ABSTRACT
In this chapter we consider situations where we have experimental units of different sizes (e.g., large plots of land containing multiple small plots of land) with one treatment factor being randomized and applied to the large units, while another treatment factor is being randomized and applied to the small units. This leads to so-called split-plot designs which are (most often unintentionally) very common in practice. We show how they can be easily fitted in R using a mixed model approach. In addition, we give an intuition about the precision and power at the different levels and illustrate what goes wrong if such data is analyzed as coming from a completely randomized design.
