I have a cell culture, 5 Block DOE with 6 input process parameters. The first block showed temperature (one of the 6 input process parameters) to be off from the intended setpoint, but I know the actual setpoint the conditions were run at. Does this invalidate my entire model? Or, can I continue? Or, is it better to do a 4 block design and lose some two-factor interactions?
My initial thought is to recreate a new and seperate design with the executed runs, then use the JMP Compare Designs platform to see what the consequences are from a model evaluation point of view wrt to the planned original design. I'm not sure of the exact model you specificed so some terms may not be estimable or at the least have different parameter estimate correlations. But my initial thought is you are not in an "all is lost" scenario. Once done with this step, I would proceed with your analysis as you would normally with an eye towards the inadvertent levels within the one block. Again, some terms may be problematic, but you may still be able to realize your primary experimental goals with this new design.
Thank you for your reply @P_Bartell.
I compared the two designs (the current 5 block run with the four block run without a few of the secondary interactions, but it just shows the 5 block design is more optimal, which makes sense because it includes all the secondary interactions. Or, are you suggesting I make an entirely new design?
I am also confused with your last comment, what do you mean by "inadvertent levels within one block?"
@CricciWhat I was suggesting in my initial reply was to compare your originally intended design with the actually executed design. I don't think you'll see dramatic design evaluation characteristics between the two, 5 'block' designs. Then as my former colleague @markbailey suggests, analyze the actually executed design by substituting the actual levels into the design matrix. My comment around 'inadvertent levels' refers to the single 'block' that had the temperatures run at unintended levels. But @markbailey 's questions about a 'whole plot' are well taken.
This aberration is not a problem. If I understand, the temperature level in the first block was not set correctly but is a known value. Correct? If so, simply update the temperature values for that block in the data table before performing any analysis.
Odd that the temperature was supposed to be the same level for the entire block. That design is very unusual. Did you make temperature a hard to change factor? Do you mean whole plot, not block?
Thank you for your reply @markbailey
"the temperature level in the first block was not set correctly but is a known value. Correct?" -- Yes.
"the temperature was supposed to be the same level for the entire block." -- No. The temperature had three levels (high, low, and control), but after the experiment completed, we realized the temperature was off on all of them.
"Did you make temperature a hard to change factor?" -- No.
"Do you mean whole plot, not block?" -- I am not sure what you mean by whole plot. We're running five blocks, but this one block had an error in all the temperature probes. Therefore, the temperature we thought we were running at (no matter the condition, high, low, or control) was not the actual temperature. We can back-calculate the actual temperature because we know how much the temperatures were off for each condition. This was our first block out of the five blocks, but we are on a tight timeline and only have time for four more blocks, whether they are within this DOE, or in a new DOE that reduces a few interactions. Did that help clear up our situation?
There is no issue here. I just wanted to be sure I understood your design and how you actually ran the experiment. Thank you for the clarification.
So simply update the factor levels in the data table before your regression analysis and model selection.