Hi, The design is a Full Factorial(4-Factors) + 1 Center Point + 2 special combos-19 runs-1 Response-4 missing
The 4 factors are continuous (level +1= presence at the highest dose; 0= presence at half dose; -1=absence=dose 0).
In a first stage, I expected to evaluate the importance of interactions and main factors to select the best components to make a product. This is not a mixture design as more than 99.9% of the product is a solvent(only additives without a fixed quantity)
Final Product= combo of 1 to 4 A (A6,A8,A9,A10). Unfortunately the response for 4 runs were invalidated after QC and therefore I don't have sufficient degrees of freedom to estimate the 3-way interactions and the 4-way interaction without simplification of the model. In a first time, I estimate only the 4 main factors and the 6 two-way interactions.
Sorted Parameter Estimates
Term Estimate t-ratio Prob>|t|
(A8-0.11667)*(A9+0.01667) -0.007222 -3.41 0.0271*
A8 -0.006444 -3.10 0.0361*
A9 0.01211 5.90 0.0041*
I have attached the data in a .jmp file and output, profiler and interaction graphs and comments in a word doc.
The interaction between the 2 continuous factors A8 and A9, centered respectively by their means in the design(0.11667 and -0.01667), was significant indicating that the effect of A8 was clearly negative when A9 was present at its maximum dose(level +1,=higher dose), near zero when A9 was absent(level-1,=dose 0) and intermediate when A9 was at half dose(level 0-half dose).
The A8 parameter was estimated for A9 at its mean value. Is it possible, with JMP, to have an interaction not centered and respective tests of A8 parameter when A9 was absent(at level -1,=dose 0) or present at the higher dose(at level +1,=higher dose).
