iSixSigma

DOE P values

Six Sigma – iSixSigma Forums Old Forums General DOE P values

Viewing 11 posts - 1 through 11 (of 11 total)
  • Author
    Posts
  • #30789

    billybob
    Participant

    Hello folks,
    Whats up when all the interaction “P” values on my DOE are all over .1?
     
    Thanks,
    Billybob

    0
    #80717

    Mikel
    Member

    It means they are not important. Drop them from your model.

    0
    #80718

    Erik L
    Participant

    Woa Trigger,
    I think you first need to understand what the acceptable risk is for this DOE, and even more important what level the DOE is being run against.  If a DOE is in the screening, or even into the characterization stage, you could be perfectly fine with keeping a value that came back with a higher P value.  We’d leave the next iterations of experimentation to knock out the less significant factors.
    Regards,
    Erik    

    0
    #80719

    Erik L
    Participant

    Woa Trigger,
    I think you first need to understand what the acceptable risk is for this DOE, and even more important what level the DOE is being run against.  If a DOE is in the screening, or even into the characterization stage, you could be perfectly fine with keeping a value that came back with a higher P value.  We’d leave the next iterations of experimentation to knock out the less significant factors.
    Regards,
    Erik    

    0
    #80724

    Mikel
    Member

     
     
     
     
     
    Hi ho silver.
    Assuming he set up the experiement correctly including proper sample size, you eliminate the two ways with p > .1. Where are you getting your advice? Mine is from BHH.
     
     
     
     

    0
    #80727

    CT
    Participant

    that is correct. drop them and perform another regression. you will see an improvement in R sq and R sq (adj).
    CT.

    0
    #80732

    rams
    Participant

    Just don’t forget to use good engineering judgement, prior knowledge and common sense when assessing the importance of each effect.
    rams

    0
    #80737

    zhou
    Participant

    Erik,
    i agree with you 100%

    0
    #80745

    Mikel
    Member

    I agree – use good engineering judgement, prior knowledge and common sense when assessing the importance of each effect.
    If your data teels you something that does not match your preconcieved notions, ignore the data.

    0
    #81296

    Kim Niles
    Participant

    Dear Billybob:
    I like your posts. Regarding this one, globally and generally it means that you can’t say the interactions are significant with more than 90% confidence. This may mean that:

    The test ranges you used were not large enough to show greater differences that might exist relative to error.
    The interaction really isn’t important.
    Your alpha limit is set too high.  When making a binary decision (i.e. turning a knob left vs. right), anything over 0.5 is good information assuming no other information exists.  If your p value = 0.2 then you have 80% confidence that your interaction is significant and the remaining 20% confidence could go either way (assuming an equal distribution of error).  Therefore, 80% +10% = 90% confidence that you should move in one of the two directions as directed by the DOE.   
    Other things that you didn’t control changed during the experiment and affected your results. This would be reflected in the amount of error in the experiment.
    The data was not very normally distributed to the point that it adversely affected your results. You can check for normality as well as the R^2 value shows you how well the data fit the mathematical model used.  You can also increase your sample size per run to take advantage of the Central Limit Theorem.  
    The matrix you used didn’t have enough degrees of freedom to isolate significance relative to error / residual variation. You would have seen more significance if you added center points, replicated or augmented the experiment, or picked a matrix with more power.
    I hope that helps.
    Sincerely,
    KN – https://www.isixsigma.com/library/bio/kniles.asp

    0
    #81322

    Anonymous
    Participant

    Kim,
    I always find it useful to assume that the DOE was not planned correctly, therefore you can second guess the data itstead of using it.

    0
Viewing 11 posts - 1 through 11 (of 11 total)

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