Feature request a bona fide repeated-measures ANOVA (not just a linear-mixed-effects substitute)

andersony3k

My understanding is that in JASP the "repeated-measures" analysis is really a linear-mixed-effects analysis. Given that the results of a linear-mixed-mixed-effects model can sometimes substantially disagree with those of a repeated-measures ANOVA, it would be good to have JASP offer users the capability to conduct a bona fide repeated-measures ANOVA.

Thanks.

patc3

There is a bona fide RM-ANOVA under the ANOVA menu, select repeated-measures ANOVA (with post-hoc comparisons and effect sizes and the whole shebang):

https://i.imgur.com/TugifUG.gif

https://i.imgur.com/TugifUG.gif

andersony3k

Hi Patc3. This is my mistake. I was mis-remembering slightly. It's the *Bayesian* "repeated-measures ANOVA" that's really a linear mixed-effects model: https://forum.cogsci.nl/discussion/8568/repeated-measures-anova-with-missing-values

patc3

Hmm, I don't know any other way than through a hierarchical model to estimate a RM-ANOVA in the Bayesian framework (that's how I learned it anyhow, through the Kruschke book). Hierarchical models are much more natural (and prevalent) in the Bayesian framework than in the frequentist framework, so I think it's quite normal that the Bayesian ANOVA is a hierarchical model...

Not an expert in Bayesian ANOVA, just saying I don't know any other way than through a multilevel model to do an ANOVA as a Bayesian model.

andersony3k

OK then. How about just renaming it to: "Bayesian ANOVA-Like Repeated Measures"?

patc3

What I meant was I'm pretty sure it actually is equivalent to an ANOVA, not just "like" an ANOVA. That's how an ANOVA is done in the Bayesian framework. If you compare results, pretty sure they'll be basically identical (within Monte Carlo error if there's sampling involved).

andersony3k

That's surprising to hear, since a frequentist ANOVA sometimes gives substantially different results than mixed model, even with no Monte Carlo simulation.

See the examples in this JASP file:

https://falconbgsu-my.sharepoint.com/:u:/g/personal/randers_bgsu_edu/ERf4LSTqX6RLrzQTtdnw_YUB8Fmcx0UMwDfxw4HCIMZzzw?e=HVMta7

EJ

The ANOVA is traditionally based on a partitioning of the variance, which is not what the Bayesian "ANOVA" does (it is a linear mixed model). Note that we have recently changed the methodology to be more in line with the frequentist approach. In particular, see the blogpost here:

https://jasp-stats.org/2022/07/29/bayesian-repeated-measures-anova-an-updated-methodology-implemented-in-jasp/

and the associated paper here:

https://psyarxiv.com/fb8zn/

van den Bergh, D., Wagenmakers, E.-J., & Aust, F. (in press). Bayesian repeated-measures ANOVA: An updated methodology implemented in JASP. Advances in Methods and Practices in Psychological Science.

EJ