A simplex centroid design for a mixture with q components consists of 2**q - 1 points. Design points are as follows:
Lattice of degree (m) | Number of components (q) |
---|---|
1 | 2 to 20 |
2 | 2 to 20 |
3 | 2 to 17 |
4 | 2 to 11 |
5 | 2 to 8 |
6 | 2 to 7 |
7 | 2 to 6 |
8 | 2 to 5 |
9 | 2 to 5 |
10 | 2 to 5 |
Minitab generates the extreme vertices of the constrained design space using the XVERT algorithm, and then calculates the centroid points up to the specified degree using Piepel's CONAEV algorithm. See Cornell1 and St. John2 for more information.
Minitab allows you to analyze data from three types of experiments:
Type | Response depends on… |
Mixture | the relative proportions of the components only. |
Mixture-process variables | the components and the process variables. Process variables are factors in an experiment that are not part of the mixture but may affect the response. |
Mixture-amounts | the relative proportions of the components and the total amount of the mixture. |
Minitab augments (or adds points to) the design as shown below. Each added point is half way between a vertex and the center of the design.
By augmenting a design, you can obtain information on the responses in the interior of the design, instead of just relying on points on the edges.
In Minitab you can create designs and analyze data in amount, proportion, and pseudocomponent units.
To convert data from amount to proportion units, the formula is:
If the total = 1, then the proportion units = amount units.
To convert proportion to pseudocomponent units, the formula is:
If all lower bounds = 0, then pseudo units = proportion units
Mixture designs include many types of model terms. The terms and their representations are:
Model | Terms |
Linear | A B C |
Quadratic | Linear + AB AC BC |
Special cubic | Quadratic+ ABC |
Full cubic | Special cubic + AB(A-B) AC(A-C) BC(B-C) |
Special quartic | Quadratic + AABC ABBC ABCC |
Full quartic | Special quartic + AB(A-B) AC(A-C) BC(B-C) AB(A-B)2 AC(A-C)2 BC(B-C)2 |
Note that there is no constant term in mixture models. Inverse terms include 1/A, 1/B, 1/C, etc.