Classification of urn models with multiple drawings

Abstract

We consider multicolor urn models with multiple drawings. An urn model is called linear if the conditional expected value of the urn composition at time $n$ is a linear function of the composition at time $n-1$. For four different sampling schemes - ordered and unordered samples with or without replacement - we classify urns into linear and non-linear models. We also discuss representations of the expected value and the covariance for linear models.