How to set different priors for Sensitivity analysis in Bayesian RM ANOVA
Hi,
Thank you for your hard work! I've observed a trend where journal reviewers are increasingly leaning towards Bayesian-oriented approaches. When I incorporate a Bayes Factor analysis alongside my frequentist analyses, they often request a sensitivity analysis.
In JASP, implementing sensitivity analysis for pairwise comparisons (e.g., Bayesian t-test) is straightforward. However, I've encountered difficulty in applying it to RM ANOVA models, specifically when testing the effect of a specific factor (without interactions or posthoc comparisons). How should I set different priors for this purpose? What criteria should be employed to determine a wide prior, ultrawide prior, and so forth?
Your insights on this matter would be greatly appreciated!
Thank you so much!
P.S. On a side note, whenever I've raised this question before, I've been informed that the computation time is high, and implementing this analysis isn't deemed feasible. In my experience, processes such as MVPA in fMRI and MEG can take days, even weeks. Therefore, having JASP run an analysis for several hours shouldn't be a significant concern.
Comments
Dear JLeborgne,
Adding a robustness check for these more complex tests is definitely a good idea -- we just haven't gotten round to it yet. In the mean time, what I suggest is pragmatic: just halving and doubling the default value.
Cheers,
E.J.