Prior model probabilities

cblack3

Dear JASP community,

I am writing to gain clarity about setting prior model probabilities. In Wagenmakers et al. (2018), the authors state "P(M) indicates prior model probabilities (which the current version of JASP sets to be equal across all models at hand)." However, in the current version, when I run my own Bayes regression for the Auction data, I find that P(M) has variable values

QUESTION: Do users have the ability to set P(M) or is it automatically generated by JASP? How were the priors set to have variable values?

Many thanks,

Caroline

EJ

Hi Caroline,

Basically, there are two default ways to set the prior model probabilities. One is uniform (basically assuming that every predictor has probability 0.5 of being included; this does lead to a prior preference for models with 50% of the predictors included), the other is based on a more complicated method by Scott and Berger (2006, 2010) which results in a uniform prior across model classes of different numbers of predictors. The model probabilities you see are the Scott and Berger ones. You can specify which one you prefer in the JASP GUI. We hope to present a paper shortly that will explain this in more detail.

Cheers,

E.J.

alexa

Hi E.J.,

I have the same problem as I re-ran my analyses with the new JASP version and now the prior model probabilites are not the same anymore. Did you publish the paper meanwhile? I am a little bit unsecure which method is the best for my data. Overall, both analysis yield the same results.

Thanks

Alexa

EJ

van den Bergh, D., Clyde, M. A., Raj, A., de Jong, T., Gronau, Q. F., Marsman, M., Ly, A., and Wagenmakers, E.-J. (2020). A tutorial on Bayesian multi-model linear regression with BAS and JASP. Manuscript submitted for publication. 

https://psyarxiv.com/pqju6/

Cheers,

E.J.

alexa

Thank you E.J.,! The paper is very informative and helps me choosing a uniform prior over the number of included predictors.