iSixSigma

DOE Questions

Six Sigma – iSixSigma Forums Old Forums General DOE Questions

This topic contains 5 replies, has 5 voices, and was last updated by  fake accrington alert 12 years ago.

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #48444

    annon
    Participant

    Is is fair to say that using the ANOVA method of DOE assessment is unnecessary?  All I need are the effects and the SE and I can make the determination of statistical significance, right?
    If I want to avoid wasting runs on replicates, can I use any column of signs that proves inert as a reference set for my error calculation?  I am aware of the scarcity of effects in higher order interactions, but technically I should be able to use any inert factor or interaction as a reference to noise, yes or no?
    Thanks!!

    0
    #163305

    Mikel
    Member

    No

    0
    #163306

    Taylor
    Participant

    I would suspect you want to use the p-value to determine significance which is derived from the ANOVA table
    second part, I dont know. I think it depends on the type of interactions you have, similar to screening. With screening you tend to ignore most of the higher interaction effects. So am going to say Yes.

    0
    #163307

    Erik L
    Participant

    Annon,
    1.)    Yes and no.  If you are speaking to determining which factors are going to be statistically significant; a significant ANOVA simply means that at least one of the parameters is significantly better than the overall average in explaining the response.  In that scenario, I would say yes that the top-level ANOVA is somewhat superfluous.  The effects and SE of effects would really be the key pieces of data.  
    However, there is additional value from the ANOVA analysis and this will come from partitioning the MSerror into the two variance components of Pure Error and Lack of Fit.  In that case, the ANOVA provides us a glimpse of the response surface and factors that may be significant that the straight report of effects and associated SEs does not provide.
    2.)     Yes.  As you reduce a model to the statistically significant few, we would use those associated dof from the terms, which are not significant, as a pseudo approximation modeling random noise than as a statistically significant effect.  Another technique would be to use 3-4 center points, if this is a viable option.
     
    Regards,
    Erik

    0
    #163347

    annon
    Participant

    Thanks for the response EL.  As to this statement;
    In that case, the ANOVA provides us a glimpse of the response surface and factors that may be significant that the straight report of effects and associated SEs does not provide
    Can you point me towards a reference of provide a brief example? 
    Thanks for your time.

    0
    #163351

    fake accrington alert
    Participant

    SE stands  for  standard  error

    0
Viewing 6 posts - 1 through 6 (of 6 total)

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