iSixSigma

what is 1.5 signa drift

Six Sigma – iSixSigma Forums Old Forums General what is 1.5 signa drift

This topic contains 47 replies, has 18 voices, and was last updated by  Jim Ace 12 years, 3 months ago.

Viewing 48 posts - 1 through 48 (of 48 total)
  • Author
    Posts
  • #47510

    Ranjan
    Participant

    Hi Experts will appreciate if some one explain with example what is 1.5 sigma drift .
    Regards
    Rahul

    0
    #158423

    Jim Shelor
    Participant

    Rahul,
     
    The 1.5 Sigma Drift is a hotly debated area in Six Sigma.  There are those who fervently disagree with the concept and those who agree with it and the split appears to be about 50% – 50%.
     
    The concept was originally developed by Motorola and Mikel Harry.  It was developed through multiple studies of many processes that revealed the mean of the processes drifted by 1.5 sigma during very long term operation of the system.
     
    For example, the sigma for a process is calculated after a short time collecting data in a manner that minimizes the effect of special cause variation.  Assume the short term sigma is 6.0.  Now we run the process over a long period of time and special cause variation is now in the process.  The special cause variation causes the process mean to drift such that the difference between the current process mean and the original process mean in terms of sigma is 1.5 sigma.
     
    The process that started out operating at a short term process sigma of 6.0 is now operating at a long term process sigma of 4.5.
     
    Motorola and Mikel Harry made a fatal mistake when introducing this concept by calling it a “fudge factor” and by insisting it was empirically derived.  Further examination would have revealed that the 1.5 sigma drift is actually a real effect, not a fudge factor, that is related to special cause variation and the inseparable effect of control charting operations.  This effect applies to virtually all processes both manufacturing and service, but is restricted primarily to processes being controlled using variables flow charts.
     
    Now, stand by for the torrential flood of very charged disagreements with this post.
     
    Best regards,
     
    Jim Shelor
    PMP, CSSBB

    0
    #158426

    Omashi Sabachi
    Participant

    Stan
    Please  explain  your  point  of  view

    0
    #158425

    Mikel
    Member

    Jim,
    You once again show you are light on real experience. Please tell us about YOUR experience with the shift – how big, can it be improved or eliminated, etc?

    0
    #158427

    Jim Shelor
    Participant

    Stan,
    You are correct, I do not have a lot of experience dealing with the 1.5 sigma drift, but then neither do you.
    The answer to you other questions are ~1.0 – ~1.8, improved yes, eliminated no.
    Now, if I am so wrong, prove it.
    Regards,
    Jim Shelor

    0
    #158428

    Mikel
    Member

    You are wrong about my experience.
    I have data on many processes with shifts of less than .5. Many with no use of SPC.

    0
    #158429

    Jim Shelor
    Participant

    Stan,
    So you agree there is a shift, the only contention is how much.
    Congratulations for catching it at 0.5.  Was that by some specific monitoring technique, or luck?
    Jim Shelor

    0
    #158430

    lin
    Participant

    Jim,
    Anybody who has ever done real work with a control chart knows that processes shift over time.  That’s not the point.  The idiocy of the “1.5 Shift” argument is the claim that it is anything other than a generalized fudge factor to use when long term instability is not understood.   Your prior post admitted seeing shifts ranging from 1.0 to 1.8.  So why use 1.5?  The whole concept is noise – a distraction best ignored since it has no bearing on the actions that must be taken to actually improve a process, in my opinion.

    0
    #158434

    pedro
    Participant

    Hi Jim.,,
    I remember doing myself some math and found that 1.5 sigma is only being affected in 6 sigma process, this according with sigma calculators.
    Processes with less than 6 sigma were affected with shifts smallers that 1.5…. in other words the smaller the sigma level the smaller the shift (in terms of 1.5 sigma).. I couldn’t do it more accurate as I did not have a z-table with values up to z = 6.
    I will appreciate if you give me your input on this, is my assumtion correct ?   it would help me if I would get a Z table up to 6 sigma.
    I welcome your comments.
    ps – I am not an expert on this,, just doing my first researchs.. so need orientation.
     
    Pedro

    0
    #158435

    mand
    Member

    The 1.5 “drift”, “shift” or “correction” don’t exist.  This basis for Six Sigma is total nonsense. 
    Read “Sick Sigma, Part Two” in “Quality Digest” for the full history of how this ridiculous number came about.
    Do not use Six Sigma tables or calculations.

    0
    #158438

    RDogg
    Participant

    Enough with the shi(f)t already…I read the Sick Sigma articles and agreed with many points. The 1.5 Shift, the kajillion tools, and belts are not a sufficient substitute for good (process) management. That said, I think any one should be wary of any professional who endorses such an approach as a guaranteed roadmap to increased profitability and culture change and blah blah blah. It’s worth noting that the author’s bio markets his own lame software for quality improvement and names his company….I’m sure there’s no self-interest there….

    0
    #158439

    Jim Shelor
    Participant

    Bill,
     
    I like your post, you are absolutely correct, 1.5 is not a magic number, the real issue is that processes do drift over time and our job is to keep that drift as small as possible.
     
    I only used 1.5 because that is the number the sigma calculators use, the number that is taught in training, and that is even the number used when being tested.
     
    The point I was trying to make, is if a control chart is the only tool being used to monitor the process, the process can drift by 1.0 sigma and there is approximately a 30% probability that the control chart (X-bar, R) will signal that shift happening.  Further, at 1.5 sigma, there is approximately a 60% probability that the control chart will signal the shift.
     
    All of us who use control charts need to be aware of the effect on the ability of detecting this shift when we make decisions about how we are going to operate, as I am sure you are.
     
    For example, some authors recommend using only 4 control chart rules to reduce the number of false alarms.
     

    1 point > 3 sigma
    2 of 3 points > 2 sigma
    4 of 5 points > 1 sigma
    8 points in a row on the same side of the mean (some say 9).
     
    Limiting the rules to those 4, while it lessens the probability of a false alarm, also significantly weakens the probability of detecting a mean drift.
     
    For example, some authors recommend recalculating mean and control limits “periodically”.  One of the worse things you can do to your control chart from the perspective of detecting a mean drift, is to routinely recalculate control limits.
     
    One trick that makes the drift easy to spot is to plot the current actual process mean on the chart along with the target mean.
     
    Calculating the capability of the process routinely is another way to detect the drift early.
     
    My only point is, there is a reason to be concerned about the mean drift, it is real, and control charts alone are ill-equipped to detect this drift.  I am also concerned that publications that “teach” control charting do not pay enough attention to this significant weakness and the additional tools that should be used to guard against it.
     
    It is unfortunate that Motorola and Mikel Harry locked in on 1.5 as such a solid number because you are right; there is nothing magic about 1.5.  As you said, understanding process drift and how to detect it is the important thing.
     
    Sincere regards,
     
    Jim Shelor

    0
    #158441

    Amandeep S Munial
    Participant

    The 1.5 sigma shift should not be used to convert long term sigma level to short term sigma level and vice versa.
    All processes behave differently and have differnt magnitude of control.  All processes do not behave in the manner Motorola’s processes did when this number was given to the world.
    My view is that a proper capability study with rational subgrouping be carried to determine the actual short term and long term sigma levels.  You process could actually be better in control and hence, the shift will be small.  On the other hand, if the controls are not good, the shift can be larger than 1.5.  Hence, using the 1.5 shift should be avoided. 
    Use the sigma level which you calculate from your data and do not try to estimate the short term or long term sigma level from it.
    I am really concerned that all who train on Six Sigma do not pay enough attention to preparing a subgrouping strategy for capability analysis and I have seen most people recommending a subgroup size of 1, irrespective of that fact whether subgrouping is possible or not.  This, to me, is ridiculous.

    0
    #158443

    JM
    Participant

    Hello Experts,
    As I was reading the thread I got educated in an way or so regarding the shift issue.  Per my experience, I cannot prove that there is a 1.5sigma shift in the process, though part of my training was the conclusion made by Motorolla.  Maybe it was their experience back then.  Maybe, things have changed from then.  I agree on some of the posts that this is indeed an argument meant to be unclosed.  But for whatever is worth we all shared the experiences and validated some of the articles and materials imparted to us during trainings.  I just hope the person who post this inquiry did not have a hard time understanding what to pick from your posts.
    Thanks for your inputs.

    0
    #158444

    Omashi Sabachi
    Participant

    Nonsense

    0
    #158447

    The Badger
    Member

    There is another good reason we need a 1.5 shift. If we didn’t have it we’d all be working for ‘four and a half sigma’ organisations, and that just doesn’t sound that cool!
    For what its worth, I think we need a method of ‘tracking’ drift, not simply allowing for it with a fixed value. I saw some nice control charts that applied control limits to a linear trend in the average. If there were something out there that tracked the average drift up and down, that would be really useful?
    The Badger

    0
    #158449

    JM
    Participant

    thanks omashi. how come i cant see you comment or explain your view on the forum? other than asking questions.
    you should share to us your inputs.

    0
    #158451

    Mikel
    Member

    Of course I agree that no process stays exactly centered with no change in variation over time.
    What I object to is your first post where you stated things as factual that you have no real knowledge of. If you want to do a literature review, say that is what you are doing.
    How to control a process to better than  .5 shift over time? Look to the tools of TPS, most of their supply chain does it every day – without a SPC chart to be found anywhere in their facility.

    0
    #158452

    Mikel
    Member

    Jim, you stated –
    “The point I was trying to make, is if a control chart is the only tool being used to monitor the process, the process can drift by 1.0 sigma and there is approximately a 30% probability that the control chart (X-bar, R) will signal that shift happening.  Further, at 1.5 sigma, there is approximately a 60% probability that the control chart will signal the shift.”
    What nonsense. Go do a few simulations of a 1 sigma shift and use some common sense starting rules as found in Juran or the QS / TS documents. It is virtually impossible to start a process with a shift of 1. When do most shifts occur? At start up.
     

    0
    #158457

    Jim Shelor
    Participant

    Stan,
    I used 14 rules for the calculation of probabilities that resulted in those numbers.
    The rules were all the rules found in Juran, Breyfogle, Grant & Leavenworth, NIST, and WECO.
    I did 15 simulations, all of which were consistent with the probabilities.
    I read the first post as a question from a student who is trying to understand the so called 1.5 sigma drift and I tried to answer him in a way he could use the information for his training.
    There is nothing magic about 1.5, but it is clearly possible to get a shift that large if control charts ALONE are used to monitor the process.
    I do not, and have never, recommended the use of control charts ALONE to perform process monitoring because control charts are ill-equipped to detect this condition.  I am simply indicating my concern that many Black Belts do not understand the weakness of control charts for detecting this condition and how to handle it.
    Jim Shelor

    0
    #158459

    Jim Shelor
    Participant

    Stan,
    If you change the conditions of the question, the answer is different.
    Of course the drift over time can be controlled to < 0.5, but that is unlikely if CONTROL CHARTS ARE THE ONLY TOOLS BEING USED to monitor the process.
    There are several other tools and techniques to use that can easily detect this condition and allow us to control it.
    If we go strictly by the rules associated with control charting, that is, do not adjust the process or control chart unless a signal of a possible out of control condition is provided by the control chart, as stated by several authors, this condition becomes difficult to detect.
    One author even said process adjustments and control chart adjustments should be made “if and only if you see a point beyond the plus or minus 3 sigma control limits”.
    Have a nice day,
    Jim Shelor

    0
    #158460

    Mikel
    Member

    Jim,
    You are falling into the trap of a shift has to be detected in a single sample. Your percentages will vary dramatically based on your subgroup size and how quickly you want to know a change has occurred.
    So let’s set some simple rules and see what really happens –
    – What subgroup size do you want?
    – How often do you sample?
    – Do you have rules for starting a process anytime it has been disrupted? (tool change, material lot change, after breaks, shift changes, …)

    0
    #158465

    Jim Shelor
    Participant

    Stan,
    Where are you getting this stuff from?
    I never said, nor implied, that a drift can be detected on one sample.  To think that a drift can be detected in one sample is rediculous!
    Jim Shelor

    0
    #158471

    Mikel
    Member

    You did imply it with your 30% number (depends on sample size).
    What sample size do you want?
    How often are you going to sample?
    How many samples are acceptable before a shift is detected?
    Do you have starting rules?
    Answer the questions and let’s see what the probabilities are.

    0
    #158472

    Chad Taylor
    Participant

    There is a saying among Engineers and Scientist: All models are wrong, but some are useful. The traditional normal model is certainly wrong, but is still often times useful. 1.5 sigma shift adjustments create a model that’s more useful than the traditional model. While all models simplify reality, the traditional model oversimplifies reality. It makes things look much better to us than they look to our customers.
    That said the reality is most likely in the middle. How you report that to your customer is up to you.
    Just my 2 cents worth
    Chad Taylor

    0
    #158473

    Jim Shelor
    Participant

    Stan,
    I calculated those probabilities through use of the formulas for a normal distribution with a process drift reaching 1.0.
    I performed simulations for the model and the process took approximately 560 data points to reach an offset of 1.0 sigma.
    At a 1.0 offset, I calculated the probability of each of the 14 rules providing a signal and then added all the detect probabilities together.  The overall probability of a detect was 30%.
    If you want to pick some rules and prove that number is wrong, be my guest, by using the formulas rather than an example distribution, I was able to eliminate the “but what is the distribution was this” questions.
    Have a good day.
    Jim Shelor

    0
    #158474

    Jim Shelor
    Participant

    Isn’t that the truth!!
    Thanks Chad.

    0
    #158475

    Mikel
    Member

    Jim,
    Let me speak slowly this time.
    You said –
    At a 1.0 offset, I calculated the probability of each of the 14 rules providing a signal and then added all the detect probabilities together.  The overall probability of a detect was 30%.
    That 30% is for the next sample of some subgroup size. What was the subgroup size?
    For a subgroup of 4, for example, the average number of data points to see the “shift” is less than 3! And that’s just using the 4 rules you suggested in a different post.
    The statements that you are making are misleading. A shift of 1 is readily detected. And if you follow sound advice on starting a process, you would almost never start a process with a 1 sigma shift.

    0
    #158478

    Chad Taylor
    Participant
    #158480

    Jim Ace
    Participant

    Here is what my research has uncovered on this subject:
    1.  The 1.5 sigma shift is not a constant.  No where can I find any literature that says the 1.5 shift is a constant.  To the contrary, the shift is simply a logical way to adjust the long-term report of capability for the effect of process shifts and drifts.  Even Dr. Harry says the shift is not a constant.  Moreover, it was Bill Smith that introduced the 1.5 sigma shift at Motorola, not Dr. Harry.
    2.  The 1.5 sigma shift factor has nothing to do with how quickly a process shift can be detected.  Your discussion is based on the use of SPC methods and ASSUMES the process is being monitored using SPC methods and those methods are being used properly.  MOST processes do not employ SPC charts.  This means that process variations are not monitored or analyzed over time.  Therefore it is entirely possible (and perhaps probable) that such a process would experience shifts and drifts.
    3. The exact magnitude of a shift is not at the center of this priniciple.  What is important is to recognize that processes will shift and drift over time, especially those processes that are not controlled through SPC methods.  One should attempt to quantify the magnitude of shift on a case-by-case basis.  However, until that time, using the 1.5 approximation is far better than assuming no shift or drift.
    4. Whether the shift is 1.5, 1.2, 1.8, 1.0 does not matter.  What does matter is that one accounts for shifts and drifts.  Using any shift magnitude that is greater than zero is much better than assuming a perfectly centered process.
    5.  There are several technical papers on the internet that shows real processes and how much they shift and drift.  These papers do not show a shift and drift of EXACTLY 1.5 sigma, but close enough to conclude that the shift and drift is NOT zero.

    0
    #158481

    Jim Shelor
    Participant

    Stan,
    I started out the mathematical proof I am doing for the purpose of demonstrating the 1.5 sigma drift was not possible given even the most rudimentary monitoring.
    When I saw the “low probability of detect” numbers, I got myself turned around and neglected to take the next simple step.  If I have a 30% probability of detect, I can expect to get a detect in the next 3 samples but certainly within the next 6.
    Boy do I feel stupid.  But now at least I can go back and continue my mathematical proof that the 1.5 sigma drift is the crap we all think it is.
    The answer to the first post is, in my opinion, what a Black Belt in training needed to hear because the training community and testing community still teach that concept, although none of us actually use it in the real world.
    Just as a suggestion, if your posts were a little less of an attack and degrading to the person you are speaking to, technical disagreements like this would get resolved faster.
    I do thank you for your help.  You actually helped me go exactly where I wanted to go, but your posts are so disgusting to read and just full of questions instead of information that the message is generally not there.  I expected the proof to tell me the 1.5 shift was crap, but I got so wrapped up in the numbers I forgot to really look at the implications of the numbers (forest for the trees thing I guess).
    Have a nice day,
    Jim Shelor

    0
    #158483

    Jim Ace
    Participant

    Jim Shelor,
    You are right on!  It is absurb to believe that a shift can be detected in one sample.  I believe that’s the main theme behind the 1.5 sigma shift.  For some period of time a shift can exist without being detected and for that period of time it can cause harm.

    0
    #158485

    Mikel
    Member

    Jim,
    You call my posts disgusting, so be it.
    You’re the same guy who had a hissy fit over me challenging your “real world example” that turned out to be trumped up data. Your claiming to have seen shifts of 1 – 1.8 is the same kind of trumped up data. If you’ve got experience, you’ve seen shifts >> 1.8.
    I assume when I see a post from you that it is theory, not experience, talking.

    0
    #158489

    Randy Galyon
    Participant

    Guys, Finally Sigmeister came with the BRILLIANT TRUTH about the 1.5 sigma shift, drift …Check this link: http://www.ascsixsigma.com/Sigmeister.html
    Cheers…

    0
    #158492

    Lebowski
    Participant

    Amandeep,So you are concerned that people who trained on Six Sigma didn’t pay enough attention ……yadda yadda yadda because they won’t get a good capability analysis. Very magnanimous. When was the last time you fixed anything with a Capability Study? You must be one of those people who loves to attend ASQ meetings and craps all over every presentation because they ran a capability study on non normal data. Not because you care or actually have a normality test of your own but mostly because you like to hear yourself talk.Every point you made was already made by someone previously. We have our eyes on you Butchie.Lebowski

    0
    #158493

    Lebowski
    Participant

    Stan,Have you noticed when people duck behind the “real world” tag it is the first sign that they do not have a clue what the real world is.He couldn’t even do the basic research to figure out Harry didn’t introduce the 1.5 sigma shift he just made a career out of selling it and whatever image he was into at the time. Have you caught the latest Tommy Bahama bling bling thing from Harry? How do you spell male menopause? H A R R Y Great job on exposing this guy one more time.Lebowski

    0
    #158494

    Lebowski
    Participant

    Badger,You started out strong with the 4.5 sigma comment. Very funny.WTF is the tracking drift thing got to do with anything. You must have been hanging out with the pygmy potentate that is concerned over subgrouping. Go fix something.OK its bowling night. We are out of here.Lebowski

    0
    #158495

    Drift, shift and other nonsens
    Participant

    does anybody even care about the “sigma shift/drift etc.” these days any more? it’s the 21st century …

    0
    #158496

    Lebowski
    Participant

    Drift S__t,Naw nobody cares. Thats why nobody responded to Rahuls question since July 10. I don’t suppose you noticed the other posts? Very happy you are aware of what century it is. That will be a useful piece of information for several more decades.Lebowski

    0
    #158498

    Jim Ace
    Participant

    Mr./Ms./Mrs Lebowski,
    This post is offered with all due respect to to the self-prescribed mastery of Lebowski.
    I just just put “Lebowski + Six Sigma” into Google and could not find any meaningful professional contributions or papers related to the subject of six sigma.  With such expertise at hand I could hardly believe such a thing.
    Could you please provide additional key words to further refine my search so that your many contributions can be examined.  I would really like to study your fine contributions.  After your last several posts to various individuals I feel certain you are among the highly esteemed that skilfully stalks the six sigma community and, therefore, surely have many documented offerings. 
    At this point I can only conclude one of two things.  Either my search criteria is severley lacking or you’ve never done much except beat up on novices at this site.  Please confirm this by providing me a nasty posting.
    Jim Ace

    0
    #158501

    Drift, shift and other nonsens
    Participant

    what a self-defeating joke to demonstrate such poor communication skills and lecture others on this site on getting projects done. bad manners and poor communication are not a sign of superior intelligence, professionalism or expertise. tout au contraire! have a nice evening.

    0
    #158502

    Married with Children
    Participant

    Jim Ace,
    Lebowski gets these uncontrolled outbursts of emotions and random attacks on an anonymos site as an attention catcher ever so often. There is no substance to him or his ranting. Imagine, the guys does bowling for a hobby … what expert in this field would ever engage or even have the time to engage in Al Bundy’s favorite “sport”, and publicly announce it? He’s one of those married with children kind of guys … he has never been very bright and at this point has become really old news on this site. The best thing you can do is smile and politely move on. What’s the value add of an “intellectual” engagement at his level anyway? Time has passed him by. There are more enlightening things to discuss at this site. Cheers.

    0
    #158512

    The Badger
    Member

    Its no wonder six-sigma gets such a poor review these days, no one can agree how it should be done. In the mean time, those customers keep on walking.
    The Badger

    0
    #158513

    The Badger
    Member

    Ah yes, sub-grouping, I’m off to get a coffe and study the independence between the beans.
    The Badger

    0
    #158515

    focus on the cash
    Participant

    Well said The Badger. I can hear the boardroom doors slamming now

    0
    #158570

    Lebowski
    Participant

    Jim Ace,
    This is how I judge the contribution of people is by how many papers that they have published. I am afterall the star of a cult film.
    How would I know who is a novice and who is not? It is actually indiscriminate beatings.
    Lebowski

    0
    #158571

    Lebowski
    Participant

    Drift S__t,
    Interesting comments concerning communication, superior intelligence, and professionalism. Here is a news flash for you, take a look at your sentence structure and capitalization.
    Lebowski

    0
    #158573

    Jim Ace
    Participant

    Lebowski, thank you for validating our assertions.  Best of luck to you and hope you find peace in all your endeavors.  Jim Ace.

    0
Viewing 48 posts - 1 through 48 (of 48 total)

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