An approach to choosing statistical models for data when the purpose is simply to describe the given data rather than to estimate the parameters of some hypothetical population.