iSixSigma

Box Cox

This topic contains 4 replies, has 3 voices, and was last updated by  lin 12 years, 11 months ago.

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #44259

    lin
    Participant

    Is there a range that lamda can vary over when applying a Box-Cox tranformation? 

    0
    #141523

    Hans
    Participant

    Minitab uses the following lambda’s for its transformations:
    lambda = 2 => squared transformation
    lambda = .5 => square root transformation
    lambda = 0 => log transformation
    lambda = -.5 => 1/square root of y transformation
    lambda = -1.0 => 1/y transformation
    lambda = 1.0 => no transformation needed
    To my knowledge this is the range of transformations that the Box Cox procedure uses. However, I depend on text books rather than the original literature, so someone who has read the original texts may jump in and correct me.
     

    0
    #141531

    Quainoo
    Member

    Hello,
    I have tried an experiment with box cox transformation.
    I used an exponential set of numbers.
    Minitab ‘suggested’ to use 0,33 as a value for Lambda in order to make the data look normal.
    Then I tried to find the mathematical meaning of this Lambda value without success.
    Is there a way to find the transformation factor ?
    I guess you always need to know this factor to keep on going with the analysis.
    Vincent
     

    0
    #141532

    lin
    Participant

    The Box-Cox power transformation is given by (L = lambda):
    x(L)=(x^L-1)/L for L0
    x(L)=ln(x) for L = 0
    Then you find the value of L by maximizing the logarithm of the likelihood function (which I won’t give here).
    I just wondered if there were any bounds on lambda.  I don’t think so.
    Thanks,
    Bill

    0
    #141533

    lin
    Participant

    Here is a link that explains the transformation. 
    http://www.itl.nist.gov/div898/handbook/pmc/section5/pmc52.htm
     

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

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