In a series of laboratory and online experiments, we investigate the perceived and actual influence, direct and indirect, that early movers have over later movers in a sequential-move collective action game played on a network. Each human subject is assigned to a particular node in a specified 5 or 10 node network, and chooses “In” or “Out” knowing how many neighbors have previously chosen “In”. First-mover influence is measured as the proportion of network members who choose “In” following a first mover’s “In” versus “Out” choice. Influence is quite heterogeneous across nodes. Human subjects are aware of the influence their network position conveys and act on it, but tend to overestimate influence at poorly connected nodes. A simple adaptive learning model explains entry and perceptions of influence reasonably well, especially at well connected nodes.