iSixSigma

How to Identify Significant Factors Using DOE

Six Sigma – iSixSigma Forums General Forums Methodology How to Identify Significant Factors Using DOE

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #55379

    Yeshwanth
    Participant

    I have about 3000 data points for nine factors affecting the product under consideration. I would like to identify the most significant/ critical factors affecting the product. Each factor has more than 2 levels. How should I proceed to identify the significant factors? Can I use DOE?

    Thanks

    0
    #199751

    Chris Seider
    Participant

    Use graphical and statistical analysis tools to see what factors might have a cause and effect relationship. Then use a DOE to prove cause and effect.

    0
    #199754

    Susan McDermott
    Participant

    You will only be able to use DOE if you actively manipulate the levels of your factors. Since you already have your data points, you may not have the right combinations available to run a DOE analysis. Presuming that your output variable is continuous, GLM may be a good option. Start with graphing options to look for significant relationships. Eliminate factors that are not significant and remember to include at least second-level interactions. If it is feasible, it is a good idea to run a DOE with these filtered variables to confirm cause and effect.

    0
    #199760

    Robert Butler
    Participant

    What you have is commonly called happenstance data. You can’t plug that data into a DOE design matrix and expect to get much of anything of value. To the points made by the other posters – in addition to plotting your response against your 9 factors you will also want to plot your 9 factors against one another and examine the correlations between them.

    This will give you some sense of their possible confounding with one another. Ideally, you would want to put the 9 factors through a formal co-linearity analysis – eigenvalues and condition indices – but that is not an option that is available on very many statistics packages.

    Even if you do this it still won’t address (and cannot be made to address) the issue of variables-of-interest confounding with important unknown/undocumented process variables and the possibility that the significant correlations you might note between your 9 variables and the outcome are really due to some/all of your nine variables in combination with the these unknown/undocumented process variables.

    0
Viewing 4 posts - 1 through 4 (of 4 total)

You must be logged in to reply to this topic.