Network diagram by Alden Klovdahl, Australian National University
Social Network Analysis: Overview of the Field Today
Steve Borgatti
MB 874 Social Network Analysis
September 6, 2006
Agenda
SNA as a discipline Introduction to the field Critical assessment Frontier
Painting by Idahlia Stanley
Presentation @ National Academy of Sciences 2005 Steve Borgatti
Formal Organization of the Field
Professional association (since 78)
Int'l Network for Social Network Analysis [Link] Incorporated 1993
Sunbelt annual conference (since 79)
2001: Budapest, HUNGARY 2002: New Orleans, USA 2003: Cancun, MEXICO 2004: Portors, SLOVENIA 2005: Los Angeles, USA 2006: Vancouver, CANADA 2007: Corfu, GREECE
No Department of Social Network Analysis
But some centers for complexity and networks
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Resources of the Field
Specialized journals
Social Networks, (since 79) CONNECTIONS, official bulletin of INSNA Journal of Social Structure (electronic) CMOT
Software
UCINET 6/NETDRAW; PAJEK STRUCTURE; GRADAP; KRACKPLOT
Listservs
SOCNET listserv (1993) REDES listserv UCINET users group
Textbooks
Kilduff & Tsai, 2004 Scott, John. 1991/2000. Degenne & Fors. 1999. Wasserman & Faust. 1994.
Regular Training Workshops
Sunbelt social networks conference Academy of Management University of Essex, UK ICPSR-Michigan
2005 Steve Borgatti
Presentation @ National Academy of Sciences
Explosive Growth
4500 4000 3500 3000 2500 2000 1500 1000 500 0 1988
600
Google Scholar entries by year of publication
Google page rank Social networking software Management consulting Network organizations
1990
1992
1994
1996
1998
2000
2002
2004
2006
500
Social Network
Population Ecology
400
Articles w/ social network in title
300
Embeddedness, social capital, structural holes, alliances TCE, RD, Inst theory, SRT, etc
200
100 y = 0.001e
2
0.134x
R = 0.917 0 1960 1970 1980 1990 2000 2010
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Development of the Field
1900s
Durkheim Simmel
1970s Rise of Sociologists
Modern field of SN is established (journal, conference, assoc, etc) Milgram small-world (late 60s) White; Granovetter weak ties
1930s Sociometry
Moreno; Hawthorne studies Erdos
1980s Personal Computing
IBM PC & network programs
1940s Psychologists
Clique formally defined
1990s Adaptive Radiation
UCINET IV released; Pajek Wasserman & Faust text Spread of networks & dyadic thinking; Rise of social capital,
1950s Anthropologists
Barnes, Bott & Manchester school
1960s Anthros & graph theorists
Kinship algebras; Mitchell Harary establishes graph theory w/ textbooks, journals, etc
2000s Physicists new science
Scale-free Small world
Presentation @ National Academy of Sciences
2005 Steve Borgatti
What is distinctive about the field?
The phenomena we study i.e., the data
The observations (cases) are dyads, not individual actors Fundamental variables are social relations (e.g., friendship) rather than attributes of individuals (e.g., education, personality) Theoretical constructs like centrality, structural equivalence or network shape
The methodology
Dyadic, autocorrelated data require different statistical methods
Theoretical perspective
Not a single theory across all disciplines, but some common principles and perspectives
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Introduction to the Field
Overview of Basic Concepts
A Multi-layered Enterprise
Conceptual Layer
Deepest metaphors Taken for granted axioms
Technical Layer
Graph theory Theoretical vocabulary network constructs Methodology Network antecedents Network consequences Interface with other research streams
Substantive Layer
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Conceptual Layer
Actors do not act independently
Have ties of various kinds with other actors
Actors and ties link together to form networks
Whether actors are aware of it or not Pattern / arrangement of ties is discernable
Ties as pipes
Connectionist or flow-based axiom
Diffusion and influence across links: actors affect each other Access to resources through ties: social resource theory
Ties as scaffolding
Structuralist or topology-based axiom
Structure of ties in the network has profound effects on the capabilities, constraints and ultimately outcomes of the network and its constituents Bavelas-Leavitt work (1950s) on centralization of work teams
Presentation @ National Academy of Sciences 2005 Steve Borgatti
GUIDING THEORETICAL PRINCIPLES
Relations vs. Attributes
PA 1 GA 2 GA 1 FL 2 TX 1 FL 1 LA 4
LA 5
LA 1
LA 2
Traditional soc sci focuses on actor attributes as explanatory variables Sexual relations among Network science focuses patients with rare cancers --- Bill Darrow, CDC on relations among the actors Influences & flows of Connectionist view
Tell each other information Provide material aid Copy attitudes & behavior Transmit diseases
NY 5 NY 21 NJ 1 NY 16 NY7 NY 2 NY 12 LA 7 NY 13 SF 1 LA 6 LA 8 LA 9
LA 3
NY 10
NY 3
0
NY 15
NY 4
NY 19
NY 9
NY 11 NY 8 NY 14
NY 17 NY 22
NY 1
NY 18
NY 20 NY 6
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GUIDING THEORETICAL PRINCIPLES
Its not just the parts but the structure
Emergent, non-reductionist, non-individualist, holistic, structuralist flavor to some of the research
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GUIDING THEORETICAL PRINCIPLES
Opportunities & Constraints
A persons position in a social network (i.e., social capital) determines in part the set of opportunities and constraints they will encounter
Maire Messenger Davies
Presentation @ National Academy of Sciences 2005 Steve Borgatti
GUIDING THEORETICAL PRINCIPLES
Rate of return on human capital
Burt: A persons connections determine the rate of return on human capital
Human capital
rate of return
profit
social capital
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GUIDING THEORETICAL PRINCIPLES
Universal network laws?
Similar network properties observed in
300 250
Gene interaction networks World wide web links Sexual partners
200
150
100
50
y = 185.98x-1 .51 5 2 R 2 = 0.8699
0 0
1000
20
40
60
80
100
120
One of natures standard solutions?
Or just a popular lens for understanding nature? (cf power laws)
100
10
y = 275.81x-1 .7 1 47 2 R = 0.9287 1
10
100
0.1
Warning: different social relations have different characteristic structures
2005 Steve Borgatti
Presentation @ National Academy of Sciences
Technical Layer
Key Constructs that are good to think with
What is a Network?
A set of actors (nodes, points, vertices)
Individuals (e.g., persons, chimps) Collectivities (e.g., firms, nations, species)
A set of ties (links, lines, edges, arcs) that connect pairs of actors
Directed or undirected Valued or presence/absence
Set of ties of a given type constitutes a social relation Different relations have different structures & consequences
1000 scientists
Presentation @ National Academy of Sciences 2005 Steve Borgatti
Types of Tie Among Persons
Social relations
Kinship Other role-based Cognitive Affective
Interactions
Sent email to, had sex with Communicated with
Flows
Personnel Goods Ideas/information Infection
Correlations
Co-membership Similarity Proximity
Influence
Roads
Traffic
Presentation @ National Academy of Sciences
Each kind of tie (i.e., social relation) defines a different network 2005 Steve Borgatti
Simple Answers
Who you ask for answers to straightforward questions.
HR Dept of Large Health Care Organization
Recent acquisition Older acquisitions Original company Cross, R., Borgatti, S.P., & Parker, A. 2001. Beyond Answers: Dimensions of the Advice Network. Social Networks 23(3): 215-235
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Problem Reformulation
Who you see to help you think through issues
Recent acquisition Older acquisitions Original company Cross, R., Borgatti, S.P., & Parker, A. 2001. Beyond Answers: Dimensions of the Advice Network. Social Networks 23(3): 215-235
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Hawthorne Games & Conflicts
I3 S2 S4 S1 S1 I3 S2 S4
W4 W2 W5 W1 W7 W9
W2
W4 W9 W5 W7 W8
W1 W8
I1 I1 W3 W6
W3
W6
Presentation @ National Academy of Sciences
2005 Steve Borgatti
S2
Combining Games & Fights
S4 S1
W4 W2 W5 W1 W7 W8 W9
I1
W3 W6
2005 Steve Borgatti
GREEN = games only RED = fights only BLACK = games & fights Presentation @ National Academy of Sciences
Relations Among Organizations
As corporate entities
sells to, leases to, lends to, outsources to joint ventures, alliances, invests in, subsidiary regulates ex-member of (personnel flow) interlocking directorates all social relations
Through members
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Internet Alliances
Microsoft AOL
Yahoo
AT&T
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Co-Membership > 27%
PN
OCIS CM HCM OM
GDO
ODC HR
MC
TIM
MH
CAR
OB
OMT
BPS
IM
MSR MED MOC SIM ONE
RM ENT
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Mainstream Logical Data Structure
2-mode rectangular matrices in which rows (cases) are entities or objects and columns (variables) are attributes of the cases Analysis consists of correlating columns
Typically identify one column as the thing to be explained We explain one characteristic as a function of the others Variables (attributes)
Age Sex Education Income 1001 1002 1003 1004 1005
Cases (entities)
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Network Logical Data Structures
Adjacency matrices Incidence matrix
Friendship Jim Jill Jen Joe Jim - 1 0 1 Jill 1 - 1 0 Jen 0 1 1 Joe 1 0 1 Proximity Jim Jim Jill 3 Jen 9 Joe 2
Jim - Jill Jim - Jen Jim - Joe Jill - Jen Jill - Joe Jen - Joe
Friendship Proximity 1 3 0 9 1 2 1 1 0 15 1 3
Jill Jen Joe 3 9 2 - 1 15 1 3 15 3 -
Multiple relations recorded for the same set of actors Each relation is a variable
variables can also be defined at more aggregate levels
Values are assigned to pairs of actors Hypotheses can be phrased in terms of correlations between relations
Dyadic-level hypotheses
2005 Steve Borgatti
Presentation @ National Academy of Sciences
Ego Network Analysis
Mainstream Social Science data Network Analysis perspective
Ego Networks
Combine the perspective of network analysis with the data of mainstream social science
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Ego Network Data Collection
(Random) survey of members of a population Ask respondents (egos) about their contacts (alters)
E.g., who they confide important matters with
Characterize relationship with each alter Obtain attribute data about each alter (egos perception) Optionally obtain egos perception of which alters have ties with which other alters
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Ego Network Analysis
Network composition assessments
E.g., % women in each persons network
Selection: Investigating homophily / heterophily
Do races prefer to marry endogamously? Does eye color matter?
Network homogeneity / heterogeneity assessments
How diverse is each persons network?
Network quality assessments
Do entrepreneurs vary in their social access to resources?
Structural holes & other local density assessments
Are my friends friends with each other?
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Levels of Analysis
Dyad (relationship) level
Network data is fundamentally dyadic
Who is friends with whom in an office Distance in meters between peoples desks Marriage ties among families in Renaissance Florence Business ties among the same families
Node (actor) level
Can aggregate to the node level
The number of friends each person has
Or measure aspects of a nodes position in the network
Betweenness centrality of each node
Network (group) level
Aggregation to the group or whole network level
Density of ties within a group
Measure aspects of the networks structure
How centralized the network is; how concentrated the ties are around small set of actors
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Families of Network Concepts
degree distribution clumpiness core periphery avg distance density
Shape
Cohesion
degree
Group level
faction
clique
closeness
block
Centrality Subgroup identification Proximity
adjacency
Presentation @ National Academy of Sciences
Node level
Role identification Equivalence
geodesic distance structural equivalence
Dyad level
regular equivalence
2005 Steve Borgatti
simmelian tie
GROUP level of analysis
Density of ties
Density = proportion of pairs of actors that are actually tied In some contexts, could be thought of as measure of social capital
Low Density (25%)
Presentation @ National Academy of Sciences
High Density (39%)
2005 Steve Borgatti
GROUP level of analysis
Help With the Rice Harvest
Village 1
Data from Entwistle et al
Presentation @ National Academy of Sciences 2005 Steve Borgatti
GROUP level of analysis
Help with the rice harvest
Village 2
Data from Entwistle et al
Presentation @ National Academy of Sciences 2005 Steve Borgatti
GROUP level of analysis
Graph-Theoretic Distance
AKA degrees of separation
10
The graph-theoretic distance between two nodes is the number of links in the shortest path that connects them
Distance from 4 to 10 is 3 links
12 11 8 9 2 7
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GROUP level of analysis
Core/Periphery Structures
Core/Periphery
Network consists of single group (a core) together with hangers-on (a periphery),
Core connects to all Periphery connects only to the core
C/P
Short distances, good for transmitting information, practices Identification with group as whole E.g., structure of physics
Clique structure
Multiple subgroups or factions Identity with subgroup Diversity of norms, belief E.g., structure of social science
Clique
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GROUP level of analysis
On Innovation and Network Structure
I would never have conceived my theory, let alone have made a great effort to verify it, if I had been more familiar with major developments in physics that were taking place. Moreover, my initial ignorance of the powerful, false objections that were raised against my ideas protected those ideas from being nipped in the bud.
Michael Polanyi (1963), on a major contribution to physics
Presentation @ National Academy of Sciences
2005 Steve Borgatti
GROUP level of analysis
C/P Structures & Morale
50 45
Study by Jeff Johnson of a South Pole scientific team over 8 months
40
Core/Periphery-ness
35
C/P structure seems to affect morale
30
25
Group Morale
20
15
10
Caution: this is an n of 1
Presentation @ National Academy of Sciences
Month
2005 Steve Borgatti
NODE level of analysis
Node Level Variables
Year 1
White House Diary Data, Carter Administration
Presentation @ National Academy of Sciences
Year 4
Data courtesy of Michael Link
2005 Steve Borgatti
NODE level of analysis
Centrality
Betweenness Closeness Degree
Eigenvector
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Information flow in a virtual group
Cultural interventions, relationship building
New leader
Data warehousing, systems architecture
Cross, Parker, & Borgatti, 2002. Making Invisible Work Visible. California Management Review. 44(2): 25-46
Presentation @ National Academy of Sciences 2005 Steve Borgatti
Changes Made
Cross-staffed new internal projects
white papers, database development
Established cross-selling sales goals
managers accountable for selling projects with both kinds of expertise
New communication vehicles
project tracking db; weekly email update
Personnel changes
Presentation @ National Academy of Sciences
2005 Steve Borgatti
9 Months Later
Note: Different EV same initials.
Cross, Parker, & Borgatti, 2002. Making Invisible Work Visible. California Management Review. 44(2): 25-46
Presentation @ National Academy of Sciences 2005 Steve Borgatti
Substantive Layer
Antecedents of network variables Consequences of network variables Relations with other schools of thought
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Causality and Network Research
Mathematicians, methodologists, network priesthood How density relates to distance Most common area of research Appropriate for young field
Antecedents
Network variables
Consequences
Rare in sociology, more common in psych, physics Developing in management
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Types of hypotheses involving network variables
Dyad (relationship) level
Likelihood of office friendships increases as distance between offices decreases Marriage ties between families in Renaissance Florence facilitate business ties between the same families
Node (actor) level
centrality in interaction network leads better immune system Self-monitoring personality leads to higher betweenness centrality
Network (group) level
groups with c/p structure in affective network perform better Compared to advice relations, affective relations will contain more transitive triples
Mixed dyad-node (autocorrelation)
Members of org units interact more members of same units (homophily) Interaction leads to similarity in attitudes (influence)
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Antecedents of Network Variables
Homophily
Propinquity Common affiliation Socially significant attributes
Prob of Daily Communication
Dyad level who has ties with whom?
0.4 0.3 0.2 0.1 0 0 20 40 60 80 100
Distance (meters)
Triadic balance theory
AB and AC tends to lead to BC Strength of tie
Multiplexity
Cross-sectional Longitudinal Male Male Female 1245 970 Female 748 1515
Node characteristics
Personality centrality
Network (group) characteristics
Small world networks (clumpy networks with short distances) Scale-free networks (skewed degree distributions)
Presentation @ National Academy of Sciences
2005 Steve Borgatti
Consequences of Network Variables
Ends Means Connectionist mechanisms
(flows thru ties)
Explaining Variance in Performance (social capital)
Success comes from obtaining resources through social ties; Its who you know Network positions /shapes provide opportunities for exploitation; Its how you know others
Explaining Social Homogeneity (adoption)
People have same behavior because they directly influence each other & transmit ideas, beliefs, etc. People have same behavior because their network positions are similar (and affect them similarly); same social environment
Structuralist mechanisms
(emergent properties of topology)
Borgatti, S.P. and Foster, P. 2003. The network paradigm in organizational research: A review and typology. Journal of Management. 29(6): 991-1013
Presentation @ National Academy of Sciences 2005 Steve Borgatti
Critical Assessment
Have we accomplished anything? Where is the field going?
Changes in the Field
25 years ago
Descriptive, methodological Small datasets (< 100 nodes) Structuralist cast Focus on the consequences of network characteristics
Network is fixed Cross-sectional data
Now
Theory testing in soc sci Large datasets 00s 000s Increasing attention to agency Increasing attention to causes of network variables
Network change Longitudinal data
Focus on the pattern of ties Deterministic & analytical models Inter-network comparisons
Presentation @ National Academy of Sciences
Increasing interest in what flows through networks Increasing interest in stochastic models & simulations Comparison with theoretical baselines
2005 Steve Borgatti
Trends & Buzzwords
Is the field getting too popular too fast?
Small worlds Scale-free Communities?
Social Capital
Dangers of trademarked concepts
# of Papers
Embeddedness Weak ties
Networking
Network ties
1975
1985
1995
Time
WARNING: Totally made-up data! Do not take seriously!
Do fads sweep out equal areas under the graph?
Presentation @ National Academy of Sciences 2005 Steve Borgatti
& New!
Traditional Criticisms of Network Research
Not Theoretical
Just descriptive Just methodological; Too mathematical Not process-based Flow
Mechanisms \
Goals
Explaining Variance in Performance (social capital)
Explaining Social Homogeneity (adoption)
Static
Ties dont change Flows through ties arent considered
Connectionist mechanisms (flows through ties) Structuralist mechanisms (emergent properties of topology)
Lack of agency
Actors dont act
Trendy Unethical / exploitative
Presentation @ National Academy of Sciences
New!
Agency
2005 Steve Borgatti
Theoretical Perspectives
Presentation @ National Academy of Sciences
2005 Steve Borgatti