iSixSigma

Is There a Tool That Will Work?

Six Sigma – iSixSigma Forums General Forums General Is There a Tool That Will Work?

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #54027

    Randy Eccleston
    Participant

    My organization collects data on a variety of independent variables, but only when there’s a failure. The dependent “variable” is discreet; the independent variables are all attribute data. Is there an analysis tool I could use – without changing our data collection process – that would show me which independent variables might be better predictors of a failure?

    0
    #193019

    Robert Butler
    Participant

    Given the situation as described the short answer is – no, there isn’t. Specifically, what is there to predict? If all you have is data on failure then you don’t have anything for purposes of comparison so there’s nothing to provide a means of assessing variable effect.

    For example – say you have two variables A and B and you have them as either present or absent and you run 4 experiments.

    Expt A B Result1 Result2
    1 -1 -1 F F
    2 1 -1 F S
    3 -1 1 F F
    4 1 1 F S

    No matter how you vary the combinations of A and B result1 is always a failure whereas result2 has a mix of success and failure. It looks like B has no correlation with changes in result2 whereas every time A is absent (-1) there is a failure and every time it is present (1) there is a success. This would suggest A might be a good predictor of failure – every time it is absent there is a failure.

    0
    #193041

    Chris Seider
    Participant

    @reccleston
    It looks like the sampling plan needs rework. Sampling should encompass good and bad results.

    Even basic problem solving techniques need a way to compare good vs. bad. If the organization says there is too much data to gather, remind them that sampling techniques can be periodic, random, or even concentrated for a short period of time.

    0
    #193046

    Randy Eccleston
    Participant

    Thanks for the responses that confirm what I suspected. I really appreciate the advice regarding different sampling techniques.

    0
Viewing 4 posts - 1 through 4 (of 4 total)

You must be logged in to reply to this topic.