Comparison of two recently proposed expressions for partial least squares regression prediction error

Two recently proposed expressions for partial least squares (PLS) prediction error are compared. Using extensive Monte Carlo simulations, it is found that the expression based on the so-called errors-in-variables approach yields prediction intervals with coverage probabilities close to their nominal value, whereas the expression, which is implemented in the latest version of Unscrambler (7.0), is found to behave unsatisfactorily. The difference between the two approaches is illustrated on a real near-infrared data set taken from the literature.