topic Re: Lack of fit test in Discussions

Lack of fit test

juvelva — Wed, 19 Oct 2016 03:10:44 GMT

Hi all JMPers,

Does anyone know what saturated model is used in the LACK OF FIT test? More specifically, when running Fit Model, say by Standard Least Squares, I get the table of Lack Of Fit as following:

My understanding is that:

1) p-value gives me that the model can be improved by adding interaction terms,

2) Max RSq is obtained by the saturated model.

Is my understanding correct? If so, is there a way I can get the saturated model in JMP?

Thank you all!

Re: Lack of fit test

jiancao — Wed, 19 Oct 2016 03:11:46 GMT

Yes, your understating is correct.

If so, is there a way I can get the saturated model in JMP?

Yes, you can run such model with some data prep.

(Caution-a saturated model is over-parameterized to the point that it is essentially interpolating the data. It is not a sound modeling practice.)

(1) The screenshot shows a linear regression fit to predict HEIGHT with SEX and WEIGHT as predictors. Although this main effect model doesn't appears to be under-fit, Max R Sq indicates a saturated model would achieve R Sq at 0.8872.

(2) To do this Combine predictor variables to form a grouping variable, SEX_WEIGHT, so that I can assigns a parameter to each unique combination of the predictors. Use the Combine Columns to get it

(3) Refit the model using SEX_WEIGHT as the predictor . As shown, R Square is indeed 0.8872. There are 32 parameter estimates (plus intercept) in the model.

Re: Lack of fit test

juvelva — Thu, 23 Jun 2016 22:19:54 GMT

Thanks a million!

This is a great answer, although I didn't find the Combine Column, I am using JMP Pro 11.0, but I got the idea.

Re: Lack of fit test

jiancao — Fri, 24 Jun 2016 12:32:35 GMT

Yes, Combine Columns along with many other column utilities were added in JMP 12.