Magnetic systems with several stable states characterized by distinct magnetic orderings are particularly important for the technological applications as they can in principle be used to code, store and process data. However, the preparation of a magnetic system in a particular state can be destroyed by spontaneous, thermally activated transitions to other available states. These transitions are typically rare events on the time scale of oscillations of the magnetic moments, making direct simulations of spin dynamics an impractical way to calculate transition rates. This separation of time scales, however, makes it possible to apply statistical approaches to study long time-scale spin dynamics. Development and implementation of such statistical methods will be presented in the lecture. Applications of the methods to various systems, including thermally-active artificial spin ice, magnetic nanoclusters on a surface and skyrmions, will be discussed.