Statistical Graphics and Communication
Stefany Coxe
September 12, 2019
Visual Perception
Visual objects
Information in graphics is encoded in visual attributes
Position
Size
Color
Angle
Area
Etc.
Comparisons, estimates of magnitude, and patterns are determined by the visual features
Encoding and decoding information
To use a graphic
Data is encoded into the graphic
A person views the graphic
The person decodes the visual elements to get information
We should try to make this last step as easy as possible, given the limitations of a human
Perception
Basic research on perception
- Generally from cognitive (and related) areas
Applied research on perception
- Often comes out of “human factors” areas
Bottom-up versus top-down perception
Bottom-up processing
Start with basic perceptual features (e.g., edges, color)
Use those to work up to shapes and patterns
Uninfluenced by expectations
Pre-attentive
Top-down processing
Driven by expectations, experiences, knowledge
Filters what we see through that
Bottom-up example
How many 8s?
11768583633938959279
Bottom-up example
How many 8s?
11768583633938959279
How many 8s?
11765363393959279
Bottom-up example
In the first set, you have to read and process the numbers
When the numbers are a different color, you notice the color before you can even tell that they’re 8s
Color is processed bottom-up (pre-attentionally)
- This is the same effect you see with the Stroop task (but in the opposite direction, where you read the word easier than you name the color) https://en.wikipedia.org/wiki/Stroop_effect
Top-down example
What is this an image of?
Top-down example
It depends on the context
Top-down processing helps you decide whether it’s a letter or numbers
Gestalt
Type of top-down processing
We tend to see things that are
near one another
similar to one another
continuations of one another or connected to one another
in a common region
moving together
as being a distinct unit (even if they’re not)
Figure-ground
The figure is the thing you’re focusing on
The ground is the rest – think, “background”
Many common illusions based on figure-ground ambiguity
e.g. Rubin’s vase: https://en.wikipedia.org/wiki/Rubin_vase
Can lead to ambiguous or misleading graphics
Figure-ground in graphics
Original from: https://www.businessinsider.com/gun-deaths-in-florida-increased-with-stand-your-ground-2014-2
Visual illusions
- Muller Lyre illusion:
https://en.wikipedia.org/wiki/M%C3%BCller-Lyer_illusion
- Ebbinghaus or Tichener illusion:
https://en.wikipedia.org/wiki/Ebbinghaus_illusion
- Ponzo illusion:
https://en.wikipedia.org/wiki/Ponzo_illusion
All of these illusions exist because of top-down processing / heuristics / context
What are we good at?
We’ve seen a bunch of examples of what we’re bad at and how perception can go wrong
How can it go right?
What are we good at?
It turns out, quite a few things
Judging whether a line is straight or not
Q-Q plot: quantile-quantile plot
https://en.wikipedia.org/wiki/Q%E2%80%93Q_plot
Judging magnitude (sometimes)
Cleveland & McGill (1984) and Heer & Bostock (2010) had people judge magnitude based on different visual representations
From highest accuracy to lowest:
Position
Angle
Circular area
Rectangular area
Position = good
Area = bad
Effectiveness of different representations for comparisons
Munzner (2015) book: Visualization Analysis & Design
Conclusions
We are good at comparing position and length
- We are bad at comparing areas, volumes, curvature
Use bar plot or dot plot (length and position, respectively)
- Instead of circles or squares (areas, volumes)
Attributes that are processed bottom-up (pre-attentionally) are better
e.g., color for categorical variables
Color on a continuum is trickier
We’ll talk about color more in a second
Be aware of potential illusions
Color
Use of color
We use color in plots to
Highlight
Identify
Group
the elements in a graphic
Physiology of color perception
In the retina, there are 3 types of cones
Each cone is sensitive to a specific range of light wavelengths
- red, green, or blue
There are also rods that help distinguish black and white
The color we perceive is determined by how much (relatively) each of the cones is activated by visible light
All red cones activated + no blue or green = we see red
Some blue, some green = we seen blue-green
There are many theories of exactly how we perceive color, but for our purposes, those details don’t really matter
- Focus on how we define color for the purposes of graphics
Hue, saturation, lightness
Hue
What we would colloquially call the “color”
Red, blue, yellow, etc.
Saturation
- Intensity of the color, from very little color (close to grey) to pure color
Lightness (or brightness)
Relative lightness or darkness
Black = no lightness
White = full lightness
Colors can vary along any or all of these dimensions
Categorical, sequential, diverging color schemes
Categorical: different hues,
similar saturation
Sequential: similar hues of varying
saturation
Diverging: two sequential schemes
From: http://www.perceptualedge.com/articles/visual_business_intelligence/rules_for_using_color.pdf
Color vs black and white
What will your plot look like if it’s converted to black and white?
Incorrect interpretations due to common illusions
e.g., https://en.wikipedia.org/wiki/File:Gradient-optical-illusion.svg
Saturation is very important
Categorical scheme has the same saturation for all colors
- You can’t see any differences between colors
Sequential scheme varies saturation / intensity
- You can still see all the differences
Diverging scheme varies saturation, but two ends are similar
- You can still see which are in the middle and which are at the ends, but probably can’t tell the two ends from one another
Alpha
Adjusting the transparency / opacity (alpha) of colors can enhance their use
The points for different groups are different colors, but they may be obscured due to overplotting
Color plus reduced alpha lets you use color to distinguish groups while still being able to see all the points
Adjusting alpha may look similar to adjusting saturation
- But stacked points with reduced alpha create high saturation areas
Color creation schemes
RGB: red, green, blue
Mixture of all = white
Computer monitors, film
FIU blue = (8, 30, 63)
CMYK: cyan, magenta, yellow (and black)
Mixture of all = black
Physical printing
FIU blue = (100, 87, 42, 52)
Hexadecimal
Web-based purposes (i.e., HTML)
FIU blue = #081E3F
Colorblindness
Genetic loss of one (or more) cone type
Most common is red-green color blindess
- Red and green are indistinguishable
Choose a color scheme that doesn’t depend on people needing to distinguish between these colors
- Remember to consider other properties of the color besides the color (hue), like saturation
Test your plot to see how it looks to a person who is color blind
Colors in ggplot2
We will have most of a lecture on the details of using color in ggplot
Changing the overall color theme for a plot
Changing the color of individual parts of a plot
Today, the focus is on color as a perceptual concept
Other design decisions
Wind frequency and direction
Radial bar chart is a good choice to see the prevailing wind direction
- But not to be able to compare wind direction frequencies
What is the purpose of the graphic?
https://medium.com/nightingale/breaking-the-rules-bd212fecd045
Tips for making good graphics
In terms of small design decisions (based on cognition and perception)
High contrast color choices
Minimize distractions
Story of the graphic should jump out at you – squint at it (or take off your glasses)
Some advice based on Tufte and Robbins
Data-to-ink ratio
- Remove extra or decorative lines
Reduce the mental processing needed
- Don’t make your viewer do more mental work than they need to
Does the Y axis need to start at 0?
- Yes, for some types of charts
https://flowingdata.com/2015/08/31/bar-chart-baselines-start-at-zero/
- Maybe
https://byjustinfox.com/2014/12/14/the-rise-of-the-y-axis-zero-fundamentalists/
(But some of the people quoted here say YES)
- No, shut up about the y-axis.