RM ANOVA frequentist/bayesian: opposite conclusion ?

johan_achard

Hello,

I worked on an RM ANOVA on JASP. The conclusions were opposite between frequentist and Bayesian.

I used the latest version of JASP.

What to do ?

Thanks in advance for your help.

Best regards,

Johan

---------- Files -------------

JASP : file

https://github.com/JohanACHARD/jasp_RM_anova_difference_freq_bayesian/blob/7820da3867dccfd656f2dc17e8d923d77e9ac400/slope_data_norm_exp1_onset.jasp

Csv File :

https://github.com/JohanACHARD/jasp_RM_anova_difference_freq_bayesian/blob/7820da3867dccfd656f2dc17e8d923d77e9ac400/slope_data_norm_exp1_onset.csv

andersony3k

While there could be an error in the Bayesian repeated measures routine, here's another possibility: It's known that JASP's "Bayesian repeated measures" routine doesn't actually perform a repeated measures analysis. Instead, it performs a linear mixed effects analysis (but with dropping of participants who have any missing data). This is significant because, especially when there are multiple factors in the analysis, a linear mixed effects model won't necessarily produce the same result as a repeated measures ANOVA. So I think that if you want to compare frequentist to Bayesian results you might want to restructure your data to long format and then compare a Classical Linear Mixed Effects model to a Bayesian Linear Mixed Effects model.

EJ

Also:

1. The discrepancy is not glaring; the BF indicates anecdotal evidence, whereas the p-value is near .03, which is not that compelling
2. The inclusion BF is based on taking all models into account, which is fundamentally different from the frequentist approach.

EJ

johan_achard

Thanks for the explanation, for your expertise and your help.

It's clearer now.

Johan