In addition to the restricted covariance structure assumed, RM-ANOVA can’t handle missing values (need to impute or drop cases), observations need to be at the same time points for all subjects, and results are not that informative (a significant treatment x time interaction, great). This paper has a nice discussion.
2 Likes
I agree that OLS repeated measures ANOVA is a mess, and something I haven’t done in over 30 years. I even remember learning split-plot in time, and having an instructor point out that unless some really restrictive assumptions are made (Huynh-Feldt), the analysis of the subplot was grossly over-powered, and the analysis of the whole plot needed a special calculation to get the proper denominator for an F test.
As a result of that, when I hear RMANOVA, I always think of the mixed model approach. Missing values are not an issue, so long as they are not MNAR, you can handle different time points for subjects in a lot of different ways (splines, for instance),
1 Like
I would modify that only slightly. Whenever I hear of serial data in the absence of a nested structure otherwise, I think of serial correlation models, not random effects. Serial data models are particularly called for when the time span of the longitudinal study is interestingly long.
Dear prof,
what to use in place of RM-ANOVA in cases where number of serial observation few with no nested structure
Generalized least squares or Markov models come to mind.
1 Like