draft

draft is an R package that launches a local Shiny app to help you write traceable inline results in R Markdown and Quarto documents. It reads live objects from your current R session: models, data frames, and single values: and gives you:

• A browsable view of model parameters and data summaries
• Copyable inline reference paths (e.g. {params_$wt$estimate})
• A live-rendering text editor where you type prose and see actual values
• A one-click setup chunk to paste into your .qmd or .Rmd file

The app does not write to your document or interpret results. You still have to do the thinking; draft removes the friction. When combined with automatic table creation (e.g., https://gist.github.com/rbcavanaugh/be76e983fdf3a1920fe1f354d73a5a78) your results section should be built entirely from your R objects and not manually created.

---

Installation

# Install from GitHub
remotes::install_github("rbcavanaugh/draft")

---

Quick start

library(draft)

# 1. Run your analysis in the normal way
m1     <- lm(mpg ~ wt + cyl, data = mtcars)
params <- parameters::model_parameters(m1)

# 2. Launch the app: it reads everything in your global environment
launch_app()

In the app:

1. Environment panel (left): click params to inspect it

2. Object inspector (center): see formatted parameter estimates with copyable values

3. Text editor (right): write prose like:

   Weight was negatively associated with fuel economy
   (b = {params_$wt$estimate}, 95% CI {params_$wt$ci},
   p = {params_$wt$p}).
   

   The app suggests params_ as the list name (object name + trailing underscore).

4. Click Copy inline code: the copied text uses `r expr` syntax, ready to paste into your .qmd or .Rmd.

5. Click Copy setup chunk: copies the prep_params() call to put in your document's setup chunk.

Always re-render your full document to confirm results are reproducible.

---

Standalone helpers

You can also use these functions directly in a .qmd setup chunk without launching the app:

Model parameters: prep_params()

Converts a parameters_model object into a nested named list for inline reporting:

# Setup chunk
params_m1 <- parameters::model_parameters(m1)
results   <- draft::prep_params(params_m1)

# Inline usage
# Weight: b = `r results$wt$estimate` (95% CI `r results$wt$ci`, p = `r results$wt$p`)

Each parameter is a sub-list containing all available formatted fields: estimate, se, ci, ci_low, ci_high, p (frequentist) or pd, rope_pct (Bayesian). Works with any model class supported by parameters::model_parameters(), including correlation matrices.

Data frame summaries: prep_data()

Summarises a data frame into a nested named list for demographics reporting:

# Setup chunk
stats <- draft::prep_data(demo_data)

# Inline usage
# Age: M = `r stats$age$mean` (SD = `r stats$age$sd`)
# Female: `r stats$sex$female$pct`%

Continuous columns return mean, sd, median, min, max, n, n_missing. Categorical columns return n, n_missing, and per-level n and pct.

Marginal means and contrasts: prep_modelbased()

Converts estimate_means, estimate_contrasts, or estimate_slopes objects from the modelbased package:

means_ <- draft::prep_modelbased(modelbased::estimate_means(m1, "cyl"))
# `r means_$x4$mean`, `r means_$x6$mean`, `r means_$x8$mean`

Model performance: prep_performance()

Converts model_performance or compare_performance output from the performance package:

perf_ <- draft::prep_performance(performance::model_performance(m1))
# R2 = `r perf_$r2`, RMSE = `r perf_$rmse`

comp_ <- draft::prep_performance(performance::compare_performance(m1, m2))
# `r comp_$m1$aic` vs `r comp_$m2$aic`

Effect sizes: prep_effectsize()

Converts an effectsize_table object (e.g. cohens_d(), eta_squared(), cramers_v()) into a named list:

es_ <- draft::prep_effectsize(effectsize::cohens_d(mpg ~ am, data = mtcars))
# `r es_$cohens_d`, 95% CI [`r es_$ci_low`, `r es_$ci_high`]

eta_ <- draft::prep_effectsize(effectsize::eta_squared(aov(mpg ~ cyl + gear, mtcars)))
# `r eta_$cyl$eta2`, `r eta_$gear$eta2`

Single-effect objects return a flat list; multi-effect objects (e.g. eta_squared with several predictors) return a nested list keyed by the Parameter column.

Distribution summaries: prep_distribution()

Converts datawizard::describe_distribution() output into a named list for inline reporting:

dist_ <- draft::prep_distribution(datawizard::describe_distribution(mtcars))
# `r dist_$mpg$mean`, SD = `r dist_$mpg$sd`

# With grouping (by argument):
dist_ <- draft::prep_distribution(datawizard::describe_distribution(iris, by = "Species"))
# `r dist_$setosa$sepal_length$mean`

Supports both flat (no by argument) and stratified (with by argument) shapes.

---

Recommended workflow

1. Open your .qmd / .Rmd and run all chunks so objects are in globalenv()
2. Call draft::launch_app()
3. Browse parameters in the inspector; note list paths
4. Write prose in the text editor; see live-rendered values
5. Copy the setup chunk into your document's setup block
6. Copy inline code and paste into your document
7. Re-render the full document to confirm reproducibility

---

Dependencies

Package	Purpose
shiny	App framework
bslib	UI layout
parameters	model_parameters() and value formatting
insight	Model detection and value formatting
dplyr	Data frame manipulation
htmltools	Safe HTML rendering
effectsize	Effect size objects (suggested)
datawizard	Distribution descriptions (suggested)
modelbased	Marginal means and contrasts (suggested)
performance	Model performance metrics (suggested)

---

Roadmap

easystats ecosystem support

Package	Status	Notes
parameters	Supported	Core: all model types with a parameters_model class
correlation	Supported	correlation() output inherits parameters_model
modelbased	Supported	estimate_means(), estimate_contrasts(), estimate_slopes()
performance	Supported	model_performance() and compare_performance()
effectsize	Supported	cohens_d, eta_squared, cramers_v, and all effectsize_table subclasses
datawizard	Supported	describe_distribution(): flat and stratified (grouped) output

---

License

MIT