Reporting Bayesian ANOVA comparing 3 groups
Hello all
I'm new in the Bayesian and JASP world and currently navigating through reporting my analysis using JASP, but the more I delve into it, the more confusion seems to crop up. It's as if some crucial values needed for reporting aren't readily available within JASP (for what I want to do).
My study involves comparing nutrient content in food items classified into three distinct groups. After running the analysis, I obtain a model comparison yielding a Bayes Factor of BF10=9.421xE31.
To lok deeper into the differences between each group (1 vs 2, 1 vs 3, 2 vs 3), I performed model-averaged posterior summaries, distributions, and post-hoc tests. However, when it comes to reporting, I encounter a hurdle. Some literature, like Keysers et al. 2020, recommends reporting 'median posterior δ and 95% CI', but I've hit a dead end in finding this information within the software. I've attempted various methods, including using R packages like brms, yet the challenge persists, likely due to the presence of three levels rather than two in my analysis.
So, what am I overlooking? Is there a misstep in my approach? My goal is to report both the Bayes Factor from the ANOVA and the posthoc tests, specifically for each comparison between the groups.
Comments
So you'd like the posterior median and 95% CI for the differences between the three groups, right?
If you click "Estimates" you'll find the posterior means and 95% credible intervals -- is the problem that you would like to report the median instead of the mean? Usually, in these models, the difference is minute; on our GitHub page you can request that we add the median (for details see https://jasp-stats.org/2018/03/29/request-feature-report-bug-jasp/)
EJ
Hello EJ
Happy new year!
Yes, It seems that it is recommended to report it but I might be wrong, this is just what the paper I mentionned (https://doi.org/10.1038/s41593-020-0660-4) recommended. I Just want to make sure I follow a good protocol so it is easy to readers to interpret my results.
I see so, you think reporting the posterior median instead of the mean is not that necessary?
Thanks for the link I check it out !
No I don't think it will matter much (I do personally prefer the median though :-))
EJ