I'm using JMP Pro PLS, but my response is binary. I've seen an approach, PLS-DA that does the math in a different way recognizing that the predictors and the responses are not of the same type and so the notion of covariance between them is different than if they were both numeric. I have only skimmed the literature before taking a deeper dive. I thought JMP would use an approach specifically designed for binary responses. However, in the notes for PLS in JMP Pro, when the response is binary the notes indicate that the response is simply coded to 0/1 and then the algorithm proceeds as if it were a continuous response. Is the solution impacted by this method? That is, is it close to the same solution as PLS-DA?
At this point it appears that at least some of the papers about PLS-DA treat binary responses the same as continuous. They code to 0/1 and move on. They state that no assumptions are made about the distribution of the responses, so, in effect, no assumptions are explicitly violated by coding. I suppose there would be an additional step to set a threshold for classification.