iSixSigma

Stratified or Bimodal?

Six Sigma – iSixSigma Forums Old Forums General Stratified or Bimodal?

Viewing 12 posts - 1 through 12 (of 12 total)
  • Author
    Posts
  • #47955

    Mike Archer
    Participant

    Hello.  I am looking over my companies MSA teaching material.  It shows an I chart with several data points between 15 and 16.  There are also several data points between 20 and 22.  But there are no data points between 16 and 20.  The particular slide is calling this condition to have stratified data.  To me, the correct termonology is that the data is bimodal.  Is the slide in error?
    I will be teaching the MSA modual next week to green belts and so I want to know what I’m talking about here.
    Thanks,
    Mike

    0
    #160456

    Mike Archer
    Participant

    Any help would be greatly appreciated… even opinions.
    Thanks,
    Mike

    0
    #160458

    Robert Butler
    Participant

      I’d say within the context of an I chart one would state the data appears to be stratified. One would also state that a histogram of the I chart data is bimodal.

    0
    #160459

    Erik L
    Participant

    Mike,
     
    I guess I would need to know more about the plotted data itself.  Are the levels in the chart the result of changes in a specific factor?  For instance, is it showing the effect of moving between operators, instruments, locations, batches etc. etc.  If it is, then I would most likely lean more toward describing the resulting pattern in the data as a result of stratification factors. 
     
    Or, is the data giving some baseline performance on the process itself, with no known process changes, and what we are seeing is normal patterns in the data and we would normally expect to have data falling over the entire interval?  If that is true, then I would say that it is exhibiting bi-modality.  If we would expect to normally see a unimodal distribution we would still look to identify stratification factors that would explain the signals in the data and to bring it into alignment with the hypothesized shape of the distribution.  Additionally, that graph could also be explained through numerous other MSA concepts (e.g. bias, resolution, rounding measures.)   Don’t know if this helped at all, but some thoughts…
     
    Regards,
    Erik    

    0
    #160461

    GrayR
    Participant

    In the classic process improvement definition, stratified data would be the result of a stratification process.  Stratification means that the data is divided into categories to see if the categories have an impact on the result. One example would be to startify results by male vs. female; different operators; different facilities; different part numbers, etc.  To me, your data seem to be bimodal; and would be considered stratified if the high values have different source categories than the low values.

    0
    #160462

    Ron
    Member

    Nice explanation of stratification here: https://www.isixsigma.com/dictionary/Stratification-176.htm

    0
    #160467

    Craig
    Participant

    This can get confusing….but
    I would go with bi-modal. When you think of data stratification in the measure phase, it is generally in terms of how you will slice the data.
    If your data shifted from the 15-16 range to the 20-22 range, it could have been due to an unexplained mean-shift due to the measurement equipment. You would then have a bimodal distribution. If you plot the data by operator in Minitab or JMP , see if the shift was due to “operator”. If so, one of your stratification factors cause a bimodal distribution.
    I hope this makes sense!
    HACL

    0
    #160475

    BTDT
    Participant

    Mike:Your data shows a bimodal distribution because it come from a stratified process.The bimodal distribution shows the process is not undergoing pure “common cause” variation. It is the result of the superposition of two subprocesses. The idea is that since the data shows this unusual distribution, you should look to find the “special cause”. Use a run chart to help you see if it is time related.The “special cause” could be different machines, shifts, operators, material suppliers, times of day, sales regions, product types, etc., etc. etc.We once saw this for a process where the semiconductors produced were for two different customers with different specifications. The components very close to the target were removed and sold to the fussy customers. The remainder showed a bimodal distribution clustered near the high side and the low side of the specification window.Cheers, BTDT

    0
    #160687

    Wiz of Oz
    Member

    BTDT,
    Excellent response – and totally correct.
    It is important that people start to understand what is meant by the terminology, and your succinct point that “your data shows a bimodal distribution because it comes from a stratified process” is wonderful in its simplicity.

    0
    #160688

    John W
    Participant

    Bimodal distribution: A statistical distribution having two separated peaks.
    The term bimodal distribution, which refers to a distribution having two local maxima (as opposed to two equal most common values) is a slight corruption of the definition of “bimodal,” posessing two modes.
    You might wish to spend some time here: http://mathworld.wolfram.com/letters/

    0
    #160745

    Steven B
    Member

    The data is stratified, indicating that it is showing two or more distinct groups, so the slide is not incorrect.  You are correct is saying that the information is bimodal (a frequency distribution with two distinct peaks).  I don’t think the slide is telling the whole story. 
    For the MSA training, the slide is there most likely to illustrate a bimodal situation.  A good possibility exists that the two peaks represent two processes that are being incorrectly analyzed as a single process, something the MSA should define more clearly.

    0
    #160852

    GrayR
    Participant

    Unfortunately the response was not correct . . .
    Stratification is something that you do with data to understand causes for average and variation.  Stratification does not necessarily show any difference in the data groupings — bimodal data distributios are not the result of a stratified process.  More correctly, a bimodal distribution could represent a process having multiple causes for the results.  By stratifying the data into groups, you may, or may not, be able to find out the cause for the non-normality.

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

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