iSixSigma

DOE Non-Normal Distribution

Six Sigma – iSixSigma Forums Old Forums General DOE Non-Normal Distribution

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #32217

    Chris Ely
    Participant

    Hi,
    I am new to DOE and have a question about running a DOE without a process producing a normal distribution. I have used Full Factorials at the beginning of a tool/die development project to reduce variation by comparing UCLsubF to F. This seems to work well. However, a coworker pointed out that without normal distribution I am getting skewed results. Can I trust and use my method with data that does not have normal distribution? Thanks for your help!!

    0
    #85750

    Jamal
    Participant

    Chris,
    Most parametric statistical tests are robust to normality.  Even if you don’t have normally distributed data, the test might work.  My recommendation is to check for the normality assumption using residual analysis.  If the test shows that the data are non-normal, some data transformation is needed.  What kind of data do you have?
    Jamal  

    0
    #85751

    Chris Ely
    Participant

    Jamal,
    Thank you for the quick response. We use CMM analysis of points on formed metal parts using certified gages. I have found that my run parameters or settings which might include tool speed or pneumatic and hydraulic psi change very little throughout development of the tool/die if I start development by comparing F to UCLsubF. In fact, final testing and approval of the tools is usually done at the same settings determined in the original DOE. I would like to run a capability study (Pp and Ppk) at the beginning of a project but if I can avoid doing so I need to. Our timelines are very tight and I usually eat up a whole day running a 30 piece capabilty study and then waiting for CMM results. My final acceptance of tooling is based on Pp and PpK measurements of 1.67 or greater. I have never done a residual analysis. What is the best method?

    0
    #85755

    Robert Butler
    Participant

      Neither your X’s nor your Y’s need to be normal.  Issues surrounding normality only arise when attempting to assess the significance of the X’s relative to the Y’s and then these issues only apply to the distribution of the residuals.  The first chapter of Applied Regression Analysis by Draper and Smith (pp.17 of the 1st edition) has a very good summary of this point.

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

The forum ‘General’ is closed to new topics and replies.