The Heiss plot, developed by the statistician Andrew Heiss, is a plot of the predicted proportions of components on a bounded scale that are grouped by the unique levels of a grouping variable or factor (such as a random effect) in the model. The plot excels at showing how the scale components--that is, the bottom, middle continuous, and top ends of the scale--vary with a discrete variable while also capturing posterior uncertainty. This plot was the winner of the 2023 ordbetareg Visualization Prize.
plot_heiss(
object,
grouping_fac = NULL,
recode_group_labels = NULL,
ndraws = NULL,
show_category_perc_labels = TRUE,
category_label_font_size = 3,
category_label_accuracy = 1,
strip_text_font = element_text(face = "plain", size = 9),
plot_title = "Predicted Proportions of Bounded Scale Components",
plot_subtitle = paste0("By Unique Values of ", grouping_fac),
plot_caption = NULL,
plot_caption_width = 70,
calc_func = mean,
lb = 0.05,
upb = 0.95,
plot_font_size = 11,
plot_font = "",
y_axis_label = "Predicted Proportions",
legend_name = "Scale Components",
component_colors = c("#ef8737", "#bb292c", "#62205f"),
component_labels = c("0", "(0-1)", "1"),
...
)A ggplot2 object representing the predicted proportions of the components.
A fitted ordbetareg() model object.
A character string indicating the name of the discrete column in the data used for grouping predictions. Must be a valid column name that was passed to ordbetareg().
Optional. A character vector of new labels for the grouping factor levels. Must match the number and order of unique levels/values in grouping_fac.
Optional. The number of posterior draws to use for predictions. If NULL, all available draws are used.
Logical. Whether to display category percentage labels on the plot. Defaults to TRUE.
The ggplot2 font size for the labels on the
scale components (if show_category_perc_labels is TRUE). Defaults to 3.
The accuracy, or amount of rounding,
for component label ranges on the plot (if show_category_perc_labels is TRUE).
Default is 1. See scales::label_percent() for more info on meaning of
accuracy parameter.
A ggplot2::element_text object defining the font style for facet strip text. Defaults to element_text(face = "plain", size = 9).
Title of the plot. Defaults to "Predicted Proportions of Bounded Scale Components".
Subtitle of the plot. Defaults to a message indicating the grouping variable.
Caption text for the plot. If NULL, the default, will use a detailed but static description of the plot contents.
Width (in characters) at which the caption is wrapped. Defaults to 60.
A function used to calculate the central tendency of predictions. Defaults to mean.
Lower bound for uncertainty intervals. Defaults to 0.05 (5th percentile).
Upper bound for uncertainty intervals. Defaults to 0.95 (95th percentile).
Base font size for the plot. Defaults to 11.
Base font family for the plot. Defaults to an empty string (uses system default).
Label for the y-axis. Defaults to "Predicted Proportions".
Legend title. Defaults to "Scale Components".
A character vector of colors for the plot components (bottom, continuous, top). Defaults to c("#ef8737", "#bb292c", "#62205f").
A character vector of labels for the scale/outcome components (bottom, continuous, top). Defaults to c("0", "(0-1)", "1").
Additional arguments passed to [posterior_epred_ordbeta())].
[posterior_epred_ordbeta())]: R:posterior_epred_ordbeta())
For more details of the plot, see:
Heiss, Andrew and Ye, Meng. "Enforcing Boundaries: China's Overseas NGO Law and Operational Constraints for Global Civil Society." Working Paper, 2023. https://stats.andrewheiss.com/compassionate-clam/notebook/manuscript.html.
# Load a fitted model object and create a plot for
# distinct values of the factor education
#
# data('ord_fit_mean')
#
# plot_heiss(ord_fit_mean,ndraws=100)
#
# See introductory package vignette for more information on function options
Run the code above in your browser using DataLab