Following up an interaction in Bayesian RM ANOVA

Comments

• PM_Stat

  April 2020

  Hi !

  
  

  I have the same problem... And I asked this question last week without answer.

  I asked in this forum : https://stats.stackexchange.com/questions/459608/bayesian-anova-simple-main-effect-and-post-hoc-analysis

  An answer suggested bayesian t-test to analyze simple main effect...
• JohnnyB

  April 2020

  Hi rohanp16,

  The simple main effects analysis is essentially running ANOVA's for different subset of the data (conditioning on each level of the "moderator variable"). In the frequentist ANOVA this is automated for you, but unfortunately this feature is still missing in the Bayesian ANOVA.

  What you can do in the meantime, is to manually subset the data yourself with the column filtering option (see the gif illustration here: https://jasp-stats.org/wp-content/uploads/2018/06/01_click_filter.gif)

  You can repeat this for each level of the moderator variable. This is a crude method though, as it will not take into account the full ANOVA (which is what the simple main effects analysis does), but it can already give you an indication for which level of the moderator variable there is a difference between the levels of the variable of interest.

  The t-test suggestion from PM_Stat is due to the fact that running an ANOVA on a factor with 2 levels is essentially a t-test. However, that suggestion does not include the subsetting of the data set, which is what the simple main effects analysis is about, and is more related to running post hoc tests.

  I hope this clarifies the issue a bit, I will discuss with the relevant team member about possible implementation of Bayesian simple main effects.

  Kind regards,

  Johnny
• rohanp16

  April 2020

  Thanks for the clarification Johnny, it all makes sense.

  I'm glad you agree that the method suggested is a crude method and while not overly cumbersome for a 2 x 2 ANOVA, it would start to get complex as the number of factors and levels in an ANOVA increase.

  Thank you for discussing it with a relevant team member, I look forward to it being implemented in JASP.

  On a related note, I believe JASP uses the 'BayesFactor' R package. I'm relatively comfortable with R so might give it a try there. If there are any websites/tutorials/pieces of code that you are aware of that would help explain it conceptually, I'd be grateful if they could be shared.

  Cheers :)
• JohnnyB

  April 2020

  Hi Rohanp,

  
  

  Ah yes that makes life a lot easier! In that case you can do the subsetting and ANOVA in R. JASP indeed uses the BayesFactor package for the Bayesian ANOVA's. Here you can read more about using BayesFactor: https://cran.r-project.org/web/packages/BayesFactor/vignettes/manual.html

  
  

  Cheers,

  Johnny