where Y is each response, the first summation is the linear blending
portion, the second represents the excess response from the quadratic
model over the linear model (30), and represents the error of model.
This model was fitted using the PLS (partial least squares projections
to latent structures) regression technique. PLS has been extensively
described in the literature (28).
Although the second-degree model provided information on each
of the components individually (main effects) as well as on pairs of
components (secondary effects), to locate stationary points, data were
fitted by PLS regression to a special form of polynomial equation
developed by Scheffe and generally known as the special cubic model
(31), which includes a third-order term (X1 × X2 × X3) to reveal the
three-component interaction, if any, according to the following model