`ggscatite` package

Natalia da Silva, Ignacio Alvarez-Castro, Dianne Cook & Jayani P., Gamage.

Overview

ggscatite is an R package that extends ggplot2 to create bivariate jittered scatterplots. This package provides specialized functionality for adding controlled random noise in two dimensions, making it easier to visualize overlapping data points in scatterplots where both x and y variables may have discrete or semi-discrete values.

Description

When creating scatterplots with discrete or rounded data, points often overlap, making it difficult to assess the true density and distribution of observations. While base R and ggplot2 provide one-dimensional jittering (typically along one axis), ggscatite extends this concept to apply jittering simultaneously to both x and y coordinates.

Currently there are two implemented methods:

geom_jitter_gauss: Adds Bivariate Gaussian random noise
geom_jitter_quasi: Adds Quasi-random noise based on Sobolev sequences. If loc = TRUE a local sobol sequence is generated and if loc = FALSE the sequence is generated for the complete data set.

Installation

You can install the development version of ggscatie from GitHub:

# Install devtools if you haven't already
install.packages("devtools")

# Install ggscatite from GitHub
devtools::install_github("natydasilva/ggscatite")

Usage

Basic Example

ash	beg	count
1.5	1.5	5
3	2	3
3	4	2
4.5	5	1
5.5	6	1


data(dayles)

base <- ggplot(dayles, aes(x = ash, y = beg)) +
  geom_point(col = 'red', size = .8)

p1 <- base + geom_jitter() + labs(title = 'Jitter') + theme(aspect.ratio = 1)
p2 <- base + geom_jitter_gauss() + labs(title = 'Gaussian')  + theme(aspect.ratio = 1)
p3 <- base + geom_jitter_quasi(loc = FALSE) + labs(title = 'Sobol seq.')+ theme(aspect.ratio = 1)
p4 <- base + geom_jitter_quasi(loc = TRUE) + labs(title = 'Local Sobol seq.') + theme(aspect.ratio = 1)

(p1 + p2) / (p3 + p4)

Description of small example

mpg Example

library(ggplot2)
library(ggscatie)
library(patchwork)

data(mpg)

p <- mpg |> ggplot(aes(x = cty, y = hwy))

p0 <- p + geom_point() + theme(aspect.ratio = 1) + labs(title = 'Data')


p2 <- p +
  geom_jitter_gauss() +
  theme(aspect.ratio = 1) +
  labs(title = 'Gaussian')

p1 <- p + geom_jitter() + theme(aspect.ratio = 1) + labs(title = 'Jitter')

p3 <- p + geom_jitter_quasi(loc = FALSE) + theme(aspect.ratio = 1) + labs(title = 'Sobol seq.')

p4 <- p + geom_jitter_quasi(loc =TRUE) + theme(aspect.ratio = 1) + labs(title = 'Local Sobol seq.')
  


(p0 + p1) / (p2 + p3+p4)

Example ggscatie plots for mpg dataset

To Do list

Include other extensions: nonparametric instead of Gaussian

- kde did'n work as we expected
- should use hdr_2d() from hdrcde package

Work in the pkg documentation, webpage, sticker
Experiment to evaluate different methods
Find an interesting real data example

ggscatite package