DOE: Significant factor or not


Michel Saad


I am looking at a 2x2 DOE. When I analyze it, I have no terms that are significant. I then analyze again by removing the interaction term to see the R squared and the term A is now significant and I have 99.7% R squared.

How can a term be non-significant and them become significant by analyzing less terms?

Thank you.


Forum Moderator
When you reduce your model by removing the non-significant interaction term, the degrees of freedom and variance attributable to that term are reallocated to the residual error, which is used to test the significance of the remaining model terms. The additional degrees of freedom make this test more powerful meaning it is better able to detect real effects.

You can further reduce your model by removing the B term as well. This would improve the predictive ability of your model.

Michel Saad

Those darn degrees of freedom!

It makes a lot of sense. Thank you.
Top Bottom