A Brief Introduction to Social Network Analysis Jennifer Roberts
Outline Description of Social Network Analysis Sociocentric vs. Egocentric networks Estimating a social network TRANSIMS A case study
What is Social Network Analysis (SNA)? An analysis technique which studies Relationships between people and groups How those relationships arise Consequences of the relationships (Christopher McCarty, Director of the UF Survey Research Center) Nodes = people Ties = relationships or interactions
Two Types of SNA Egocentric Analysis Focuses on the individual Studies an individual s s personal network and its affects on that individual Sociocentric Analysis Focuses on large groups of people Quantifies relationships between people in a group Studies patterns of interactions and how these patterns affect the group as a whole From redwing.human.net/~mreed/ warriorshtm/ego.htm From www.plant.uga.edu/dblmajor.htm
Egocentric SNA Examines local network structure Describes the network around a single node (the ego) Number of other nodes (alters) Types of connections Extracts network features Uses these factors to predict health and longevity, economic success, levels of depression, access to new opportunities From www.stop.hu/ showcikk.php?scid=1005217
Sociocentric SNA Quantifies relationships and interactions between a group of people Studies how interactions, patterns of interactions, and network structure affect Concentration of power and resources Spread of disease Access to new ideas Group dynamics Raines Laboratory Research Group, U. Wisconsin-Madison http://www.biochem.wisc.edu/raines/people.html
Survey and Interview Data Collection Techniques Name generator/name interpreter questions Questions to elicit list of names: Who do you discuss important matters with? Follow-up questions: Reports about that persons attributes, type of tie, ties between pairs of contacts Inflow/outflow of resources Expensive, subject to human error, influenced by the nature of the questions asked
Other Data Sources Indirect measures Corporate records, event attendance, co-citations, citations, co- authorship, trading patterns, shared affiliations, email, phone calls, computer conferencing Non-invasive, inexpensive Not obvious how indirect measures relate to actual interactions Small scale methods Observation, diaries Experimental
Accuracy Surveys accessing validity of people s reports Recall and observation do not match up -- people do not know, with any acceptable accuracy, to whom they talk over any given period of time Evidence that people are good at recalling typical interactions, bad at answering about specific time scales
TRANSIM A Method for Estimating Large Social Networks Assumes transportation network constrains interactions Creates a synthetic population Models large-scale human interactions through simulations Used to study transportation planning, disease propagation, mobile communications, and demand within the electric power grid From www.environ.org.uk/energy/vehicle%20fuels/
Creating a Synthetic Population Data Land use and demographic census data Survey data about daily activities Information used to create a synthetic population with List of activities consistent with survey data Access to transportation consistent with survey data
Activity Planning and Simulation Creates schedule based on activity lists Optimizes sequential plan based on transportation mode Updates schedule based on current congestion levels Simulation Bipartite graphs contains people nodes and location nodes Simulation updates people s s locations every second
Graph Structure Set P of people nodes Set L of location nodes Edge (p, l) indicates p visits l on a normal day L has a power law distribution with = 2.8 Number nodes with degree i equals β P s distribution is concentrated around a small average value
CL-Model Theoretical Model Generates graphs based on expected degree sequences Each node is assigned a weight, d(u),, equal to its expected degree Edge probabilities are proportional to node weights and edge assignments are independent = σ σ = =
Projection Graphs
Results CL, fastgen approximation create networks with similar properties to actual data (fig 1) Also include approx algorithm description, table, graphs of results, projection graphs