Interpreting PLS regression coefficients for a binary categorical variable

[#1]

I used Fit Model with the Partial Least Squares personality to create a model for a continuous dependent variable. Explanatory variables include several continuous variables and a categorical variable with two options, A or B. I understand that option A and B are treated as separate dummy variables. When I run the fitting, the regression coefficient for option A is about -0.4 and the regression coefficient for option B is a similar value. How could this be? I assumed that the coefficient for option A was determined relative to option B, and expected the two coefficients to be different, at the very least.

[#2]

The parameterization of the linear predictor in JMP is such that the coefficients for a categorical term must sum to zero. Here is an example of a PLS regression using the Fitness data set. The dependent variable is Oxy. Notice that the two-level predictor Sex has coefficients of equal magnitude but opposite sign.

Is that what you see in your case?

[#3]

Hi Mark, thanks for your reply. My two-level predictor actually has coefficients -0.3160 and -0.2881. I've centered and scaled my data and also double checked that there were only two levels for this predictor in the whole dataset.

[#4]

I don't know. Maybe another member has the answer.

Otherwise, contact JMP Technical Support (support@jmp.com). Please reply with their answer if you do seek their help.