When performing the optimization through a DoE approach, and the results of some runs are null what should be put as result in databook?
Hi Joao, is there a reason for not using the actual value (zero)? What kind of design are you using? Depending on the design, such a value may be undesirable (despite accurate) as it may, among others, impair a suitable surface fit. Best regards, Luis
As suggested by Luis there are 2 things to consider. First, if the response is truly zero (i.e., a good data point, not some mistake or error), then it must be included. Second, if the value is a zero then the difficulty is the design space being used. Check your design space to ensure that there are not substantial regions where the response is zero. In my work in plant tissue culture I call these regions "zones of death". They do not model well because they are flat and polynomials work best on smooth curves.
If the "zero" space is required, then one possible approach is to divide the space into 2 regions and consider these 2 regions as 2 different states. Then model each region separately to characterize these states. For example, something like this occurs in the design space when treating the PO3(3-) ion as a statistical factor. There are 3 distinct "states", and they are because PO3 is a triprotic acid with 3 pKa's (pH 2.12, 7.21, and 12.44). I can model PO3 very well by DOE as long as I confine the design space to a single pKa at a time.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running