Visualizations

Introduction

• This presentation will cover two topics:

1. A few principles for making effective choices in the visualizations you construct
2. A variety of visualizations that you likely didn’t encounter in your intro stats class, but you can create in R! (with the midterm project in mind)

Principles

• Visualizations encode data using visual “cues”, or mappings of the data to variations in size, shape, color, etc.
• Not all cues are equally effective in accurately conveying the quantitative information contained in data
• In the 1980s, two statisticians, William Cleveland and Robert McGill, ran a series of experiments measuring how accurately human subjects preceived quantitative information from different visual cues.

Visual Cues

• Cleveland and McGill’s experiments found the following hierarchy of visual cues:
  • Subjects most accurately preceived quantitative characteristics from judgements of position and length
  • Judgements of angle and slope were almost as accurate
  • Judgements based upon area and color intensity were less accurate
  • Judgements based upon volume were even less accurate

Visual Cue Hierarchy - Example #1

Does the top representation (length) or bottom representation (area) better encode the underlying numeric differences in these five categories?

Source: Peter Aldhous https://paldhous.github.io/ucb/2016/dataviz.html

Visual Cue Hierarchy - Example #2

Notice how each chart communicates the same information, but draws your attention to different aspects.

• The column chart empthasizes each year’s value as a discrete point in time.
• The line chart empthasizes trajectory
• The dot-and-line chart and dot-column chart are compromises between these two approaches

Source: Peter Aldhous https://paldhous.github.io/ucb/2016/dataviz.html

Takeaways

• To maximize how well your viewers preceive numeric values, use length-based visuals
  • Barcharts for category percentages, stacked barcharts for conditional relationships, etc.
• Piecharts aren’t useless, as people are decent at perceiving information through angles, but they also don’t scale as well as bar charts do
• Avoid forcing your viewers to make important judgements using area, color scales, or volume
  • Instead use these aesthetics for secondary comparisons, or for variables where quantitative judgements are less important
  • For example, if you use the area of the points on a scatterplot to depict a third variable, you should recognize that differences in this variable will be less acurately preceived than the variables shown on the x and y axes

Radar Charts

• Radar charts (sometimes shown as spider charts) are a way of depicting scores for numeric variables for a small number of groups

##   group      var1      var2      var3      var4     var5
## 1     A 0.0000000 0.1894722 0.9777146 1.0000000 1.000000
## 2     B 1.0000000 1.0000000 0.0000000 0.1419618 0.802495
## 3     C 0.3637406 0.0000000 1.0000000 0.0000000 0.000000

Radar Charts - Example Code

# devtools::install_github("ricardo-bion/ggradar") # This package must be installed from github using the "devtools" package
library(ggradar)
library(scales)
data <- data.frame(group = c("A", "B", "C"),
                   var1 = rescale(rnorm(3)),
                   var2 = rescale(rnorm(3)),
                   var3 = rescale(rnorm(3)),
                   var4 = rescale(rnorm(3)),
                   var5 = rescale(rnorm(3)))

## Note: the rescale function in the "scales" package is used to map values onto a [0,1] scale

ggradar(plot.data = data,
        group.colours = c("red", "green", "blue"))

Radar Charts - Application

• Comparing the performance of two players on the Memphis Grizzlies (2013-14)

Source: https://www.researchgate.net/figure/a-Team-radar-chart-and-box-score-b-Player-radar-charts_fig2_310661881

Treemaps

• Treemaps depict the composition of data with a nested structure (ie: each data-point has a main category and one or more subcategories)

Treemaps - Example Code

library(treemap)
data <- data.frame(group = c(rep("group-1",4),rep("group-2",2),rep("group-3",3)), 
                   subgroup = paste("subgroup" , c(1,2,3,4,1,2,1,2,3), sep="-"),
                   value = c(13,5,22,12,11,7,3,1,23))

treemap(data, 
        index=c("group","subgroup"), 
        vSize="value", type="index",
        fontsize.labels=c(15,12),           # size of labels (ie: size for group, size for subgroup, sub-subgroups...) 
    fontcolor.labels=c("white","orange"),   # Color of labels
    fontface.labels=c(2,1),                   # Font of labels: 1 = normal, 2 = bold
    bg.labels=c("transparent"),               # Background color of labels
    align.labels=list(c("center", "center"), 
                      c("right", "bottom")),    # Group label = center-center, subgroup labels = right-bottom
    overlap.labels=0.5, 
    inflate.labels=F                        # If true, labels are bigger when rectangle is bigger.
)

Treemaps - Application

• Displaying the exports of France

Source: https://iscpif.fr/2018/01/postdoctoral-researcher-position-mapping-economics-in-france-isc-pif/

Stacked Area Charts

• Stacked Area Charts represent changes in counts or proportions over time

Stacked Area Chart - Example Code

data <- data.frame(time = as.numeric(rep(seq(1,7),each=7)), 
                   group = rep(LETTERS[1:7],times=7),
                   value = runif(49, 10, 100))

# Compute percentages 
data_per <- data  %>%
  group_by(time, group) %>%
  summarise(n = sum(value)) %>%
  mutate(percentage = n / sum(n))

# Plot #1 - Counts
p1 <- ggplot(data, aes(x=time, y=value, fill=group)) + 
    geom_area(alpha=0.6 , size=1, color = "black") + labs(title = "Frequencies")

# Plot #2 - Percentages
p2 <- ggplot(data_per, aes(x=time, y=percentage, fill=group)) + 
    geom_area(alpha=0.6 , size=1, color="black") + labs(title = "Conditional Proportions")

grid.arrange(p1, p2)

Stacked Area Charts - Application

• Carbon emissions over time

Source: https://www.nytimes.com/2019/02/28/learning/teach-about-climate-change-with-these-24-new-york-times-graphs.html

Additional resources

I recommend you browse these repositories when considering what graphics to use in your midterm project:

• Graph types and example code: https://www.r-graph-gallery.com
• 50 ggplot visualizations with code: http://r-statistics.co/Top50-Ggplot2-Visualizations-MasterList-R-Code.html