iSixSigma

DOE for attribute

Six Sigma – iSixSigma Forums Old Forums General DOE for attribute

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #41779

    che
    Participant

    hello,
    how do i analyze the result of my doe study when my response is attibute data.. pls help.
    thanks!

    0
    #131412

    jediblackbelt
    Participant

    Looks like nobody else is jumping in so I will break the ice and start to take the heat because I also want to know a better way of doing it.
    What I have done is to perform repeats within the run.  So my response would be a percentage of failures.  So I may run one sample run and within that run sample 100 parts and find that 40 are bad. 
    The next run I would show 100 parts and 80 are bad.  Use those percentages as my output and then see what the DOE shows to give me the best result.
    I have seen several articles that float around for this, but have not read any of them and unfortunately there wasn’t anything else that I was taught through the training that works much better.
    Good luck and let the responses flow…
     

    0
    #131418

    CB
    Participant

    You might try investigating nonparametric methods. Do a search of review your stats package help.Be careful using rates as a response, the data may require tranformation to do a proper analysis. Check your residuals etc. Most of the time you’ll be ok.
    Don’t underestimate the value of doing sorts on the response and performing end counts on the factors or graphing the data.

    0
    #131427

    Craig
    Participant

    Did you run your experiment already? This is a question you need to ask in advance of the DOE.
    If you can elaborate on what your response variable is, you might get the answer you need.  Sometimes data that is not continuous can be analyzed using ANOVA / Regression.  I have seen Likert scale data used in this type of situation.  Do your model adequacy checks to validate! A study I performed was on solderball formation where there were a limited number of defects that could be formed.  The response was #solderballs on each unit.  The data ranged from 0 to a max of 100 or so.  I validated residuals versus run order, residuals versus the factors, I did a probabliity plot of the residuals, etc, etc.  All looked OK.
     
     

    0
    #131438

    Ø6 Sigma BB Coordinator
    Participant

    I suggest 2 ways :
    1. Simple one — Use % as response as Jedi Black Belt propose.
    2. More complicate one —-Use Logistic Regression to analyze the data
     

    0
Viewing 5 posts - 1 through 5 (of 5 total)

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