Epsilon and Mauchly interaction in rm_anova2 differ from JASP

[raphaelvallat]

[raphaelvallat]

• pg.epsilon with two-way design gives slightly different but consistent GG and HF values
• pg.sphericity do not work with two-way design: values are completely inconsistent.

[raphaelvallat]

Fixed for two-way interaction where one factor has only two values in e734df5

[raphaelvallat]

In two-way repeated measures where each within factors have more than two levels, I think the regression approach is the only way. Need to create the design matrix using Patsy.