Model comparison vs Output Effects in RM ANOVA
Hi, I like the Bayesian approach to data-analysis, but I'm wondering about the best way of doing this in JASP, given my research:
The design of the experiment: 2 (condition: level 1 vs level 2; within-participants) x 2 (condition_order; first level 1, first level 2; between-participants) with an outcome variable. My hypothesis: outcome variable for condition level 2 > condition level 1. Condition_order is just to check for order effects, in my field it is typical to test and report these.
I would normally conduct a repeated measures ANOVA (which shows a main effect of condition F = 32, and no effects for condition_order F < 1 or condition*condition_order F < 1.2). When I conduct a Bayesian RM ANOVA, and look at the analysis of effects: there's very strong evidence for an effect of condition and no evidence for an effect of or interaction with condition_order (but also no strong evidence for no effect of order, that's fine).
When I look at model comparison, the model with only condition is much better than the null model. And, there is anecdotal/barely any evidence that the model with only condition is better than the model that also includes a main effect of condition_order (3515/1591 = 2.21). But, I'm not really interested in comparing models. Condition_order is a factor, but not a variable of interest, and I can't set it to nuisance because I need to test and report it's effects. I don't really see why this model comparison is interesting for this research paradigm.
I'm primarily interested in the effect of condition, and so I think it would be best to report the BF's of the analysis of effects, and then conclude that there is very strong evidence for my hypothesis, and no evidence for main or interaction effects of order. Because it's the first time I'm using JASP, I was hoping there would be someone to tell me if this is a right approach or not. Thank you 
!
Comments
Yes, I think you're spot on.
Cheers,
E.J.
Thanks, also for the quick response!