iSixSigma

Common causes Vs Special causes

Six Sigma – iSixSigma Forums Old Forums General Common causes Vs Special causes

Viewing 12 posts - 1 through 12 (of 12 total)
  • Author
    Posts
  • #44642

    Neupane
    Participant

    Hi
    can any one explian why six sigma is applied only to common cause variations & not to special causes variation.

    0
    #143497

    LBJ
    Participant

    Special cause variation is caused by a known factor, but results in non normal distribution.  So what does that mean.  A good example of special cause variation… Say you are measuring a facilities throughput.  One day the power goes out for 6 hours.  Your output is much lower on this day.  This would be special cause.  You know the reason for the reduced throughput and won’t count the skewed data.  Now if your power goes regularly then you could factor that in.  We don’t focus special cause because we can’t know when it will happen.  Normally it is unpredictable and unrepeatable. 

    0
    #143499

    Brit
    Participant

    It’s not focused on just commn cause variation.  It focuses on all variation and defects.  It is simply more productive from an improvement standpoint to focus on the commn variation in the process.  Improving that provides the more sustainable and often more profitable gain.  If anyone told you that six sigma or SPC ignores special variation, they were mistaken.

    0
    #143522

    Amit Dhar
    Participant

    Hi,
    I agree to previous illustrations. Another perspective to the discussion, you would not like to focus on the common causes as these are the causes which don’t occur again & again. They are exceptions. Efforts to control this variation becomes out of scope as gain is not significant. Not to say that Six sigma completely discards the root causes leading to exceptions. If the reason of variation is known, you may not like to dig it further to reach “No Conclusion mode”.
    On the other hand, common causes gives you higher variation which you are forced to control to acheive a significant shift of the mean value to the target performance.   
     

    0
    #143524

    Amit Dhar
    Participant

    First half talks about the special causes…..not the common causes….
     
    Hope this clarifies.

    0
    #143525

    Anonymous
    Guest

    Amit,

    With all due respect your statement is in error. Common causes are called common because they are common to similar processes. The are not exceptions because they are present all the time!!!
    For example, after studying a furnace and removing the cause of variation you would probably find the same source of variation on other similar furnaces. I certainly would not describe them as exceptions.
    Special causes on the other hand are localized, or special, to a particular operation, such as a furnace operation where an operator has run the wrong program.
    Andy

    0
    #143527

    chhabra
    Participant

    Hi Andy,
    I realised that, zif you check my next post, you would realize I corrected my typo error…
    :)
     
     
     

    0
    #143528

    Hans
    Participant

    The following outlines the origins of the difference between assignable cause and common cause. Note that Shewart does not use the term common cause, but the term “a constant system of a large number of chance causes in which no cause produces a predominating effect”. He goes on by demonstrating in subsequent chapters how to handle these “common causes”, once you have been assured that the maximum level of control has been achieved. (The index of subjects does not list the term “common cause”. By Deming using the simpler term “common cause” a lot of information about the original ideas of Shewart became more chewable, but it also lost some of Shewart’s original thoughts which are now haunting us in our discussions in for example the role of normality, when and how to remove assignable causes, why this is so important etc. Just a reminder from the original literature.
    Shewart (1931). Economic Control of Quality of Manufacturing Product. 50th anniversary edtion 1980 (pp. 145 – 151):
    If there is a causal orderliness in events and phenomena as we postulate, then it follows that, to one with perfect knowledge, everything is predictable and therefore controlled. However, for practical purposes the quality of product is controlled only tothe extent that we know the laws that make prediction possible. (…), it is a significant fact, as we have seen in the previous chapter, that empirical laws do not make possible the prediction erratic fluctuations upon the basis of probability theory. (…). If a cause system is not constant, we shall say that an assignable cause of Type I is present. Assignable cuases of this type in an economic series are such things as trends, cycles, and seasonals; and in a production process, they are such things as differences in machines and in sources of raw material.
    Stated in terms of efects of a cause system, it is necessary that differences inthe qualities of a number of pieces of a product appear to be consistent with the assumption tht they arose from a constant system of chance causes. (…) We start with a sequence of observed values of the fraction defective, and from this we try to determine whether or not the quality as measured by fraction defective is statistically controlled. (…) This kind of experience leads us to postulate that it is not feasible to explain in terms of specific causes those phenomena which are attributable to a very large number of causes such as the throw of a head or a coin, the motion of molecules, the daily fluctuations in in the price of a stock, hereditary influences and so on.
    Therefore maximum control for our purpose will be defined as the condition reached when by a constant system of a large number of chance causes in which no cause produces a predominating effect.

    0
    #143530

    kishore
    Participant

    Previous posting is really informative.
    Can we say that
    – An assignable cause leads to variation and this can be assigned or attributed to a source that is generating this cause. Assignable cause disturbs the Normality.
    – A common/chance cause can not be attributed to a single source and will not disturb normality but increeases the variation.
    Is it Ok

    0
    #143536

    Hans
    Participant

    Kishore,
    The fist thing to realize is that Shewart relates assignable cause and constant system of chance causes. His logic is that if there is a constant system of chance causes, assignable causes are not present. A constant system of chance is defined as “numbers being drawn from a bowl in the ideal-bowl experiment (sic)”.
    Thus, in the words of Deming (1950 see below), the control chart tests “the hypothesis that the quality is varying like numbers drawn from a bowl in the ideal-bowl experiment. If they (i.e. the numbers) do so behave, the manufacturing process (including the sampling and measuring) is said to be in statistical control. The alternative hypothesis H1 is that, occasionally at least the level of quality slips away to one side or the other below or above its average value, and slips far enough to cause concern. To test the hypothesis the control chart is plotted (…). If a point falls beyond the xbar control limits, an assignable cause of variability is indicated, and the hypothesis H null that statistical control exists is rejected, and the alternative hypothesis accepted. (…) On the other hand, if the points all fall within the control limits and if no suspicious runs, trends, or other patterns can be discerned, Ho is accepted and the process is said to exhibit statistical control or stability. (…) The purpose of quality-control is to find assignable causes and (if practicable) to eliminate them one by one until statistical control is exhibited. The distribution of the particular quality-characteristic under consideration is then definite, predictable and stable (NOTE that there is no allusion to normal distribution even thought that may be desirable!).
    Remark 4 (p. 570) in very small print continues: “Control charts direct attention to time-trends of the points and to the extreme value of xbar, R, s or p (…). The analysis of variance on the other hand, directs attention to the general spread of the points through a comparison of the variance of the plotted  points (xbar, R, s, p) with the variance as estimated internally”.
    Deming continues with the following: The challenge with the usage of control limits to test a hypothesis is that we are dealing with “fixed interval prediction” (Remark 1, p. 567. “It is customary to start a control-chart for xbar on the basis of 25 samples, but fewer may be used if there is a paucity of good data. Why specify 25? This is the same question as asking how accurately the control limits need to be fixed (…). Deming goes on in Remark 3, p. 569 that “If more power is needed to detect assignable causes (lack of control) than is provided by the 3-sigma limits, as when it is extremely costly not to detect assignable causes, and when it is not costly to look for trouble occasionally when there is none, limits arbitrarily narrower than 3 sigma may be used, such as 2 or 2.5 limits, or “probability limits” may be assigned if there is any basis for arriving at appropriate values of alpha and beta”.
    To wrap this up: We are dealing with a series of hypothesis tests that are related to each other in a logical way. While control charting started out with a clear logic leading to guidelines to be followed by practitioners the theory behind it has the potential of deteriorating into iron-clad rules that are prescribed by recipes in a cook book. I hope this clarifies your question.
    Deming (1950). Some Theory of Sampling”. Dover Publications, New York, Chapter 16).

    0
    #143796

    Kishore EN
    Participant

    Thank you Hans. Your explanation got really some depth.-Kishore

    0
    #143811

    Orang_Utan
    Participant

    Now I have a practical issue. Power blackout is a commone or special cause?

    0
Viewing 12 posts - 1 through 12 (of 12 total)

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