Many social-science questions ask how individuals' influence shifts over time, but answering them is hard when network data are sampled. Sampling removes ties and can bias inference, and frequent snapshot data are usually required to observe change. This paper introduces a practical method that recovers influence rankings from sampled, longitudinal network data and makes daily tracking of influence feasible.
📊 How Influence Is Measured From Incomplete Networks
- Introduces "neighbor cumulative indegree centrality": the sum of an individual's connections' indegrees (their connections' connections).
- Shows that ranking actors by this measure preserves the rank ordering of influence that would appear with complete network data, even when the network is sampled.
- Emphasizes rank preservation (who is more influential than whom) rather than claiming exact recovery of absolute centrality scores.
🧪 Validated on 2011 Twitter Networks in Bahrain and Egypt
- Demonstrated and validated the method using longitudinal observations of 21 Twitter accounts from early 2011.
- Uses these real-world sampled network data to show the method's ability to reproduce influence orderings and to detect changes over time.
📈 Detecting Daily Changes in Influence
- Shows how the neighbor cumulative indegree centrality can be computed day-by-day to analyze when and why an individual's influence rises or falls.
- Enables researchers to move from static snapshots to event-driven, temporal analyses of social influence using sampled data.
🔑 Key Findings
- Ranking by neighbor cumulative indegree centrality preserves influence orderings that complete data would produce.
- The method lowers the barrier to accurately measuring influence from sampled networks.
- It supports daily, longitudinal analysis of influence dynamics without full-network observation.
🌍 Why It Matters
- Provides a practical tool for fields that rely on influence measurement under data limitations, including voter mobilization, marketing, and digital political communication.
- Makes it possible to study who matters in a network and when they matter, even when full network enumeration is infeasible.
