Understanding Bi-partite Networks
Bi-partite (or two-mode) networks consist of two distinct sets of nodes, with edges only connecting nodes from different sets. This structure is fundamental in organizational research for analyzing affiliations, memberships, and participation patterns.
Bi-partite Network Structure:
- Two Node Types: Network contains two disjoint sets of nodes (e.g., people and events)
- Cross-Type Edges Only: Edges connect nodes of different types, never within the same type
- Affiliation Matrix: Represented as rectangular matrix A where A_ij = 1 if person i attends event j
- Common Examples: People-events, authors-papers, directors-boards, actors-movies
Network Projections:
- Type 1 Projection: Creates network among Type 1 nodes based on shared Type 2 connections
- Type 2 Projection: Creates network among Type 2 nodes based on shared Type 1 connections
- Edge Weight: Number of shared connections determines edge weight in projection
- Information Loss: Projections simplify structure but lose some relationship details
Key Metrics for Bi-partite Networks:
- Density: Proportion of possible cross-type connections that exist
- Degree Distribution: Patterns differ between node types (e.g., popular events vs. active participants)
- Co-occurrence: Frequency that two Type 1 nodes share Type 2 connections
- Nestedness: Extent to which specialist nodes connect to subsets of generalist connections
Organizational Applications:
- Board Interlocks: Directors serving on multiple corporate boards create firm-to-firm ties
- Team Membership: Employees participating in projects create collaboration networks
- Event Participation: Attendance at meetings reveals informal communication structures
- Knowledge Sharing: Experts and topics create expertise networks in organizations
- Resource Allocation: Departments and budgets reveal organizational priorities
Analytical Approaches:
- Keep Bi-partite: Analyze full two-mode structure to preserve complete information
- Project and Analyze: Create one-mode projections for standard network analysis
- Compare Projections: Different insights emerge from each projection type
- Weighted Edges: Use co-occurrence counts as edge weights in projections
- Community Detection: Identify groups in both node types simultaneously
Interpretation Guidelines:
- High-degree Type 2 nodes are central venues or popular events that integrate the network
- High-degree Type 1 nodes are active participants who bridge different contexts
- Strong ties in projections indicate frequent co-participation or shared membership
- Isolated components may reveal distinct organizational subcultures or specializations
