Order : Compare to best model vs. compare to null model
Hi everyone,
I was wondering if there are some recommendations about selecting "Compare to best model" or "compare to null model" when conducting a bayesian Anova with JASP. The default option is "compare to best model". I understood that this will result in comparing all the models with the best model. Some colleagues said to me that choosing one option or another will lead to very different results. Thank you very much for your help. Best regards. Nico
Comments
Hello Nico,
These options define the model to which all other models are compared. The fundamental information in the table does not change, however. Because the null model is often very poor, it is usually more interpretable to compare all models against the best one.
Suppose you select "Compare to null model" and the ordering of the models in the table is
H0
HB
HA
Suppose HB is the best model, and you are interested in the BF between HB and HA. Then you can select "best model on top" and set the Bayes factor display option to BF_01 (instead of the default BF_10) and you see immediately how much better HB predicted the data. Alternatively, you could take that initial table and manually divide the two Bayes factors for HB and HA versus H0 -- by transitivity the p(data|H0) term drops out.
Cheers,
E.J.
Hi E.J,
thanks for your explanation. When working with BFincl/BFexcl this option doesn't change anything because it doesn't change p(M) and p(M|D) in the model comparison table, is it right?
Best,
Nico
Correct
E.J.
thanks
Nico
Hi
I know its an old tread, but maybe I can still ask. Why does JASP add random slopes and subject to bayesian rmANOVA? Thanks!
Hi Katharina,
I am not sure how your question relates to this tread :-) However, we do have a recent paper submitted on this topic: https://psyarxiv.com/fb8zn/
Cheers,
E.J.