The 'ggcorset' package introduces a visualization technique coined the corset plot. It is used strictly to visualize repeat measures data at 2 time points (such as pre- and post- data). The distribution of measurements at each time point is visualized using a half violin, whilst the trajectories of individual change are visualized via spaghetti plots by connecting these two points linearly. These lines can be filled to visualize the magnitude of change or other user-defined observed value. This method of visualization is ideal for showing the true heterogeneity of data, including differences by sub-groups.
The package relies on the 'ggplot2' package to produce the visualizations. As such, the corset plot allows for easy integration with 'ggplot2', so that users can customize their visualizations as required. This package is geared towards users with limited experience in R, creating corset plots using data in either wide or long format using the functions gg_corset() or gg_corset_elongated(), respectively.
The 'ggcorset' package allows for easy customization by offering different descriptive features by group. Specifically, 'eyelets' can be used to show either standard error means (SEM) or the mean +/- 1 standard deviation by group, and 'faceted' corsets can be used highlight the distributions of each group.
Install the package directly from CRAN using:
install.packages("ggcorset")
The latest version of this package can also be installed using install_github() from the 'devtools' package to download from this GitHub respository directly:
devtools::install_github("kbelisar/ggcorset")
Examples of corset plots in use, can be seen in the following publications:
Minhas, M., Belisario, K., González-Roz, A., Halladay, J., Murphy, J.G. and MacKillop, J. (2021), COVID-19 impacts on drinking and mental health in emerging adults: Longitudinal changes and moderation by economic disruption and sex. Alcohol Clin Exp Res, https://doi.org/10.1111/acer.14624
González-Roz, A., Haik, A. K., Rahman, L., Todi, A. A., Kane, C., Walji, A., Dickerman S. R., Scarfe, M., Levitt E. E., Belisario, K., Kelly, J. F., & MacKillop, J. (2022) Impacts of the COVID-19 public health restrictions on substance use, mental health, and psychosocial functioning among individuals with alcohol use disorder, The American Journal of Drug and Alcohol Abuse, 48:6, 712-723, https://doi.org/10.1080/00952990.2022.2134021
You, Y., Liu, J., Yao, Z., Zhang, S., Chen, K., & Ma, X. (2023). Neural mechanisms of long-term exercise intervention on cognitive performance among short-sleep young adults: a hemodynamic study. Sleep Medicine, 110, 7-16, https://doi.org/10.1016/j.sleep.2023.07.020
This function is used to create a corset plot with data in the wide format. It takes the following arguments:
data = the name of the data frame
y_var1 = the name of the measured variable at the first time point
y_var2 = the name of the measured variable at the second time point
group = the name of the units measured at each time point (such as id)
c_var = the name of the variable to visualize by line colour
eyelets = optional argument (defaults to FALSE) which creates either standard error means for the c_var (default) or 1 standard deviation above/ below the mean as defined by the 'e_type` argument
e_type = optional argument when eyelets is set to TRUE. One of "SE" (default) or "SD"
faceted = optional argument (defaults to FALSE) which facets corset plots by c_var
facet_design = defaults to "original" when faceted is set to TRUE. One of "original", "line", or "group"
vio_fill = optional argument to change the fill colour of the half violins (defaults to a soft black)
line_size = optional argument to change the size (thickness) of the lines (default is 0.25)
line_col = optional argument when facet_design is set to "line" to change the colour of the background group lines
line_dodge = optional argument to change the amount of vertical dodge of the lines (default is 0.1)
This function is used to create a corset plot with data in the long format. It takes the following arguments:
data = the name of the data frame
x_var = the name of the x-axis variable (time variable)
x_vals = the values representing the two time points (in chronological order)
y_var = the name of the repeated measure
group = the name of the units measured at each time point (such as id)
c_var = the name of the variable to visualize by line colour
eyelets = optional argument (defaults to FALSE) which creates standard error means for the c_var (default) or 1 standard deviation above/ below the mean as defined by the 'e_type` argument
e_type = optional argument when eyelets is set to TRUE. One of "SE" (default) or "SD"
faceted = optional argument (defaults to FALSE) which facets corset plots by c_var with soft grey lines denoting entire sample trajectories
facet_design = defaults to "original" when faceted is set to TRUE. One of "original", "line", or "group"
vio_fill = optional argument to change the fill colour of the half violins (defaults to a soft black)
line_size = optional argument to change the size (thickness) of the lines (default is 0.25)
line_col = optional argument when facet_design is set to "line" to change the colour of the background group lines
line_dodge = optional argument to change the amount of vertical dodge of the lines (default is 0.1)
This is a 'ggplot2' theme to quickly polish visualizations, and can be added to any ggplot2 object.
By creating a corset plot as an R object, 'ggplot2' customizations can easily be made. Here are a few changes to get started:
xlab() - Change the x-axis title
ylab() - Change the y-axis title
ggtitle() - Title of the plot
scale_x_discrete() - Change the labels of time1 and time2 (particularly helpful when using the gg_corset() function)
scale_colour_manual() - Custom colours as specified by the user (via values = argument)
scale_colour_gradientn() - Custom gradient of colours for continuous outcomes
scale_colour_viridis() - Custom colour-blind friendly colours for the c_var variable using the 'viridis' package
guides(colour = guide_legend(override.aes = list(linewidth = 3))) - Overrides thickness of legend so colours are more easily seen
The example data set included in this package (named 'drinkdays') is in wide format, and has been simulated with mock data. This data set consists of 300 individuals, with 1 individual per row, as identified by a distinct 'id'. The repreated measure is the number of drinking days per week, measured at 'time1' and 'time2'.
The variable 'change', is calculated by subtracting 'time1' from 'time2', and is used for the c_var argument:
library(ggcorset)
data("drinkdays")
#Calculating Change
drinkdays$change <- drinkdays$time2-drinkdays$time1
Since this data is in wide format, the function gg_corset() will be used to create the corset plot using the following code:
plot <- gg_corset(drinkdays, y_var1 = "time1", y_var2 = "time2",
group = "id", c_var = "change", eyelets = F) +
theme_ggcorset() # A ggcorset theme for polished corset plots!
plot # see the initial visualization
Changes to the corset plot can be made by using 'ggplot2':
library(ggplot2)
plot +
scale_colour_gradientn("Direction of Change",
colours = MetBrewer::met.brewer("Troy")) + # Changes legend title, and selects a colour-palette
ggtitle("Change in Drinking Days") + # Changes the plot title
ylab("Number of Drinking Days per Week") + # Changes the y-axis title
xlab("") + # Changes the x-axis title (removes in favour of the 2 time point labels below)
scale_x_discrete(labels = c("Pre","Post")) + # Changes the labels of the 2 time points (on the x-axis)
guides(colour = guide_legend(override.aes = list(linewidth = 3))) # Makes the legend lines thicker
Example Corset Plot:
This plot was solely created using the example data set and the code used in the example above:
Additional corset plot examples and uses can be found in detail in the CRAN vignette.