Skip to content

A set of functions to facilitate manipulation of tibbles produced by broom

License

Notifications You must be signed in to change notification settings

larmarange/broom.helpers

Repository files navigation

broom.helpers

Lifecycle: stable R-CMD-check Codecov test coverage CRAN status DOI

The broom.helpers package provides suite of functions to work with regression model broom::tidy() tibbles.

The suite includes functions to group regression model terms by variable, insert reference and header rows for categorical variables, add variable labels, and more.

broom.helpers is used, in particular, by gtsummary::tbl_regression() for producing nice formatted tables of model coefficients and by ggstats::ggcoef_model() for plotting model coefficients.

Installation & Documentation

To install stable version:

install.packages("broom.helpers")

Documentation of stable version: https://larmarange.github.io/broom.helpers/

To install development version:

remotes::install_github("larmarange/broom.helpers")

Documentation of development version: https://larmarange.github.io/broom.helpers/dev/

Examples

all-in-one wrapper

mod1 <- lm(Sepal.Length ~ Sepal.Width + Species, data = iris)
library(broom.helpers)
ex1 <- mod1 |> tidy_plus_plus()
ex1
#> # A tibble: 4 × 17
#>   term              variable  var_label var_class var_type var_nlevels contrasts
#>   <chr>             <chr>     <chr>     <chr>     <chr>          <int> <chr>    
#> 1 Sepal.Width       Sepal.Wi… Sepal.Wi… numeric   continu…          NA <NA>     
#> 2 Speciessetosa     Species   Species   factor    categor…           3 contr.tr…
#> 3 Speciesversicolor Species   Species   factor    categor…           3 contr.tr…
#> 4 Speciesvirginica  Species   Species   factor    categor…           3 contr.tr…
#> # ℹ 10 more variables: contrasts_type <chr>, reference_row <lgl>, label <chr>,
#> #   n_obs <dbl>, estimate <dbl>, std.error <dbl>, statistic <dbl>,
#> #   p.value <dbl>, conf.low <dbl>, conf.high <dbl>
dplyr::glimpse(ex1)
#> Rows: 4
#> Columns: 17
#> $ term           <chr> "Sepal.Width", "Speciessetosa", "Speciesversicolor", "S…
#> $ variable       <chr> "Sepal.Width", "Species", "Species", "Species"
#> $ var_label      <chr> "Sepal.Width", "Species", "Species", "Species"
#> $ var_class      <chr> "numeric", "factor", "factor", "factor"
#> $ var_type       <chr> "continuous", "categorical", "categorical", "categorica…
#> $ var_nlevels    <int> NA, 3, 3, 3
#> $ contrasts      <chr> NA, "contr.treatment", "contr.treatment", "contr.treatm…
#> $ contrasts_type <chr> NA, "treatment", "treatment", "treatment"
#> $ reference_row  <lgl> NA, TRUE, FALSE, FALSE
#> $ label          <chr> "Sepal.Width", "setosa", "versicolor", "virginica"
#> $ n_obs          <dbl> 150, 50, 50, 50
#> $ estimate       <dbl> 0.8035609, 0.0000000, 1.4587431, 1.9468166
#> $ std.error      <dbl> 0.1063390, NA, 0.1121079, 0.1000150
#> $ statistic      <dbl> 7.556598, NA, 13.011954, 19.465255
#> $ p.value        <dbl> 4.187340e-12, NA, 3.478232e-26, 2.094475e-42
#> $ conf.low       <dbl> 0.5933983, NA, 1.2371791, 1.7491525
#> $ conf.high      <dbl> 1.013723, NA, 1.680307, 2.144481

mod2 <- glm(
  response ~ poly(age, 3) + stage + grade * trt,
  na.omit(gtsummary::trial),
  family = binomial,
  contrasts = list(
    stage = contr.treatment(4, base = 3),
    grade = contr.sum
  )
)
ex2 <- mod2 |>
  tidy_plus_plus(
    exponentiate = TRUE,
    variable_labels = c(age = "Age (in years)"),
    add_header_rows = TRUE,
    show_single_row = "trt"
  )
ex2
#> # A tibble: 17 × 19
#>    term   variable var_label var_class var_type var_nlevels header_row contrasts
#>    <chr>  <chr>    <chr>     <chr>     <chr>          <int> <lgl>      <chr>    
#>  1 <NA>   age      Age (in … nmatrix.3 continu…          NA TRUE       <NA>     
#>  2 poly(… age      Age (in … nmatrix.3 continu…          NA FALSE      <NA>     
#>  3 poly(… age      Age (in … nmatrix.3 continu…          NA FALSE      <NA>     
#>  4 poly(… age      Age (in … nmatrix.3 continu…          NA FALSE      <NA>     
#>  5 <NA>   stage    T Stage   factor    categor…           4 TRUE       contr.tr…
#>  6 stage1 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#>  7 stage2 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#>  8 stage3 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#>  9 stage4 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#> 10 <NA>   grade    Grade     factor    categor…           3 TRUE       contr.sum
#> 11 grade1 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 12 grade2 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 13 grade3 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 14 trtDr… trt      Chemothe… character dichoto…           2 NA         contr.tr…
#> 15 <NA>   grade:t… Grade * … <NA>      interac…          NA TRUE       <NA>     
#> 16 grade… grade:t… Grade * … <NA>      interac…          NA FALSE      <NA>     
#> 17 grade… grade:t… Grade * … <NA>      interac…          NA FALSE      <NA>     
#> # ℹ 11 more variables: contrasts_type <chr>, reference_row <lgl>, label <chr>,
#> #   n_obs <dbl>, n_event <dbl>, estimate <dbl>, std.error <dbl>,
#> #   statistic <dbl>, p.value <dbl>, conf.low <dbl>, conf.high <dbl>
dplyr::glimpse(ex2)
#> Rows: 17
#> Columns: 19
#> $ term           <chr> NA, "poly(age, 3)1", "poly(age, 3)2", "poly(age, 3)3", …
#> $ variable       <chr> "age", "age", "age", "age", "stage", "stage", "stage", …
#> $ var_label      <chr> "Age (in years)", "Age (in years)", "Age (in years)", "…
#> $ var_class      <chr> "nmatrix.3", "nmatrix.3", "nmatrix.3", "nmatrix.3", "fa…
#> $ var_type       <chr> "continuous", "continuous", "continuous", "continuous",…
#> $ var_nlevels    <int> NA, NA, NA, NA, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, NA, NA, NA
#> $ header_row     <lgl> TRUE, FALSE, FALSE, FALSE, TRUE, FALSE, FALSE, FALSE, F…
#> $ contrasts      <chr> NA, NA, NA, NA, "contr.treatment(base=3)", "contr.treat…
#> $ contrasts_type <chr> NA, NA, NA, NA, "treatment", "treatment", "treatment", …
#> $ reference_row  <lgl> NA, NA, NA, NA, NA, FALSE, FALSE, TRUE, FALSE, NA, FALS…
#> $ label          <chr> "Age (in years)", "Age (in years)", "Age (in years)²", …
#> $ n_obs          <dbl> NA, 92, 56, 80, NA, 46, 50, 35, 42, NA, 63, 53, 57, 90,…
#> $ n_event        <dbl> NA, 31, 17, 22, NA, 17, 12, 13, 12, NA, 20, 16, 18, 30,…
#> $ estimate       <dbl> NA, 20.2416394, 1.2337899, 0.4931553, NA, 1.0047885, 0.…
#> $ std.error      <dbl> NA, 2.3254455, 2.3512842, 2.3936657, NA, 0.4959893, 0.5…
#> $ statistic      <dbl> NA, 1.29340459, 0.08935144, -0.29533409, NA, 0.00963137…
#> $ p.value        <dbl> NA, 0.1958712, 0.9288026, 0.7677387, NA, 0.9923154, 0.1…
#> $ conf.low       <dbl> NA, 0.225454425, 0.007493208, 0.004745694, NA, 0.379776…
#> $ conf.high      <dbl> NA, 2315.587655, 100.318341, 74.226179, NA, 2.683385, 1…

fine control

ex3 <- mod1 |>
  # perform initial tidying of model
  tidy_and_attach() |>
  # add reference row
  tidy_add_reference_rows() |>
  # add term labels
  tidy_add_term_labels() |>
  # remove intercept
  tidy_remove_intercept()
ex3
#> # A tibble: 4 × 16
#>   term              variable  var_label var_class var_type var_nlevels contrasts
#>   <chr>             <chr>     <chr>     <chr>     <chr>          <int> <chr>    
#> 1 Sepal.Width       Sepal.Wi… Sepal.Wi… numeric   continu…          NA <NA>     
#> 2 Speciessetosa     Species   Species   factor    categor…           3 contr.tr…
#> 3 Speciesversicolor Species   Species   factor    categor…           3 contr.tr…
#> 4 Speciesvirginica  Species   Species   factor    categor…           3 contr.tr…
#> # ℹ 9 more variables: contrasts_type <chr>, reference_row <lgl>, label <chr>,
#> #   estimate <dbl>, std.error <dbl>, statistic <dbl>, p.value <dbl>,
#> #   conf.low <dbl>, conf.high <dbl>
dplyr::glimpse(ex3)
#> Rows: 4
#> Columns: 16
#> $ term           <chr> "Sepal.Width", "Speciessetosa", "Speciesversicolor", "S…
#> $ variable       <chr> "Sepal.Width", "Species", "Species", "Species"
#> $ var_label      <chr> "Sepal.Width", "Species", "Species", "Species"
#> $ var_class      <chr> "numeric", "factor", "factor", "factor"
#> $ var_type       <chr> "continuous", "categorical", "categorical", "categorica…
#> $ var_nlevels    <int> NA, 3, 3, 3
#> $ contrasts      <chr> NA, "contr.treatment", "contr.treatment", "contr.treatm…
#> $ contrasts_type <chr> NA, "treatment", "treatment", "treatment"
#> $ reference_row  <lgl> NA, TRUE, FALSE, FALSE
#> $ label          <chr> "Sepal.Width", "setosa", "versicolor", "virginica"
#> $ estimate       <dbl> 0.8035609, NA, 1.4587431, 1.9468166
#> $ std.error      <dbl> 0.1063390, NA, 0.1121079, 0.1000150
#> $ statistic      <dbl> 7.556598, NA, 13.011954, 19.465255
#> $ p.value        <dbl> 4.187340e-12, NA, 3.478232e-26, 2.094475e-42
#> $ conf.low       <dbl> 0.5933983, NA, 1.2371791, 1.7491525
#> $ conf.high      <dbl> 1.013723, NA, 1.680307, 2.144481

ex4 <- mod2 |>
  # perform initial tidying of model
  tidy_and_attach(exponentiate = TRUE) |>
  # add variable labels, including a custom value for age
  tidy_add_variable_labels(labels = c(age = "Age in years")) |>
  # add reference rows for categorical variables
  tidy_add_reference_rows() |>
  # add a, estimate value of reference terms
  tidy_add_estimate_to_reference_rows(exponentiate = TRUE) |>
  # add header rows for categorical variables
  tidy_add_header_rows()
ex4
#> # A tibble: 20 × 17
#>    term   variable var_label var_class var_type var_nlevels header_row contrasts
#>    <chr>  <chr>    <chr>     <chr>     <chr>          <int> <lgl>      <chr>    
#>  1 (Inte… (Interc… (Interce… <NA>      interce…          NA NA         <NA>     
#>  2 <NA>   age      Age in y… nmatrix.3 continu…          NA TRUE       <NA>     
#>  3 poly(… age      Age in y… nmatrix.3 continu…          NA FALSE      <NA>     
#>  4 poly(… age      Age in y… nmatrix.3 continu…          NA FALSE      <NA>     
#>  5 poly(… age      Age in y… nmatrix.3 continu…          NA FALSE      <NA>     
#>  6 <NA>   stage    T Stage   factor    categor…           4 TRUE       contr.tr…
#>  7 stage1 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#>  8 stage2 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#>  9 stage3 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#> 10 stage4 stage    T Stage   factor    categor…           4 FALSE      contr.tr…
#> 11 <NA>   grade    Grade     factor    categor…           3 TRUE       contr.sum
#> 12 grade1 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 13 grade2 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 14 grade3 grade    Grade     factor    categor…           3 FALSE      contr.sum
#> 15 <NA>   trt      Chemothe… character dichoto…           2 TRUE       contr.tr…
#> 16 trtDr… trt      Chemothe… character dichoto…           2 FALSE      contr.tr…
#> 17 trtDr… trt      Chemothe… character dichoto…           2 FALSE      contr.tr…
#> 18 <NA>   grade:t… Grade * … <NA>      interac…          NA TRUE       <NA>     
#> 19 grade… grade:t… Grade * … <NA>      interac…          NA FALSE      <NA>     
#> 20 grade… grade:t… Grade * … <NA>      interac…          NA FALSE      <NA>     
#> # ℹ 9 more variables: contrasts_type <chr>, reference_row <lgl>, label <chr>,
#> #   estimate <dbl>, std.error <dbl>, statistic <dbl>, p.value <dbl>,
#> #   conf.low <dbl>, conf.high <dbl>
dplyr::glimpse(ex4)
#> Rows: 20
#> Columns: 17
#> $ term           <chr> "(Intercept)", NA, "poly(age, 3)1", "poly(age, 3)2", "p…
#> $ variable       <chr> "(Intercept)", "age", "age", "age", "age", "stage", "st…
#> $ var_label      <chr> "(Intercept)", "Age in years", "Age in years", "Age in …
#> $ var_class      <chr> NA, "nmatrix.3", "nmatrix.3", "nmatrix.3", "nmatrix.3",…
#> $ var_type       <chr> "intercept", "continuous", "continuous", "continuous", …
#> $ var_nlevels    <int> NA, NA, NA, NA, NA, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 2,…
#> $ header_row     <lgl> NA, TRUE, FALSE, FALSE, FALSE, TRUE, FALSE, FALSE, FALS…
#> $ contrasts      <chr> NA, NA, NA, NA, NA, "contr.treatment(base=3)", "contr.t…
#> $ contrasts_type <chr> NA, NA, NA, NA, NA, "treatment", "treatment", "treatmen…
#> $ reference_row  <lgl> NA, NA, NA, NA, NA, NA, FALSE, FALSE, TRUE, FALSE, NA, …
#> $ label          <chr> "(Intercept)", "Age in years", "Age in years", "Age in …
#> $ estimate       <dbl> 0.5266376, NA, 20.2416394, 1.2337899, 0.4931553, NA, 1.…
#> $ std.error      <dbl> 0.4130930, NA, 2.3254455, 2.3512842, 2.3936657, NA, 0.4…
#> $ statistic      <dbl> -1.55229592, NA, 1.29340459, 0.08935144, -0.29533409, N…
#> $ p.value        <dbl> 0.1205914, NA, 0.1958712, 0.9288026, 0.7677387, NA, 0.9…
#> $ conf.low       <dbl> 0.227717775, NA, 0.225454425, 0.007493208, 0.004745694,…
#> $ conf.high      <dbl> 1.164600, NA, 2315.587655, 100.318341, 74.226179, NA, 2…