THURSDAY, NOVEMBER 23, 2017
Font Size
Topic Why Have Normal Data When Using a Control Chart?

Why Have Normal Data When Using a Control Chart?

Home Forums General Forums Tools & Templates Why Have Normal Data When Using a Control Chart?

This topic contains 7 replies, has 5 voices, and was last updated by  Chris Seider 3 months, 2 weeks ago.

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #670418 Reply

    Tom

    It is my understanding that to use a control chart (IMR chart, for example), the data I use must be normal, as the control chart tool assumes normality.

    1. Why? Can someone please explain why this is? Why does the data need to be normal in order to draw conclusions from a control chart?

    2. If my data is not normal, which seems to be the case most often, should I be using a Run chart instead? (I realize I could also try to transform the data)

    #670456 Reply

    The problem is that your understanding is incorrect. The control chart does not assume normality. Get a copy of Understanding Statistical Process Control 2nd Edition by Wheeler and Chambers through inter-library loan and read sections 4.3 and 4.4 of Chapter 4. The results are too long to attempt to quote here but the key sentence is this “Control charts work well even if the data are not normally distributed. This issue was addressed by Shewhart in his first book, and it should never have been an issue.”

    #672895 Reply

    I think Robert summed it up nicely. The only thing I would add is I am a big fan of Wheeler and would recommend his books to anyone working with statistics in industry. I have several of his books and they have all been helpful.

    #672952 Reply

    Myth two points out what Robert and John have already pointed out. Their reference is also a primary reference for Minitab’s control charts.

    Attachments:
    1. DJW238.pdf
    #673102 Reply

    @petertibbetts
    thanks for this reference…now I have my hands on it again.

    I do, not without controversy, contend because I’ve experienced it–in implementation. IF one is fairly confident in the distribution of data AND the process is homogeneous and one sees the underlying distribution (yes, some probability of being wrong!), one COULD expand the control limits to beyond 3 sigma to keep false indications of out of control to a minimum. Nothing is more frustrating in those rare circumstances when something is in control and out of control signals are given because of a weibull or other distribution–then one could use wider control limits to mimic the typical 0.3% indication of out of control. Yes it may seem that 98% vs 99.7% (odd extremes I realize) wouldn’t seem that much to worry about but when you increase chance of false indications by a good factor–then credibility of the tool must be considered. Yes, I’m considering the case of individual charts where this should appear even more.

    I’d only advocate such an approach after understanding the underlying process and source of variation, potential natural distribution, etc.

    Before arrows and darts come…I’m not disputing the greatness of the Shewhart charts or the logic within. Remember, improving the processes is the goal. :)

    #673156 Reply

    Tom, Am I right in assuming that you ask this because you’re seeing many outliers (variation of special cause), some of them extreme? If so, you should look into the special cause(s) of each of these and address them.

    #675209 Reply

    Chris – I am not sure of who you are any more! Do transofrmations work? Sure. I consider transofrmations a last resort, becasue now, not only do I have to make sure users understand SPC, but now I need to get them to understand transformed data. In most cases, I would rather risk Type 1 errors, than having to explain log-normal transformations.

    Always a pleasure to hear from you (seriously),
    Pete

    #675444 Reply

    Peter…..I don’t believe in transformations! ;)

    I was just saying if a known distribution (I’d never advocate a transformation–or still not seen a good reason to do one), and the tail was long and wide–to minimize investigations of why something is out of control for a maintained process with SPC (some could give 4X or more extra work for no reason), I would advocate expanding the control limit beyond 3 SL. These cases are more nuance and not a regular use of this technique.

    Again, love having that electronic document now–the internet is good overall LOL.

    Miss seeing you also, Mr. @petertibbetts

Viewing 8 posts - 1 through 8 (of 8 total)

Register Now

  • Stop this in-your-face notice
  • Reserve your username
  • Follow people you like, learn from
  • Extend your profile
  • Gain reputation for your contributions
  • No annoying captchas across site
And much more! C'mon, register now.

Reply To: Why Have Normal Data When Using a Control Chart?
Your information:






5S and Lean eBooks
GAGEpack for Quality Assurance
Six Sigma Statistical and Graphical Analysis with SigmaXL
Six Sigma Online Certification: White, Yellow, Green and Black Belt

Lean Six Sigma Project Tracking

Lean and Six Sigma Project Examples
Six Sigma Online Certification: White, Yellow, Green and Black Belt

Find the Perfect Six Sigma Job

Login Form