Regression analysis: dichotomous covariate and reporting results
Hi there,
I recently got the chance to revise a paper in which I used JASP. To “control” for covariates, the reviewer asked me to include some covariates in my models (of 5 covariates, one covariate is gender). I computed simple regression analyses and multiple regression analyses. Now some issues came up and I am wondering if somebody could help me out.
1) In Bayesian regression analysis, it does not make any sense to include dichotomous covariates, right? I saw that there is the opportunity to perform an ANCOVA instead: https://forum.cogsci.nl/discussion/4288/bayesian-linear-regression-no-categorical-variables However, I also report results of classical regression analysis (NHST). I would like to avoid comparing Bayesian ANCOVA results and frequentist regression results. Therefore, I think I have to decide to exclude the dichotomous covariate (there is also a theoretical argument for that). Or is there another solution for this issue? Related to this thread https://forum.cogsci.nl/discussion/4433/suddenly-i-cant-add-nominal-vars-in-bayesian-regression-models, I would write something like: “In Bayesian regression analysis, the prior structure of the regression coefficients does not allow factors as variable values.” Is that OK?
2) Just to make sure: the best way is to add the covariates to the null model, right? Related to: https://forum.cogsci.nl/discussion/3413/should-i-tick-the-variable-which-i-want-to-control-namely-a-covariate-as-nuisance
3) When reporting the results of the regression analyses, is it sufficient to report the model parameter ( P(M), P(M|data), BFM, BF10, % error) or is it necessary to include the posterior summary?
I hope that my explanations are understandable!
Thank you very much everybody!
Alexa
Comments
Cheers,
E.J.
Thank you very much EJ for the helpful feedback!!