Derringer/Suich case

This case leads to models containing all interactions of the main effects relevant for any of the $Y_i,\, i = 1, 2, \ldots, Z$ .

$\begin{eqnarray*}D^{DS} & = & \sqrt{ d^{DS}_1(Y_1) \cdot d^{DS}_2(Y_2) }\\ & =... ... \gamma_{2, 1} X_1 + \gamma_{2, 2} X_2 + \epsilon_2) )^{0.5} \\ \end{eqnarray*}$

A model for (D^DS)² would have to be of order $\prod_i (\text{order}(d_i))$ .

Bottom line: direct modelling of desirability indices is not an option.

Dipl.-Stat. Detlef Steuer
1999-10-04