TUESDAY, OCTOBER 21, 2014
Font Size
Topic Control chart outliers

Control chart outliers

Home Forums General Forums Methodology Control chart outliers

This topic contains 14 replies, has 9 voices, and was last updated by Profile photo of SixSigmaGuy SixSigmaGuy 1 year, 2 months ago.

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
    Posts
  • #167555 Reply

    Is there a thumb rule which talks about how many data points(outliers) can be removed to establish the control limits using a control chart?

    #167556 Reply

    zero

    #167557 Reply

    Can you elaborate your answer a bit? What do you mean by zero?

    #167558 Reply

    by zero I mean ‘0’

    #167562 Reply

    Remove outliers only when root cause is known and has been eliminated.

    #167563 Reply

    I think one can remove 20% of the Total Data Points, when Control Limits are arrived for the first time.
    Please correct me, if I am wrong.

    #167574 Reply

    The rule is very simple:
    – If you know the root cause and understand that it was a one time occurance and would not be repeated; you can remove the outlier
    – If you know the root cause and can’t decide if it is a special cause or not; mostly it is your process which is behaving as such; you should not remove the outliers.

    Theoratically; you can remove a maximum of 20% data points as outlier; however that should not be taken as a rule and convienently remove outliers. The above should be exception not rule.

    ALso another one thing: NEVER REMOVE OUTLIER CONDUCT A STATISTCAL TEST AND THEN REMOVE OUTLIERS AGAIN. The reason is tendency to stop when you want to see what you like.

    #167575 Reply

    The only rule I have been taught is

    “Thou shalt not tamper with data” .

    The data belongs to the process , not to you :S

    #167577 Reply

    Great quote! Do you know its origin?

    #167578 Reply

    I should clarify my question. I am not asking about the tamper quote. I am inquiring about “The data belong to the process, not to you.” Is that of your origin, or did you hear it from somebody else? I’d like to use it, and I’d like to give credit where due.

    Thanks!

    #167579 Reply

    Thanks . The quote is very much my own. :cheer: :cheer:

    #193717 Reply

    Alan

    @gsb
    I’m very glad I found this page. Do you know of an academic source that I can reference that backs up what you said about when to include outliers?

    I agree with your statements, I just need to persuade some colleagues.

    Here’s a quote from your comment:
    “The rule is very simple:
    – If you know the root cause and understand that it was a one time occurance and would not be repeated; you can remove the outlier
    – If you know the root cause and can¬ít decide if it is a special cause or not; mostly it is your process which is behaving as such; you should not remove the outliers.”

    #193839 Reply

    Quality Control and Industrial Statistics 4th edition – Duncan – Chapter 31, section 1.4, Outliers, page 706.
    “Occasionally in a set of individual values or a set of means there is one value that appears to differ considerably from the others. The tendency among those not trained in statistics is to discard these exceptional values as not belonging to the set. The statistician prefers to follow a policy in such cases for which he can calculate the probability of wrong decisions. Of course, if a value is KNOWN to have some characteristics that differentiates it sharply from the rest other than being simply an extreme value of the set, then this value can be discarded on the grounds that its inclusion will destroy the homogeneity of the set. Such decisions must be based on sound knowledge and must be exercised with great caution.”

    #194273 Reply
    Profile photo of Brian Leonard
    Brian L
    Reputation - 15
    Rank - Aluminum

    Great discussion. It is so frustrating to hear people automatically claim a data point is an outlier simply because they don’t like the data point. I try to never throw out any data, but I do know may use the old outlier formula if they use box and whisker plots (Mean +/- 1.5*IQR)with anything falling outside of that range considered an outlier. But even with the formula, I tend to include all data, period. But if anyone were to make a claim a point is an outlier make them use the statistical formula to prove it. Better yet, convince them to do as was suggested earlier…keep it all because it was an observation of the process.

    #201305 Reply

    Great discussion; I have to agree that it wouldn’t make sense to pull a data point out of your set because you think it’s an outlier. It’s mainly outliers that cause processes to be out of control.

Viewing 15 posts - 1 through 15 (of 15 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: Control chart outliers
Your information:




Login Form