6.1 Binwidths

Always change the binwidths on a histogram. Sometimes the default in ggplot works great, sometimes it does not.

Super narrow:

#2 x 2 histograms in ggplot
ggplot(four, aes(x = vals)) + 
  geom_histogram(binwidth = .1, fill = "white", colour = "black") + #super narrow bins
  coord_cartesian(xlim = c(-5, 5)) + 
  facet_wrap(~ dist)

Super wide:

#2 x 2 histograms in ggplot
ggplot(four, aes(x = vals)) + 
  geom_histogram(binwidth = 2, fill = "white", colour = "black") + #super wide bins
  coord_cartesian(xlim = c(-5, 5)) + 
  facet_wrap(~ dist)

Just right? Pretty close to the default for this data.

#2 x 2 histograms in ggplot
ggplot(four, aes(x = vals)) + 
  geom_histogram(binwidth = .2, fill = "white", colour = "black") + 
  coord_cartesian(xlim = c(-5, 5)) + 
  facet_wrap(~ dist)

6.2 Add a rug

#2 x 2 histograms in ggplot
ggplot(four, aes(x = vals)) + 
  geom_histogram(binwidth = .2, fill = "white", colour = "black") + 
  geom_rug() + 
  coord_cartesian(xlim = c(-5, 5)) + 
  facet_wrap(~ dist)

7.1 Add notches

ggplot notches: “Notches are used to compare groups; if the notches of two boxes do not overlap, this is strong evidence that the medians differ.” (Chambers et al., 1983, p. 62)

ggplot(four, aes(y = vals, x = dist)) + 
  geom_boxplot(notch = T) + 
  scale_x_discrete(name = "") + 
  scale_y_continuous(name = "") + 
  coord_cartesian(ylim = c(-4,4))

7.2 Add summary statistics

Here we add a diamond for the mean (see other possible shape codes here).

ggplot(four, aes(x = dist, y = vals)) + 
  geom_boxplot() +
  stat_summary(fun.y = mean, 
               geom = "point", 
               shape = 18, 
               size = 4, 
               colour = "lightseagreen") + 
  coord_cartesian(ylim = c(-4, 4))

8.1 Stripchart

Combining geom_jitter() + stat_summary() is the ggplot corollary to a stripchart.

ggplot(four, aes(x = dist, y = vals)) +
  geom_jitter(position = position_jitter(height = 0, width = .1), 
              fill = "lightseagreen", 
              colour = "lightseagreen",
              alpha = .5) + 
  stat_summary(fun.y = median, 
               fun.ymin = median, 
               fun.ymax = median, 
               geom = "crossbar", 
               width = 0.5) +
  scale_x_discrete(name = "") + 
  scale_y_continuous(name = "") + 
  coord_cartesian(ylim = c(-4, 4))

8.2 Dotplot

This is a beeswarm-like ggplot- not exactly the same, but gives you the same idea.

ggplot(four, aes(x = dist, y = vals)) +
  geom_dotplot(stackdir = "center", 
               binaxis = "y", 
               binwidth = .1,
               binpositions = "all",
               stackratio = 1.5, 
               fill = "lightseagreen", 
               colour = "lightseagreen") + 
  scale_x_discrete(name = "") + 
  scale_y_continuous(name = "") + 
  coord_cartesian(ylim = c(-4, 4))

8.3 Beeswarm

https://github.com/eclarke/ggbeeswarm

install.packages("ggbeeswarm")
library(ggbeeswarm)

ggplot(four, aes(x = dist, y = vals)) +
  geom_quasirandom(fill = "lightseagreen", 
               colour = "lightseagreen") + 
  scale_x_discrete(name = "") + 
  scale_y_continuous(name = "") + 
  coord_cartesian(ylim = c(-4, 4))

ggplot(four, aes(x = dist, y = vals)) +
  geom_quasirandom(fill = "lightseagreen", 
               colour = "lightseagreen", 
               method = "smiley") + 
  scale_x_discrete(name = "") + 
  scale_y_continuous(name = "") + 
  coord_cartesian(ylim = c(-4, 4))

Note that these recommendations do not apply if your data is “big”. You will know your data is too big if you try the below methods and you can’t see many of the individual points (typically, N > 100).

10.1 Density plots

ggplot(four, aes(x = vals)) + 
  geom_density(fill = "lightseagreen") +
  coord_cartesian(xlim = c(-5, 5)) +
  facet_wrap(~ dist)

Instead of doing a facet_wrap, I could make just one plot showing all four distributions. To make each distribution a different color, set the fill within the aes, and assign it to the factor variable dist. Since now all four will be plotted on top of each other, add an alpha level to make the color fill transparent (0 = completely transparent; 1 = completely opaque).

# Density plots with semi-transparent fill
ggplot(four, aes(x = vals, fill = dist)) + 
  geom_density(alpha = .5)

ggplot(four, aes(x = vals)) + 
  geom_histogram(aes(y = ..density..), 
                 binwidth = .5, 
                 colour = "black", 
                 fill = "white") +
  geom_density(alpha = .5, fill = "lightseagreen") + 
  coord_cartesian(xlim = c(-5,5)) + 
  facet_wrap(~ dist)

10.2 Violin plots

My advice: always set color = NA for geom_violin. For fill, always set alpha.

ggplot(four, aes(x = dist, y = vals)) +
    geom_violin(color = NA,
                fill = "lightseagreen",
                alpha = .5,
                na.rm = TRUE,
                scale = "count") + # max width proportional to sample size
  coord_cartesian(ylim = c(-4, 4))

ggplot(four, aes(x = dist, y = vals)) +
    geom_boxplot(outlier.size = 2, 
                 colour = "lightseagreen",
                 fill = "black",
                 na.rm = TRUE,
                 width = .1) + 
    geom_violin(alpha = .2, 
                fill = "lightseagreen",
                colour = NA,
                na.rm = TRUE) +
  coord_cartesian(ylim = c(-4, 4))

14.1 Means and error bars

The simplest summary function is mean_se, which returns the mean and the mean plus its standard error on each side. Thus, stat_summary will calculate and plot the mean and standard errors for the y variable at each x value.

The default geom is “pointrange” which places a dot at the central y value and extends lines to the minimum and maximum y values. Other geoms you might consider to display summarized data:

• geom_errorbar
• geom_pointrange
• geom_linerange
• geom_crossbar

There are a few summary functions from the Hmisc package which are reformatted for use in stat_summary(). They all return aesthetics for y, ymax, and ymin.

• mean_cl_normal()
  • Returns sample mean and 95% confidence intervals assuming normality (i.e., t-distribution based)
• mean_sdl()
  • Returns sample mean and a confidence interval based on the standard deviation times some constant
• mean_cl_boot()
  • Uses a bootstrap method to determine a confidence interval for the sample mean without assuming normality.
• median_hilow()
  • Returns the median and an upper and lower quantiles.

ggplot(four, aes(x = dist, y = vals)) +
  stat_summary(fun.data = "mean_se")

ggplot(four, aes(x = dist, y = vals)) +
  stat_summary(fun.y = "mean", geom = "point") +
  stat_summary(fun.y = "max", geom = "point", shape = 21)

ggplot(four, aes(x = dist, y = vals)) +
  stat_summary(fun.data = median_hilow)

You may have noticed two different arguments that are potentially confusing: fun.data and fun.y. If the function returns three values, specify the function with the argument fun.data. If the function returns one value, specify fun.y. See below.

x <- c(1, 2, 3)
mean(x) # use fun.y

[1] 2

mean_cl_normal(x) # use fun.data

  y       ymin     ymax
1 2 -0.4841377 4.484138

Confidence limits may give us a better idea than standard error limits of whether two means would be deemed statistically different when modeling, so we frequently use mean_cl_normal or mean_cl_boot in addition to mean_se.

14.2 Connecting means with lines

Using the ToothGrowth dataset

data(ToothGrowth)
tg <- ToothGrowth

# Standard error of the mean
ggplot(tg, aes(x = dose, y = len, colour = supp)) + 
  stat_summary(fun.data = "mean_se") +
  ggtitle("Mean +/- SE")

# Connect the points with lines
ggplot(tg, aes(x = dose, y = len, colour = supp)) + 
  stat_summary(fun.data = "mean_se") +
  stat_summary(fun.y = mean, geom = "line") +
  ggtitle("Mean +/- SE")

# Use 95% confidence interval instead of SEM
ggplot(tg, aes(x = dose, y = len, colour = supp)) + 
  stat_summary(fun.data = "mean_cl_normal") +
  stat_summary(fun.y = mean, geom = "line") +
  ggtitle("Mean with 95% confidence intervals")

# The errorbars overlapped, so use position_dodge to move them horizontally
pd <- position_dodge(0.1) # move them .05 to the left and right

ggplot(tg, aes(x = dose, y = len, colour = supp)) + 
  stat_summary(fun.data = "mean_cl_normal", position = pd) +
  stat_summary(fun.y = mean, geom = "line", position = pd) +
  ggtitle("Mean with 95% confidence intervals")

Not the best example for this geom, but another good one for showing variability…

# ribbon geom
ggplot(tg, aes(x = dose, y = len, colour = supp, fill = supp)) + 
  stat_summary(fun.y = mean, geom = "line") +
  stat_summary(fun.data = mean_se, geom = "ribbon", alpha = .5)

14.3 Bars with error bars

If you must…

# Standard error of the mean; note positioning
ggplot(tg, aes(x = factor(dose), y = len, fill = supp)) + 
  stat_summary(fun.y = mean, geom = "bar", position = position_dodge(width = .9)) +
  stat_summary(fun.data = mean_se, geom = "linerange", position = position_dodge(width = .9)) +
  ggtitle("Mean +/- SE")

# Use 95% confidence interval instead of SEM
ggplot(tg, aes(x = factor(dose), y = len, fill = supp)) + 
  stat_summary(fun.y = mean, geom = "bar", position = position_dodge(width = .9)) +
  stat_summary(fun.data = mean_cl_boot, geom = "linerange", position = position_dodge(width = .9)) +
  ggtitle("Mean with 95% confidence intervals")

More help here: http://www.cookbook-r.com/Graphs/Plotting_means_and_error_bars_(ggplot2)/