iSixSigma

Trying to convert attribute data in to variable

Six Sigma – iSixSigma Forums Old Forums General Trying to convert attribute data in to variable

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #49480

    mcleod
    Member

    I have an associate who is trying to convert his attribute data in variable.  He is making proportions from his daily runs. (Defects/opps)  Then he is stacking them into capability charts as variable data.  My skills are rusty, but I thought that was a no, no.  He is also taking different process off the same line and combining all those proportions on one capability chart to make it normal. I am not sure that you can do that either.  Any advice would be helpful.
    Thanks.

    0
    #169560

    Jen
    Participant

    Scott,
    The process of how your collegue is making variable data is not wrong.  He can even look at the capability as long as he realizes that he needs to look at capability by Operation and decide what percentage (since that is what he is analyzing) the upper and lower spec limits.  If the data is not normal then that needs to be accounted for either through transformation or through the more advanced Minitab programs that deal with non-normal data.  Just make sure he picks the appropriate non normal distribution.  What is wrong is the combining of data.  It would need to be broken out by operation, line, etc but to lump all operations as one capability that will not really tell him anything and he will end up working on the wrong thing or possibly not working on anything at all.

    0
    #169563

    Sloan
    Participant

    Whoa, hold on a minute! Changing discrete data into a proportion does not make it continuous data! There are tools to use to measure the performance of a process using discrete data (c-chart, p-chart, u-chart, np-chart) and there are statistical tools to determine significant differences between proportions (1-proportion test, 2-proportion test, chi-square, etc). But you should not just mangle the data and try to transform it to “normality” in order to use tools meant for truely continuous data in a normal distribution, that’s just wrong! Your results will be meaningless at best and misleading at worst.
    I do agree with Jen that you should not combine all of the operations together just to make the data look better though. You need to look at each operation on its own.
    Analyze independent processes seperately, use the tools for discrete data that you have available (there are plenty of them) and stop trying to insist that everything has to fit into a nice neat “normal” distribution.
    Outlier

    0
    #169576

    Swaggerty
    Participant

    Just wondering…..in what scenarios can a Box-Cox transformation be used????

    0
    #169577

    Swaggerty
    Participant

    sorry….forgot it was to convert non-normal to normal data. Slip of mind…guess i ought to start havin a more nutritious brkfast

    0
    #169590

    BTDT
    Participant

    Scott:You can not convert discrete data to continuous – this is a no, no. Combining different processes is also wrong.Your associate may be able to get away with it if his subgroups are large and all about the same size. When he calculates a proportion, he will have only one mean without a standard deviation and will only be able to chart with an Xbar chart.If he really wants a control chart, he can use one of the ones designed for discrete data; n, p, c, or u.
    He should also have a separate chart for each process. Whether he has distinct subgroups can be tested independently.https://www.isixsigma.com/forum/showmessage.asp?messageID=137128Cheers, Alastair

    0
Viewing 6 posts - 1 through 6 (of 6 total)

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