ABSTRACT
There may be covariates at several levels of nesting. This chapter illustrates some issues which arise with covariates in a split plot design and with random effects. The split plot design embodies most of the key features that arise when there is more than one random error, associated with more than one size of experimental unit. A design which is superficially similar to the split plot is the strip plot in which plots are divided into subplots in different ways for different factors. A covariate which is measured on the split plot can affect all comparisons and tests for both split plot and whole plot. In split plot designs, part of the variation 'explained' is the whole plot error. Blocking and whole plot error are mechanically considered part of the 'model' for most packages, even if they allow subsequent tests using different errors.
