Different sized whole plots in Split Plot DOE

Phil_Kay

Hi,

A user told me they have experiments with Hard To Change factors and they sometimes get split plot designs where the whole plots don't all have the same number of runs allocated, even when the total number of runs is divisible by the number of whole plots.

I don't have an illustrative example from them. Has anyone seen examples like this?

My simple answer is that the algorithm has found that to be the optimal design.

I don't see any way to specify that whole plots should all be a fixed size. I imagine that most experimenters would want to be able to do this when they have HTC factors.

Thanks,

Phil

DonMcCormack

I see this happen often, even after I up the time or number of starts. Not sure what's going on, my suspicion is that it's algorithmic. I would think a balanced design would be more optimal. Depending on the complexity of the design, you could disregard the hard to change factor, randomize the run order, assign block numbers sequentially based on block size, then randomly assign hard to change factor levels to the blocks. To get the correct analysis, you'd have to add block to the model and make it random along with the hard to change factor. You might also have to set the Factor Changes column property to Hard to Change for the hard to change variable. I'd be curious to see how the diagnostics for this would compare to the original design.