Data visualization

Day 4, A

Michael C Sachs

General graphics principles

1. Encode quantitative data as linear lengths or distances.

They are easier and more accurate to interpret than angles, area, volume, or color.

2. Place things that are meant to be compared next to each other.

3. If no comparison is relevant, sort by a quantitative variable

4. Show the raw data or some display of uncertainty rather than only summary statistics.

5. Carefully choose appropriate color schemes.

Be aware that 5 - 10% of the population is color blind, but color deficiencies or not, the scheme strongly affects the interpretation of the graphic.

Deciding what to plot is often more difficult than making the plot itself

Graphics in R

R has 2 systems for graphics generation:

  1. base graphics
    • “Just” draws things
    • Figures are created by overlaying drawings in one or more steps
    • Includes functions for “standard” statistical graphics
  2. The grid system
    • A low-level graphics system to create and arrange graphical output
    • ggplot2 and lattice are built on top of grid
    • ggplot2 uses a coherent grammar to describe figures, and gets grid to do the drawing

base graphics basics

  • The primitive types are points, lines, polygons, text, and raster images (aka bitmaps)
    • These are the basic drawing tools that are combined to create a figure
  • Graphical parameters are all documented in par. These include things like point shape, line type, color, margins, font, etc.
par(mar=rep(0.5, 4))  # small plot margins (bottom, left, top, right)
plot.new()  # start a new plot
plot.window(c(0, 6), c(0, 2), asp=1)  # x range: 0–6, y: 0–2; proportional
x <- c(0, 0, NA, 1, 2, 3, 4, 4, 5,    6)
y <- c(0, 2, NA, 2, 1, 2, 2, 1, 0.25, 0)
points(x[-(1:6)], y[-(1:6)])  # symbols
lines(x, y)   # line segments
text(c(0, 6), c(0, 2), c("(0, 0)", "(6, 2)"), col="red")  # two text labels
rasterImage(
    matrix(c(1, 0,  # 2x3 pixel "image"; 0=black, 1=white
             0, 1,
             0, 0), byrow=TRUE, ncol=2),
    5, 0.5, 6, 2,  # position: xleft, ybottom, xright, ytop
    interpolate=FALSE
)
polygon(
    c(4, 5, 5.5,    4),  # x coordinates of the vertices
    c(0, 0,   1, 0.75),  # y coordinates
    lty="dotted",   # border style
    col="#ffff0044"  # fill colour: semi-transparent yellow
)

Colors

You can refer to colors in several ways:

  • by name, e.g., "red", see colors()
  • by hex code: "#dd3333"
  • by specifying them in a color space with one of the functions: rgb, hsv, hcl.

More often, you want to choose a good color palette .

In base graphics, the palette() function is used to view and modify the current color palette.

The default is both ugly, and can be poorly perceived when data are mapped to color values.

Choosing a palette

Use the colorspace package to find a good palette to meet your needs. It uses the Hue-Chroma-Luminance colorspace

library(colorspace)
swatchplot(
  "Hue"       = sequential_hcl(5, h = c(0, 300), c = c(60, 60), l = 65),
  "Chroma"    = sequential_hcl(5, h = 0, c = c(100, 0), l = 65, rev = TRUE, power = 1),
  "Luminance" = sequential_hcl(5, h = 260, c = c(25, 25), l = c(25, 90), rev = TRUE, power = 1),
  off = 0
)

You can also use it to simulate color blindness

library(palmerpenguins)
par(mfrow = c(1, 2))
palette("R3")
plot(bill_length_mm ~ body_mass_g, 
     col = island, data = penguins, pch = 20, 
     main = "Default palette")
legend("bottomright", fill= palette(), legend = levels(penguins$island))
palette(deutan(palette()))
plot(bill_length_mm ~ body_mass_g, 
     col = island, data = penguins, pch = 20, 
     main = "Deuteranope")
legend("bottomright", fill= palette(), legend = levels(penguins$island))

ggplot2

GG stands for “Grammar of Graphics”, and this is actually the 2nd iteration of the package. Hadley Wickham started from scratch in 2005 with ggplot2, abandoning the original ggplot1

Grammar of Graphics

Introduced in the eponymous book by Leland Wilkinson, Hadley adapted a bit

Hadley Wickham. A layered grammar of graphics. Journal of Computational and Graphical Statistics, vol. 19, no. 1, pp. 3–28, 2010.

  • The main idea is to concisely describe a graphic using a set of fundamental rules and concepts
  • In the background, this also facilitates the creation of the graphic by the software

The building blocks:

  1. Data and aesthetic mappings (ggplot(data, aes(x = x, y = y)))
  2. Geometric objects (e.g., geom_point())
  3. Statistical transformations (e.g., stat_smooth())
  4. Scales
  5. Facets
  6. Coordinate systems

With ggplot2, we describe the building blocks, and combine them to construct a graphic

ggplot2 basics

  1. Start with the data, tidy data
library(ggplot2)

head(penguins)
# A tibble: 6 × 8
  species island    bill_length_mm bill_depth_mm flipper_length_mm body_mass_g
  <fct>   <fct>              <dbl>         <dbl>             <int>       <int>
1 Adelie  Torgersen           39.1          18.7               181        3750
2 Adelie  Torgersen           39.5          17.4               186        3800
3 Adelie  Torgersen           40.3          18                 195        3250
4 Adelie  Torgersen           NA            NA                  NA          NA
5 Adelie  Torgersen           36.7          19.3               193        3450
6 Adelie  Torgersen           39.3          20.6               190        3650
# ℹ 2 more variables: sex <fct>, year <int>
  1. Specify aesthetic mappings, i.e., how variables in the data are associated to visual elements (e.g., x position, y position, color, size, transparency, …)
step1 <- ggplot(penguins, 
                aes(x = body_mass_g, y = bill_length_mm, color = species))
step1

  1. Add geometric elements
step2 <- step1 + geom_point()
step2

  1. Adjust the scales
plot1 <- step2 + scale_color_brewer("Species", type = "qual")
plot1

Notes and details

  1. Layers/components of the plot are distinct functions
  2. Layers are combined with the +
  3. The plot can be saved as an object (what class?)
  4. The plot is displayed/rendered when printed

How does the adding of layers work?

pgeom <- geom_point()
class(pgeom)
[1] "LayerInstance" "Layer"         "ggproto"       "gg"           
class(step1)
[1] "gg"     "ggplot"
ggplot2:::`+.gg`
function (e1, e2) 
{
    if (missing(e2)) {
        abort("Cannot use `+.gg()` with a single argument. Did you accidentally put + on a new line?")
    }
    e2name <- deparse(substitute(e2))
    if (is.theme(e1)) 
        add_theme(e1, e2, e2name)
    else if (is.ggplot(e1)) 
        add_ggplot(e1, e2, e2name)
    else if (is.ggproto(e1)) {
        abort("Cannot add ggproto objects together. Did you forget to add this object to a ggplot object?")
    }
}
<bytecode: 0x62abfe3f5b90>
<environment: namespace:ggplot2>

More complex plots

Just keep adding layers:

plot1 + stat_smooth(method = "lm")

plot1 + stat_smooth(method = "lm") + 
  facet_grid(sex ~ year)

plot2 <- plot1 + stat_smooth(method = "lm") + 
  facet_grid(sex ~ year) + 
  ggtitle("A nice figure", subtitle = "made by ggplot2")
plot2

Changing how things look

  1. Change titles and axis labels with ggtitle(), xlab(), ylab(), limits with xlim(), ylim()
  2. Themes can be adjusted with theme(), changing the appearance of plot elements. The elements are all documented fairly well. There are some nice built-in themes like theme_bw(). Save your theme as an object for reuse.
  3. Annotations can be added with annotate(), but that does not play nicely with facets. Instead use geom_text()
mytheme <- theme(strip.background = element_rect(fill = "steelblue"), 
        text = element_text(family = "Comic sans"), 
        plot.background = element_rect(fill = "grey81"), 
        legend.background = element_rect(fill = NA), 
        legend.position = "bottom") 

plot2b <- plot2 + mytheme + 
  geom_text(data = data.frame(body_mass_g = 3000, 
                              bill_length_mm = 55, 
                              year = 2009, sex = NA,
                              label = "Some missing data here"), 
            aes(label = label, color = NULL), 
            hjust = 0)
plot2b

Tidy data is key

Most plotting problems are actually data problems.

Set up the thing you want to add to the plot in a tidy data frame. Then add the geometric element with the mappings.

plot2 +  
  geom_text(data = data.frame(body_mass_g = 3000, 
                              bill_length_mm = 55, 
                              year = 2009, sex = NA,
                              label = "Some missing data here"), 
            aes(label = label, color = NULL), 
            hjust = 0)

Another example

plot3 <- ggplot(penguins, aes(x = species, y = body_mass_g)) + geom_jitter()
plot3

These are alphabetical, how do I change the order?

Reordering things

Again, this is a data problem. The order is determined by the levels of the factor

class(penguins$species)
[1] "factor"
levels(as.factor(penguins$species))
[1] "Adelie"    "Chinstrap" "Gentoo"

When R converts a character to a factor, and by default the levels are determined by alphabetical order. Change the order using factor() or reorder()

penguins$species2 <- factor(penguins$species, 
                            levels = c("Chinstrap", "Gentoo", "Adelie"))

ggplot(penguins, aes(x = species2, y = body_mass_g)) + geom_jitter()

penguins$species2 <- reorder(penguins$species2, 
                             penguins$body_mass_g, 
                             min, na.rm = TRUE)
ggplot(penguins, aes(x = species2, y = body_mass_g)) + geom_jitter()

Challenge/exercise

Add solid horizontal lines at the mean and dashed horizontal lines at 95% confidence intervals for the mean

Solution

Step 1: assemble the data

Step 2: add to the plot using the desired geometric elements

Creating reusable elements

It is possible to make a function that adds layers to a plot.

Code 
ggmean_ci <- function(data, yname, groupname, buffer = .45) {
  
  means_and_cis <- split(data[[yname]], data[[groupname]]) |> 
  lapply(\(bm) {
    ttestbm <- t.test(bm)
    data.frame(mean = ttestbm[["estimate"]], 
               lower = ttestbm[["conf.int"]][1], 
               upper = ttestbm[["conf.int"]][2])
  })

  mcidf <- do.call("rbind.data.frame", means_and_cis)
  mcidf[[groupname]] <- as.factor(names(means_and_cis))

## get the x position with a buffer

  mcidf$xmin <- as.numeric(mcidf[[groupname]]) - buffer
  mcidf$xmax <- as.numeric(mcidf[[groupname]]) + buffer
  
  list(
  geom_linerange(data = mcidf, aes(xmin = xmin, 
                                   xmax = xmax, 
                                   y = mean, 
                                   color = .data[[groupname]]), linewidth = 1),
  geom_linerange(data = mcidf, aes(xmin = xmin, 
                                     xmax = xmax,
                                   y = lower, color = .data[[groupname]]), linetype = 2),
  geom_linerange(data = mcidf, aes(xmin = xmin, 
                                     xmax = xmax,
                                   y = upper, color = .data[[groupname]]), linetype = 2) 
  )

}
plot3 + ggmean_ci(penguins, "body_mass_g", "species")

ggplot(penguins |> subset(!is.na(sex)), aes(x = sex, y = bill_length_mm)) + 
  geom_jitter() + 
  ggmean_ci(penguins, "bill_length_mm", "sex")

Using ggplot2 in functions

  1. Note the use of .data[[[groupname]]. This says to use the layer data and not some other global object.
  2. Layers cannot be added by themselves without a call to ggplot(). This is why we return the elements in list. Themes are an exception to this.

If you want a function that takes a name or expression as a variable (instead of a string), you can do the following:

ggboxplot <- function(data, yvar, xvar) {
  
  ggplot(data, aes(x = {{ xvar }}, y = {{ yvar }})) + 
    geom_jitter() + 
    geom_boxplot(fill = NA)
  
}

ggboxplot(penguins |> subset(!is.na(sex)), bill_length_mm, sex)

ggboxplot(penguins |> subset(!is.na(sex)), sqrt(bill_length_mm), sex)

Combining figures

Use the patchwork package

library(patchwork)

(plot1 + plot3) / plot2

There are many options for layouts and adding different elements

Interactive graphics

Why

  • Animation/interactivity should only be used when it enhances understanding, condenses information, makes something more accessible
  • Sometimes it is only distracting

What

Interesting applications:

  • Summarizing complex data/results of analysis
  • Illustrating statistical concepts for education
  • Let readers play the what-if? game

How

  • One easy tool is the plotly package

Example

library(plotly)

ggplotly(plot1 + facet_wrap(~ year))

Summary

  1. Remember tidy data
    • Keep data defining groups you want to compare (colors, facets) as variables
    • Distinct geometric elements can be separated into different columns or different datasets
  2. Choose good color scales
    • colorspace is a great resource for this
    • qualitative (discrete data), sequential and diverging (continuous data)
  3. Save images at the correct size and format
  4. Create reusable elements
    • Functions for special types of plots
    • Themes and groups of geoms as lists
  5. Define your own Stats and Geoms
    • See “Extending ggplot2” vignette

Try it yourself

Practical

Link to lesson

Link home