Blocked DSD results in quadratic effect confounding

Question

I created a 20 run blocked 7-factor DSD with 5 continuous & 2 categorical effects. There are five blocks with 4 randomized runs/block; as advised by JMP experst in each block there are center runs, in this case two/bock; I get a DSD where confounding of the quadratic effects is acceptable. However by adding dummy factors in order to increase runs and factor estimation power I get blocked DSD's with significantly confounded quadratic effects! Inspecting the 2 center points in each block I can see that they always are at categorical level L1L1 and L2L2 (and never at L1L2 and L2L1). Is there something wrong? How to solve this confounding problem? Thanks.

cwillden · Accepted Answer

You shouldn't need to manually add dummy factors any more with the updated DSD platform.  That's what the "Extra Runs" option does for you, but that way you don't have the extra dummy factors in your outputted table.  I think this is what is actually creating your problem, because when I ask for 7 blocks with the option for centerpoints to be added, and use Extra Runs = 8 instead of having X6 and X7, I don't end up with nearly as many centerpoints as you did.  It looks like dummy factor approach to adding runs is having a bad interaction with the blocking functionality.  When I look at the Design Evaluation with quadratic terms specified for the Extra Runs approach, the correlations are reasonable.  I also have 2 fewer runs, which is a bonus I guess.

Did you really want 7 blocks? The description says 5, so I'm confused about that one.

frankderuyck · Accepted Answer

OK thanks for input! 7 blocks with 4 runs/block is fine. I did not know about these extra run option. In quite some demo's by Brad Jones & Peter Goos dummy variables are used to increase doe power also in a discovery summit, think 2 years ago, Brad claimed that in a blocked DSD in every block center runs are necessary, confusing..