Currently JMP PLS can include categorical variable in PLS model by setting multiple dummy variables for each categorical variable. For example, if I have a three level variable, chassis type, low level, middle level and high level, JMP will generate 3 dummy variables, low chassis type (1 for chassis type =low level, 0 for others), middle chassis type (1 for chassis type=middle level, 0 for others) and high chassis type (1 for chassis type =high level, 0 for others), and then include those variable in PLS to get final model. In this way, the VIP value will calculate for each level of original chassis type variable but not one VIP value for whole chassis type variable. There is a new method called PLS-CAP (PLS algorithm for Categorical Predictors) which can generate a single VIP value for one categorical variable regardless how many levels the categorical variable has (see attached PDF file). There is very useful for our consumer analysis because we normally want to understand one variable impact rather than each level of the variable impact to response. Hope JMP can add this feature in PLS module.
Zhiwu Liang liang.zw@pg.com Kobe Technical Center Japan
