead the
is
way
in
problem choices and unrelenting
new
creating a
isolated early entrants for in-depth examination, in
doing
underestimate the significance of contributions to a
research specialty.
so
field
'VCTiile
we nonetheless do not
made by
scientists
who
follow afterward.
Institutional stratification within the scientific
with respect
to scientists
of a universit>' has
who
enter a field early.
some relevance
in
communit)'
It
may have
take chances to
institutions
more
the resources that
readily explore
may tend
to reinforce
ver>' well
the pioneering behavior of
what might be called the "backwater hypothesis."
universities
may
raises
new
On
among
On
be that the relative stature
its
faculty
scientists
who
are inclined to
the other hand, the prominence of such
their scientists a
more cautious
doing science that extends rather than challenges conventional thinking.
that in the early stages of emergence, radically
and students
the one hand, prestigious research
would enable those
fields.
an interesting question
new streams
It is
attitude toward
not
uncommon
of research can lack legitimacy
within the scientific community. Unable to convince their mainstream colleagues,
scientists
may
seek haven at lesser
research. In the
same
vein,
known
institutions in order to pursue their unconventional
one might extend the argument
pursue pioneering research agendas are any more or
prestigious universities
The
some
less
to ask
whether students who
likely to obtain
positions at
upon graduation.
case of "cold fusion" research provides a recent illustration (MaJlove 1991). Setting
aside the issue of whether or not cold fusion has merit, the events surrounding this discovery
exhibit
how
institutional
prestige
may
play
a
role
in
the
way
scientists
approach
—
unconventional research.
confirm
it,
who found
Through
The
The remarkable
claims ol cold fusion and the subsequent effons to
quickly degenerated into a major scientific controversy pitting those scientists
evidence of
effect against those
its
who saw
it
a5
spurious
the course ot the debate, undercurrents of elitism emerged
lesser
known
institutions, while
known
"schools not
for
perceptions did not actually
fit
unconfirmed reports came from,
What
team."
football
their
the reality.
A
one
as
come from
is
put
it,
that such
shows
little if
confirm cold fusion
to
Cold fusion
a cautionarv tale
is
can become muddled
issues ol institutional prestige
scientists.
scientist
close examination ot the record
research. Nevertheless scientists are mindful ot perceptions.
how
likely to
most interesting
is
any correlation between institutional prestige and the propensity
that underscores
among some
more
suggestion was that reports which confirmed cold fusion were
not scandalous.
\i
in
scientific
debates.
THE NEURAL NETWORK RESEARCH COMMUNITY
In order to
examine these questions empirically, we take
international survey
we conducted
of
more than seven-hundred
development of neural networks. A neural network
system that
is
inspired by models of the
human
scientists
By using
design, a neural network system has certain features that
a biological
make
performance of
information
Another
task.
Unlike
in
a neural
it
is
a task,
it
as
it
not
processes
becomes increasingly more accurate
distinctive feature of a neural
a
is
model
in
in
its
form and
programmed
in
trained with data. This implies that the computational
network improves with experience:
performing
unique
it
function from conventional computers. For example, a neural network
the usual sense, but rather
working on the
a type of information processing
is
brain.
paper a recent
as the basis of this
normal computer with
network
is
its
its
degree of parallelism
a single or small
processing units, a neural network has a very large
in
number
number
more and more
response.
in
processing a
of sophisticated central
of simple processing elements that
operate simultaneously on a computational problem. These features allow
might be very
certain tasks that otherwise
Neural networks are also referred to
neurocomputers
(see
DARPA
Neural networks have
computer technology.
difficult using existing
connectionist systems, adaptive systems, or
as
1988).
a
considerable history of development, stretching back to
theoretical explanations of the brain
and cognitive processes proposed during the 1940s. In
the early years, scientists formulated and elaborated basic models of neural
they then used to explore
perform
to
it
phenomena such
a5
computing
adaptive stimulus-response relations
in
that
random
networks. By the 1960s there were several efforts to implement neural networks, the most
notable being the single-layer "perceptron.
was considered a watershed, but
from
scientists
more
at
the
"
Among
same time
neuraJ network scientists the perceptron
it
served as a lightning rod for criticism
interested in the burgeoning field of anificial intelligence.
neural networks, as exemplified by the perceptron, quickly
to
became seen
as
The
idea of
almost antithetical
the symbolic reasoning principles of artificial intelligence. Critical analysis of the
perceptron led some highly respected Al scientists to proclaim that the concept was
fundamentally flawed, and
casting
doubt
as
to
its
as
such, inappropriate for scientists to waste
legitimacy, antagonists of neural networks
dissuaded other scientists from entering the
field in larger
much
effort on.
may have
By
effectively
numbers (Minsky and Papert
1988; Paperr 1988).
The
controversy surrounding neural neuvorks notwithstanding, work continued during
the early 1970s with perhaps no
Undeterred
in their belief
more than
a few
hundred
scientists
in the field.
of the potential of neural networks, their persistence over the next
decade eventually paid-off By the 1980s, neural networks began
by
worldwide
scientists in a variety of disciplines,
such that the
legitimacy within the scientific community.
A
field
to
be viewed
soon achieved
in a
new
light
a position
of
professional society for neural network
scientists
was tormed, specialized journals and books were published, and the
it
is
difficult
a series
were held.
ot internationaJ conferences
While
first in
why
explain exactly
to
perceptions of the
changed
field
so
dramaticaJly, at least tour important technical events can be discerned: (1) the evolution of
the single-layer perceptron into a multi-layer system; (2) the rapid development of related
technologies that enabled scientists to develop, simulate, and diagnose neural networks of
greater sophistication;
biological processes;
significant progress
(3)
and
(4)
in
theoretical
understanding of neuro-
the contributions ot scientists pursuing the idea of parallel
distributed processing, the so-called PDP-group. In light ot these developments, as well as
others, interest in the field
became widespread, such
neural networks expanded rapidly.
membership from
The
a
By
the
ot scientists
size
of the
working on
field
swelled in
few hundred to several thousand scientists worldwide.
be typical oi emerging fields in
it
number
end of the decade the
evolution of the neural network research
have found that
that the
is
fairly
some
of- its
common
for
community
is
not unusual and
sociological characteristics.
new
fields to lack
From our
may
even
research,
we
widespread acceptance for long
periods, sometimes attracting controversy, other times simply being ignored by scientists. But
when
they do catch on, fields tend to
lesser extent, in several fields
grow
we have examined
Given the recent experience within the
opponunity
numbers
to
examine
of scientists
field
in great detail the
who
rapidly.
— including one
field that
ultimately faltered.
of neural networks, this case presents us with an
experience of pioneering scientists relative to large
follow in their footsteps.
METHOD AND
Through an
This pattern has occurred, to greater or
analysis of published
sources,
DATA.
including books, journal
conference proceedings for the two-year period from 1988 to 1989,
we
articles,
identified
and
more than
3,000
scientisti
were able
to
determine the exact address
Given the scope
countries.
determmed
in
worldwide working on the subject of neuraJ networks. From
to
oi'
of 2,037 scientists
for each
community,
the research
English was sent to scientists inquiring about
decision to begin working on neural networks,
neural network research in favor ot another
questionnaire was pretested
in the
United
factors that
(c)
problem
of the neural network research community, and
States.
a
is
in thirry-five different
twelve-page questionnaire
network
activities, (b) their
might lead them
to cease their
area, (d) their interaction
(e) their
demographic
Additional
to reduce potential interpretational difficulties arising
English
A
their neural
(a)
tests
among
we
survey questionnaire was
a
be the most appropriate method oi investigation.
this material,
with the
characteristics.
were conducted
m
those respondents for
rest
The
Europe
whom
second language.
Since thirty-seven scientists had more than one address during the time period
considered, a total ot 2,074 questionnaires were mailed
week
of-
data collection,
computer
we mailed
a follow-up letter
bulletin boards to alert neural
network
in
February 1990. After the third
and posted e-mail messages on
scientists ot the survey.
questionnaires, 162 were returned as undelivered bv the post office.
scientists
with more than one address were
among
in
Of
None of the
the 2,074
thirt\'-seven
the undelivered questionnaires. At the
completion of the data collection period approximately ninety days
later,
720 of the 1,8^5
questionnaires presumed to be delivered were completed and returned, yielding a final
response rate of 38.4 percent.
Some of
the factors that
may have
affected the response rate
include: the length of the questionnaire, the global scope of the survey,
and the
institutional
mobility of scientists.
The
representativeness of the respondent sample was evaluated in three ways: by
contrasting respondents with the original survey population in terms of their geographic
distribution, their institutional distribution,
and the distribution of
their disciplinary
backgrounds. In each instance, there
no
is
berween the
statistically significant difference
respondent sample and the survey population.
INSTITUTIONAL PRESTIGE
We
rank-order the universities
institutional prestige.
We
our survey database according
in
to
an index of
use as the basis oi our index the citation and publication data on
U.S. universities, which was compiled by Small (1990) and his colleagues at the Institute for
Scientific
Information and recently used by the Office of Technology Assessment (1991) to
rank U.S. universities. Citation impacts scores
(i.e.,
the ratio ol total citations to total papers
published') have been implemented in a variety ol studies to measure the relative eminence
of a scientist (Myers 1970) and prestige of academic depanmenrs (Roche and Smith 1978), a
laboratory's research performance (Mullins 1987; Narin 1987),
ot a country's scientific
community (Narin and Frame
and the competitive
stature
1989).
In our study the citation impact score ol publications lor each university over the fifteen-
year period between
1973 and 1988
institutional prestige.
An examination
is
used
as
a (continuous)
of the rank-order ol the top-100 U.S. research
universities suggests that citation impact scores have
institutional prestige (see Small 1990). Nonetheless,
measure pertains
to
it
the university as a whole and
departments, which can vary widely
institutional prestige that
may
arise
proximate measure of
good
is
criteria other
important
measure oi
to recognize that this
not to the prestige of individual
in a given university.
from
face validity as a
The
score also does not reflect
than research performance such
as
excellence in teaching, for example.
'
The continuous
prestige index
is
computed
as
follows (with ?,= prestige score for academic institution
1988
^
(
Pi=-
citations
)[
19^3
1988
(
X
publications
)[
1973
10
i):
Because the validity of making internationaJ comparisons with citation impact scores
not vvell-estabhshed, the present anaJysis includes only those respondents
trom U.S. academic
from 720
to
Most
longer reside there,
large majority
the countr)'
twenty-two of them held posts
educated and reside
of
whom (N
of the twenty-five respondents
left
in
graduated
Excluding non-U. S. universities reduces the original sample
institutions.
3^3 respondents, the
within the U.S.
who
is
=
who were
upon graduation. At
educated
the time the survey was conducted,
employed
in
academic
who were
— primarily government
103 (30-
labs,
percent) reside in industrial laboratories, and 38 respondents (11-percent) are
non-academic, not-for-profit institutions
U.S. but no
in the
For the 348 respondents
at foreign universities.
the U.S., 207 (59-percent) are
348) are currently employed
employed
laboratories.
The
in
sector
distribution oi respondents does not differ significantly from that of the original sample
(X-=3.35,
n.s.).
For each ot the 373 U.S. -educated respondents
impact score
for their
sample. Using the
graduate school. There are a
ISl
data,
we
also
appointments
at
86 different U.S.
scores for industrial employers.
all
cases
universities. \X'e
a total of
measures both tor their graduate school and
The continuous measure
universities (graduate schools
thereby creating
citation
five
impact scores
ot prestige
in excess
The 25
of 16.31.
exist, its
which
for their current
measure
tor
in
the
each
The respondents hold
tor
is
adequacy
as a
measure of
beyond the scope of the
whom we
calculate prestige
academic employer.
used to create an ordinal variable.
and current employers)
equal ranks.
but two).
205 respondents
is
a citation
do not calculate institutional prestige
Although industry data
is
we compute
institutional prestige
prestige for industrial labs requires closer inspection,
present study. As a result, there
the sample
of 104 universities represented
total
compute an
respondent's currerit academic employer (in
in
The 125
are divided into 20-percent intervals,
institutions in the rop-20 percent inter\al have
The 20-to-40
11
percent interval have scores ranging
from 13.55
to 16.17.
The 40-to-60
The 60-ro-80 percent
percent
interval have scores
have scores berween 10.64 and 13.48.
between 8.10
to
10.52.
The 25 remaining
below 8.10.
institutions have citation impact scores
The
inter\'al
distribution ot respondents by prestige of graduate school and by prestige of current
academic employer are shown
this table
have a
modus
in
in
Table
Both
1.
sets ot
academic institutions considered
the top-20 category. Inspection ol the
median values
in
for both
distributions further indicates that the majority ot respondents are in the top categories as far
as institutional prestige
is
concerned.
—
Insert
Table
1
—
about here
AND MINOR UNrV'ERSITIES
AND EARLY ENTRY TO THE FIELD
SCIENTISTS AT .MAJOR
According
respondents
scheme previously developed and reported, we
to a classification
as early
or late entrants depending
early entrants are those scientists
who
upon when they
begin research in a
field
enter the
before
it
field.
classify
In short,
obtains widespread
legitimacy within the scientific community. After a careful historical and statistical analysis of
the field ot neural network, examining
many
different factors
divided the sample into early and late entrants using 1984
nothing inherently significant about
any year between 1980 and 1984.
as
this year, in particular.
We
and testing
is
Indeed,
is
we could have chosen
tested the sensitivity of selecting
The
1984
as
the cut-off
results indicate that
robust.
By demarcating the sample
into
two periods, we do not mean
transition to legitimacy was instantaneous;
analysis. Nonetheless,
we
the transitional year. There
year by performing a discriminant analysis on the core survey items.
the categorization scheme
lor sensitivity,
something
—
or,
we do
perhaps,
12
to
imply that the
field's
so simply to preserve cases for the statistical
many
things
— unmistakably happened
in the
and the object of skepticism
early 1980s that transformed neuraJ networks
from
a major interdisciplinary stream of research.
An examination of
a curiosity
to
the scientific literature and
discussions with neural network scientists also supports our selection oi 1984. Prior to 1980
there were no
more than
neural network
scientists
few hundred neural network
a
community grew many
working
told,
scientists
worldwide; after 1985, the
such that today there are several thousand
in the field.
Table 2 shows the distribution ot respondents by rank (graduate school and current
employer) according to our classification o\ early and
late entry.
Among
the 373 respondents
present in the sample, 76 (21-percent) entered the field ol neural networks prior to 1984.
The 287 respondents who
entered the
field since
1984 (79-percent). Ten respondents did
not specif.' the year they started their neural network activities.
No
statistically significant
differences are apparent between early and late entrants as far as the distributions of graduate
school rankings and current academic institution rankings are concerned.
—
We
Table 2 about here
Insert
—
further classify respondents according to their educational status at the time they
began neural network research: that
For simplicity, we
will
refer
to
is,
pre- or post-receipt of their highest
pre-
and post-degree respondents
as
academic degree.
"students" and
"graduates," respectively. In the present sample, 162 respondents (45-percent) are classified
as
students
when they
entered the
field;
graduates. In order to avoid ambiguities,
169 respondents (47-percent) are classified
we omit 29 respondents
(8-percent)
who obtained
their highest degree in the year they entered the field of neural networks. Sixty-six percent
the early entrants were students (principally pursuing doctoral degrees)
in neural
networks,
in
comparison
to
about 44-percent of
distribution of respondents by rank of graduate school,
according to their educational status when entering the
13
when
late entrants.
comparing
field.
as
initiating
of
work
Table 3 shows the
early
and
late
entrants
—
Mann-Whitney
tests,
Insert
comparing the distribution
group, indicate significant differences both
oi early entrants
trom
who
entered the
among
is
early
and
who
About 48-percent
highest degree have graduated
who
entered the
field
once
entrants are concerned, however,
far as late
the respondents
and graduates within each
late entrants.
not true for early entrants
they obtained their highest degree (p<0.05). As
slightly less than 40-percent
ot students
field prior to receiving their
top-ranked university. This
a
—
Table 3 about here
entered the
field after
graduation
obtained their highest degree from a top-ranked university. About 16-percent of
late entrants
who
oj-
entered prior to graduation hold degrees from institutions with the lowest rank
(p<0.05).
It
is
become
interesting to note that the lower-ranked institutions (fourth-20
sample only
visible in the
analysis of the students
among
late
20 rank, 80-percent were students
after the field attains
and
widespread legitimacy.
fifth-20)
A
further
entrants shows that, of the 15 respondents in the fourthat
the time of the
sur>,'ey.
For the 18 respondents
in to
the
fih:h-20 rank, 56-percent were in the process of obtaining their highest degree at the time of
the survey.
The
disproportionate representation of students
universities in the respondent
processes.
sample may
Given the time span
educational status
at
among
also be the
of the field's
early entrants at top-ranked
consequence of time-dependent
emergence,
scientists
the time they began neural networks research)
universities of lesser rank during the early years
to other research
may
have led to their
exclusion from the survey population as early entrants. If this
produce
scientists
with a higher
who graduated from
may have moved-on
agendas. As a consequence, their lack of persistence in the
universities
(regardless of their
commitment
field
is
the case, then top-ranked
to their
chosen research agenda
than lower-ranked insritutions. But to be certain, we must
longitudinaJ research design, which
Additional Mann-W'hitney
now
is
tests
we
hypothesis using a
being conducted.
comparing
and
early
late
entrant graduates do not yield
However, when we compare
a statistically significant difference (z=0.63, n.s.).
entrant students,
test this
early
and
late
find the difference to be highly significant (z=3.08, p<0.01): earl entrant
student are more likely to graduate from top-ranked universities than are students
who
are
late entrants.
A
final
word
ot caution
entrant students relative to graduates
may
processes. Since graduate early entrants are,
early entrants
when
sample population)
graduate
who
would be
on average, about ten years older than student
field, their
are likely to be
diminished
may
numbers (and hence representation
to
some degree by
in
the
retirement. As a result,
be slightly underrepresented in the respondent sample.
fairly large disparity in
difficult to explain
disproportionate representation of early
be another consequence of time-dependent
they enter the
entered early
Nonetheless, the
The
warranted.
is
representation between students and graduates
by retirement alone.
WHAT HAPPENS AFTER GRADUATION.'
Now
that
we have an
insight into the graduate school distributions of early
entrants, the next question
degree?
Do
What happens
community? Furthermore, what
of institutional prestige, and
what
to secure an initial
are the
how
is
employment
in
another sector of
the nature of mobiliry along the
continuum
does mobility relate to the conduct ot pioneering research?
consequences of entering a
appointment
late
respondents aftet they receive their highest
to
they pursue academic careers or do they seek
the scientific
Specifically,
is:
and
after
graduation?
15
field early in
terms of a scientist's ability
We
use the ordinal prestige rankings to investigate the inter-sector mobiUty of graduates
from major and minor
mto
universities. In order to facilitate the analysis,
three categories: those
second-20 percent
who graduated trom
and
interval,
(3)
to alleviate the potential tor cell size
used
in
this section.
(1)
we
top-20 institutions,
collapse respondents
(2) universities in the
other graduate schools. This aggregation
all
problems
some
in
at the
necessary
of the non-parametric statistical tests
Furthermore, to avoid any ambiguity, respondents
obtaining their highest degree or graduating
is
in
the process of
time of the survey are omitted from the
analysis.
As demonstrated
in
Table
4,
no
to the respondents' current sector
sectoral distribution pattern
universities.
are based
Furthermore,
on the rank of
for
statistically significant
diHerences are found with respect
of employment: graduates from major universities show a
which
is
highly similar to that ot their colleagues from minor
each sector of employment, the respondent distributions which
their graduate school are not significantly different.
—
Insert
Table 4 about here
—
Introducing the early/late entrant dichotomy does not modify the conclusions discussed
in
Table
4.
Detailed contingency table analyses do not allow us to reject the null-hypothesis
of independence between graduate school prestige and current sector of employment for
both early and
linear
model
This
late entrants.
was
result
fijrther
confirmed by
fitting
an unsaturated log
to the data using the three-way sectoral classification, the three-way ordinal
prestige classification,
and the dichotomous
early/late entry classification as parameters. If the
variables are independent, they can be represented by a log-linear
any interaction terms (Knoke and Burke 1980). Thus,
in
model
that does not have
our case the independence model
looks as follows:
,
_
,
entrv
logF,,k=(i+A.,
•
-
sector
+^,
16
.
prcsnec
+A[;
,
,
(eq.
1)
,
where
Fijk
is
the expected frequency in the
are required for convergence.
The
based on the model.
{i,j,l<)th cell
Two
iterations
standardized residuals are well below ±1.96, indicating no
substantial discrepancies between the
model and the
data. Furthermore, inspection of the
normal probability plot does not show the distribution of the standardized residuals
deviate substantially from a normal distribution.
independence model
p=0.33).
The
results
is
15.66 (d.f = 12, p=0.21).
do not allow us
The
likelihood-ratio X" statistic for the
The Pearson X~
to reject the
to
statistic
is
13.55 (d.f = 12,
independence model and thus confirm the
contingency table analyses.
Due
to
empty
time of entry
cells,
we can not include
the educational status of the respondent at the
fourth parameter in the independence model.
as a
We
do, however, repeat the
analysis with the three-way sector classification, the three-way ordinal prestige classification,
and the dichotomous student/graduate
classification as parameters
independence model similar
1.
to
Equation
The
result
is
comparable
in
an unsaturated
to the
one obtained
with the previous model: likelihood-ratio X"=15.42 (d.f = 12, p=0.22) and Pearson X^=13.15
(d.f = 12, p=0.36).
Once
again,
we
are unable to reject the
independence model. Inspection
of the standardized residuals reveal no problems related to normality. This result
expected trom detailed contingency table analyses: the sectoral patterns shown
remain consistent when studying respondents
respondents
To
who
who
in
is
to be
Table 4
enter the field as students versus
enter after graduation.
conclude, the prestige of one's graduate school does not appear to be an important
determinant of a respondent's current sector of employment. Whether a respondent
graduate from a top-ranked university or not, or whether a respondent
not, does not lead to significantly different
employment
sector patterns
is
is
a
an early entrant or
upon graduation. For
instance, the empirical evidence presented here does not suggest that a graduate from a top-
ranked institution
is
more
likely to stay in
academia than
17
a
graduate from an institution of
This
lesser rank.
result holds tor early as well as late entrants.
warrant further scrutiny.
respondents
who
More
specifically,
stayed in academia: what
.MOBILIPt'
In this section
we examine
graduate school and
respondents
who
his or
is
we
These findings, of course,
are interested to see
what happened
their mobility along the prestige
to those
continuum?
ALONG THE PRESTIGE CONTINUU.M
the relative difference in prestige ranking for a respondent's
upon graduation.
her employer
hold academic appointments.
The
We
limit the analysis to the
205
Pearson correlation coefficient between
the prestige ot one's graduate school and the prestige ot one's current academic employer
0.56 (p<0.001;
N' =
205). This finding reaffirms prior sociological research
on the
relationship
between the prestige of one's graduate institution and the chance oi becoming employed
prestigious academic institution. After adjusting the data by
graduates, the remaining sample has a 0.40 (p<0.001;
is
at a
removing students and recent
N=139)
correlation between graduate
school prestige and the prestige ol current academic affiliation.
In order to study mobility along the
we compute
institutional prestige in greater detail,
the change in institutional prestige between one's current academic employer
and
his or her
(see
Table
much
continuum of
5).
graduate school for each respondent using the continuous prestige measure
In
comparison
to late entrants, the data indicate that early entrants realize a
greater decrease in their institutional prestige ranking (a marginal
—
Insert
Table
In order to understand the possible
5
about here
meaning
mean of-3.37).
—
we employ
a
two-factor
analysis of co-variance with the continuous differential prestige variable as a
dependent
variable
Table
and the
6).
Our
early/late
ol this result,
and student/graduate dichotomies
as
independent variables
choice of independent variables follows from the previous analysis.
IS
(see
The time
elapsed since the respondent's graduation
covariate.
would be preferable
(It
about the date ot
yielding
respondent's year of
this covariate.
employment, there
initial
sizes that are
cell
to use the
compute
current academic affiliation to
years of professional experience-^)
(i.e.,
.AJthough
are a large
inquired
number of missing
assumed
By
as a
at his
the survey
in
values thereby
too small tor statistical analysis. As a consequence,
professional experience at the time oi the survey as a proximate covariate.)
it is
used
employment
initial
we
is
we
use
this definition,
that students have not yet accumulated prolessionaJ experience.
—
As shown
in
Tables
6,
Insert
changes
in
Table 6 about here
—
institutional prestige
between graduate school and
current academic employer can largely be explained as a function of the time elapsed since
obtaining one's highest degree.
The raw
regression coefficient for the covariate
(p=0.005), which suggests a decrease in institutional prestige
experience increases.
The
respondent's educational status
networks does not exert any main
significant interaction effects. In
more
early are
likely to be
is
comparison
employed
graduate schools were. However,
difference
effects,
when
at
as
is
-0.155
the respondent's professional
when
entering the field ot neural
nor do there appear
to
to late entrants, scientists
be any statistically
who
entered the
field
institutions that are less prestigious than their
controlling for professional experience, the firsr-order
not significant. Thus, the early/late entry dichotomy does not help us to explain
differences in institutional prestige: early
and
late
entrants to the field experience similar
decreases in institutional prestige as they progress in their career.
Repeating the two-factor ANCOV.'\ with the respondents' age
similar to that reported in Table 6.
Professional experience
experience
at
entry
is
is
defined
then measured
The independent
as the
as the
variables
as a
covariate yields results
do not show any
statistically
time elapsed since the receipt of one's highest degree. Professional
number of years between
the receipt of one's highest degree and the
year he entered the field of neural networks. Professional experience at the moment of the survey is measured as
the number of years elapsed since the receipt of one's highest degree in the year the survey took place, i.e. 1990.
19
•
main
significant interaction etfects nor
(p=0.004) and has
The
age covariate
a negative regression coefficient (-0.165).
professional experience and age
The
effects.
is
is
statistically significant
The
correlation between
0.89 (p<0.001).
relationship between graduate school and the prestige of one's current employer
is
funher investigated by classifying respondents into four cohorts based on professional
experience: l-to-5 years, 6-to-lO years,
us to test the relationship
model
l-to-15 years, and
a regression
to test the relevance of a
year of
employment
respondent's status
at
ANCOVA,
at
A dummy
as
a
as a
institutional prestige of graduate school
Beyond
five years,
of the institutional prestige of
a
entrant
compute
the covariate.
first
year of
use professional experience at the
is
for the
first
cohort
is
highly significant for respondents within l-to-5
graduate school prestige
is
no longer
dummv
variable
is
always negative,
thus confirming the previous .'KNCOV.A results.
—
statistically significant:
a
good predictor
respondent's current employer. Although the regression
coefficient of the early/late entrant
statistical significance,
the
preferable to use the respondent's
respondent's graduation and
we
late
in
proximate covariate.
Table 7 shows that only the model
years of graduation.
included
is
an early (value=l) or
it is
the current academic employer. Instead
time of the survey
variable
the current academic affiliation to
Cohorts could then be based on the time between
employment
This enables
years.
model. The prestige of the respondent's
the dependent variable.
is
(value=0). Admittedly, as with the previous
initial
more than 15
between prestige oi graduate school and prestige of current
employer within each cohort with
current academic affiliation
1
Insert
Table " about here
20
—
it
never attains
DISCUSSION
The
of
relative prestige
consideration
when
it
comes
choosing
a doctoral
program. While
enter into their decision, the reputation of a university
of prospective doctoral students.
human and
is
likely to
institutional prestige
'VC'ith
community
within the scientific
a university
to
AND CONCLUSION
a
is
an important
number of factors may
weigh heavily on the minds
comes
access to an
abundance of
physical resources necessary to conduct leading-edge research. Moreover, the
centrality of prestigious universities provides a level ot visibility to scientists within the
scientific
community
that can be instrumental to establishing the legitimacy of a research
agenda. Institutions also benefit from their relative standing precisely because they are able to
attract highly qualified students,
university.
One
need only
listen
the weight of a school's ranking
Clearly,
institutional
who
in
turn reinforce the overall research capabilities of a
momentarily
among
its
—
students, to
to
administrators. Nonetheless, the benefits of prestige
scientific innovation, since the next
maintaining one. However, when
it
fruit, that
no other
comes
faculty
may come with
most imponant objective
necessary for scientists to take career
bear
dean or provost before
realizing
peer institutions.
matters
prestige
to a university
to pioneering
to
new
risks: to risk that their
and
to
university
a cost in terms of
having a good reputation
fields of science,
it
unconventional ideas
scientists will follow their lead, or that their efforts will
is
is
often
will
not
be seen
as
misguided by colleagues. Does the pressure of protecting an institution's standing reduce the
incentives to scientists for pursuing unconventional research direction? Or, conversely,
among
lesser
known
institutions, does the desire to attain a higher standing lead scientists to
take risks that others might not otherwise consider?
In
the case of neural
network
scientists
distribution of respondents across the prestige
between
scientists
who
entered the
field earlv
the evidence
continuum we
is
find
mi.xed.
no
Comparing
the
significant difference
and those who followed them. However, when
21
we
divide the sample according to whether or not the respondent initiated
of neural networks prior to receiving
among
differences. First,
neural
to those
who
among
degree. Second,
their graduate
initiated neural
respondents
who
(at
find
some
interesting
when they staned
are students
more
at
prestigious
universities
in
after receiving their highest
when
entering the field of neural
are students
Thus, we find pioneering behavior
were students
work
we
in the field
network research
networks, early entrants do their graduate work
entrants.
who
early entrants, respondents
network research do
comparison
his or her highest degree,
work
to
at
more
prestigious universities than
be most prevalent
among
the time they started neural networks research) at the
do
late
who
respondents
more highly ranked
universities.
Examining the
move from
relatively
occurs regardless
student
when
seen in the
career progress ot respondents,
ol'
more
prestigious universities to
whether a respondent
premium
who
an early or
is
prestigious universities. This pattern
late
The
entrant or whether he or she
entered the
When comparing
early
field early are less likely to receive
their graduate school.
experience the difference
is
If institutional prestige
we examine
not
matters
at all,
it
and
on
late entrants,
first
longer significant.
The
find that
for professional
.
appears to matter most early-on in a scientist's
the data bv cohorts
years ot a scientist's career.
we
an appointment at an institution
we
find that graduate school prestige
significant determinant oi the prestige of a respondent's subsequent
during the
a
starting their career at a
However, when controlling
statistically significant
is
relevance of such a finding can be
that prospective doctoral students place
matching the prestige of
career. ^X'^^en
find that over time scientists tend to
less
entering the field of neural networks.
highly ranked graduate school.
scientists
we
Beyond
five years,
is
a
academic appointment
graduate school prestige
is
no
cohort model thus supports Finkelstein's (1984) contention that
graduate school prestige matters most during the first years of an academic
22
career.
^X^ether or
not a respondent
is
an early entrant,
is
of no consequence
in
explaining the prestige of his or
her current university.
Thus, the neural network community does not provide evidence
"backwater" hypothesis. Instead, we fmd that early entrants
come from
interesting
laboratories at the
is
more
that early entrants are
it
is
support the
persisted in neuraJ
prestigious graduate schools. Nonetheless,
more
likely to
be students
already hold their doctorate. In light oi this finding
whether
who
to
as
opposed
we must consider
network
what
to scientists
is
who
the question of
the length of a scientist's professional experience rather than institutional
prestige that has a bearing
on pioneering behavior.
23
References
Paul D. and j. Scott Long. 1987. "Interuniversity mobility
American Sociological Review 52: 643-652.
i-\Jlison,
Allison, Paul D.
and
Scott Long. 1990. "Departmental effects
J.
on
ot
academic
scientists."
scientific productivity."
American Sociological Review 55: 469-478.
Cole, Jonathan R. and Stephen Cole. 1973. Social Stratification in Science. Chicago:
University ok Chicago Press.
major and minor universities:
recognition." American Sociological Review iO: 699- 713.
Crane, Diana. 1965. "Scientists
at
A
study of productivity and
Crane, Diana. 1970. "The academic marketplace revisited: A study of faculty mobility using
the Cartter Ratings." American Journal of SocioIoqi 7 5: 953-964.
DARPA.
1988. Neural
Finkelstein, Martin
J.
Network Study.
Fairfax,
VA:
AFCEA
International Press.
1984. The American Academic Profession. Columbus:
Ohio
State
University Press.
Giddens, Anthony. 1989.
Sociology.
Cambridge, UK:
Polity.
Hagstrom, Warren O. 1971 "Inputs, outputs, and the prestige of university science
depanments." Sociology of Education 44: 375-397.
.
Hargens, Lowell
1969. "Patterns of mobilitv of
L.
new
Ph.D.'s." Sociology of Education 42:
18-37.
L. and Warren O. Hagstrom. 1967. "Sponsored and contest mobility of
American academic scientists." Sociology ofEducation 40: 24-38.
Hargens, Lowell
L., Peter D. Tessner and Anhur M. Diamond.
102 (17 November): 717-723.
Hull, David
Science
Knoke, David and Peter
Long,
J.
J.
Burke. 1980. Log- Linear Models. Beverly Hills: Sage Publications.
and academic position
Review 43: 889-908.
Scott. 1978. "Productivity
Sociological
Long,
J.
19''8. "Planck's Principle."
in
the scientific career." American
D. ;\llison and Scott McGinnis. 1979. "Entrance into the academic
American Sociological Review 44: 8 6-830.
Scott, Paul
career."
Mallove, E.F.
1
1
99
I
.
Fire from
ice:
Searching for the truth behind the cold fusion fixror.
New
York: John Wiley and Sons.
Minsky, Marvin and Seymour Papert. 1988.
Perceptrons.
Cambridge, \LA:
MIT
Press.
Mullins, Nicholas C. 1987. "EvaJuating research programs:
Science
and
Measuremenr and data
Myers, C.R. 1970. "Journal citations and the scientific eminence
psychology." American Psychologist 15: 1041-1048.
Narin,
F.
sources."
Public Poliq 14:91 -98^
1987. Bibliometnc techniques
in
in
contemporary
the evaluation ot research programs." Science
and
Public Policy \4: 99-106.
F. and J.D. Frame. 1989.
245: 600-604.
Narin,
The growth of Japanese
science
and technology." Science
Office of Technology Assessment, U.S. Congress. 1991. Federally funded research: Decisions
for a decade.
OTA-SET-490. Washington D.C.: U.S. Government Printing Office.
Papert, Seymour. 1988.
"One AI
or
Many'" Daedalus
117: 1-14.
Rappa, Michael A. and Raghu Garud. 1992. "Modeling Contribution-Spans of Scientists
a Field: The Case of Cochlear Implants." R6^D Management, 22 (3): in press.
Roche, Thomas and David Lewis Smith. 1978. "Frequency of citations as criterion for the
ranking of departments, journals, and individuals." Sociological Inquiry 48: 49-57.
Sanders, Jimy
M. and Herbert
Sociological Inquiry 55:
1
Y.
Wong.
1985. "Graduate and
initial
job training."
54-169.
Small, Henry. 1990. Bibliometrics of Basic Research. Washington,
25
DC: NTIS.
in
EARLY EN IR^ANTS
STL' DENT
-3.03(N=24)
GR-^DL'ATE
3.98 (N= 13)
5
:
Average changes
ENTRANTS
o.y-'tNrig)
0.96 (N=64)
- 3.37
MARGINAL MEANS
Table
lv\TE
-0.56
.MARGINAL MEANS
- 1.33
1.47
-1.43
m presrige ofacademic affiliation for early and late
entrants according to their educational status at the time they entered the
field
Note: presnge chinge
ib
calcuUced
as:
prestige of current
school prestige.
30
employer minus graduate
VARIABLES IN THE ANALYSIS
279S
Date Due
Lib-26-67
MIT LIBRARIES DUPl
3
TDflO
D071'=m3D
fl
'AliaillBllBllC
'
"^
.iiiilPiP§ii