Data Visualization

“….That which gives the viewer the greatest number of ideas in the shortest time with the least ink in the smallest space” - Edward Tufte

“...graphical excellence requires telling the truth about the data.” - Edward Tufte

​

​

​

​

​

edwardtufte.com

Data Visualization

• Title
• Abstract
• Introduction
• Methods
• Results
• Discussion
• Citations
• Tables
• Figures

1. Summarize the data
1. tell us what the important trends are

​

2. Illustrate and support
• details, statistics
• tables, figures

What is a graph

A graph/figure/plot maps values in a data set to visual coordinates. It reveals patterns that cannot be understood just from the data.

​

​

What is a graph

A graph/figure/plot maps values in a data set to visual coordinates. It reveals patterns that cannot be understood just from the data.

​

​

What is a graph

A graph/figure/plot maps values in a data set to visual coordinates. It reveals patterns that cannot be understood just from the data.

​

​

Components of a graph

y-axis - Vertical axis. Often used to represent the response variable.

x-axis - Horizontal axis. Often used to represent the predictor variable.

coordinates - The x and y axis together form the coordinates of the plot.

data - The values of y for each value of x. Commonly shown as points.

colors - Used to distinguish different groups of data. Can be used to represent a separate response variable.

shapes - Used to distinguish different groups of data. Can be used to represent a separate response variable.

lines - Often used to show trends in the data.

Components of a good graph

1. It shows all of the data
2. It shows the data summaries
3. The text is meaningful
4. The text is legible
5. It has a good figure caption
6. All of that looks good in the paper

What can you learn about the study just from the graph?

• It shows all of the data

​

​

iris %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_jitter(width = 0.05)

iris %>%

filter(Sepal.Length <= 6) %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_jitter(width = 0.05) +

ylim(4, 8)

​

2. It shows the data summaries

​

​

iris %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_boxplot(aes(group = Species)) +

geom_jitter(width = 0.05)

​

iris %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_jitter(width = 0.05)

3. The text is meaningful

​

​

iris %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_boxplot(aes(group = Species)) +

geom_jitter(width = 0.05) +

labs(x = "Plant Species",

y = "Sepal Length (cm)")

​

​

4. The text is legible

​

​

iris %>%

ggplot(aes(x = Species, y = Sepal.Length)) +

geom_boxplot(aes(group = Species)) +

geom_jitter(width = 0.05) +

labs(x = "Plant Species",

y = "Sepal Length (cm)")

​

​

5. Has a good figure legend or caption.

​

​

Figure 1. Results from data.

Figure 1. Comparison of sepal length among three species of iris plants.

6. All of it looks good in the paper

​

How to do this poorly

1. Deception
2. Not deceptive, just confusing
3. Not deceptive, not confusing, but also not revealing the data

How to do this poorly

• Deception
• Not deceptive, just confusing
• Not deceptive, not confusing, but also not revealing the data

It’s all there, but it hurts to look at. Low resolution, corners stretched.

It’s all there, but it hurts to look at. Low resolution, corners stretched.

Life expectancy

GDP

How to do this poorly

• Deception
• Not deceptive, just confusing
• Not deceptive, not confusing, but also not revealing the data

How to do this goodly

• Show the full scale
• Consider what you want to communicate
• Show all of the data
• Test it out on a friend
• Describe magnitudes in the text

​

Some tricks

Use the log scale

Sort axis by value

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = year, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)")

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = year, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)") +

scale_y_log10()

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = year, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)") +

scale_y_log10() +

geom_boxplot()

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = year, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)") +

scale_y_log10() +

geom_boxplot(aes(group = year))

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = country, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)")

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = country, y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)") +

geom_boxplot(aes(group = country))

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = reorder(country, gdpPercap), y = gdpPercap)) +

geom_point() +

labs(y = "Gross Domestic Product (US dollars 2007)",

x = “Country”) +

geom_boxplot(aes(group = country))

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = reorder(country, gdpPercap), y = gdpPercap)) +

geom_point(size = 0.2) +

labs(y = "Gross Domestic Product (US dollars 2007)",

x = “Country”)) +

geom_boxplot(aes(group = country))

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = reorder(country, gdpPercap), y = gdpPercap)) +

geom_point(size = 0.2) +

labs(y = "Gross Domestic Product (US dollars 2007)",

x = “Country”)) +

geom_boxplot(aes(group = country)) +

coord_flip()

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = reorder(country, gdpPercap), y = gdpPercap)) +

geom_point(size = 0.2) +

labs(y = "Gross Domestic Product (US dollars 2007)",

x = “Country”) +

geom_boxplot(aes(group = country)) +

coord_flip() +

theme(axis.text.y = element_text(size = 2))

​

​

Use the log scale

Sort axis by value

d %>%

ggplot(aes(x = reorder(country, gdpPercap), y = gdpPercap)) +

geom_point(size = 0.2) +

labs(y = "Gross Domestic Product (US dollars 2007)",

x = “Country”)) +

geom_boxplot(aes(group = country)) +

coord_flip() +

theme(axis.text.y = element_blank())

​

​

Use the log scale

Sort axis by value

What are we supposed to say about this graph?

Results

???

• How much does y change across groups?
• What is the minimum change?
• What is the maximum change?
• What is the most important result?

Tables

Tables

Three parts:

1. The data summaries

Tables

Three parts:

• The data summaries
• The lines (visual guides)

Tables

Three parts:

• The data summaries
• The lines (visual guides)

Tables

Three parts:

• The data summaries
• The lines (visual guides)

Tables

Three parts:

• The data summaries
• The lines (visual guides)

Tables

Three parts:

• The data summaries
• The lines (visual guides)
• The description (aka legend/caption/title)

Tables

Table legends go on Top of the table

Three parts:

• The data summaries
• The lines (visual guides)
• The description (aka legend/caption/title)

Table legends go on Top of the table ????

Table legends go on Top of the table ????

Three problems

And a solution

Spurious correlations

​

Ecological fallacy

​

Statistical significance

​

​

Spurious correlations

Shockingly easy to find

Reveals no useful information

No plausible underlying biology

Spurious correlations

Shockingly easy to find. Can you guess the predictor variables?

Spurious Correlations

• Shockingly easy to find

​

• Your brain is a pattern finding machine

​

• You are not alone

Ecological fallacy

Using data from populations to infer something about individual risk

Underlying mechanisms can be worth studying

Can generate useful future hypotheses

​

​

https://www.scientificamerican.com/article/graphics-that-seem-clear-can-easily-be-misread/

https://www.scientificamerican.com/article/graphics-that-seem-clear-can-easily-be-misread/

Spurious correlations

​

Ecological fallacy

​

Statistical significance

​

​

P-values and statistical significance

P-value

​

P-values and statistical significance

P-value

probability of observing a given value of a test statistic, or a more extreme value, under the assumption that the null hypothesis is exactly true

​

​

​

P-values and statistical significance

P-value

probability of observing a given value of a test statistic, or a more extreme value, under the assumption that the null hypothesis is exactly true

measure of importance

probability result is due to chance alone

probability of your hypothesis (or the null)

​

​

​

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

P-value measure of importance

1. Keep it simple
2. Show the raw data
3. Label axes clearly
4. Use ggplot()

​

​

ggplot() stands for the Grammar of Graphics. It’s a built in set of design principles that make it harder to make bad graphs (but certainly not impossible)