17.4 Regression Models Including Nominal Variables with Three or More Levels

Regression methods for nominal variables with two levels can be easily extended to nominal variables with three or more levels. As a rule, a nominal variable with k levels requires (k -1) coding variables. In JMP, each coding variable will then have (1, 0,-1) values. In the loan amount to sales revenue example, suppose that Region of Branches is a nominal variable with k = 3 levels. A third region called South Wells is added. In this case, because there are three branch regions, i.e., k = 3 levels, we need (k - 1) = 2 coding variables. The three levels will be identified by two coding variables called Region of Branches [Aderly] and Region of Branches [Purth] with the internal coding shown in Table 17.1.

Table 17.1. Effect Coding of Variable with Three Levels

Level of Region of Branches	Region of Branches [Aderly]	Region of Branches [Purth]
Aderly	1	0
Purth	0	1
SouthWells	-1	-1