Abstract

We measure the “information content” of online economic networks – sets of connected entities where links are created by realizations of shared prior outcomes. We conjecture that such electronic networks contain information about similarity in latent preferences across actors that are not captured by observable product or consumer features. We provide a methodology for measuring this information content in a rigorous and outcome-driven manner using matchedsample estimation techniques to mimic the optimal use of all observable non-network data. Using detailed transaction-level data about 257,851 contributions to 95,684 charitable projects by 99,720 donors on an leading online giving web site, we show that co-donors in an economic network have an 80-fold higher overlap in future choice than a random benchmark, the network outperforms even matched sample alternatives based on sophisticated feature-based predictive models 5-fold to 23-fold, and this inferred overlap in latent preferences persists with local network traversal.

Download

Share

COinS