An approach to evaluating screening experiments when several responses are measured

Abstract Strategies are discussed for the evaluation of orthogonal screening experiments when many responses have been measured. By using principal components (PC) decomposition of the response matrix, the systematic variation of the responses in the individual experiments over the entire set of responses can be described by the PC scores. As the score vectors are orthogonal, they can be used as independent response variables to account for the systematic gross variation. Five chemical examples are used to demonstrate how these score responses can be used to identify significant experimental variables in screening experiments when factorial, fractional factorial or D-optimal designs have been used. It is shown that the significant experimental factors are easily distinguished from experimental noise by using cumulative normal probability plots. The use of partial least squares modelling to further probe the experimental variables is discussed.