Minitab Lack-of-Fit Test in DOE (Design of Experiments)

[knagus]

Hello,

I created a Central Composite Design in Minitab; only center points are replicated. Residual error is divided into Pure Error and Lack-of-fit in ANOVA table.
How does Minitab calculate lack-of-fit in such cases? Is it calculated using so-called "data subsetting lack-of-fit test", because the design is not replicated, or this procedure is based on "pure error lack-of-fit test"?

Regards,
knagus

 

[Statistical Steven]

Re: Minitab lack-of-fit test in DOE

Looks like the pure error method using the center point repeats to estimate the pure error

 

[knagus]

Re: Minitab lack-of-fit test in DOE

Thank you Steven. So, it seems that the "data subsetting lack-of-fit" is never used in DOE in Minitab, because if there is only one center point in the design, Minitab does not calculate lack-of-fit. This is not said clearly in Help.

If I understand correctly, lack-of-fit represents variation due to removed terms from the model (so also linear term, if it is removed) AND lack of higher order terms in the model (so cubic terms and higher in case of CCD). Am I right? According to Minitab support "lack of fit represents the total effect of all estimable interaction terms omitted from the model".

Thank you in advance.

knagus (not a statistican)

 

[Miner]

See Minitab's Knowledgebase ID 1697 How and when Error is partitioned...

Lack of Fit is what is left over once curvature and pure error have been backed out of the residual error.

 

[knagus]

I had read this before I asked my questions.

In case of CCD error is not partitioned in curvature, but only Pure Error and lack-of-fit. Therefore I asked if lack-of-fit represents variation due to removed terms from the model and lack of higher order terms in the model.

If CCD includes one center point only, Minitab does not calculate lack-of-fit test, while there is a suggestion in Minitab's Help that data subsetting lack-of-fit test is used in such cases.