Académique Documents
Professionnel Documents
Culture Documents
1999
Author(s): James Moody
Source: American Sociological Review, Vol. 69, No. 2 (Apr., 2004), pp. 213-238
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/3593085 .
Accessed: 02/12/2014 05:35
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
American Sociological Review.
http://www.jstor.org
Science, carved up into a host of detailed studies that have no link with one another, no longer
forms a solid whole.
Durkheim, 1933 [1984] p. 294
SOCIOLOGICAL
REVIEW,2004, VOL.69 (April:213-238)
AMERICAN
214
SOCIOLOGICAL
AMERICAN
REVIEW
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
SOCIALANDTHEORETICAL
INTEGRATION
NETWORKSTRUCTURESAND IDEA SPACES
215
is negative,andseveral
5Not all of thisliterature
what
is
comment
on
people
goodaboutthediscipline,
includingthe diversityof ideasandinterestof topinteics, andmanyattemptsaremadeat theoretical
grationor reformulations
(cf LiebersonandLynn
2002;Skvoretz1998).
216
AMERICAN SOCIOLOGICALREVIEW
collaborators.Graduatestudentswill be trained
within particularspecialties and a shop-productionmodel shouldbuild distinctcommunities surroundingparticulartopics.The resulting
networkwill admit to clear clusters with little
collaborationcrossing specialty boundaries.
The social networkmodel that best fits this
descriptionis the small-worldmodel (Milgram
1969; Watts 1999; Watts and Strogatz 1998).
Intuitively,a small-worldnetworkis anynetwork
where the level of local clustering (one's collaboratorsarealso collaboratorswith eachother)
is high,butthe averagenumberof stepsbetween
actors is small. An archetypical small-world
networkwill have many distinct clusters, connectedto each otherby a small numberof links.
Distinct research clusters will likely inhibit
broadtheoreticalintegration,since theory will
progresslargelywithindistinctresearchgroups.
Past researchon stratifiSTARPRODUCTION.
cation in the sciences has identified large
inequality in the returns to scientific labor
(Allison, Long andKrauze1982;Cole andCole
1973; Merton 1968). Although most scientists
laborin obscurity,a small numberof scientists
receive disproportionaterecognition.This has
been clearlydemonstratedfor indicatorssuchas
citations, number of publications, or grants.
However,researchsuggests that collaboration
is also unequallydivided. Crane(1972) found
thata smallnumberof very prominentscientists
form the core of each specialty'scollaboration
networkandthatmost otherswere connectedto
the rest of the communitythroughthese highly activeindividuals.This centralpositionhelps
explainwhy core scientistswere able to so rapidly diffuse theirideas throughthe community,
and we would expect thatthose with many collaboratorsare likely to be influential (at least
locally). Newman (2001) turns collaboration
itself into a statusmarkerand asks, "Whois the
Best ConnectedScientist?"6
The large inequalityin numbersof collaborators can be explained through a process of
preferentialattachment.High-statusscientists
make attractivecollaboratorssince one's own
overhaving
Competition
amongmathematicians
to thestathesmallestErddsnumberspeakssimilarly
nettusattachedto one'spositionin a collaboration
work.
METHODS.
Abbott (2001) also argues
GENERIC
that the social sciences have little theoretical
consensus, but he does not suggest that this
generatesa clustereddiscipline.Instead,he suggests that the natureof sociology creates permeable theoretical boundaries that make it
impossible for sociology to exclude ideas from
the disciplineonce they areintroduced(p. 6; see
also Daipha2001). Moreover,the processof theoretical developmentis not linear, but instead
follows a "fractalwalk" throughthe available
idea space. Pushed by competition for status,
proponentsof one set of ideas attemptto van-
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
217
SCIENTIFIC
TRENDS
COLLABORATION
The probability of coauthoringdiffers across
disciplinesandovertime. Coauthorshipis more
common in the natural sciences than in the
social sciences, but has been increasingsteadily acrossall fields (Endersby1996; Fisheret al.
1998; Hargens 1975; Laband and Tollison
2000). The changing likelihood of coauthorship is evidentin Figure1, which showstheproportion of all articles coauthoredin ASR from
inception and in Sociological Abstracts from
1963 to 1999.
Severalexplanationshavebeen given forthe
increasein coauthorshipovertime (Labandand
Tollison 2000; McDowell and Michael 1983).
Fundingrequirements,particularlyin largelab
settings, might induce collaboration(Laband
and Tollison 2000; Zuckerman and Merton
1973). While social scientists are rarely as
dependenton labs, the rise of large-scale data
collection efforts suggests a similar team-production model. Training differences between
disciplinesmight also accountfor coauthorship
differences.Advancedworkby PhD studentsin
the naturalsciences is usuallyclosely relatedto
an advisor'swork,and commonlyresultsin collaboration.Social science students,in contrast,
tend to work on projects that are more independent.
Otherexplanationsfocus on the division of
labor among scientists. In high-growth, fastchanging specialties, we would expect to see
more coauthorshipbecause it is easier to bring
in a new authorthan it is to learnnew material
oneself. Hudson(1996) arguesthatthe increase
REVIEW
AMERICAN
SOCIOLOGICAL
218
0.75
0 0.6
A
m
S0.45
FL
0.
CL
C
0.3
0
0.
0
a.
ASR
Sociological Abstracts
0.15
0
1930
1940
1950
1970
1960
1980
1990
2000
Year
Trendsin Sociology
Figure 1. Coauthorship
DATAAND METHODS
My primaryinterestis to identify the observed
structureof the social science collaborationnetwork to distinguishbetween the three models
suggested by commentaries on sociological
practice. Because participationis a necessary
minimumrequirementfor influence in the network,I first model participationin the network,
and then examine the structureof the network
among those who have coauthored.
SAMPLEAND SOURCE
To examine networkparticipationand embeddedness, I use all English journal articles listed in SociologicalAbstractsthatwerepublished
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
9A cutoffof 15 appearances
of a namewasused
to distinguishcommonfromuncommon.
workon largecollaboration
networks
has
10 Prior
not attemptedto identifythesetypesof errors.An
alternative
sourcefornamecleaningwouldbeto use
authoraffiliation. Unfortunately,Sociological
forthefirstauthor,
and
Abstracts,
onlylistsaffiliation
authorsmay have multipleaffiliations(suchas a
researchcenterandanacademicdepartment).
While
suchcorrectionsareimportant
to helpensureaccuratemeasures,thegeneralgraphfeaturesexamined
heredonotdiffersignificantlyif I use thecorrected
versusthenon-corrected
data.
219
enteringearlier.Starproductionmodels suggest
thatthosewho havebeen in the disciplinelonger
should be more deeply embedded than those
who just entered.Finally,by linkingfirstnames
to the Census'genderdistributionof firstnames,
we can estimate the effects of gender on collaborationand position in the network.11
Every article in Sociological Abstracts is
assigned to one of 149 detailed subjectcodes,
which are nested within 36 broad specialty
areas. These 36 areas are used to capture
research specialties.12Sociological Abstracts
also lists the numberof tables in every article,
which providesa simple proxy for whetherthe
To controlfor
paperuses quantitativemethods.13
in
covered
changes journals
by Sociological
Abstracts,I include an indicatorfor how completely SociologicalAbstractsindexesthejournals wherepeoplepublish.Coverageis indicated
at three levels: complete (100% of the articles
in thatjournalare indexed),priority(morethan
50% of the articles are indexed), and selective
(less than 50% are indexed).
I construct the collaboration network by
assigning an edge betweenany two people who
wrote a paper together,regardlessof the how
often they have coauthored.Figure 2 demon-
scorebasedonthe
1 Thisresultsin a probability
of peoplewitha givenfirstnamewhoare
proportion
male.Sincenotallfirstnamescanbe matched,
using
thismeasure
resultsin missingdatafor32%of cases,
likelypredominantly
amongrareandforeignnames.
Thesubstantive
modelresultsfortheothervariables
do notdifferwiththeinclusionof thegendermeasure.Tablesnotusinggenderareavailablefromthe
authoruponrequest.
12 The sociologicalabstractareacategoriesare
ideal,beingsubjectto both
likelynot substantively
errorsof misclassification
andinternalheterogeneity. However,theyremainthe only tractableinformation on substantivearea.While each record
containskeywords
content,thesheernumdescribing
ber of such words(over 8000 uniquekeywords)
wouldrequire
somesortof categorization
the
routine,
of whichis nottransparent.
development
13 Thisis alsoan imperfect
measure,sincewhile
allquantitative
papersincludetablessomenon-quantitativepapersalso includetextualtables,so this
measureover-estimates
thenumberof quantitative
papers.At the aggregatelevelusedhere,the small
numberof nonquantitative
paperswithtableswashes outrelativeto differencesacrossspecialties.
220
REVIEW
SOCIOLOGICAL
AMERICAN
a) IndividualPublications
QLN
Authors
1D
K
O Papers
Network
b) Collaboration
[]I
-B-]"-L
v
U
Distribution
Degree
Degree Freq
0
R.1
.
2
3
4
5
10
2
0
J.5
Networks
Collaboration
Figure2. Constructing
strateshow the networksare constructedfrom
the authorshipdata.14
The top panel of figure 2 is a schematicrepresentation of data as given in Sociological
Abstracts,with authors(squares)connected to
the papers(circles)they write.The datainclude
single authoredpapers (personsA,B,C and D)
as well as those with more authors.The structureon the toprightof figure2 representsa large
connected set of authors, each of whom has
coauthoredwith someone who has coauthored
with someone else. The bottompanel of figure
2 providesthe resultingcollaborationnetwork.
Those who have written only single authored
papers do not participatein the collaboration
network,butcanbe representedas structuralisolates. Pairsof people who have only coauthored
with eachotherarerepresentedas isolateddyads
{EF, GH}.
14Thanksto a reviewerforsuggestingthisfigure.
The largestconnectedcomponentis the maximal set of people who are connectedby a chain
of any length to each other.The large structure
at the bottom right of figure 2 is the largest
connectedcomponent.Nested withinthis componentis a bicomponent(circled).Whilea component requires only a single traceable path
between each actor, a bicomponent requires
that there be at least two node-independent
pathsconnectingeverypairof actorsin the network. Simmel (1950) arguedthatthe necessary
condition for a group is that a supra-individual
body remains even if a person leaves.
Bicomponents meet this criterion, since the
group remains connected even if a single person is deleted (Moody and White 2003). This
conception scales, as tricomponents (3-node
independentpaths),4-components,and higher
orderk-componentsidentifyincreasinglycohesive subgroupsin a network.
The degree distribution of the network is
used to test the preferentialattachmentmodel.
NETWORK 221
STRUCTURE
OF SOCIALSCIENCE
COLLABORATION
An actor'sdegree is the numberof uniquepeople they are directly connected to, in this context the number of unique collaborators.The
degree distributionfor the example networkis
given in the lower right of figure 2. Geodesic
pathsdefine networkdistance,as the numberof
intermediarieson the shortest path connected
two nodes in a network.So, for example,nodes
L and S are 3 steps apart.
PUBLICATIONTRENDS
The primaryconstraintson the shape of a collaborationnetworkare the distributionsof the
numberof paperspeople publish and the number of authorson a paper.Table 1 below gives
these distributionsfor all papers in the dataset
(includingthose with only a single author).
Of all authors that appear in Sociological
Abstracts,66% appearonly once, and an additional 15%appearonly twice, with the number
of publications dropping quickly after that.
Publicationvolume has increasedslightly over
time. The percentof authorswith only one publication has dropped,from 71% in the 75-85
period to 67% in the 89-99 period and the tail
of the distributionis a little fatter.'5About 67%
AREAAND NETWORK
SPECIALTY
PARTICIPATION
Having ever coauthoreda paper is a necessary
conditionfor being embeddedin the largercollaboration network. If the collaboration network shapes commitmentto particularways of
doing science, then identifying systematicdifferences in who collaborateswill identify key
differencesin those exposed to the information
andNumberof
Table 1. SociologyPublicationPatterns:Distributionsof Publications,Coauthorship
Collaborators
PublicationsperAuthor
Authorsper Paper
1989-1999
62.57%
22.79
8.47
3.41
1.45
.66
.31
.14
.08
.04
.02
.04
141,497
UniqueCollaboratorsa
Total 1975-1985 1989-1999
35.27% 41.06%
23.88
27.55
14.48
14.53
8.73
7.40
5.34
3.74
3.53
2.18
2.22
1.14
1.54
.67
1.11
.45
.81
.36
.24
.59
.48
.22
2.02
.46
197,976 59,567
Onlypeoplewith coauthorships.
32.35%
22.99
15.01
9.73
6.19
4.04
2.67
1.77
1.30
.92
.63
.47
1.93
123,766
222
AMERICANSOCIOLOGICALREVIEW
Areasof Sociology
Growth
All Areas
IndividualAreas
Marxist
Radical
Knowledge
History&Theory
CultureandSociety
281,090
1,044
908
3,406
17,231
7,040
Papers(%)
100
.37
.32
1.21
6.13
2.50
Paper
Growth
Coauthored
(%)
Coauthorship
346.68
33.2
.013
1.66
2.82
1.91
17.03
3.67
8.0
8.0
8.2
12.9
14.3
-.009a
-.001c
.000O
.001C
.003
141
.05
.25c
14.8
-.009b,c
5,673
14,412
4,518
6,632
5,569
8,611
2.02
5.13
1.61
2.36
1.98
3.06
7.44
15.93
5.28
7.12
4.84
13.48
17.8
18.1
20.0
20.5
23.6
24.4
.008
.002
.007
.007a
.009a
.006
Community Development
4,444
1,694
1.58
.60
25.6
26.0
.004
.009
FeministGenderStudies
Social Development
Social Control
Policy& Planning
7,225
9,805
7,804
3,243
2.57
3.49
2.78
1.15
11.9
15.93
6.16
6.75
27.2
27.9
28.4
28.6
.003a,c
.016
.009
.008
.000b,c
Visual
LanguageandArts
Political
Science
Change& EconomicDevelopment
Religion
GroupInteractions
Urban
Mass Phenomena
Rural
Education
Interactions
Environmental
Methodology
Studiesin Violence
Demography
Social Differentiation
Studiesin Poverty
Social Planning/Policy
ComplexOrganizations
Business
SocialPsychology
Problems& Welfare
The Family
Health/Medicine
SocialWelfare
.10
-. 15c
29.8
12,069
4.29
14.98
30.0
.006
3,746
1.33
30.5
.008
10,628
3,102
8,897
1,521
6,542
9,769
1,393
12,232
13,986
3.78
1.10
3.16
.54
2.33
3.48
.50
4.35
4.98
31.9
32.1
32.1
33.4
33.6
34.4
34.9
35.8
37.4
.010
.012
.009
.010
.010
.011
.014
.014
.012
280
Clinical
.06c
-. 39c,d
.65c,d
8.71
7.93
6.68
3.11
4.38
-.58c
2.06
20.21
20.22
195
.07
1.79
40.0
.054b,c
13,527
10,674
19,806
14,634
28,689
4.81
3.80
7.05
5.20
10.21
4.67
16.35
20.13
23.16
90.58
44.2
45.4
46.1
49.5
53.2
.013
.019
.019
.025
.045
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
Welfare.16The numberof articles in the database has increasedby about350 papersper year
over the past 30 years (column 4, row 1), and
overall growth by specialty can be seen in the
rest of the table (the sum of the specialty cells
equals the total). Cultureandtheorypapersare
among the least likely to be coauthored,as are
papersin the sociology of knowledge, sociology of science, language and arts, radical sociology, Marxist sociology, political sociology,
and visual sociology. Paperson social welfare
are the most likely to be coauthored,followed
closely by those in social psychology, the family, sociology of healthandmedicine,and social
problemsandwelfare.Workin these areasoften
involves largedatasetsand cumbersomeanaly-
16 Social Welfareis a
largecategoryincluding
0.562
48
0.52
17Previousversionsof thispaperincludeddataon
a limitedsetofjournalsandexamined
funding,menforcoauthorship.
torshipandlocationexplanations
Thesetablesareavailableon request.
.*"*
1975-1985
Soc.welfare
1989-19S.Health
0.8
-a
0
S00Ee0
0.32-1.
S0.24
- 0.20
Sof Science -
0.1
0Econ Change
politicalSc.
0.2
O
Demography
.anS
FernmGender
RadicalSoc.
0.0
Complex
S.
e
RuralS.
M2
Methdology Educationo
0 . 36
0 2
&Medicine
Soc. Psych.
0.44
S0.40rgs
223
0.3
&Deelop.
0.4
0.5
0.6
Proportionof paperswithtables
andQuantitative
Work1975-1999by Specialty.
Figure3. Coauthorship
0.1
0.8
224
AMERICANSOCIOLOGICALREVIEW
18Thecorrelation
in theearlyperiodis .82, inthe
lateperiodit is .89.Thechangein slopebetweenthe
twoperiodsis statistically
significantatp = .017.To
avoidclutterin the figure,each specialtyis only
to using
labeledforone timeperiod.An alternative
the specialtyareaswouldbe to aggregatewithin
journals,becausewiththeexceptionof a fewgeneral journals,journalsspecializein particular
topics.
Thisfigureis availableon request,andsubstantive
resultsarethesame.
19Sincethecategoriesareexhaustive,
atleastone
reasonto
mustbe omitted.Thereis no substantive
one
area
over
another
as
the
reference
categopick
ry,buta consistentreferencecategoryaidsin evaluating changeovertime. In all modelsI used the
sociology of education as the reference category,
becausetheoddsof coauthorship
wereverycloseto
portionof an author'spapersthathavetables,to
model 2, allowing us to identify specialty difference net of researchmethod.
The models show thatthose with greatertime
in the discipline (exposure) are slightly more
likely to coauthor,thoughthe magnitudeof this
effect is small.20The cohort effects are consistent, with those publishinglaterhaving a higher likelihoodof havingcoauthored,thoughagain
the effect is relativelysmall.The strongestpublication effect is the simple numberof publications. Each additionalpublicationincreasesthe
odds of coauthoringby 1.28. Consistent with
prior researchin economics, the odds of men
coauthoring are about .64 times the odds of
women coauthoring,thoughthis effect decreases (to .73) once you controlfor specialtyarea.21
The actor-attributeeffects remain largely
constantwhencontrollingfor specialtyor examined overtime.The clearestexceptionis thatthe
effect of the numberof publicationsincreases
when controlling for specialty area, and is
strongerin the latterperiodthanin the earlyperiod. Similarly,cohorteffects areless pronounced
in the laterperiodthan in the earlyperiod,likely reflectingthe greaterubiquityof coauthorship
over time.
There is a clear effect of specialty on the
likelihood of having coauthored.Authorswho
write in historical,qualitative,radicaland interpretive specialties are less likely to coauthor
thanthose writingin more positivist and quantitative specialties. For example, the odds of
coauthorshipcontrollingfor specialtyoverlapin
Marxistsociology are abouthalf (.57) those in
sociology of education, and those writing in
social historyandtheory are about.58 times as
likely. In contrast,those writing on the family
averagein bothtimeperiods,the rateof changein
mirrorstherateof changeoverall,and
coauthorship
theareais largeenoughto providea stablereference
category.
20 To avoidredundancy
in the text,I will focus
commentsmainlyonthepooled1963-1999models,
andmentiontheothermodelsonlytotheextentthat
thepatternsdifferfromthismodel.
21Inaddition,
forSociological
thecontrolvariables
Abstractscoveragearealsosignificant,showingthat
(theomittedcatjournalsincludedonlyincidentally
egory) are slightlymorelikely to be coauthored,
thoughthis effect eitherbecomesinsignificantor
areais included.
changesdirectionwhensubstantive
STRUCTURE
OFSOCIALSCIENCE
COLLABORATION
NETWORK 225
Table3. LogisticRegressionof HavingEverCoauthoredon PublicationCharacteristics
1963-1999
Mod 1 Mod2 Mod 3
Variable
Exposure
Numberof Publications
Yearof 1stPublication
MaleAuthor(probability)
CompleteCoverage
1.02
1.28
1.03
.64
.92
.87
Priority Coverage
Quantitative Work
.26
.32
.24
.49
.52
.47
.56
.58
.64
.63
.74
.64
.89
1989-1999
1.01
1.45
1.02
.73
1.22
1.01
1.44
1.01
.73
1.19
1.03
1.33
1.04
.70
1.16
1.00a
.94a
5.45
.98a
.48
.57
.38
.58
.56
.57
.59
.68
.69
.77
.72
.68
.91a
.61
.62
.42
.64
.62
.69a
.60
.69
.73
.76
.72
.71
.89
SpecialtyAreaof Sociology(code)
Radical(25)
Marxist(30)
Knowledge(22)
History& Theory(2)
Culture& Society(5)
Visual(33)
Language& Arts(13)
Political(9)
Science(17)
SocialChange(7)
Religion(15)
GroupInteraction(4)
Urban(12)
1975-1985
1.01a 1.02a
1.31 1.29
1.03 1.01a
.74
.76
1.32 1.22
1.02
1.34
1.01
.64
.66
.99a
1.61
1.01
.76
1.02a
1.00a
1.56
1.01a
.75
1.03a
.94a
4.17
.80
.98a
.91a
6.38
.35
.58
.34
.50
.26
.37
.54
.68
.42
.49
.44a .59a
.61
.51
.60
.75
.81
.67
.65
.87a
.85
.77
.79
.90a
1.06a 1.20
.70a
.66
.48
.81
.52
.86a
.69
.74
.92a
.93a
.92a
.89a
1.14a
.20
.20
.16
.35
.37
.32
.57
.48
.58
.61
.76
.58
.74
.34
.35
.26
.42
.44
.35
.59
.59
.64
.75
.73
.62
.78
1.02a
1.16a
1.01a
.47
.49
.40
.54
.57
.48a
.63
.63
.75
.75
.72
.66
.73
.85
.90a
.94a
.85a
.91a
1.04a
1.14a
FemaleGender(29)
SocialDevelopment(83=36)
.68
.82
.69
.86
.71a
.76
.55
.78
.58
.86
.66
.79
.88
.81
.82
.87
.84
.95a
.96a
.79
.87
.74
.81
.79
.86
.91a
1.34
1.24a
1.21a
.93a 1.01a
1.09a
.76a
.83a
.98a
.90
.99a
.95
.97a
1.07a
.95a
1.01a
.98a
.86a
.89
1.08a
.96a
.82
1.07a
.93
.87
1.08a
.88a
.98a
Methodology(1)
Environmental
(26)
1.07
.95a
1.15
1.03a
1.10
1.02a
.84
1.17a
1.24a
1.14a
.93a
1.02a
1.21
1.17
1.29
1.28
1.04a
1.13a
1.03a 1.01a
1.18 1.22
Clinical(31)
1.02a
.91a
1.13a
1.02a
1.07a
.96a
.89a
Demography(18)
SocialDifference(10)
1.07a
1.14
Poverty (27)
1.07a
1.01a
.94a
1.08a
1.30a
1.05a
1.12a
1.06a
.96a
SocialPlan/Policy(72)/35
(6)
ComplexOrganizations
1.19
1.17
1.21
1.19
1.13
1.06a
1.23
1.25
1.35
1.38
1.34
1.23
1.05a 1.10
1.17 1.20
1.13
1.03a
Business (32)
1.39a
1.45a
1.26a
.69a
.81a
.59a
1.76
1.99
1.80a
SocialPsychology(3)
SocialProblems(21)
Family(19)
Health(20)
SocialWelfare(61)
1.88
1.76
1.71
2.02
2.35
1.91
1.65
1.61
1.92
2.19
1.63
1.46
1.35
1.64
2.04
2.21
1.53
1.68
1.53
1.56
2.34
1.37
1.72
1.62
1.66
1.95
1.46
1.46
1.48
1.80
1.60
1.90
1.83
2.24
2.67
1.77
1.77
1.75
2.13
2.40
1.50
1.49
1.37
1.72
2.24
.212
.322
.154
.254
.215
.352
R-Square
.096b
.067b
.73
.75
.88
1.09
1.12
Violence (28)
.067b
.75
.83
.97a
82,475
130,141
41,386
Note: Datashownas oddsratios.Unless otherwisenoted,all cell valuesaresignificantatp ? .01 (tableswith
detailedsignificancelevels areavailablefromthe author).Mod = Model.
aValueis not significantat the < .01
p
b Pertainsonly to the publicationand
demographiccharacteristics.
226
AMERICAN SOCIOLOGICALREVIEW
STRUCTURE
OF SOCIALSCIENCE
COLLABORATION
NETWORK 227
1,000,000
1963-1999
100,000
/
10,000 t
ComponentSensitivity
Degree
Distribution
.
1,000
100
10
o 0
o
o
100
0
o0
10
Degree + 1
100
1,000,000
1975-1985
100,000
10,000
,
Component Sensitivity
1,000
100
10
00
10o0
10
100
Degree + 1
1,000,000
1989-1999
100,000
Component Sensitivity
101,000
)
1,000
100
10
1
1
10
Degree + 1
Networks
Figure 4. Scale-Freepropertiesof Coauthorship
100
228
SOCIOLOGICAL
REVIEW
AMERICAN
A SOCIOLOGICAL
SMALLWORLD?
STRUCTURE
OFSOCIAL
SCIENCE
COLLABORATION
NETWORK229
Table4. Comparisonof ObservedCoauthorship
Structureto EquivalentRandomNetworks
1963-1999
1975-1985
1989-1999
128,151
35,109
87,731
.194
(.207)
9.81
(7.57)
.306
(.312)
12.26
(8.31)
.266
(.302)
11.53
(8.24)
68,285
95,078
(169)
.72
78,753
(132)
.87
7,492
16,736
(131)
.45
15,378
(90)
.49
36,772
59,736
(145)
.62
49,061
(120)
.75
29,462
.43
47,339
(166)
.50
.62
48,769
(153)
.62
.60
2,034
.27
4,774
(94)
.28
.43
7,882
(78)
.51
.26
15,281
.42
29,738
(186)
.50
.51
31,806
(123)
.65
.48
Nodes (n)a
Small-worldParameters
ClusterCoefficient
(Randomexpected)
AveragePathLengthb
(Randomexpected)
Size of LargestComponent
Observed
RandomPaperAssignment
(SD)
Ratioof Observedto Random
RandomPaper+ One Publication
(SD)
Ratioof Observedto Random
Size of LargestBicomponent
Observed
Ratioof Bicomponentto Component
RandomPaperAssignment
(SD)
Ratioof Bicomponentto Component
Ratioof Observedto Random
RandomPaper+ One Publication
(SD)
Ratioof Bicomponentto Component
Ratioof Observedto Random
a Excludespeoplewithoutcoauthors.
bAppliesonly withinthe
largestconnectedcomponent.
STRUCTURAL
COHESION?
23o
AMERICAN
SOCIOLOGICAL
REVIEW
24
Additionalcontrolswere checked,including
onmixingbynumber
of authors
conditioning
beyond
isolateddyads(equivalent
to fixingthe diagonalof
the mixingmatrix,while the 'one-pub'condition
only fixes the 1,1 cell), and constrainingmixing
withinareas,whichaccountsforchangesin specialovertime.Neitherof theserestricty representation
tionshaveas stronganeffectontheexpectedvalues
astheone-pubrestriction.
onrequest.
Tablesavailable
Becausebicomponents
mustbe nestedwithin
components,andbecauseourobservedcomponent
is smallerthantherandomcomponent,
directlycomto the
paringthe size of the observedbicomponent
randomgraphsomewhat
underestimates
therelative
cohesionintheobservednetwork.
Toaccountforthis
I presentthe ratioof the size of the
underestimate,
to the size of the largestcomlargestbicomponent
ponentforboththeobservedandtherandomgraph.
Thesefiguresalsoshowthatcohesionhasincreased
overtime,from.51 in the earlyperiodto .65 in the
laterperiod.
26 Estimatesare basedon the numberof nodeindependentpathsconnectingrandomlysampled
thesizeof thekpairsof nodes.I thenback-estimate
of node-independcomponentfromthedistribution
ent paths. I estimate confidence intervals by
theresultingdistribution,
thenusing
bootstrapping
distribution
meansat5%and95%.Becausethisestimateis basedon a sample,it is impossibleto identify thesetsof nodesthatcomprisethehigher-order
Theseshouldprobablybe takenas
k-components.
high-endestimates,sincepeoplecanbelongto differentk-components,
thoughmy informalexplorationsof thesedatasuggestthatthis is unlikelyat
theselowerk-levels.
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
27Tomyknowledge,
thisis thefirsttimenetworks
havebeenrepresented
withthistypeof figure.
28
of pointsis oftenveryuneven,
Theconcentration
resultingin veryjaggedcontourplots.I haveuseda
kernaldensityestimationtechnique
non-parametric
231
232
AMERICANSOCIOLOGICALREVIEW
3. 7
2.74
4.0
3. 7
2.74
2.11
1.48
0.84
0.21
ContourSociogram.Contourvaluesdenote bivariatedensityestimates,indicating
every15thpercentof the rangebetweenthe 5th and 95th percentiles.
NetworkLargestBicomponent(n = 29,462)
Figure 5. Social ScienceCoauthorship
STRUCTUREOF SOCIALSCIENCECOLLABORATIONNETWORK
233
Variable
Exposure
Numberof Publications
Yearof 1st Publication
MaleAuthor(probability)
AuthorsperPaper
UniqueCoauthors
Mentorship
CoauthorDiversity
CompleteCoverage
PriorityCoverage
1963-1999
1975-1985
1989-1999
1.01
1.17
1.01
.81
.97
1.90
.85
.78
1.41
1.16
Quantitative Work
SpecialtyArea(code)
Radical(25)
Marxist(30)
.70
.57
1.01
1.12
1.01
.83
1.01a
1.81
.85
.86
1.44
1.17
.95a
.76
1.20
History& Theory(2)
Culture& Society(5)
Visual(33)
Language& Arts (13)
PoliticalSociology(9)
Science(17)
Social Change(7)
Religion(15)
1.05a 1.01a
.97a .80
.94a 1.02a
.87
.80
.88
.99a
.81
.89
1.02a .97a
.95
1.01a
Methodology(1)
Environmental
(26)
1.02a
1.32
1.03
.93a
.82
1.63
.74
1.0a
2.55
1.42
.86
1.02a
1.34
1.02
.93a
.81
1.62
.75
1.09a
2.49
1.38
1.55
1.05
1.14
1.02
.85
1.03
1.61
.73
1.08a
1.31
1.11
1.06
1.16
1.02
.85
1.02a
1.59
.73
1.14
1.35
1.08
.77a .88a
.87a 1.04a
1.83
.32
.56
.43a
.62
.47a
.65
.49
.55a
.87a
.75a
.89a
.90a
.59
.94a
1.06a
.84
.64
.72a
.78
.80
.68
.72
1.00a
1.06a
.81
1.30a
.95a
1.05a
.87
1.00a
1.19
1.09a
.88a
1.49a
.95a
1.04a
.92a
.97a
1.15
1.21
1.21
.95a
.89a
.94a
.93a
1.19
1.07a
.02
1.15
1.20
.92a
.88a
.95a
.91a
1.20
1.11a
.03a
1.14
1.18
1.02a
.82
1.11a
.80
.98a
.90
.97a
1.00a
.97a
1.07a
1.07a
1.14a
1.17a
.87
.68
.95a
.74
.95
.83
.57
.81
1.07
.96
1.15a
.96a
.82
1.11
.91a
1.15a
1.09
1.04a
.95a
1.17a
.96a
.81
1.11
.94a
1a1.
1.09
1.03a
1.05a
1.52
1.15
.80
1.03a
.94a
1.13a
.80
1.53
1.05a
1.50
1.12a
.80
1.04a
1.05a 1.14
.92 1.03a
1.13
1.05a
.98a
1.12a
1.05a
1.51
1.13a
.81
1.04a
.96a 1.02a
1.13a 1.12a
.83
.82
1.57 1.53
-
.96a
.71
.65
.78
.73
1.01a
.82
.04
.92
.92a
.81
1.19
.85
.96a
.98a
1.15a
1.12a
1.14a
1.13a
Demography(18)
SocialDiffer.(10)
.98a 1.06a
.88
1.14
1.04a
1.12
1.30
1.13
1.30 1.21
1.15a 1.11a
1.00a 1.20
.96a 1.21
1.15
1.16
.96a
1.03a
1.02a
1.08a
1.19a
1.15a
1.03a
1.26
1.25
1.01a
.73
1.12a
.91
1.20
1.12
1.13
1.08
1.10
1.11
1.29a
1.19
1.23
1.18
1.19
1.13
1.10
1.09
1.28a
1.17
1.21
1.16
1.18
1.14
.83
1.06a
1.51a
.96a
1.23
1.07
1.12
.86
.86
1.08a
1.46a
.97a
1.25
1.07a
1.13a
.88a
.85
1.07a
1.21a
.94a
1.24
1.04a
1.12a
.88a
.95a
.90
.99a
1.07
1.13
1.18
1.21
1.16
1.16
1.10
1.38a
1.33
1.31
1.40
1.43
1.35
1.19
1.06a
1.25a
1.29
1.28
1.35
1.40
1.35
.506
.511
.418
.422
.489
.497
SocialPlan/Policy(72/35)
ComplexOrgs(6)
Business(32)
SocialPsychology(3)
SocialProblems(21)
Family(19)
Health(20)
SocialWelfare(61/34)
R-Square
.499b
.404b
1.10a
86,498
24,897
Note: Datashownas oddsratios.Unlessotherwisenoted,all cell valuesaresignificantatp
detailedsignificancelevels areavailablefromthe author).Mod= Model.
a Valueis not
significantat thep < .01
b Pertains
only to the publicationanddemographiccharacteristics.
.98a
1.48 1.45
1.08a 1.06a
Violence (28)
Poverty (27)
.98a
1.04
1.33
1.02
.82
1.00a
1.69
.72
.97a
1.17
1.09
.99a
.77
.92
Education (14)
1.02a
1.36
1.02
.91
.82
1.62
.74
1.02a
2.94
1.54
1.58
Knowledge (22)
Urban(12)
CommunityDevelopment(23)
FemaleGender(29)
SocialDevelopment(83 = 36)
SocialControl(16)
Policy& Plan(24)
Clinical(31)
MassPhenomena(8)
Rural(11)
1.01
1.13
1.01
.83
1.00a
1.80
.85
.88
1.46
1.14
.478b
56,632
.01 (tableswith
234
AMERICAN
SOCIOLOGICAL
REVIEW
31
trollingfordatabasecoverage.Thoseauthorswho
publishin completeor priorityjournalsaremore
likely to be in the core of the network.
STRUCTURE
OF SOCIALSCIENCE
COLLABORATION
NETWORK 235
ASR are coauthored.Coauthorshipis not evenly distributedacross sociological work.As predicted by others,coauthorshipis more likely in
specialties that admit to an easier division of
labor. Research method seems particularly
important, showing that quantitativework is
more likely to be coauthoredthan non-quantitativework.Whilethereis a specialtygap in network participation, among those who have
participated,specialty area is only a weak predictor of networkembeddedness.Thus,just as
heterogeneity provides only limited information on social integration(Moody 2001), observations aboutfractionalizationin the discipline
based on increasing numbers of specialties
might be misleading.The coauthorshippattern
shows a steadily growing cohesive core, suggesting that while authors might specialize,
their skills marrywell with others creatingan
integratedcollaborationnetwork.
How do we account for the observed collaborationpattern?Twocomplimentaryimages
of science productionare suggestive. Abbott's
(2001) descriptionof social science as having
permeabletheoreticalboundariessuggests that
specialization within the social sciences does
not necessarilygeneratedivisionsbetweenspecialists. Instead, competitors actively borrow
ideas from each other (even if under new
names), to cover the availableidea space. This
free mixing means that one's coauthors need
not coauthorwith each other,and thus the network as a whole admitsto little clusteringand
few schisms, insteadspreadingquicklyoverthe
relevant idea spaces representedin the discipline. Friedkin's(1998) work suggests a specialty analogueto Abbott'scompetitivemixing
model. Friedkinfound thatwhile contactclustered within specialties, these clusters where
strongly connected to each other,creatingnetwork conduits throughwhich ideas and information flow. Tie heterogeneity within groups
means that groups can act as bridges between
other groups but still maintain internal cohesion (Paxtonand Moody 2002). Fleshingthese
hypotheses out will require examining the
internalstructureof the collaborationnetwork
in more detail, though preliminarywork suggests that Friedkin'smodel fits for short-run
images of later periods of the collaboration
graph.
What do these findings suggest for the
prospects of scientific consensus in sociolo-
divisionthat
weaklywith the humanist-positivist
troubles
buildingnecessaryambimanysociologists,
ignores
guityintothesefindings.Third,thedatabase
book publications,which might systematically
excludesomeareasmorethanothers.
236
REVIEW
AMERICAN
SOCIOLOGICAL
REFERENCES
Abbott,Andrew.2000. "Reflectionson the Futureof
Sociology 29:296-300.
Sociology."Contemporary
. 2001. Chaos of Disciplines. Chicago, IL:
University of Chicago Press.
Allison, Paul D., Scott J. Long, and Tad K. Krauze.
1982. "CumulativeAdvantage and Inequalityin
Science." American Sociological Review
47:615-25.
Babchuk,Nicholas, KeithBruce,andPetersGeorge.
1999. "Collaboration in Sociology and Other
Scientific Disciplines: A Comparative Trend
Analysis of Scholarship in the Social, Physical
and Mathematical Sciences." The American
Sociologist 30:5-21.
STRUCTURE
OF SOCIALSCIENCE
COLLABORATION
NETWORK 237
vol. 2, J. Berger, M. Zelditch, and B. Anderson.
Boston, MA: HoughtonMifflin.
Durkheim, Emile. [1933] 1984. The Division of
Labor in Society, translatedby W. D. Halls. New
York:The Free Press.
Endersby,J.W. 1996. "CollaborativeResearchin the
Social Sciences: Multiple Authorship and
Publication Credit." Social Science Quarterly
77:375-92.
Ennis, James G. 1992. "The Social Organizationof
Sociological
Knowledge: Modeling the
Intersectionof Specialties."AmericanSociological
Review 57:259-65.
Fisher, Bonnie S., Craig T. Cobane, Thomas M.
VanderVen, and FrancisT. Cullen. 1998. "Trends
andPatternsin PoliticalScience."PoliticalScience
Online:847-56.
Friedkin,Noah E. 1998.A StructuralTheoryofSocial
Influence. Cambridge, England: Cambridge
UniversityPress.
Gould, Rodger. 2002. "The Origins of Status
Hierarchies:A FormalTheoryandEmpiricalTest."
AmericanJournal of Sociology 107:1143-78.
Hagstrom, Warren 0. 1965. The Scientific
Community.New York:Basic Books.
Harary, Frank, Robert Z. Norman, and Dorwin
Cartwright. 1965. Structural Models: An
Introductionto the Theory of Directed Graphs.
New York:JohnWiley and Sons.
Hargens, Lowell. 1975. Patterns of Scientific
Research. Washington D.C.: The American
Sociological Association.
Holland, Paul W. and Samuel Leinhardt. 1971.
"Transitivity in Structural Models of Small
Groups."ComparativeGroupsStudies 2:107-24.
Hudson, John. 1996. "Trends in Multi-authored
Papers in Economics." Journal of Economic
Perspectives 10:153-58.
Jones, James and Mark Handcock. 2003. "Sexual
Contacts and Epidemic Thresholds." Nature
423:605-6.
Kuhn, Thomas. 1970. The Structure of Scientific
Revolutions. Chicago, IL: University of Chicago
Press.
Laband,D. N. and R. D. Tollison.2000. "Intellectual
Collaboration." Journal of Political Economy
108:632-62.
Leifer, Eric M. 1988. "InteractionPreludesto Role
Setting: Exploratory Local Action." American
Sociological Review 53:865-78.
Lieberson, Stanley and Freda B. Lynn. 2002.
"Barking Up the Wrong Branch: Scientific
Alternativesto the CurrentModel of Sociological
Science."Annual Review of Sociology 28:1-19.
Mannheim,Karl. 1936. Ideology and Utopia. New
York:Harcourt,Brace & Company.
Markovsky, Barry. 1998. "Social Network
Conceptions of Solidarity."Pp. 343-72 in The
ProblemofSolidarity: Theoriesand Models, edit-
238
SOCIOLOGICAL
REVIEW
AMERICAN