https://serialmentor.com/dataviz/proportional-ink.html

• We are visual beings
• Graphics are effective forms of storytelling
• Statistics can be deceptive when presented alone
  • For example, Anscombe’s Quartet:

Grammar of Graphics—Wilkinson (1999)

A framework for describing components of a figure and the structure that underlies statistical graphics.

• DATA — variables from datasets
• TRANSformations — variable transformations
• FRAME — a set of variables, combined with operators, defining a space
• SCALE — scale transformations
• COORDinates — a coordinate system
• GRAPH — graphs and their aesthetic attributes
• GUIDE — axes, legends

Hadley Wickham built on Wilkinson’s grammatical framework, describing a layered system that:

• Has a different arrangement of grammatical components
• Has a hierarchy of defaults
• Can be easily implemented in a programming language (in our case R)

https://byrneslab.net/classes/biol607/readings/wickham_layered-grammar.pdf

Most commonly as a data frame (ggplot requires this structure)

data(anscombe) # let's continue with the Anscombe example
library(stargazer)
stargazer(anscombe, type = "html", summary = FALSE)

Every aspect of a given graphic element

library(ggplot2)
ggplot(data = anscombe, aes(x = x1, y = y1)) + geom_blank()

The mapping between data values and aesthetic values

We can scale any aesthetic component to our liking based on our data:

• Position: X/Y axes limits and breaks? Continuous? Discrete?
  • e.g., scale_y_discrete(), scale_y_continuous(), scale_y_date()
• Shape: what shapes?
  • e.g., scale_shape()
• Size: what sizes?
  • e.g., scale_size()
• Colors: which colors? Continuous? Discrete?
  • e.g., scale_color_manual(), scale_fill_manual()
• Line width/type: what widths, what types?
  • e.g., scale_size(), scale_linetype()

library(ggplot2)
ggplot(data = anscombe, aes(x = x1, y = y1)) + geom_blank() +
  scale_x_continuous(limits = c(0,20), breaks = seq(0,20,5)) +
  scale_y_continuous(limits = c(0,20), breaks = seq(0,20,5))

What physical features we actually put in the plot area.

The most vocabulary-rich layer.

• Points
  • geom_point()
• Lines
  • geom_line(), geom_hline(), geom_vline(), geom_abline(), geom_contour(), geom_path(), geom_segment()
• Bars
  • geom_bar() or geom_col()
• Polygons
  • geom_polygon(), geom_rect(), etc.
• Tiles
  • geom_tile(), geom_rect(), geom_raster(), etc.
• Text, annotations
  • geom_text(), geom_label(), annotate()

ggplot(data = anscombe, aes(x = x1, y = y1)) + 
  geom_point()

Geoms with statistical properties.

• Boxplots
  • geom_boxplot()
• Histograms
  • geom_hist()
• Distributions
  • geom_density(), geom_rug(), geom_violin()
• Error
  • geom_errorbar(), geom_ribbon(), stat_summary()
• Rugs
  • geom_rug()
• Other statistical summaries
  • geom_smooth(), stat_summary()

lm(y1~x1, data = anscombe)

## 
## Call:
## lm(formula = y1 ~ x1, data = anscombe)
## 
## Coefficients:
## (Intercept)           x1  
##      3.0001       0.5001

ggplot(data = anscombe, aes(x = x1, y = y1)) +
  geom_point() +
  geom_abline(intercept = 3.0001, slope = 0.5001)

Small multiples!

Going beyond two dimensions.

• Small exercise: how many dimensions can you plot at once?

  • Answer: You can easily plot SEVEN dimensions in R and still have a reasonably clear figure if you include two axes (X, Y), two facet structures, one color aesthetic, one shape aesthetic, and one size aesthetic.

  • Eight dimensions if you do all of the above with some time animation (ggannimate).

  • Nine dimensions if you do all that add a third axis (e.g., a surface plot with a Z variable), but doubtful this will be useful to many people.

  • Ten dimensions if you use the secret spacetime() function in developer mode to distort general relativity in your data visualization and open a portal to graphics hell.

How can we reproduce Anscombe’s original quartet figure?

# first we need to reshape the data
anscombe.quartet <- data.frame(X = c(anscombe$x1, anscombe$x2, anscombe$x3, anscombe$x4),
                               Y = c(anscombe$y1, anscombe$y2, anscombe$y3, anscombe$y4),
                               dataset = c(rep("I", 11), rep("II", 11), 
                                           rep("III", 11), rep("IV", 11)))
head(anscombe.quartet)

##    X    Y dataset
## 1 10 8.04       I
## 2  8 6.95       I
## 3 13 7.58       I
## 4  9 8.81       I
## 5 11 8.33       I
## 6 14 9.96       I

Small multiples with facet_wrap() or facet_grid()

ggplot(data = anscombe.quartet, aes(x = X, y = Y)) +
  geom_point(color = "midnightblue") +
  geom_abline(intercept = 3.0001, slope = 0.5001) +
  facet_wrap(~dataset)

Set of position scales combined with their geometrical arrangement

• Cartesian: 2 dimensional (X and Y), linear
  • coord_cartesian() (default), coord_equal(), coord_fixed()
• Nonlinear (e.g., a log transformed axis)
  • coord_trans(), coord_munch()
• Curved (e.g., polar coordinate system) - good for periodic data, geospatial data
  • coord_polar()

Yes, that includes spatial data (we can spend a whole day on spatial data).

To include or not?

• When is it redundant, when is it useful?

• Where to put them?

• Not always necessary to have a standalone legend (for example, below).

In ggplot we can modify virtually any thematic plot component with theme() and labs()

• Axes, ticks, background grids—theme()

• Axis labels, plot title—labs()

newthemes <- 
  ggplot(data = anscombe.quartet, aes(x = X, y = Y)) +
  geom_point(color = "midnightblue") +
  geom_abline(intercept = 3.0001, slope = 0.5001) +
  facet_wrap(.~dataset) + 
  labs(x = "new x-axis label", y = "new y-axis label", title = "Fancy title") +
  theme(axis.title = element_text(size = 16),
        axis.text = element_text(size = 14))

newthemes

We can also use some pre-installed themes from ggplot and other packages like ggthemes.

theme_gray() (default)

default <- 
  ggplot(data = anscombe.quartet, aes(x = X, y = Y)) +
  geom_point(color = "midnightblue") +
  geom_abline(intercept = 3.0001, slope = 0.5001) +
  facet_wrap(.~dataset) + ggtitle("default") + theme_gray()

default

Other ggplot2 pre-installed themes.

bw <- default + theme_bw() + ggtitle("theme_bw()")
classic <- default + theme_classic() + ggtitle("theme_classic()")
minimal <- default + theme_minimal() + ggtitle("theme_minimal()")

library(ggpubr)
theme.plots <- ggarrange(default, bw, classic, minimal, nrow = 2, ncol = 2)

theme.plots

Some themes from the ggthemes package.

library(ggthemes)
economist <- default + theme_economist() + ggtitle("theme_economist()")
wsj <- default + theme_wsj() + ggtitle("theme_wsj()")
fivethirtyeight <- default + theme_fivethirtyeight() + ggtitle("theme_fivethirtyeight()") 
tufte <- default + theme_tufte() + ggtitle("theme_tufte()")

ggthemes.plots <- ggarrange(economist, wsj, fivethirtyeight, tufte, 
                            nrow = 2, ncol = 2, align = "hv")

ggthemes.plots

You can even make an xkcd style figure:

days <- seq(0,10,1)
stench <- days - 2.31*I(days^2) + 0.15*I(days^3) + 11*days
library(xkcd)
library(extrafont)

stench.plot <- 
  ggplot() + geom_line(aes(x = days, y = stench), color = "steelblue", lwd = 2) + 
  xkcdaxis(xrange = c(0,10), yrange = range(stench)) + 
  labs(x = "Days since last shower", y = "Stench", title = "The stench curve") +
  geom_segment(data = NULL, aes(x = 8, y = 25, xend = 8, yend = 40), lty = "dashed") +
  geom_segment(data = NULL, aes(x = 8, y = 25, xend = 10, yend = 25), lty = "dashed") +
  xkcdline(aes(x=4.5,y=22,xend=4.5,yend=24), data = NULL, xjitteramount = 0.12) +
  xkcdline(aes(x=9,y=24,xend=9,yend=20), data = NULL, xjitteramount = 0.12) +
  scale_x_continuous(breaks = 0:10) +
  annotate("text", x = 4.5, y = 26, label = "poop plateau", family="xkcd") +
  annotate("text", x = 9, y = 18, label = "point of no return", family="xkcd") +
  theme(text=element_text(size=16, family="xkcd"),
        axis.text.y = element_blank(), axis.ticks.y = element_blank()) 

stench.plot

The empirically-driven hiker stench curve (Goldspiel & Fuller, 2018).

http://www.stat.columbia.edu/~tzheng/files/Rcolor.pdf

• How to use effectively

• Opacity is everything (alpha = ...)

• Colorblind considerations

  • ColorBrewer: http://colorbrewer2.org/

• Essentials:
  • Scico
  • Viridis
• Other fun ones
  • Wes Anderson
  • Ghibli
  • Lisa
• The master reference
  • R color palette aggregator

• Or make your own palette with RColorBrewer!

library(RColorBrewer) 
display.brewer.all(colorblindFriendly = TRUE) # colorblind friendly palettes

# the good old iris data (default colors)
def <- ggplot(iris, aes(Species, Sepal.Length)) + geom_boxplot(aes(fill = Species)) + 
  theme_minimal() + theme(legend.position = "top") + ggtitle("default")
# max ernst colors
library(lisa)
dali <- def + scale_fill_manual(values = lisa$MaxErnst) + ggtitle("Woman, Old Man, and Flower")
ggarrange(def, dali, ncol = 2)

• Only including elements you need to convey your data, and nothing more

• “Data-ink” versus “chartjunk”

• Edward Tufte, the “data-ink ratio”

Spagetti disaster.

• Things to avoid:
  • Gratuitious use of colors
  • Overuse of precision (e.g., 3.1415926535897932384626433…)
    • Don’t go beyond three decimal places
    • Be careful with number of axis ticks and labels

Climate trends in Galapagos.

Graphs are not self-promoting. They are used to convey quantitative information, not stylistic information. Effective graphs can (and should) be pretty, but focus should always be on function, not decoration. Decorative graphs are “ducks.”

• Don’t make a duck.

• When do you actually need bars?

  • Answer: Never. Bars are data summaries, and you can more efficiently (and effectively) convey data summaries with points and supplementary geoms to visualize distributions and uncertainty.

• Not distorting axes

• Bergstrom & West (2016) “The theory of proportional ink”: using responsible sizes for shaded areas (e.g., bars, polygons). In a visualization, size should be proportional to the nature of the actual data.

general figure tips

https://socviz.co/refineplots.html

a piece by piece demo of building a nuanced plot in ggplot

https://www.kdnuggets.com/2019/07/evolution-ggplot.html

sthda—great ggplot2 tutortials

http://www.sthda.com/english/wiki/data-visualization

10 rules for better figures

https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833

The ultimate guide: https://www.data-to-viz.com/

Let’s go through this guide and show some examples of how to make these figures in R.