I have a dataset with a handful of predictors and one output variable which is categorical and can only be C or N.

I am working in R, using the plsda function from the mixOmics package. When I plot the results of the PLSDA with the plotIndiv function for the first 2 variates, the separation is not great. It looks like this:

PLSDA plot for my dataset

There is an argument you can give to the plotIndiv function called rep.space, which the manual says it determines the subspace where the individuals are projected. Its possible values are X-variate, Y-variate and XY-variate. When the object being plotted is the result of a PLSDA, rep.space defaults to X-variate.

But if I set rep.space to XY-variate, then it looks like this:

PLSDA plot for my dataset with rep.space set to XY-variate

Which looks amazing. But is it too good to be true? In other words, is it using the outcome variable to artificially separate the individuals in the plot? What does XY-variate even mean? I understand that the X-variates are the new variables, analogous to the PCs in PCA. But what are the XY-variates? Are my predictors actually separating the 2 classes, or is this artificial?

  • $\begingroup$ Could it be, for example, that whereas in X-variate mode it plots the components found using deflation for each succesive component (like in PCA), in XY-variate mode it's using the important components from the first mapping only? $\endgroup$
    – Dave
    Jun 23, 2019 at 21:35


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.