Subgroups Level Subgraph: ● A graph G' = (V', E') is a subgraph of a graph G = (V, E) iff: ○ ● V' is a subset of V and E' is a subset of E A lot of network methods are designed to split a graph into subgraphs ○ Subgraphs can be used to represent the social network concepts of groups, c liques, or communities Connected subgraph:A subgraph is connected if every node within it can reach every other one via a path of some length Disconnected subgraph:If otherwise, it is disconnected. Clique:A maximal c omplete subgraph(density = 1.0). The s izeof a clique is the number of nodes in it: 3-clique, 4-clique, 5-clique … k-clique. (E.g. clique of size 4) N-cilque: ● Some subgraphs are not complete graphs (cliques) but they are almost there! ○ ● They are quasi-groups We can relax the definition to accommodate ○ An n -cliqueis a subgraphwhere each node is at least n steps away from each other node ■ 3-clique (3 steps or less), 4-clique (4 steps or less), 5-clique (5 steps or less) ■ steps or less) ...k-clique (k- Connected graph:A graph is connected if there exists a (finite) path between each pair of nodes. Disconnected graph:Otherwise the graph is disconnected. ● We can disconnect a graph by removing e dges. ○ In some graphs removing just one edge leads to disconnection ■ Edges with the property of being able to disconnect the graph upon removal are called bridges ● We can also disconnect a graph by removing n odes. ○ It can be easier to disconnect a graph this way because nodes take their links with them! ○ Some nodes are more “important” because they serve as the connectors ○ Can you guess what kind of centrality “connector” nodes are more likely to have? ■ Betweenness Component:A maximal connected subgraph is called a component. If a graph has more than one component it is by definition disconnected. Giant Component:The connected subgraph containing the l argestnumber of vertices in a graph is called the giant component. K-connectivity: A component is k -connectedif every node in the component can reach any other one via a path of length k or less. ● A graph is k-connected if it takes removing k nodes to disconnect it ● Menger’s theorem states that: If a graph is k-connected there are at least k paths that don’t repeat third nodes connecting each non-adjacent pair of nodes ( important result in graph theory) ● K-connectivity (graph theory) can represent the concept of social cohesion (network theory): ○ The larger the (node) cutset = more cohesive (integrated) groups are with one another ■ ○ More people would have to be removed to disrupt the group The smaller the (node) cutset = less cohesive and more potentially fragmented the group structure is ○ Any graph can be split into subgraphs with any given level of k-connectivity Whole Network Level Degree Sequence: ● A chart that documents the degrees (# of edges a node has) of all nodes ○ Maximum Degree:The most amount of nodes/links adjacent to a specific node. ○ Minimum Degree:The least amount of nodes/links adjacent to a specific node. ○ Average Degree:( should know how to calculate!) ■ The sum of the degrees of nodes divided by the number of nodes Degree Distribution (pk):The probability of observing a node with degree k for each value of degree in the network! Degree Centralization(should know how to calculate!): ● Compares an observed network to one that is maximally centralized (the star graph of the same order) ● It is given by the sum of the differences between maximum degree and the degree of each node in the observed network to the same number in the star graph of the same size. Path-based Network Properties: Geodesics: In a network, shortest paths between two nodes are also called geodesics. ● Graph Diameter: ● The length of the longest shortest path between two nodes Average Shortest Path (L): ● The average shortest path length between all pairs of nodes in a network; c an only be computed for a connected graph or the giant component of disconnected ones Reachability Matrix: ● The reachability matrix of a graph has a one in each cell if the corresponding pair of nodes are connected via a path of some length ● Graph Hierarchy: ● ● Example: H = 1 - (5/(5(5-1)/2) = 1 - (5/(20/2) = 1 - (5/10) = 1 - 0.5 = 0.5 Graph Reciprocity: ● Reciprocity (R) is given by counting the number of reciprocal links (V) and dividing by the total number of ties (E) : R=V/E Node Based Network Properties Average Clustering Coefficient: ● The sum of each node’s clustering coefficient divided by the total number of nodes Properties of Small World Graph (Phenomenon): ● A given (large-scale) network is a “small world” if: ● Large:The o rder of the graph (N) is large. ● Sparse: ○ The graph is s parse(any node is on average connected to only a relatively small number of other nodes k) ■ ● k ~ 102 in human social networks (Pool and Kochen 1978) Decentralization: ○ Graph is d ecentralized(max degree is much smaller than the order of the graph N) ● Clustered: ○ The network exhibits c lustering(friends of friends tend to be friends) Affiliation Networks Two mode network: A ka. Affiliation Networks: Are composed of relationships betweeen two types of entities. ● Personsand g roupsthey belong to ● Actorsand the moviesthey appear in ● Scientistsand p apersthey write Bipartite graph: Affiliation matrix: ● An affiliation matrix (A) has number of rows equals to the number of people and number of columns equal to the number of groups! Co-membership (co-affiliation) matrix: ● A X AT returns the c o-membership matrix (C) ○ Each o ff-diagonalcell (Cij) tell us the number of affiliations that person i has in common with person j ○ Each d iagonalcell (Cii) tells us the number of total memberships person i has ○ The matrix represents a weighted one-modenetwork ■ Strength of tie between i and j = number of common memberships ■ A “similarity” relation Group overlap matrix: ● AT X A returns the g roup overlap matrix (G) ○ Each o ff-diagonalcell (Gij) tell us the number of members groups i and j have in common ○ Each d iagonalcell (Gii) tells us the number of total members in group i ○ The matrix represents a weighted one-modenetwork ■ Two groups are more strongly connected if they have lots of members in common ■ This is called the o ne-mode projection of the original two-mode network Matrix transpose: The transpose of a matrix is the same matrix with the rows and columns switched around. Social Positions in Networks Relational similarity: Two actors occupy the same social position if they have the same (or similar) sets of ties to other actors in the network. Positional Analysis:is the branch of social network analysis in charge of defining different notions of network ● The position of actor b in the network is given by their pattern of relations to other actors. ● Steps in positional analysis: ○ Define a network-based notion of social position using graph theory ■ This is usually an equivalence relation ○ Translate that notion into a matrix representation ○ Cluster actors (nodes) in the network based on whether they share the same position ○ Rearrange the adjacency matrix so that actors in the same position are near one another ○ Check to see how actors in the same position are connected to: ○ ■ Other actors that share the same position ■ Other actors in different positions Come up with a simplified version of the network in which positions (cluster of actors) are the “nodes”position (using graph theory) and developing techniques (using matrix algebra) to measure them. ■ This simplified structure is a network of r elations between positions! ■ But it is still based on the original relations between actors Structural Equivalence: Two actors are similar if they have similar relationships to the sameactors or objects. Regular Equivalence:Two actors are similar if they have similar relationships to s imilaractors or objects. ● Theoretically, the notion of role implies similar patterns of connectivity to similarly positioned actors not to literally the same actors (If you lose the label, nodes that cannot be told apart are regularly equivalent. ) Image Matrix and Image Graph: Blockmodeling: ● What is a blockmodel? ○ A procedure to reorder the rows of matrix (representing nodes and their relations) such that: ■ Rows that are close together are closer to being (structural or regularly) equivalent than rows that are far apart ● The matrix can then be divided into blocks: ○ Blocks contain disjoint set of actors and thus define “positions” in the network ○ Actors in block have connections among themselves ■ ○ Actors in a block also have connections with actors in other blocks ■ ● Diagonal blocks Off-diagonal blocks We can think of blockmodeling as an attempt to reduce a matrix of N x N actors and their relations to a reduced matrix of B x B blocks and their relations Diffusion S-curve: ● Typical S-shaped pattern of cumulative adoptions ○ Initial slow uptake ○ Rapid spread ○ Final slow saturation Influence: ● Eigenvector centrality (People who can influence popular people) ● Cohesion-based network models presume that diffusion only happens between connected actors (direct influence) ● Positional equivalence (indirect influence) Comparison: ● Mechanism: Comparison to similarly positioned others Imitation: Source:Who started it. Adopter:Who adopted it. Similarity Heuristic:I should adopt the innovations that people like me adopt (homophily). Direct Cohesion Model:Focus on the connectivity of adopters to their sources via network ties. Two-Step Model:Focus on the connectivity of adopters to their sources via network sources as well as characteristics of influentials. ● Nike sends Kim K. shoes, Kim wears shoes, shows fnas, fans buy shoes. ○ Nike to Kim, Kim to fans Global Threshold Model: ● Focus on susceptibility of adopters to the behavior of multiple sources ● If 2 people start dancing, then I’ll jump in too ● Each actor (i) has a “threshold” for adoption q(i) ○ qi is between zero and one ○ 0 > q(i) < 1 ○ (i) adopts when the proportion of people who have adopted (p) is equal to q(i) Positional Equivalence Model: ● Focus on indirect influence and comparison between sources and positionally equivalent adopters ● Positionally equivalent actors don’t have to be necessarily connected to one another! Network Threshold Model: ● relaxes the assumption that persons have global information on all adoptions ○ They only know what their direct contacts are doing ○ Individuals adopt when some proportion of their contacts adopt Opinion Leaders: ● Early adopters (innovators) are well-connected ○ Central, well-integrated, prominent ○ Well connected persons more likely to adopt innovations with low cost/benefit ratios Simplex Contagion: ● Low risk: Information, gossip, disease Complex Contagion: ● High risk: social movement participation, religious conversion, migration, etc ○ ○ Strategic complementarity: (adoption cost declines as a function of existing adopters) Legitimacy: Resistance to challenge, collective acceptance, and taken-for-grantedness increases in the number of adopters ○ Credibility: Innovations appears are more trustworthy and less risky when high status members of the community also adopt ○ Emotional contagion: Cumulative interaction with a large number of adopters leads persons to adopt Tipping Point: ● Point where innovation has been adopted by enough people Rogers Classification of Adopters: ● ● ● Innovators + Early Adopters: Individuals whose time-of-adoption is greater than one standard deviation earlier than the average time-of-adoption Early Majority + Late Majority: Individuals whose time-of-adoption is bounded by one standard deviation earlier and later than the average Laggard: Individuals who adopted later than one standard deviation from the average Valente Classifications of Adopters: ● Influentials: ● Connected Bandwagon: ● Resistant Rearguard: ● Isolates: Dominance orders/ Negative Interactions: Linear hierarchy:Pecking Order Transitivity:Is a relation between node triples ● If ahas dominance over b,and b dominates c , then anecessarily dominates c. ● (a>b, b>c, c<a) Completeness: ● (either a>b or b>a) Anti-Symmetry: ● (a>b then not b>a) Cycles:A path that begins and ends with the same node (like a triangle going in 1 direction). In terms of hierarchy, this is sooo weird! Personality/Cognition Schema-based recall: ● What kinds of networks are easy for people to remember? ○ ● People store network information in a cognitive template called a s chema Schemas influence memory ● ● ○ We tend to more easily r ememberinformation that is c onsistent with stored schemas ○ We tend to forgetinformation that is inconsistentwith stored schemas ○ We tend to changeschema inconsistentinformation into a schema consistentform What kinds of networks are easy to remember? ○ Networks with h ierarchicalstructure are easier to remember ○ Networks with r eciprocity(if the relationship is friendship) are easier to remember ○ Networks with t ransitivityare easier to remember ○ Networks clustered into subgroupsare easier to remember ○ Networks featuring balanceare easier to remember The balance schema: ○ Friends of friends are friends ○ Friends of enemies are enemies ● People find it much easier to remember social structures that respect the balance schema ● People also tend to recall network as obeying the balance schema even if the original network violates the rules ○ They impose “balance” to social structures in their heads Cognitive Social Structures: ● It is the network of connections between people (including yourself but also between others) as each person perceives it to be rather than as it “really” is ● We go from asking: ○ ● To asking: ○ ● Is person i connected to person j? Does person k think that person i is connected to person j? This means that in a network of size N: ○ Each relationship in the network exists N times as recorded inside the head of each the people in the network! ● We go from a dyadic conception of relations (Rij) to a triadic one: ○ Rijk(person i is connected to person j according to person k) Self-monitoring: ● Those in high self-monitoring… ○ Regulate the way they present themselves, respond to cues (“SOCIAL CHAMELEONS”) ○ Higher indegree in sentiment (e.g. friendship) and exchange (e.g., advice) networks ○ More likely to bridge structural holes - betweenness centrality (only in sentiment networks) ○ Use of humor in interaction ○ Appropriate pace in conversation ○ Less likely to do “self-talk” ○ More likely to do “other talk” ○ Likely to collaborate and compromise Big Five Personality Traits: ● Neuroticism/Stability: ○ ○ Experiences negative emotions ■ Smaller ego-networks ■ Tend to be part of weak tie triads ■ Are less likely to introduce friends of friends Stability: ■ ● Extraversion/Introversion: ○ ○ People High in Extraversion ■ Larger networks at all layers ■ Tend to be part of a strong tie triads ■ More likely to introduce friends of friends ■ Add new contacts at a faster rate ■ Have higher indegree and outdegree centrality Introversion: ■ ● ○ Niceness, resolves conflicts ■ Have higher in-degree ■ Add new contacts as a faster rate Deta: ■ Distant/ emotionally neutral Openness/Cautiousness: ○ Like Experiencing new things ○ Cautiousness: ■ ● Prefers self-reflection Agreeableness/Detachment: ○ ● Is more emotionally “balanced” Likes consistency and predictability Conscientiousness/Carefreeness: ○ Organization, reliability ○ Carefree: ■ Spontaneous and disorganized