Gt: Feature requests: p-values and stacking spanners

[thisisnickb]

I'm already a big fan of this package, but there are a few things that would make it even more useful for the sorts of tables I generate. First off, will it be possible to stack spanners, so that the same column can have two levels of spanners? Secondly, will be possible to have a number format which automatically *'s numbers less than .05, 's numbers less than .01, and *'s numbers less than .001? I've been having trouble getting the base footnotes to work for this, so having a built-in tool would be fantastic.

[rich-iannone]

You should have a look at @vincentarelbundock 's great new gtsummary package. It uses gt and is geared toward statistical model summaries.

It's also possible to make your own formatter for this!

library(gt)
library(broom)
library(tidyverse)

# Create a simple formatter for statistical significance
# (creating stars from p values)
fmt_stars <- function(data,
                      columns,
                      rows = NULL) {

  # Capture expression in `rows`
  rows <- rlang::enquo(rows)

  # Pass `data`, `columns`, `rows`, and the formatting
  # functions as a function list to `fmt()`
  fmt(
    data = data,
    columns = columns,
    rows = !!rows,
    fns = list(
      default = function(x) {

        x_str <- 
          dplyr::case_when(
            between(x, 0, 0.005) ~ "***",
            between(x, 0, 0.01) ~ "**",
            between(x, 0, 0.05) ~ "*",
            TRUE ~ "."
          )
      }
    )
  )
}

# Create a linear model with `mtcars`, tidy the
# summary with `broom::tidy()`, add an extra
# `stars` column (with the p values), create
# a gt table, format the p values as stars
lm(mpg ~ wt + qsec, mtcars) %>%
  broom::tidy() %>%
  dplyr::mutate(stars = p.value) %>%
  gt() %>%
  fmt_stars(columns = vars(stars))

As far as the multiple spanners (level 3 and above), that adds a lot of complexity to the API and to the resulting tables. Are there some convincing use cases for going beyond 2 levels of headers?

[vincentarelbundock]

@thisisnickb if you end up trying gtsummary and feel that there are missing features, let me know. Happy to expand the scope if people find it useful.

[rasmusrhl]

I will try to answer the question:

_Are there some convincing use cases for going beyond 2 levels of headers?_

I think the following are not perfect examples, but will hint at the real utility of going beyond 2 levels of headers:

Source: https://www.researchgate.net/figure/Demographic-characteristics-of-Millennium-Cohort-Study-participants-by-smallpox_tbl1_5470825

Source: https://www.researchgate.net/figure/Characteristics-of-patients-in-the-untreated-cohort-1-and-treated-cohort-2-cohorts_tbl1_221763372

Source: https://www.nature.com/articles/nature13425/tables/4

But I understand that this probably takes a lot of work.

[vincentarelbundock]

FWIW, my gtsummary package mentioned upthread was renamed modelsummary. It now includes functions to produce crosstabs and data summaries. In that context, I constantly have to produce tables with more than one column span levels, and I always have to "flatten" headers when users want a gt table. In my view, there are many compelling use-cases for this.

You'll see several examples here:

https://vincentarelbundock.github.io/modelsummary/articles/datasummary.html

[awcm0n]

The comment by @rasmusrhl makes it clear that multiple levels of spanners/headers are a critical feature for producing scientific tables.