Details for statistical consultants

We provide details to statistical consultants in two ways.

R and SAS code that does what Prism does

We have written FAQs that contain a Prism file along with corresponding R and SAS code for one-way repeated measures anova, and for two-way repeated measures ANOVA with repeated measures in both factors.

Math overview

The fitting of the mixed model is the same whether or not you selected the Geisser-Greenhouse correction. The Geisser-Greenhouse correction only changes the degrees of freedom used when calculating P values from F ratios. So all results except the P value will be the same with and without the correction. 

Denominator degrees of freedom

Fitting the mixed model does not by itself determine the denominator degrees of freedom used to test the fixed effects. Prism offers three methods: the Legacy method used by earlier versions, Satterthwaite, and Kenward-Roger. Kenward-Roger is the default. Of the three, only Kenward-Roger changes the reported F ratios, because it also adjusts the estimated covariance matrix of the fixed effects.

The overview at denominator degrees of freedom in the mixed model describes all three, including the error stratum rule that the Legacy method uses and why it is only correct for complete, balanced data. Derivations are on the Kenward-Roger and Satterthwaite pages, along with the published sources and the R and SAS procedures that implement the same methods.