Data visualization: practices

PSYC411: Data visualization – practices

• My name is Dr Rob Davies, I am an expert in communication, individual differences, and methods

Introduction: Data visualization – perspectives

• In this PSYC411 class, we look at practices or how we get visualization done
• In the linked PSYC413 class, we focus on perspectives or how to think about visualization

Our lesson plan

1. Identify your goals
2. Think about your audience
3. Develop reflectively
4. Implement good practice

Our learning objectives: — what are we learning about?

We are working together to help you:

1. Goals — Formulate questions you can ask yourself to help you to work effectively
2. Audience — Understand the psychological factors that affect your impact
3. Development — Work reflectively through a development process
4. Implement — Produce visualizations in line with best practice

Our assessment targets: — how do you know if you have learned?

We are working together so you can:

1. Goals — Identify a set of targets for a development process in your professional teams
2. Audience — Explain what you need to do to make a visualization effective
3. Development — Locate yourself within the stages of the development process
4. Implement — Produce visualizations that look good and are useful

What are our goals – Questions to help you to work effectively

What are our goals?

• Why don’t we just use the good enough easy to produce plots in Excel? \(\rightarrow\) Why bother?
• Why don’t we just produce a summary table? \(\rightarrow\) Why bother?
• Are we engaged in making beautiful graphics or informative displays or both? \(\rightarrow\) What are we doing?
• In PSYC413, we look at \(\rightarrow\) Perspectives: the context and history of thinking about visualization

Visualize to enable comparison

What are our goals? — What are our jobs?

• Data visualization workers: we may aim to get and keep the attention of our audience, to tell a story, to persuade our viewers
• Data analysis workers: we may aim to enable our audience to understand our data, our findings, and to discover more for themselves

What are our goals? — Where or when are we in our process?

• Sometimes in a workflow, we are quickly sketching draft visualizations: exploring, for ourselves, or with others, what we can see in our data
• Sometimes, we are ready to present our visualization to a wider audience: we aim to share a polished visual object

What are our goals? — Discovery

Discovery goals

1. Do we need an overview? – To get a sense of what is in the data, and to check our assumptions
2. Are we looking for the unexpected? – Comparing groups to check for variability, exploring data open to surprises

What are our goals? — Communication

Communication goals

1. What do we need our audience to understand?
2. What story are we telling?
3. Do we need to attract attention or stimulate interest?

Think about your audience – An evidence based account of what works

Communicating uncertainty is critical

Consider accessibility from the start

Development – Work reflectively through a development process

The benefits of investing in the development process

• Identifying your goals enables you to understand what you are doing and why
• Through the development process, you may create different versions — iterations — of a plot
• This iterative work benefits both you and your audience (A. Gelman et al., 2002; Kastellec & Leoni, 2007)

The benefits of investing in the development process

Scientific thinking and data visualisation

We can use text and tables to communicate specific values but visualizations help us to:

• stimulate thinking
• discover what is unexpected
• communicate scale and complexity
• make comparisons to show how results vary
• display uncertainty about estimates

Anscombe (1973): visualizations show data features quickly and vividly

Figure 6: All 4 of the Anscombe (1973) x,y datasets are identical when examined using summary statistics but we see how they vary when we use scatterplots to visualize them

Matejka & Fitzmaurice (2017) give us the Datasaurus dozen

Figure 7: All 12 Matejka & Fitzmaurice (2017) x,y datasets (via jumpingrivers (n.d.)) have the same mean and standard deviation summary statistics but we only understand how the data are structured when we plot them and can look at the structure

Develop visualizations to discover and communicate variability in outcomes

Figure 8: In this plot we show data on the impact of sleep deprivation on reaction time, from Belenky et al. (2003; via Bates et al., 2015). We can see how reaction time slows with increasing deprivation on average (grey line) but that the rate of slowing varies between individuals

Reflect on kinds of uncertainty

• Scientists are often faced with the challenge of conveying uncertainty to their audiences (Hofman et al., 2020):

1. Inferential uncertainty — the degree to which a particular summary statistic (e.g., a population mean) is known to the scientist
2. Outcome uncertainty — how much individual outcomes vary (e.g., around the mean, regardless of how well it has been estimated)

• Inferential uncertainty can be reduced by collecting and analyzing more data, whereas outcome uncertainty cannot

As we work, reflect on the challenges of visualizing uncertainty

The challenges of uncertainty

The challenges of uncertainty

Variation and uncertainty — the importance, the challenges

Vasishth & Gelman (2021):

The most difficult idea to digest in data analysis is that conclusions based on data are almost always uncertain, regardless of whether the outcome of the statistical test is statistically significant or not

Variation and uncertainty — the importance, the challenges

a. Gelman (2015):

We must move beyond the idea that effects are ‘there’ or not and the idea that the goal of a study is to reject a null hypothesis. As many observers have noted, these attitudes lead to trouble because they deny the variation inherent in real social phenomena, and they deny the uncertainty inherent in statistical inference

We use visualizations to help us to see and understand the variation and the uncertainty in our data

• Results will vary: we should expect changes over time, or differences between individuals or between groups
• Knowledge is uncertain: outcomes will vary even when the average effect is precisely estimated
• We have the responsibility to accept and to express this uncertainty

Implement – Produce visualizations in line with best practice

{ggplot2} means: the Grammar of Graphics Plot 2

• When we use the {ggplot2} to draw plots, we are using tools developed with a philosophy of visualization in mind (Wickham, 2010; Wilkinson, 2013): The Grammar of Graphics
• A grammar is a system of rules that allows people to collaborate and individuals to create
• We do not need to think about the grammar when we produce visualizations
• But it will help you to know that when we puzzle over how we do things, there are always reasons why we do things

A simple plot has many elements

• data and aesthetic mappings
• statistical transformations
• geometric objects
• scales

A simple plot has many elements

• Plot with no objects
• Code for the plot

• When we code a plot, we tell R we want:
• to use ggplot() to create a plot
• using the data-set clearly.one.subjects
• and the variables SHIPLEY, HLVA

  ggplot(data = clearly.one.subjects, aes(x = SHIPLEY, y = HLVA))

• We bring the data-set and the variables
• We declare the aesthetic mappings:

1. SHIPLEY score \(\rightarrow\) x-axis (horizontal: left-to-right position)
2. HLVA score \(\rightarrow\) y-axis (vertical: bottom-to-top position)

A simple plot has many elements

• Plot with only objects
• Code for the plot

• When we code a plot, we tell R we want:
• to use a geometric object, like geom_point
• to display the data aesthetic mappings

  ggplot(data = clearly.one.subjects, aes(x = SHIPLEY, y = HLVA)) +
  geom_point()

• We add the geom_point() to tell R to draw the information about the SHIPLEY and HLVA scores as points
• Each point represents information about one participant in the clearly.one.subjects data-set

1. SHIPLEY score \(\rightarrow\) x-axis (horizontal: left-to-right position)
2. HLVA score \(\rightarrow\) y-axis (vertical: bottom-to-top position)

When we use {ggplot2} we work in layers

• The grammar of graphics define the components of a plot: the data, the mappings, and the geometric object
• Together, the data, mappings, and geometric object form a layer
• A plot may have multiple layers

When we use {ggplot2} we are in control and we can be creative

Plot with layers: add a smoother

• Plot with smoother
• Code for the plot

• Build a plot layer by layer
• We can begin by using points to display the vocabulary and health literacy scores for each person
• We add a layer using a smoother to show the average association between vocabulary and literacy

  ggplot(data = clearly.one.subjects, aes(x = SHIPLEY, y = HLVA)) +
  geom_point() +
  geom_smooth()

• We add the geom_smooth() to tell R to represent the average trend for the association between SHIPLEY and HLVA scores
• The line is drawn by {ggplot2} which calculates a statistical transformation
• Here, the transformation summarizes the association for different ranges of SHIPLEY vocabulary scores

Defaults and arguments

clearly.one.subjects %>%
  ggplot(aes(SHIPLEY, HLVA)) +
  geom_smooth() +
  geom_point()

• The {ggplot2} library supplies default values
• So we do not need to tell R how to do every thing
• We do not need to tell R that the points in a scatterplot:
• should represent the data aesthetic mappings in Cartesian (x-horizontal, y-vertical) 2-dimensional space
• and should be black in colour

Defaults and arguments

clearly.one.subjects %>%
  ggplot(aes(SHIPLEY, HLVA)) +
  geom_smooth() +
  geom_point(colour = "darkgrey", size = 3)

• We can over-ride the defaults by supplying arguments, entering values inside the brackets in the function calls
• geom_point(colour = "darkgrey", size = 3) tells R we want:

1. dark grey points when the default is black
2. points that are 3x larger than the default size

When we use {ggplot2} we are in control and we can be creative

We can use colour

• Using colour
• Code for the plot

• When we code a plot, we tell R we want:
• to display data about people with different education levels
• distinguishing education level by colour

clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA, 
             group = EDUCATION, colour = EDUCATION)) +
  geom_smooth(method = "lm", se = FALSE, 
              linewidth = 2, alpha = .75) +
  geom_point(size = 3)

• group = EDUCATION, colour = EDUCATION tells R to:

1. group the data by EDUCATION level
2. colour the points for people with different levels of education in different colours

Method, size, transparency

clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA, 
             group = EDUCATION, colour = EDUCATION)) +
  geom_smooth(method = "lm", se = FALSE, 
              linewidth = 2, alpha = .75) +
  geom_point(size = 3)

• method = "lm", se = FALSE tells R what method to use to draw the smoother line
• linewidth = 2 makes the width of the smoother line 2 x larger than the default
• alpha = .75 makes the line .75 x the opacity of the default (i.e. a. bit more transparent)
• Learn to edit: shape, size, transparency and colour

We facet plots to enable comparisons

• Using facets
• Code for the plot

• It is often easier to compare trends
• By presenting a separate plot for each condition or group
• Showing the separate plots in a grid side-by-side

clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA, 
             group = EDUCATION, colour = EDUCATION)) +
  geom_smooth(method = "lm", se = FALSE, 
              linewidth = 2, alpha = .75) +
  geom_point(size = 3) +
  facet_wrap(~ EDUCATION)

• facet_wrap(~ EDUCATION) tells R to split the data by EDUCATION level
• And show a separate plot for each EDUCATION level group side-by-side for easy comparison

We can guide our audience

• Labelled plot
• Code for the plot

• We do not present visualizations in isolation
• We present plots embedded in the context of labels and titles
• We use the text to guide the viewer

clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA)) +
  geom_smooth(method = "lm", se = FALSE,
              colour = "darkgreen", linewidth = 2, alpha = .75) +
  geom_point(size = 3, colour = "lightgreen") +
  labs(x = "Vocabulary (Shipley)", y = "Health literacy (HLVA)",
       title = "Scatterplot showing how higher vocabulary\npredicts higher health literacy on average")

• We use the labs() function to add: the plot title and the labels for the x-axis and y-axis
• We edit the title so that the viewer can see what we want them to see
• We use \n to make the title fit on two lines

We annotate plots to direct attention

• Annotated plot
• Code for the plot

• We can direct the attention of our audience to key features of our data
• By adding annotations like text and lines

clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA)) +
  geom_smooth(method = "lm", se = FALSE,
              colour = "darkgreen", linewidth = 2, alpha = .75) +
  geom_point(size = 3, colour = "lightgreen") +
  labs(x = "Vocabulary (Shipley)", y = "Health literacy (HLVA)") +
  geom_hline(yintercept = mean(clearly.one.subjects$HLVA),
             linetype = "dashed",
             linewidth = 2,
             colour = "grey",
             alpha = .85) +
  annotate("text", x = 27, y = 9.3, label = "Mean HLVA", colour = "grey") +
  theme_bw()

• geom_hline() adds a line to show mean health literacy
• annotate("text" ...) adds a text label

Extensions free our creativity

• Complex plot
• Code for the plot

• The power of the Grammar of Graphics lies in the rules
• Developers can use the rules to expand our capacity to visualize data
• We add marginal histograms to our scatterplot to visualize associations and distributions

plot <- clearly.one.subjects %>%
  ggplot(aes(x = SHIPLEY, y = HLVA)) +
  geom_smooth(method = "lm", se = FALSE,
              colour = "darkgreen", linewidth = 2, alpha = .75) +
  geom_point(size = 3, colour = "lightgreen") +
  labs(x = "Vocabulary (Shipley)", y = "Health literacy (HLVA)")

ggMarginal(plot, type = "histogram", fill = "lightgreen", 
           xparams = list(binwidth=2), yparams = list(binwidth=1))

• ggMarginal(plot, type = "histogram") enables us to show the distribution of scores on each variable
• This helps our viewer to process the association and information about each variable (Franconeri et al., 2021)

Choose your plot theme

• We can choose a theme to adapt the look of the whole plot to suit our needs or the needs of our audience

Summary

You start your work with these questions:

1. What are our goals?
2. What does our audience need or expect?

Summary

You develop your visualization in a reflective process:

1. Begin with a quick draft to show the distributions or make the comparisons you think about first
2. Then reflect, and edit: does this enable me to discover sources of variability in my data?
3. Then reflect, and edit: does this enable me to effectively communicate what I want to communicate?
4. Then reflect, and edit: does this look good? – do my viewers tell me this works well?

Summary