Lab 2. Fundamental Chart Types

PUBH 6199: Visualizing Data with R, Summer 2025

Xindi (Cindy) Hu, ScD

2025-05-29

Outline for today

Introducing the datasets
Review chart types to visualize basic quantitative information
Review chart types to visualize trends over time
Review chart types to visualize distribution
Review chart types to visualize relationships

Introducing the `penguins` dataset from `{palmerpenguins}`

Contains body measurements for 344 penguins on three islands in the Palmer Archipelago.

Artwork by @allison_horst

Introducing the `penguins` dataset from `{palmerpenguins}`

Rows: 342
Columns: 8
$ species           <fct> Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Adel…
$ island            <fct> Torgersen, Torgersen, Torgersen, Torgersen, Torgerse…
$ bill_length_mm    <dbl> 39.1, 39.5, 40.3, 36.7, 39.3, 38.9, 39.2, 34.1, 42.0…
$ bill_depth_mm     <dbl> 18.7, 17.4, 18.0, 19.3, 20.6, 17.8, 19.6, 18.1, 20.2…
$ flipper_length_mm <int> 181, 186, 195, 193, 190, 181, 195, 193, 190, 186, 18…
$ body_mass_g       <int> 3750, 3800, 3250, 3450, 3650, 3625, 4675, 3475, 4250…
$ sex               <fct> male, female, female, female, male, female, male, NA…
$ year              <int> 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007…

Two penguins dropped out because missing data on important variables species, flipper_length_mm, and body_mass_g.

Introducing the lyme disease surveillance data

Lyme disease has been a nationally notifiable condition in the U.S. sinde 1991. Local and state health departments collect reports of Lyme disease and shared the anonymized data with the Centers for Disease Control and Prevention (CDC). The CDC developed public use data sets to facilitate public health and reserach acccess to the data.

Download state and local data on Lyme disease case counts and incidence (cases/100k people) over time here.

Introducing the lyme disease surveillance data

Rows: 714
Columns: 4
$ state          <chr> "Alabama", "Alabama", "Alabama", "Alabama", "Alabama", …
$ year           <dbl> 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2…
$ cases          <dbl> 2, 24, 25, 24, 64, 25, 38, 41, 36, 66, 14, 51, 32, 36, …
$ rates_per_100k <dbl> 0.0, 0.5, 0.5, 0.5, 1.3, 0.5, 0.8, 0.8, 0.7, 1.3, 0.3, …

The dataset contains the number of Lyme disease cases and incidence rates (cases per 100,000 people) for 49 states (missing Hawaii) and DC in the U.S. from 2010 to 2023.

Outline for today

Introducing the datasets
Review chart types to visualize basic quantitative information
Review chart types to visualize trends over time
Review chart types to visualize distribution
Review chart types to visualize relationships

Let’s start with a simple example, how many penguins are there in each species?

Bar chart is a good choice to visualize counts.

ggplot(penguins_clean, aes(x = species)) + 
  geom_bar() +
  labs(title = "Bar Chart: Count by Species") +
  theme_minimal()

Dot plot improves upon it due to higher data-ink ratio.

penguins_clean %>%
  count(species) %>%
  ggplot(aes(x = species, y = n)) + 
  geom_point() +
  geom_linerange(aes(ymin = 0, ymax = n)) +
  labs(title = "Dot plot: Count by Species") +
  theme_minimal()

How to visualize the average body mass of each species?

The default of geom_bar() is column plot, use coord_flip() to turn it horizontal.

ggplot(penguins_clean, aes(x = species, y = body_mass_g)) + 
  stat_summary(fun = mean, geom = "bar") +
  labs(title = "Mean Body Mass by Species") +
  coord_flip() +
  theme_minimal()

Similarly, coord_flip() can be applied to dot plot

penguins_clean %>%
  group_by(species) %>%
  summarise(mean_body_mass = mean(body_mass_g, na.rm = TRUE)) %>%
  ggplot(aes(x = species, y = mean_body_mass)) + 
  geom_point() +
  geom_linerange(aes(ymin = 0, ymax = mean_body_mass)) +
  labs(title = "Mean Body Mass by Species") +
  coord_flip() +
  theme_minimal()

Add labels if the exact values are important

Use geom_text() to add labels to the bar chart and dot plot.

penguins_clean %>%
  group_by(species) %>%
  summarise(mean_body_mass = mean(body_mass_g, na.rm = TRUE)) %>%
  ggplot(aes(x = species, y = mean_body_mass)) + 
  geom_col() +
  geom_text(aes(label = round(mean_body_mass)), hjust = 1.2, color = "white") +
  labs(title = "Mean Body Mass by Species") +
  coord_flip() +
  theme_minimal()

penguins_clean %>%
  group_by(species) %>%
  summarise(mean_body_mass = mean(body_mass_g, na.rm = TRUE)) %>%
  ggplot(aes(x = species, y = mean_body_mass)) + 
  geom_point() +
  geom_linerange(aes(ymin = 0, ymax = mean_body_mass)) +
  geom_text(aes(label = round(mean_body_mass)), hjust = -0.2, color = "black") +
  scale_y_continuous(limits = c(0, 6000)) + # set y-axis limits to give room to labels
  labs(title = "Mean Body Mass by Species") +
  coord_flip() +
  theme_minimal()

When to use `geom_col()` vs `geom_bar()`?

Use geom_col() when data is already summarized, e.g., mean body mass by species.

penguins_clean %>%
  group_by(species) %>%
  summarise(mean_body_mass = mean(body_mass_g, na.rm = TRUE)) %>%
  ggplot(aes(x = species, y = mean_body_mass)) + 
  geom_col() +
  labs(title = "Mean Body Mass by Species")

Use geom_bar() when data is not summarized, e.g., count of species.

ggplot(penguins_clean, aes(x = species, y = body_mass_g)) + 
  geom_bar(stat = "summary", fun = mean) +
  labs(title = "Mean Body Mass by Species")

IMPORTANT: Bar plot axis must start at zero

Yang et al (2021) study provided empirical evidence that y-axis truncation leads to viewers to perceive illustrated differences as larger (i.e., a truncation effect).

IMPORTANT: Bar plot axis must start at zero

Yang et al (2021) study provided empirical evidence that y-axis truncation leads to viewers to perceive illustrated differences as larger (i.e., a truncation effect).

But you can cut the y-axis of dot plots!

A dot plot has has less ink and draw the eye to the end point rather than the middle of the bars. Cutting the y-axis allows easily differentiating differences in the values.

Bar plots/dot plots shine when comparing counts, but you should be careful when using them to summarize your data. Why?

Be careful when using bar plots/dot plots to summarize continuous data

Art by Allison Horst

Bar plots hide the distribution of the data
Bar plot makes readers infer that data are normally distributed with no outliers

Outline for today

Introducing the datasets
Review chart types to visualize basic quantitative information
Review chart types to visualize trends over time
Review chart types to visualize distribution
Review chart types to visualize relationships

Line plots show trends over time

Going back to the lyme disease dataset, use line plot to show changes in lyme disease incidence rates (case per 100,000 people) by state, from 2010 to 2023.

Basic line graph uses geom_line()

lyme %>%
  ggplot(aes(x = year, y = rates_per_100k, group = state)) +
  geom_line()

A line + point graph uses geom_point() to add points to the line graph.

lyme %>%
  ggplot(aes(x = year, y = rates_per_100k, group = state)) +
  geom_line()+
  geom_point()

Improve upon “Spaghetti plots”

A spaghetti plot is a line graph with many lines, which makes it hard to read. We can use {gghighlight} to draw attention to the lines of interest.

Highlight a specific state or a group of states

lyme %>%
  ggplot(aes(x = year, y = rates_per_100k, group = state, color = state)) +
  geom_line() +
  gghighlight::gghighlight(state %in% c("Michigan", "Florida")) +
  theme_minimal()

Highlight states based on the values of the incidence rate, for example, maximum rate exceeds 5 per 100k.

lyme %>%
  ggplot(aes(x = year, y = rates_per_100k, group = state, color = state)) +
  geom_line() +
  gghighlight::gghighlight(max(rates_per_100k) > 5) +
  theme_minimal()

Area chart is similar to line graph, just filled in and stacked

lyme |> 
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = year, y = rates_per_100k, group = state, color = state)) +
  geom_line()

lyme |> 
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = year, y = rates_per_100k, group = state, fill = state)) +
  geom_area()

Stacked area charts are useful for showing the evolution of a whole and the relative proportions of each group that make up the whole. But it has a few drawbacks: low data-ink ratio, using area rather than position to encode data.

A variant of area charts: proportional stacked area charts

lyme |> 
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = year, y = rates_per_100k, group = state, fill = state)) +
  geom_area(position = "fill") + # this creates the proportional stacked area chart
  scale_y_continuous(labels = scales::percent_format(scale = 100))

Which group to put on the bottom?

It is important to consider which group you want to put it on the bottom of the area chart because it is the only group where your user can easily read the values off the chart.

If you want to draw attention to “Michigan”, put it on the bottom.

lyme |> 
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  mutate(state = fct_relevel(state, "Michigan", after = Inf)) |>  # move Michigan to end
  ggplot(aes(x = year, y = rates_per_100k, group = state, fill = state)) +
  geom_area() +
  theme_minimal()

If all groups are equally important and you are not as interested in showing the whole, use a faceted line plot instead!

lyme |> 
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  mutate(state = fct_relevel(state, "Michigan", after = Inf)) |>  # move Michigan to end
  ggplot(aes(x = year, y = rates_per_100k, group = state, color = state)) +
  geom_line() +
  facet_wrap(~ state) +
  theme_minimal()

Practice makes perfect!

~ Head over to lab2 notebook! ~

Outline for today

Introducing the datasets
Review chart types to visualize basic quantitative information
Review chart types to visualize trends over time
Review chart types to visualize distribution
Review chart types to visualize relationships

Histogram

Histogram cuts a numeric variable into bins and counts the number of observations in each bin.

Too many bins!

ggplot(penguins_clean, aes(x = body_mass_g, fill = species)) +
  geom_histogram(binwidth = 30, alpha = 0.7) +
  labs(title = "Histogram: Body Mass Distribution")

A better binwidth parameter. The default is to divide data into 30 bins

ggplot(penguins_clean, aes(x = body_mass_g, fill = species)) +
  geom_histogram(binwidth = 200, alpha = 0.7) +
  labs(title = "Histogram: Body Mass Distribution")

Density plot

Density plot uses the kernel density estimate to show the probability density function of a variable. Area under each density curve sums to 1.

A smoothed version of a histogram

ggplot(penguins_clean, aes(x = body_mass_g, fill = species)) +
  geom_density(alpha = 0.5) +
  labs(title = "Density Plot: Body Mass by Species")

Why is the histogram not the same?

ggplot(penguins_clean, aes(x = body_mass_g, fill = species)) +
  geom_histogram(binwidth = 200, alpha = 0.7) +
  labs(title = "Histogram: Body Mass Distribution")

Density plot does not indicate sample size.

Box plot

Box plot shows the median, interquartile range (IQR), and outliers of a variable. Boxplot is often used with comparing the same numeric variable over multiple groups.

Boxplot shows summary statistics, which may hide the distribution of the data.

ggplot(penguins_clean, aes(x = species, y = body_mass_g)) +
  geom_boxplot() +
  labs(title = "Box Plot: Body Mass by Species")

Enhance box plot

Highlight a group if you have many groups, also flip the coordinate if your categorical variable has long labels.

lyme |>
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = state, y = rates_per_100k)) +
  geom_boxplot() +
  gghighlight::gghighlight(state == "Michigan") +
  coord_flip() +
  labs(title = "Box Plot: Lyme Disease Incidence by State") +
  theme_minimal()

Jitter raw data, but remember to remove outliers

lyme |>
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = state, y = rates_per_100k)) +
  geom_boxplot(outlier.shape = NA) +
  geom_jitter(alpha = 0.5, width = 0.2) +
  coord_flip() +
  labs(title = "Box Plot: Lyme Disease Incidence by State") +
  theme_minimal()

Violin plot

Violin plot shows the density estimate of the variable, similar to a density plot. It is often used with comparing the same numeric variable over multiple groups. It is usually a better alternative than a box plot.

lyme |>
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = state, y = rates_per_100k)) +
  geom_violin() +
  coord_flip() +
  labs(title = "Violin Plot: Lyme Disease Incidence by State") +
  theme_minimal()

Enhance violin plot

If you have many groups, consider ranking them by median values to make your readers’ brain hurt less. Recall law of continuity.

lyme |>
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |>
  mutate(state = fct_reorder(state, rates_per_100k, .fun = median, na.rm = TRUE)) |>
  ggplot(aes(x = state, y = rates_per_100k)) +
  geom_violin() +
  coord_flip() +
  labs(title = "Violin Plot: Lyme Disease Incidence by State") +
  theme_minimal()

The {see} package has geom_violindot() function that creates a half-violin half-dot plot, showing both distribution and sample size.

lyme |>
  filter(state %in% c("Michigan", "Ohio", "Iowa", "North Dakota")) |> 
  ggplot(aes(x = state, y = rates_per_100k)) +
  see::geom_violindot(size_dots = 2) +
  coord_flip() +
  labs(title = "Box Plot: Lyme Disease Incidence by State") +
  theme_minimal()

Outline for today

Introducing the datasets
Review chart types to visualize basic quantitative information
Review chart types to visualize trends over time
Review chart types to visualize distribution
Review chart types to visualize relationships

Scatter plot

Scatter plot is a good choice to visualize the relationship between two numeric variables. They help us answer questions around the effect of X on Y.

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(alpha = 0.5) +
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Add rug to visualize distribution

Rug plot uses distribution marks to visualize the distribution of the two numeric variables. Each narrow line represents one data point. It shows the density of the data along the x and y axes.

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(alpha = 0.5) +
  geom_rug() +
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Add trend lines

Trend lines are used to show the overall trend of the data. Default method for geom_smooth() is LOESS (locally estimated scatter plot smoothing), think of it as a moving average.

LOESS

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(alpha = 0.5) +
  geom_smooth() +
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Linear regression

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(alpha = 0.5) +
  geom_smooth(method = "lm", se = FALSE) +
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Add a third numeric variable with bubble chart

We can use a bubble chart to show the third numeric variable. The size of the point represents the third variable.

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(aes(size = bill_length_mm), alpha = 0.4) +
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

A few caveats:

The relationship between X and Y will be the primary focus
It may be difficult to distinguish the size of the bubbles

Adjust the size of the bubbles

Use scale_size() to adjust the size of the bubbles. Do not use scale_radius().

This is good.

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(aes(size = bill_length_mm), alpha = 0.4) +
  scale_size(range = c(1, 4), name = "Bill Length (mm)") + # adjust the size of the bubbles
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

This is misleading.

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(aes(size = bill_length_mm), alpha = 0.4) +
  scale_radius(range = c(1, 4), name = "Bill Length (mm)") + # adjust the size of the bubbles
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Add a third numeric variable with color

Recall that color hue does not natually have meaning for magnitude, consider using intensity

penguins_clean %>%
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point(aes(color = bill_length_mm)) +
  scale_color_gradient(low = "white", high = "royalblue", name = "Bill Length (mm)") + 
  labs(title = "Effect of Flipper Length on Body Mass") +
  theme_minimal()

Your turn, get on the bike!

~ Head over to lab2 notebook! ~

End-of-Class Survey

Fill out the end-of-class survey

~ This is the end of Lab 2 ~

10:00

Lab 2. Fundamental Chart Types

Outline for today

Introducing the penguins dataset from {palmerpenguins}

Introducing the penguins dataset from {palmerpenguins}

Introducing the lyme disease surveillance data

Introducing the lyme disease surveillance data

Outline for today

Let’s start with a simple example, how many penguins are there in each species?

How to visualize the average body mass of each species?

Add labels if the exact values are important

When to use geom_col() vs geom_bar()?

IMPORTANT: Bar plot axis must start at zero

IMPORTANT: Bar plot axis must start at zero

But you can cut the y-axis of dot plots!

Be careful when using bar plots/dot plots to summarize continuous data

Outline for today

Line plots show trends over time

Improve upon “Spaghetti plots”

Area chart is similar to line graph, just filled in and stacked

A variant of area charts: proportional stacked area charts

Which group to put on the bottom?

Outline for today

Histogram

Density plot

Box plot

Enhance box plot

Violin plot

Enhance violin plot

Outline for today

Scatter plot

Add rug to visualize distribution

Add trend lines

Add a third numeric variable with bubble chart

Adjust the size of the bubbles

Add a third numeric variable with color

End-of-Class Survey

Introducing the `penguins` dataset from `{palmerpenguins}`

Introducing the `penguins` dataset from `{palmerpenguins}`

When to use `geom_col()` vs `geom_bar()`?