Fig. 2: Model simulations of three learning mechanisms.

We simulated the learning model to generate zapping behavior of agents in the heterogeneous conditions which include zapper and avoider bot-players. We examine the pattern of simulated agent’s zapping behavior (right) and the underlying learning curves (left) which represent the way beliefs about players’ likelihood to zap are learned over time. A We disabled learning and attribution effects and set the group identity prior parameters to be either low (–0.9) or high (–0.1). The model predicts high zapping rates when prior is high, and low zapping rates when prior is low, and no learning about players’ behavior. Note that situational variables such as distance from target, and the tendency of zappers to be closer to other players, affected simulated zapping behavior. B We disabled the attribution effect, fixed the prior effect and varied the learning rates (LR) of zaps to be either low (0.2) or high (0.9). Learning curves illustrate faster learning of zappers behavior (left), and difference in zapping rates between zappers and avoiders (right). C We fixed zapping learning rates and priors and varied the zap attribution rate parameter to be either low (0.2) or high (0.5). High attribution rates increased the estimation of the zapping probability of zap avoiders (left) and led to more similar zapping behavior towards zappers and avoiders (right). Lines in learning curves indicate mean zapping estimation variables, shadows indicate 95% confidence intervals of the mean. Boxplots include the median in bold line, interquartile range is represented by the box, minimum and maximum range by the whiskers, and outliers by dots.