Screening Design

Screening designs are available for mixtures with 6 to 50 components. Above 24 components the components must all have about the same range with few if any additional constraints.

They allow you to look at a large number of components in a minimal number of blends. However, they won’t give much information, if any, about interactions (synergisms or antagonisms). Screening designs will reveal the important components - positive or negative. Stat-Ease offers different screening options depending on whether or not you’ve created a simplex experimental region.

For a simplex space, the program requires the vertices, but you can add axial check points and constraint plane centroids. These optional points provide lack-of-fit measures for the model. Replicates of the overall centroid provide an estimate of the pure error. The program recommends five centroid replicates, but you can change this if you like. The centroid also provides an estimate of curvature. If the response curves up or down in the middle of the space, it will be captured in the form of the quadratic terms. These may be aliased, but if significant, the quadratic terms indicate the need to do a more sophisticated design the next time around. Ideally, the screening study leads to a reduction in components so you can afford an in-depth experiment.

In the case of a constrained design space, many vertices are possible. The Vertex-Screening design uses the D-optimal algorithm to pick the best vertices to estimate a linear model. You must decide how many vertices to put in the screening design. The program recommends a number equal to two times the number of components, but you may want to scale this back somewhat. The desired number of vertices is selected D-optimally from the available set. You also should run a number of overall centroid replicates - the program recommends five.

Working with mixture screening designs