iSixSigma

Transform data or use non-parametric tests?

Six Sigma – iSixSigma Forums Old Forums General Transform data or use non-parametric tests?

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #34403

    howe
    Participant

    Curious as to what you consider to be the better method. I’ve generally transformed the data, but I’m wondering if there’s any rules to determine which is better.

    0
    #94746

    Andhale
    Participant

    In case you are looking at a completely homogenous process, and not mixture of processes you are more likely to see Normally distributed data..Seldom we want to segment the data and see if each one of them are normally distributed….even after that if we get non normal data you have 2 options
    1) Use non parametric testing and go ahead…but you have limited tools.
    2) Check what causes non normality– look for special causes–check if there are more than 1 way doing the same thing that yeild different results – treat them as 2 processes and check—look for any bias in sampling….if you still cant find a way out…use option 1 or transform the data….( I would use option 1)
     

    0
    #94748

    howe
    Participant

    Thanks for the quick reply Deepak.Assuming a homogeneous process. Using option #1 as you stated limits your toolset. Additionally, do you think that non-parametric tests more difficult for Green Belts and managers to understand? Do you think transformed data may be equally difficult to grasp (i.e. if you told someone the data they’re looking at isn’t the “real” data, but has been “transformed” so we can test it)?
    With option #2, if you look for special causes in a non-normal dataset don’t you think that there’s a higher likelihood you will see special causes where there aren’t any?

    0
    #94751

    jim jay
    Participant

    Mike,
    go to  Handling Non-Normal Data  in search option of the site and you will get your answers.

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

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