Phantom Networks: Window on Agency in Social Mechanisms

Our goal is to offer proof of concept for the idea that behavioral networks emerging with and beyond assigned treatment networks in experiments can provide a valuable window on agency. We call these often unobserved behavioral structures phantom networks. Most evidence of agency is entangled with audience — what people recall, how others attribute, or who is seen to act. We strip that away to observe agency directly in the ties participants build within and beyond what they are assigned. Their deviation from assignment is not error but structure-in-action, a behavioral trace of agency. We offer three exhibits of that “behavioral deviation” revealing agency in gendered brokerage, rival brokerage, and structural reproduction. Across the three exhibits, behavioral deviation shows how a team works (sometimes in surprising ways), not how well the team works. Teams perform as expected in theory, but the behavior responsible for their performance is made more apparent by watching behavioral deviation. We close with a summary proof of concept and discussion of next steps for particularly promising future work