Mixtures
with Gosset
Let's generate a design for a mixture problem. Let's say we have
4 factors. The ranges are, M1 from 0 to 1, M2 from 0 to 0.8, M3
from 0.3 to 0.678, and M4 from 0 to 1. We want an I-optimal design
for the full quadratic model with 3 more trials than the minimum
required. Here's the program:
10 range M1 M4 0 1
20 range M2 0 0.8
30 range M3 0.3 0.678
40 constraint M1+M2+M3+M4=1
50 model (1+M1+M2+M3)^2
"constraint" specifies that the mixture component levels
must add to 1 - a constraint imposed by nature. Also notice that
we omitted M4 from the model. The mixture constraint makes the exclusion
of one factor necessary.
Now we compile, compute moments, and generate the design as we did
in the Brief Tutorial.
|
