A linearization method for partial least squares regression prediction uncertainty

• We study a local linearization approach to provide an approximate variance for PLS predictions.
• We note and correct some problems with the original formulae.
• We develop an alternative method of computation using a parametric bootstrap.
• We study the stability of the resulting approximation using some simulations.

We study a local linearization approach put forward by Romera to provide an approximate variance for predictions in partial least squares regression. We note and correct some problems with the original formulae, study the stability of the resulting approximation using some simulations, and suggest an alternative method of computation using a parametric bootstrap. The alternative method is more stable than the algebraic approximation and is faster when the number of predictors is large.