iSixSigma

Tests for Special Causes

Six Sigma – iSixSigma Forums Old Forums General Tests for Special Causes

Viewing 13 posts - 1 through 13 (of 13 total)
  • Author
    Posts
  • #43695

    6S BB
    Participant

    When developing control charts is it necessary to perform all the tests for special causes or is there a recommended standard?
     

    0
    #138960

    Brit
    Participant

    What kind of chart are you using?

    0
    #138961

    6S BB
    Participant

    It’s a general question relating to the tests on an I&MR or Xbar-R / Xbar-S chart.

    0
    #138963

    Andejrad Ich
    Participant

    If you are truly looking to detect variation likely due to the presence of some special cause in your process, then all the tests already apply;  they are all derived to detect sample anomalies that are mathematically unlikely to occur in a truly “in-control”/”statistically constant” process.  If you pick and choose to apply some and not others, then you are actually deciding that you aren’t all that interested in sensitivity to the presence of special causes.  There is a tendency for people to think it’s all about the single point outside UCL or LCL (i.e., that it’s the most important, real rule and the other rules are…….well……”the other lesser rules”).  In reality the single point outside UCL or LCL is just one of the many equally important tools to detect the presence of special cause variation.  So it isn’t a question of whether you have to use all the rules;  it’s the question, “Are you interested in detecting the presence of special cause variation or aren’t you?”
    Andejrad Ich

    0
    #139295

    Geri
    Participant

    Special cause variation is usually defined as a data point outside the 3 sigma limit on a control chart.  If you are monitoring your process and you find variation that is outside 3 sigma, you must investigate why that is occuring and then develop a checklist to address the issue.  Hopefully, if variation (special cause) happens again your operators will know what to do to adjust the process to get it back under control with the help of the checklist.

    0
    #139300

    Andy P
    Participant

    Geri
    Sorry, I don’t agree with your response, there are more definitions for “out of control” points as stated previously.
    Also, the purpose of identifying out of control points is not to allow the operator to continually adjust the process, but to understand what caused the process to be out of control & then put in place a permanent fix to prevent it recurring.

    0
    #139310

    Geri
    Participant

    Andy,
    I used the word “usually”, the classic method is beyond 3 sigma that would indicate special cause variation is present in your process.   I realize you could also add trending examples.  The checklist is not designed to continually “tweek” the process it is designed to understand what to look for if special cause variation is present.  The process will always display variation.  If special cause variation is present corrective actions must be taken to get the process back in control. 
     

    0
    #139315

    cpiesam
    Participant

    The questions that is being asked is tests for special cause from a Statistical Process Control perspective are based off the Western Electric Rules,
    http://www.sixsigmaspc.com/spc/x-bar_and_range_charts.html
    -Sam

    0
    #139319

    Jonathon Andell
    Participant

    If you are looking at “historical” data, go ahead and invoke all the tests to see what you can learn about the process and its variation.
     
    If it’s “real-time” for SPC, select only the two or three that seem to occur most often. The reason is that each additional test increases your likelihood of a “false alarm.”

    0
    #139349

    Andejrad Ich
    Participant

    See….this is what I mean;  people think the single point beyond 3 sigma control limits has some special status among tests for special cause variation (even “classic” status).  ALL the western electric tests are equally, mathematically indicative of presence of special cause variation.  The single point beyond 3 sigma test gets the most attention because it’s the easiest for novices to recognize.  It’s not a question of “adding trending examples.”  Trending is JUST as indicative of the presence of special cause variation as is a single point outside 3 sigma limits.  You either watch the process for special cause variation by using all the tests or you don’t — that’s it — that’s how it works.
    Andejrad Ich

    0
    #139351

    Andejrad Ich
    Participant

    If you are controlling a process and are truly interested in being tipped off to the presence of assignable cause variation, then you use all the tests because they are all mathematically equivalent.  To not use all the tests is a like asking a detective to solve a murder, “but use only these 3 clues and ignore those other 5 over there.”
    Andejrad Ich

    0
    #139361

    Elbrin
    Participant

    I hate to give the default answer, but “it depends”.  How variable is your current process?  How senssitive do you want to be to “POSSIBLE” assignable cause?  How quick do you need to respond to “POSSIBLE” assignable cause?
    I would say if you are just beginnig to look at this process, start with rule #1 and if you have points that break this rule go after them first.  Then rule #4, #2 and #3 last.

    0
    #139364

    Jonathon Andell
    Participant

    I respectfully disagree. If we are exploring a process whose critical inputs are not yet known, then by all means invoke all the rules, and gather all the clues.
     
    However, by the time a process is placed under real time SPC, we should have obtained and “solved” many clues about the process. Invoking every one of the special cause rules does in fact mathematically increase the likelihood of “false alarms.” Perhaps they are mathematically equivalent, but they still add up to an unacceptable level of beta risk.
     
    Presumably we want to shift the organization away from over-reacting to common cause situations. Creating a scenario that moves us back toward that situation is less than optimal.

    0
Viewing 13 posts - 1 through 13 (of 13 total)

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