Statistics is all about randomization. As far as the maths is concerned every single run of your DOE is completely independent of the rest. The levels of every single factor are set specifically for that run and don’t inherit anything from any other run.
For most factors this is pretty reasonable: when mixing things together it’s not usually difficult to make each mixture separately, particularly in automated systems.
But for some factors it’s much harder to do this. This may be because of the way a particular machine operates, or simply because the alternative would be extremely difficult, time-consuming, or expensive.
For example, when investigating the effects of time on a reaction it would make no sense to run every reaction one after the other in many cases - for larger numbers of reactions and longer times, the total time required could run to days. Starting a block of reactions at the same time (with a 96-channel pipette for instance) is considerably more sensible!
Similarly, it’s rare to have per-well control of temperature, instead most incubators heat an entire plate at a time. Shaking a single vessel is possible if it’s a flask or falcon tube, but not if it’s a well in a plate. While you could simply fill up a single well per plate it’s unlikely you’d have enough shakers or incubators to do reasonable numbers of runs.
The solution in statistics is to define the factor as hard-to-change. Identifying this upfront means the design can account for this restriction in randomization and ensure that the experiment is as informative as it can be. This leads to the type of design known as a split-plot.
To learn more about the statistics behind hard to change factors, click here.
To learn how to define a hard to change factor in Synthace, click here.