New preprint: “Learning Social Influence with Unobserved Attention”, with Moyi Tian, Linh Huynh, Alexander Wiedemann, Vicki Modisette, Phil Chodrow, and Heather Brooks. We build a probabilistic agent-based model where agents attend to one source at a time and may ignore it, fit it by maximum likelihood with attention unobserved, and find that modeling attention usually matters more for prediction than the choice of interaction kernel.