iSixSigma

SPC on data that is not normal

Six Sigma – iSixSigma Forums Old Forums General SPC on data that is not normal

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #44488

    Json125266
    Participant

    We have a bunch of data that is slow evolving that we would like to apply SPC on. The problem is that all the measurments that we do essentially the techs are aiming for ranges and not for absolute values. I dont know if this was correct to do or not, but I tested the data to see if it was normal, using Anderson-Darling analysis. Some measurments came out normal and some did not
    My question is is it possible to apply SPC charts on measurments that are tring to hit a range rather than a number. Also, if not, then what can I do, besides run charts to try and track these points.Thanks in Advance
    Jason

    0
    #142632

    Eric Maass
    Participant

    Json,
    One possible approach is to transform to normality – using the Box-Cox Transformation, for example.  Minitab has it set up as an option within the Control Charting set-ups.
     

    0
    #142679

    Ang
    Participant

    Control charts do not depend on normality.
    Read Shewhart and Wheeler “Advanced Topics in SPC” Ch 5.
    Many theoretical text will claim that normality is required however for practical purposes, control charts work for any type of distribution.

    0
    #142682

    Chris Seider
    Participant

    We must be careful on blanket statements that control charts will work on non-normal data.  Newer folks to use of SPC need to understand the caveats….
    If the premise is to capture that 0.27% of normality curve to begin assuming special causes are in effect, then the same must be done for non-normal data (especially for individual charts) and this will actually mean the control limits won’t be equidistant from the mean.  Also, since the non-normal data (depending on the distribution) won’t be equally balanced around the mean I would hesitate to implement the special cause test of 9 points consecutively above or below the mean.
    Of course, this won’t be the first time use of SPC on non-normal distributions on this forum.

    0
    #142684

    Anonymous
    Guest

    C. Sneider,
    This is an important point – individual charts do not utilize the ‘central limit theorem.’
    On the other hand, sometimes an ‘in control’ process (n = 5) will have a non-normal distribution for individuals. It is easy to check using a spreadsheet and Minitab.
    Regards,
    Andy

    0
    #142685

    Chris Seider
    Participant

    Andy,
    Spell my name correctly the next time you agree with me. :)  Now, you shouldn’t be posting on a 3-day weekend…the only reason I am is because I’m still overseas…Enjoy Labor Day!
    Yes, I agree about being in control.  Especially those leptokurtic distributions…..!
    Cheers
     

    0
    #142688

    Anonymous
    Guest

    C. Seider,
    Apologies for my carelessness … hope you have a good trip.
    Cheers,
    Andy

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

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