Figure 2 of Hofer, Langmann, Burkart and Neubauer, 2022.
In their study, Hofer, Langmann, Burkart, & Neubauer (2022) examined who is the best judges of one’s abilities. Examining self-ratings vs. other-ratings in six domain, they found out that we are not always the best judges. They present in their Figure 2 a rain-cloud plot (Allen, Poggiali, Whitaker, Marshall, & Kievit (2019)) illustrating the ratings.
In what follow, we discuss how this plot could be customized after
its initial creation with superb.
As the six domains are within-subject ratings, the data must be composed of 6 columns (at least, there can be additional columns; they won’t be illustrated herein). In case you do not have such data, the following subsection generates mock data.
Generating mock data
We generate two sets of mock data from six sets of means and standard deviations:
Astats <- data.frame(
MNs = c(6.75, 6.00, 5.50, 6.50, 8.00, 8.75),
SDs = c(2.00, 3.00, 3.50, 3.50, 1.25, 1.25)
)
dtaA <- apply(Astats, 1,
function(stat) {rnorm(100, mean=stat[1], sd=stat[2])}
)
dtaA <- data.frame(dtaA)
colnames(dtaA) <- c("Verbal", "Numerical", "Spatial", "Creativity", "Intrapersonal", "Interpersonal")
Bstats <- data.frame(
MNs = c(3.33, 3.00, 2.50, 3.00, 2.75, 3.50),
SDs = c(0.25, 0.50, 0.66, 0.50, 0.25, 0.25)
)
dtaB <- apply(Bstats, 1,
function(stat) {rnorm(100, mean=stat[1], sd=stat[2])}
)
dtaB <- data.frame(dtaB)
colnames(dtaB) <- c("Verbal", "Numerical", "Spatial", "Creativity", "Intrapersonal", "Interpersonal")The datasets are data.frames called dtaA
and dtaB. Their columns names are the dependent variables,
e.g., “Verbal”, “Numerical”, “Spatial”, “Creativity”, “Intrapersonal”,
“Interpersonal”.
Making the top-row plot
For convenience, we make lists of the desired colors and labels we want to appear on the x-axis:
mycolors <- c("seagreen","chocolate2","mediumpurple3","deeppink","chartreuse4", "darkgoldenrod1")
mylabels <- c("Verbal", "Numerical", "Spatial", "Creativity", "Intrapersonal", "Interpersonal")We are ready to make the plot with the desired adjustments:
pltA <- superb(
crange(Verbal, Interpersonal) ~ ., # no between-subject factors
dtaA, # plot for the first data set...
WSFactors = "Domain(6)", # ...a within-subject design with 6 levels
adjustments = list(
purpose = "difference", # we want to compare means
decorrelation = "CM" # and error bars are correlated-adjusted
),
plotStyle="raincloud",
# the following (optional) arguments are adjusting some of the visuals
pointParams = list(size = 0.75),
jitterParams = list(width =0.1, shape=21,size=0.05,alpha=1), # less dispersed jitter dots,
violinParams = list(trim=TRUE, alpha=1), # not transparent,
errorbarParams = list(width = 0.1, linewidth=0.5) # wider bars, thicker lines.
)
pltA
As seen, this plot is a standard, colorless, plot. It contains all that is needed; it is just plain drab and the labels are generic ones (on the vertical axis and on the horizontal axis).
Adding a color layer to the plot
Using superb, if there is only one factor, superb will
consider that it is the one on the x-axis and there is therefore no
other layers in the plot. This is why the current plot is colorless.
It is possible, post-hoc, to indicate that we wish additional layers in the plot.
In the present, we want to add the fill and the
color of dots layers. These layers are to be “connected” to
the sole factor in the present example (that is, Domain).
Consequently, the x-axis labels, the fill color and the dot color are
all redondant information identifying the condition.
To do this, simply add an aesthetic graphic directive to
pltA with:
Adding graphic directives for fine-tuning the plot
We can customize any superb plot by adding graphic
directives one-by-one using the operator +, or we can
collect all the directives in a list, and add this list once. As we have
two plots with mostly the same directives, we use this second
approach.
Typically, a plot is customized by picking a theme. The default
theme_bw() is grayish, so we move to
theme_classic(). We also customize specific aspects of this
theme with theme() directives.
These changes are all collected within the list
commonstyle below:
commonstyle <- list(
theme_classic(), # It has no background, no bounding box.
# We customize this theme further:
theme(axis.line=element_line(linewidth=0.50), # We make the axes thicker...
axis.text = element_text(size = 10), # their text bigger...
axis.title = element_text(size = 12), # their labels bigger...
plot.title = element_text(size = 10), # and the title bigger as well.
panel.grid = element_blank(), # We remove the grid lines
legend.position = "none" # ... and we hide the side legend.
),
# Finally, we place tick marks on the units
scale_y_continuous( breaks=1:10 ),
# set the labels to be displayed
scale_x_discrete(name="Domain", labels = mylabels),
# and set colours to both colour and fill layers
scale_discrete_manual(aesthetic =c("fill","colour"), values = mycolors)
)We also changed the vertical scale (tick marks at designated
positions) and the horizontal scale with names on the tick marks (sadly,
superb replaces them with consecutive numbers…) and colors
to fill the clouds (fill) and their borders
(colour) as well as the rain drop colors.
Examining this plot with the commonstyle added, we
get
finalpltA <- pltA + aes(fill = factor(Domain), colour = factor(Domain)) +
commonstyle + # all the above directive are added;
coord_cartesian( ylim = c(1,10) ) + # the y-axis bounds are given ;
labs(title="A") + # the plot is labeled "A"...
ylab("Self-worth relevance") # and the y-axis label given.
finalpltA
Making the second row of the figure
We do exactly the same for the second plot. We just change the data
set to dtaB and in the last graphic directives, using
options tailored specifically to this second data set (smaller y-axis
range, different label, etc.):
pltB <- superb(
crange(Verbal, Interpersonal) ~ ., # no between-subject factors
dtaB, # the second data set...
WSFactors = "Domain(6)", # ...a within-subject design with 6 levels
adjustments = list(
purpose = "difference", # we want to compare means
decorrelation = "CM" # and error bars are correlated-adjusted
),
plotStyle="raincloud",
# the following (optional) arguments are adjusting some of the visuals
pointParams = list(size = 0.75),
jitterParams = list(width =0.1, shape=21,size=0.05,alpha=1), # less dispersed jitter dots,
violinParams = list(trim=TRUE, alpha=1,adjust=3), # not semi-transparent, smoother
errorbarParams = list(width = 0.1, linewidth=0.5) # wider bars, thicker lines.
)
finalpltB <- pltB + aes(fill = factor(Domain), colour = factor(Domain)) +
commonstyle + # the following three lines are the differences:
coord_cartesian( ylim = c(1,5) ) + # the limits, 1 to 5, are different
labs(title="B") + # the plot is differently-labeled
ylab("Judgment certainty") # and the y-axis label differns.
finalpltB
Combining and saving both plots
Finally, we assemble the two plots together
It can be saved with high-resolution if desired with
ggsave( "Figure2.png",
plot=finalplt,
device = "png",
dpi = 320, # pixels per inche
units = "cm", # or "in" for dimensions in inches
width = 17, # as found in the article
height = 13
)That’s it!
