Model selection and assessment of model fit

I don’t even fit univariable models anymore, never mind present them in results. These days I present only descriptive statistics and the results of a multivariable model with the adjustment variables having been listed in the methods with reasons to include them (if not obvious). Typically I also present only the parameter of interest of the multivariable model due to the dangers of the Table 2 fallacy (https://academic.oup.com/aje/article/177/4/292/147738), though I usually put a file in the Supplementary results with the full multivariable results, in case any reviewer or reader really really wants them.

I’ve never had major issues with the above approach. Although recently a reviewer wanted univariable models and I refused citing this paper: Step away from stepwise | Journal of Big Data | Full Text and the reviewer didn’t push for it anymore (paper eventually accepted after resolving other unrelated points).

2 Likes

Minor note: multivariate = multiple dependent variables. I think you mean multivariable.

2 Likes

:man_facepalming: Thx for pointing that out @f2harrell - I have a head cold, not firing all circuits currently :sweat_smile: Will edit

1 Like