Modelling with discrete qualitative variables

When one or several experimental variables are discrete, we cannot rely on a geometrical interpretation of smooth and continuous response functions. A change of catalyst, e.g. Pd on carbon to Pt on alumina, may well change the influence of other continuous variables. 

---

The randomized complete block design has provided a sensitive way of viewing the data from this set of experiments involving two qualitative factors. The linear model using dummy variables ignores much of the variation in the data by again focusing on pairwise differences associated with the different discrete levels of the factors of interest. 

---