1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
Dismiss Notice
You must be a registered member in order to post messages and view/download attached files in this forum.
Click here to register.

Johnson Tranformation Across lots for Process Capability

Discussion in 'Capability - Process, Machine, Gage …' started by Nathan Tempco, Jul 16, 2018.

  1. Nathan Tempco

    Nathan Tempco New Member

    Joined:
    Jul 13, 2018
    Messages:
    1
    Likes Received:
    0
    Trophy Points:
    1
    Location:
    Raleigh, NC
    I am conducting a capability study on a process for cutting cables to a certain length. We need 13 different lengths each with the same tolerance (+/-1mm). We are running 3 lots per our QMS with a set minimum Ppk value for each lot (Ppk of 0.74 with sample size of 30 for variable data). After running each lot through the normality test, some came back with p values above 0.05 but many did not. It was then suggested by other members of the team to use a Johnson transformation to normalize this data and then run the capability analysis. This seemed odd to me as each transformation of each data set was different from the other and failed normality test (p value below 0.05) when one equation was applied across lots. My understanding of using a transformation of this type is because the process itself has some odd characteristic that makes it non-normal, such as a bound that would skew it right or left. This approach would imply that any set of data can be normalized and then compared to different data sets to give understanding if a process is capable. This does not seem correct to me. Can anyone explain this further?
     
  2. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    576
    Likes Received:
    492
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    I never transform data because it hides important information. While some processes are inherently non normal, I have found the most common reasons are an unstable process or mixed process streams. Can you attach your data? We may have additional insights after reviewing it.
     
    Bev D likes this.
  3. _Zeno_

    _Zeno_ Member

    Joined:
    Sep 3, 2015
    Messages:
    46
    Likes Received:
    20
    Trophy Points:
    7
    I agree with Miner. When studying for my CQE, a very smart ASQ teacher of advanced statistics cautioned to only transform data when the physics of the process indicated it. The idea of randomly picking out some transformation based on how well you could make it "fit" a normal distribution when finished just doesn't pass the smell test. It would imply that a transformation algorithm could alternate from one type to another (ie: logarithmic, exponential, polynomial, ...) from a single process based on the dataset being analysed.