Question regarding format of categorical variables in Fit Model

[#1]

Good afternoon,

I am trying to model a single response with 5 continous predictors and 2 categorical predictors (one two-level and one three-level), with two-way interactions and quadratic effects. The data was collected during a experiment desgined with the "Custom Design". Using the Stepwise Regression personality in Fit Model, I've obtained a model usign the Minimize AICc/BIC stopping rules. In the Effect Summary, a series of of interactions have been deemed by JMP to be the most important. The three level categorical however, is presented in a format I do not understand and struggle to find any information about online or in these forums. The three-level categorical is "Cleaning Type" with levels 1) Soap/water, 2) Solvent and 3) Sonic. In the Effects Summary shown in the image below, the categorical variable is shown with two or more of the levels in a bracket with symbols like "-" or "&" between the levels: 

What does the brackets and symbols mean? Where can I read about this?

I've also noted that when I run the Prediction Profiler, the Cleaning Type categorcial variable is suddenly divided into two continuous variables with different combinations of the symbols inside the brackets, as shown in the following image:

Notice that this does not happen with the two-level categorical variable "Day".

I've completed the STIPS training course but I cannot remember this being covered. Is there anybody out there who can help shed some light on this?

/JF

[#2 · Accepted solution]

JMP constructs new predictors from the categorical factor levels when there are more than two levels. It finds combinations of the levels that maximize the difference in the mean response. These predictors should increase the chance of finding real effects (power).

Please read the section about categorical predictors with the Stepwise platform.

[#3]

Mark's link explains the expressions.

I'm a bit confused by your approach?  You have run an experiment designed with the factors you provided.  There is a "known model" for the design you have chosen.  Why are you running stepwise regression?  This is usually used for existing data sets to try and find an appropriate model for the data in hand (a data mining technique).  I recommend you start with a saturated model and reduce the model using the numerous statistics and graphical plots available.

Quadratic equations for a categorical factor are non-sensical.  There is no continuum between the 3 types of cleaning (You might be able to create a continuum if you could measure the differences in the cleaning types quantitatively).  JMP will essentially create models for each categorical level. 

Also, Day can't be a factor per se.  You can't get done with the experiment and conclude Day 1 is better as you can't actually set Day (or go back to Day 1) in the real world (if you can, you don't need to run experiments).  Day more than likely is a block effect.