We want to run a medium optimisation experiment in shake flasks but we have two noise factors that we need to compensate for:
- We have four incubators that can only hold four flasks (incubator noise)
- An experiment takes 24 hours and we have booked the incubators for two days (day/batch noise)
We want to get the maximum amount of information out of the experiment so we want to perform the following:
4 incubators X 4 Flasks X 2 Days = 32 runs
Ideally we'd like to test 4-5 continous medium factors but we'd settle for less if we can get more reliable results (some interactions and quadratic terms are likely). We don't care about the difference between the days or the incubators, we just want to account for the noise that they might bring. From experience we know that both the incubator and the day have an impact on the outcome.
So our question is how do we design this in JMP so that we can account for the noise while have fixed16 runs per day and 2 X 4 runs per incubator (Days X incubator capacity)?
Is this a split plot design? Can we add two random blocking factors somehow? When trying this out we never seem to be able to get both limits right (8 runs split over two days per incubator).
