Choosing a Design
main topic
      

Before you use Minitab, you need to determine what design is most appropriate for your experiment. Minitab provides simplex centroid, simplex lattice, and extreme vertices designs.

When you are choosing a design you need to

·    identify the components, process variables, and mixture amounts that are of interest

·    determine the model you want to fit - see Selecting model terms

·    ensure adequate coverage of the experimental region of interest

·    determine the impact that other considerations (such as cost, time, availability of facilities, or lower and upper bound constraints) have on your choice of a design

For a complete discussion of choosing a design, see [1].

To help you visualize a mixture design, the following illustrations show design points using triangular coordinates. Each point on the triangle represents a particular blend of components that you would use in your experiment. For simplicity, the illustrations show three component designs. The diagrams below only show a few of the mixture designs you can create. Minitab can also create simplex lattice designs up to degree 10 and extreme vertices designs. For an explanation of triangular coordinates, see Triangular coordinate systems.

 

     Unaugmented

     Augmented

 

 

Simplex
Centroid

image\SC_UNAUG.gif

permits fitting of up to a
special cubic model

image\SC_AUG.gif

permits partial fitting of up
to a full cubic model

 

 

Simplex Lattice
Degree 1

image\SL1UNAUG.gif

permits fitting of a linear
model

image\SL1_AUG.gif

permits partial fitting of up
to a quadratic model

 

 

Simplex Lattice
Degree 2

image\SL2UNAUG.gif

permits fitting of up to a
quadratic model

image\SL2_AUG.gif

permits partial fitting of up
to a special cubic model

 

 

Simplex Lattice
Degree 3

image\SL3UNAUG.gif

permits fitting of up to a
full cubic model

image\SL3_AUG.gif

permits fitting of up to a
full cubic model

Note

When selecting a design, it is important to consider the maximum order of the fitted model required to adequately model the response surface. Mixture experiments frequently require a higher-order model than is initially planned. Therefore, it is usually a good idea, whenever possible, to perform additional runs beyond the minimum required to fit the model. For guidelines, see [1].