Positive and Negative Q2

hi @bort,
There has been a few discussions about the Q2. Basically (copy pasting from the e-book we are currently preparing):

The 𝑄^2 criterion is a global measure that applies to both PLS1 and PLS2 and is calculated per dimension h (denoted 𝑄_h^2). We can decide to retain a dimension h if 𝑄_h^2 ≥ 0.975, a (somewhat) arbitrary threshold used in the SIMCA-P software (Umetri, 1996). A negative value of the 𝑄_h^2 indicates a poor fit of the PLS model on the data.

In your case, going from positive to negative indicates a poor (predictive) fit when you go from dimension 2, or vice versa. It also depends on the type of CV fold you are using, and the number of samples.

You can refer to earlier posts for some details: Q2.total negative in perf.pls

Kim-Anh