Predicting Donor Behavior Using the Dynamics of Event Co-Attendance Networks

This study investigates the dynamics of charitable donor co-attendance networks to help understand fundraising outcomes. We analyze a multi-year network of co-attendance at fundraising events, examining topological structure, node characteristics, and network properties. Key findings include a 76% increase in giving value for donors with increased centrality rank and a positive peer effect indicated by increased co-attendance with high-capacity donors, among donors who increased or maintained their giving levels. Additionally, we observe assortativity in giving patterns among co-attending donors, and identify interlinked cliques of communities with varying sizes, with larger communities having a higher proportion of wealthy donors. Furthermore, we develop a Long Short-Term Memory (LSTM) network to predict donor propensity, considering node-level features such as assortativity, centrality, clustering coefficient, and exposure to high-capacity donors. Our results demonstrate that a multi-variate model outperforms baseline statistical models, enabling outcome assessment before events. These findings have implications for budget allocation strategies and maximizing fundraising effectiveness.