In Deleetdk/kirkegaard: kirkegaard

knitr::opts_chunk$set(echo = TRUE)
library(kirkegaard); library(metafor); library(psych)

Introduction

The kirkegaard package contains a number of helper function for ggplot2. These are designed to save time, but do not genrally expand the capabilities of what a skilled ggplot2 user can do. As such, they do not constitute an extension. This documents gives some examples of the functions. All the functions begin with GG_ so they are easy to find.

GG_scatter

This is a convenience function to easily make scatterplots that add useful information such as the observed correlation in the plot and case names. To use it, give it a data.frame and the names of the two variables:

GG_scatter(iris, "Petal.Length", "Sepal.Width")

By default, the rownames are used as case names. We can turn this off with case_names = F:

GG_scatter(iris, "Petal.Length", "Sepal.Width", case_names = F)

The correlation, its confidence interval and the sample size is automatically shown in the corner where the data is least likely to be. One can control the position using text_pos:

GG_scatter(iris, "Petal.Length", "Sepal.Width", case_names = F, text_pos = "bl")

One can add weights which are automatically mapped to the size of the points using weights:

set.seed(1)
GG_scatter(iris, "Petal.Length", "Sepal.Width", case_names = F, weights = runif(150))

Note that the correlation is a weighted correlation and automatically uses the supplied weights as well.

If we want to use other case names, they can be supplied using case_names_vector:

set.seed(1)
GG_scatter(iris, "Petal.Length", "Sepal.Width", case_names_vector = sample(letters, replace = T, size = 150))

We can use another confidence interval by passing it to CI:

GG_scatter(iris, "Petal.Length", "Sepal.Width", case_names = F, CI = .99)

GG_denhist

It is frequently desired to make density or histograms of data distributions. GG_denhist makes both and combines them:

GG_denhist(iris, "Sepal.Length")

Currently, the y scale is nonsensical and only the relative differences are meaningful. In the future, the y scale will be the proportion.

A vertical one is automatically plotted for the mean. We can supply another function to vline if we want another kind of average:

GG_denhist(iris, "Sepal.Length", vline = median)

We can supply a groping variable if we want to compare groups:

GG_denhist(iris, "Sepal.Length", group = "Species")

GG_group_means

Examining group averages is a frequent task. GG_group_means makes this easier, supply a data.frame, the name of the data variable and the name of the grouping variable:

GG_group_means(iris, var = "Sepal.Length", groupvar = "Species")

There are a number of built in visualizations which can be controlled by type:

#bar (default)
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "bar")
#point
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "point")
#points
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "points")
#violin
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "violin")
#violin2
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "violin2")

Confidence intervals are automatically shown. These can be controlled with CI:

GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "violin", CI = .99999)

We can also supply subgroups using subgroup:

#make up some subgroup data
set.seed(1)
iris$letter = sample(letters[1:2], replace = T, size = 150)

#bar (default)
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "bar", subgroup = "letter")
#point
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "point", subgroup = "letter")
#points
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "points", subgroup = "letter")
#violin
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "violin", subgroup = "letter")
#violin2
GG_group_means(iris, var = "Sepal.Length", groupvar = "Species", type = "violin2", subgroup = "letter")

GG_kmeans

GG_kmeans is a simple function to perform k-means cluster analysis on a dataset and plot the results:

GG_kmeans(iris[1:4], clusters = 3)

GG_forest

The popular package metafor is used to meta-analyze data. However, the built in plotting functions are ugly and based on base-r graphics:

#load built in dataset about European genomic ancestry and socioeconomic outcomes across the Americas
data(european_ancestry)
#meta-analysis
meta = rma(yi = european_ancestry$r, sei = european_ancestry$SE_r)
#plot
forest(meta)

I really dislike base-r plots. To make a ggplot2 plot, simply call GG_forest on the analysis:

GG_forest(meta)

It doesn't seem that the names of the analyses are saved in the rma object, so we have to supply them to .names if they are desired:

GG_forest(meta, .names = european_ancestry$Author_sample)

GG_funnel

As with forest plots, metafor comes with a built in funnel plot:

funnel(meta)

GG_funnel is an intended ggplot2-based replacement of this:

GG_funnel(meta)

Outlying studies are automatically colored red. In the future, automatic labeling of (outlying/all) points will be added, but it can currently be done manually using standard ggplot2 functions:

#use ggrepel
library(ggrepel)
GG_funnel(meta) +
  ggrepel::geom_label_repel(data = european_ancestry, aes(r, SE_r, label = Author_sample), size = 2)

Deleetdk/kirkegaard documentation built on April 1, 2024, 2:23 a.m.