In this paper we present a model of impact dynamics in large dimensional systems. We describe a hybrid method, based on graph theory and probability theory, which enables us qualitatively to model the statistics of global dynamics as parameters are varied. Direct numerical simulation reveals a sudden jump from no impacts within the system to many repeated impacts at a critical value of system parameters. We show that a simple model of the most likely number of impacts also possesses a sudden jump and provides good agreement with the numerical results for large impact probability. A refinement of this model improves the agreement at lower impact probability values.