do we want df included in output where available?

bbolker / broom.mixed

tidy methods for mixed models in R

229 stars 24 forks source link

do we want df included in output where available? #3

Open bbolker opened 6 years ago

bbolker commented 6 years ago

why not:

clutters up output
not available in some cases
wasn't included in original broom tidiers (for whatever reason)

why:

might want to see it sometimes
it's easy enough to drop with a select(-df)
dplyr::bind_rows() makes incommensurate columns in different sets of output fairly harmless

bbolker commented 6 years ago

also see https://github.com/tidyverse/broom/issues/212

paul-buerkner commented 6 years ago

I think it is generally ok, not to have the same columns for all tidies. I can assume, we will run across similar issues with other special columns, which might not be available / desired in all models. Take p-values as an example (not sure if they are anywhere in the tidiers): We could easily approximate p-values in Bayesian models, but most of Bayesians statisticians would be opposed to computing them at all.

bbolker commented 6 years ago

I agree (although note that MCMCglmm does provide p-values; I agree, I don't like them ...).

I mostly raised this because it reverses a presumably deliberate design choice made in the original broom tidiers ...

bwiernik commented 4 years ago

I don't think there was a design decision here in broom—I think it's just that for many models no df is reported because they are all 1. But, for example, tidy.aov() does have a df column (and a sumsq and meansq) column.