`modelplot`: Plot model coefficients with confidence intervals

## Warning in system("timedatectl", intern = TRUE): running command 'timedatectl'
## had status 1

modelplot is a function from the modelsummary package. It allows you to plot model estimates and confidence intervals. It makes it easy to subset, rename, reorder, and customize plots using same mechanics as in modelsummary.

To illustrate how the function works, we fit a linear model to data about the Palmer Penguins:

url <- 'https://vincentarelbundock.github.io/Rdatasets/csv/palmerpenguins/penguins.csv'
dat <- read.csv(url)

# rescale mm -> cm
dat$bill_length_cm <- dat$bill_length_mm / 10
dat$flipper_length_cm <- dat$flipper_length_mm / 10

mod <- lm(bill_length_cm ~ flipper_length_cm + species, data = dat)

Then, we load the modelsummary library and call modelplot:

library(modelsummary)

modelplot(mod)

Rename, reorder, subset

modelplot uses the same mechanics as modelsummary to rename, reorder, and subset estimates. First, you can use the coef_omit argument. This will omit any coefficient that matches a string or regular expression. To omit the intercept, for example, we can type:

modelplot(mod, coef_omit = 'Interc')

Second, you can use the coef_map argument. coef_map is a named character vector, where names correspond to the original term names, and values correspond the names you want to assign. Any variable that is not included in coef_map will be excluded. Variables will also be drawn in the same order as in coef_map:

cm <- c('speciesChinstrap' = 'Chinstrap',
        'speciesGentoo' = 'Gentoo', 
        'flipper_length_cm' = 'Flipper length (cm)')

modelplot(mod, coef_map = cm)

Several models

The main modelsummary functions allows you to create a table with the results of several models side-by-side, by storing them in a (potentially named) list:

models <- list(
  "Small model" = lm(bill_length_cm ~ flipper_length_cm, data = dat),
  "Medium model" = lm(bill_length_cm ~ flipper_length_cm + body_mass_g, data = dat),
  "Large model" = lm(bill_length_cm ~ flipper_length_cm + body_mass_g + species, data = dat))

modelsummary(models, statistic = 'conf.int')

## Warning in !is.null(rmarkdown::metadata$output) && rmarkdown::metadata$output
## %in% : 'length(x) = 2 > 1' in coercion to 'logical(1)'

	Small model	Medium model	Large model
(Intercept)	−0.726	−0.344	0.984
	[−1.356, −0.097]	[−1.245, 0.557]	[0.215, 1.752]
flipper_length_cm	0.255	0.222	0.095
	[0.224, 0.286]	[0.158, 0.285]	[0.048, 0.142]
body_mass_g		0.000	0.000
		[0.000, 0.000]	[0.000, 0.000]
speciesChinstrap			0.939
			[0.867, 1.011]
speciesGentoo			0.207
			[0.088, 0.326]
Num.Obs.	342	342	342
R2	0.431	0.433	0.817
R2 Adj.	0.429	0.430	0.815
AIC	369.0	369.6	−12.6
BIC	380.5	385.0	10.4
Log.Lik.	−181.499	−180.813	12.313
F	257.092	129.365	375.333
RMSE	0.41	0.41	0.23

modelplot works the same way:

modelplot(models, coef_omit = 'Interc')

Instead of displaying results with “dodged” side-by-side lines, you can also use facet:

modelplot(models, facet = TRUE)

Customizing plots

The graphs produced by modelplot are simple ggplot2 objects. You can thus post-process them using the normal suite of functions available for all objects of this type. Here, we change the axis labels, add a title and a caption, and use a color scheme inspired by Wes Anderson’s Darjeeling Limited:

library(wesanderson)
library(ggplot2)

modelplot(models) +
    labs(x = 'Coefficients', 
         y = 'Term names',
         title = 'Linear regression models of "Bill Length (cm)"',
         caption = "Data source: Gorman, Williams & Fraser (2014), packaged for R by @apreshill and @allison_horst") +
    scale_color_manual(values = wes_palette('Darjeeling1'))

In ggplot2, some visual choices must be made when calling the “geom”, rather than with post-processing functions. For instance, the size, color, fatten, linetype arguments must all be specified inside the geom_pointrange function for them to take effect. modelplot will pass any unknown argument to geom_pointrange, so users can simply call:

modelplot(mod, size = 1, fatten = .7, color = 'darkgreen', linetype = 'dotted') + 
    theme_classic()

Conditional colors and shape

Note: This section requires a version of modelsummary greater than 1.2.0 or the development version.

In a very nice Stack Overflow answer, Allan Cameron shows how we can use the aes() function from ggplot2 to add conditional aesthetics. For example, if we want to display statistically significant coefficients in a different color:

library(ggplot2)
mod <- lm(hp ~ factor(gear) + factor(cyl), data = mtcars)

modelplot(mod, coef_rename = TRUE) +
  aes(color = ifelse(p.value < 0.001, "Significant", "Not significant")) +
  scale_color_manual(values = c("grey", "black"))

Customizing plots with ggplot2 makes modelplot very flexible. For example, imagine you want to compare the coefficients of three models with different dependent variables. First, we load the packages and estimate our models:

library(ggplot2)
library(modelsummary)

models <- list(
  lm(vs ~ carb + mpg + cyl, data = mtcars),
  lm(disp ~ carb + mpg + cyl, data = mtcars),
  lm(hp ~ carb + mpg + cyl, data = mtcars))

Then, we use the dvnames function to rename our list with names matching the the dependent variable in each model:

models <- dvnames(models)

By calling modelplot with the draw=FALSE argument, we see the raw data used to draw the plot. Here, we see that there is a model column:

modelplot(models, draw = FALSE)

##           term model     estimate    std.error      conf.low    conf.high
## 1  (Intercept)    vs   2.41742511   0.67622094    1.03224931   3.80260091
## 5  (Intercept)  disp 112.57276339 114.86315481 -122.71374324 347.85927003
## 9  (Intercept)    hp -10.56116383  68.75946117 -151.40853516 130.28620751
## 2         carb    vs  -0.06945116   0.03943402   -0.15022810   0.01132577
## 6         carb  disp -12.30144724   6.69827859  -26.02224894   1.41935446
## 10        carb    hp  17.75593287   4.00972816    9.54237706  25.96948867
## 3          mpg    vs  -0.01513960   0.01716410   -0.05029868   0.02001947
## 7          mpg  disp  -7.14964651   2.91550156  -13.12178072  -1.17751230
## 11         mpg    hp  -1.00486469   1.74527956   -4.57990780   2.57017842
## 4          cyl    vs  -0.23926135   0.05687969   -0.35577411  -0.12274859
## 8          cyl  disp  47.90105842   9.66160634   28.11015499  67.69196184
## 12         cyl    hp  20.60581208   5.78363747    8.75856779  32.45305638
##         p.value
## 1  1.296718e-03
## 5  3.354494e-01
## 9  8.790301e-01
## 2  8.912324e-02
## 6  7.692105e-02
## 10 1.320972e-04
## 3  3.852593e-01
## 7  2.068858e-02
## 11 5.693755e-01
## 4  2.410214e-04
## 8  3.111898e-05
## 12 1.338485e-03

Finally, we use the model column as our identifier in ggplot2’s facet_grid command to display models side by side:

modelplot(models, color = "black") + facet_grid(~model)

Confidence intervals: change or omit

You can change the \(\alpha\) level of your confidence intervals by changing the conf_level argument:

modelplot(mod, conf_level = .99)

modelplot(mod, conf_level = NULL)

Background annotations

Sometimes, you want to display annotations on a plot, but you would like to draw these annotations behind the geom_pointrange which displays the estimates. Since modelplot draws the geom_pointrange automatically, any ggplot2 annotation you add to the plot using + will be added on top of the existing ones.

To add your annotations in the background, you can pass them as a list of ggplot2 “geoms”:

library(ggplot2)

b <- list(geom_vline(xintercept = 0, color = 'orange'),
          annotate("rect", alpha = .1,
                   xmin = -.5, xmax = .5, 
                   ymin = -Inf, ymax = Inf),
          geom_point(aes(y = term, x = estimate), alpha = .3, 
                     size = 10, color = 'red'))

modelplot(mod, background = b)

Raw data & More customization

If you would like to customize the plots even more than what modelplot allows, you can obtain the raw data used to draw the plots by setting draw=FALSE:

modelplot(models, draw = FALSE)

##           term model     estimate    std.error      conf.low    conf.high
## 1  (Intercept)    vs   2.41742511   0.67622094    1.03224931   3.80260091
## 5  (Intercept)  disp 112.57276339 114.86315481 -122.71374324 347.85927003
## 9  (Intercept)    hp -10.56116383  68.75946117 -151.40853516 130.28620751
## 2         carb    vs  -0.06945116   0.03943402   -0.15022810   0.01132577
## 6         carb  disp -12.30144724   6.69827859  -26.02224894   1.41935446
## 10        carb    hp  17.75593287   4.00972816    9.54237706  25.96948867
## 3          mpg    vs  -0.01513960   0.01716410   -0.05029868   0.02001947
## 7          mpg  disp  -7.14964651   2.91550156  -13.12178072  -1.17751230
## 11         mpg    hp  -1.00486469   1.74527956   -4.57990780   2.57017842
## 4          cyl    vs  -0.23926135   0.05687969   -0.35577411  -0.12274859
## 8          cyl  disp  47.90105842   9.66160634   28.11015499  67.69196184
## 12         cyl    hp  20.60581208   5.78363747    8.75856779  32.45305638
##         p.value
## 1  1.296718e-03
## 5  3.354494e-01
## 9  8.790301e-01
## 2  8.912324e-02
## 6  7.692105e-02
## 10 1.320972e-04
## 3  3.852593e-01
## 7  2.068858e-02
## 11 5.693755e-01
## 4  2.410214e-04
## 8  3.111898e-05
## 12 1.338485e-03

This allows users to use external tools such as the powerful ggdist package. In this example, we use the purrr::map_dfr function to call modelplot several times with different confidence levels. Then, we draw a plot where the different confidence intervals are drawn with different thicknesses:

library(tidyverse)
library(modelsummary)
library(ggdist)

# fit
models <- list(
  lm(mpg ~ am, mtcars),
  lm(mpg ~ am + cyl, mtcars))

# summarize
dat <- map_dfr(c(.8, .9, .99), function(x) {
  modelplot(models, conf_level = x, draw = FALSE) %>%
  mutate(.width = x)
})

# plot
ggplot(dat, aes(
    y = term, x = estimate,
    xmin = conf.low, xmax = conf.high,
    color = model)) +
  ggdist::geom_pointinterval(
    position = "dodge",
    interval_size_range = c(1, 3),
    fatten_point = .1)

## Warning: Using the `size` aesthietic with geom_segment was deprecated in ggplot2 3.4.0.
## ℹ Please use the `linewidth` aesthetic instead.