iSixSigma

DOE Fractional Confounding Conclusions

Six Sigma – iSixSigma Forums General Forums Tools & Templates DOE Fractional Confounding Conclusions

This topic contains 6 replies, has 3 voices, and was last updated by  Doa 9 years, 4 months ago.

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #53481

    Doa
    Participant

    I’ve just begun DOE fractional factorial training; I have a fractional factorial design with two levels, three factors. I confounded factor C with AB interaction (C=AB), resolution III.

    I have found factor A to be significant and factor B not significant. The confounded factor C=AB is significant. How does one distinguish whether the significance is due to factor C significance or factor AB significance?

    Should a follow up be run?

    Thanks,
    Marty

    0
    #190314

    Robert Butler
    Participant

    The short answer is you don’t. When you decided to run that particular fractional factorial you made the choice with respect to the confounding. Part of that choice was that AB would not be significant. Therefore any effect found for the alias of C = AB would be attributed to C.

    If you want to separate C and AB you will have to augment you existing design with a few more specific runs so that you can examine the effect of C and AB independently.

    The above addresses the issue of C vs. AB but what I don’t understand is since C is independent of A and B why would you be concerned about the confounding of C and AB?

    0
    #190318

    MBBinWI
    Participant

    If you were suspicious about interaction effects for 2 way interactions, you never should have run a level III. Luckily, it is pretty straightforward and not very costly (relatively speaking) to add the additional runs necessary to resolve the issue (make sure to eval blocking).
    However, a bit of pre-work should have shown that factor “B” was not a significant individual actor and you could have then switched it for factor “C” and not been in this quandry.
    With 3 factors it is not usual to run a half-fraction due to the problem you have run into.

    0
    #190320

    Doa
    Participant

    Hi Robert,

    Yes this is what I thought, this was given to me as a hypotheical so I do not have the back up data. But I agree that to draw a firm conclusion as to whether the significance is due to C or the AB interaction, additional or augmented runs must be run.

    Given this scenario the choices were few; a Full factorial or a resolution III run, I’ve heard conflicting comments regarding whether a level III should ever be run. I assume that this is due to the increase in ambiguity regarding significance.

    Thanks for your assistance

    Regards,
    Marty

    0
    #190323

    Robert Butler
    Participant

    There may be conflicting comments concerning main effect designs but the reality is that in many instances the choice is that of either running a main effect design or running a series of WGP experiments (Wonder Guess Putter) which typically resolve little and still wind up costing money. Every design effort I’ve been involved in has had time and money constraints that made all but the smallest of full factorial designs impractical.

    Consider just 4 factors – that’s 16 experiments plus a replicate or two for a total of 17 or 18 experiments. That may not seem like a lot but if your base cost per experiment is in the 3 – 4 thousand dollar range (this is the low end for most of the work I’ve done) the choice of running 5 or 6 vs. 17 or18 is a fairly easy one to make.

    Given that the base effort was a main effect design it has been my experience that, if questions of the C vs. AB type arise, it is fairly easy to set up and run a small group of augmenting experiments, run those, and rerun the analysis on the total group of original and added experiments.

    To your example – if the Res II had been run with a single replicate (total of 5 design points) you would only need to augment the design with a single new design point to remove the alias between C and AB.

    0
    #190326

    MBBinWI
    Participant

    Robert: As always, good advice. In many instances, the knowledge gained and transfer function developed, can be used and reused many times. Thus the original investment gets amortized over many projects, not merely one, and is a very economical investment. Unfortunately, it is only the enlightened company and management that takes this view, and there are all too few of them.

    0
    #190328

    Doa
    Participant

    Many thanks to you both….very helpful insight. Of all the facets I’ve been exposed to DOE is by far the most interesting and defineately the most powerful, it helps that I’m actually enjoying it.

    We are dealing with the cost benefit aspects of DOE right now so your comments are very timely. The professor is devoting much time to the time/complexity/$$ constraints.

    Pretty potent or powerful tool.

    Thank you again

    Marty

    0
Viewing 7 posts - 1 through 7 (of 7 total)

You must be logged in to reply to this topic.