Intro to Data Visualization

visualization basics, ggplot2, grammar of graphics

2026-02-17

Data Visualization Basics

Data Visualization

What is data visualization?

• I’ve made charts and graphs that should finally make it clear. (I’ve prepared a lecture.)
• Graphical representation of quantitative information
• Identify, interpret, communicate patterns
• In academic research, data viz tells (not sells) a story

Core Principles

Clarity

Communicate all – and only – information necessary to tell the story

Simplicity

Minimize distraction & send one message at a time

Accuracy

Use reliable data (garbage in, garbage out) and be faithful to it

Consistency

Represent similar ideas in similar ways & meet audience expectations

Relevance

Know your audience & speak to them

Data Viz in R

Plotting with base R

• Base plots:
  • “pen and paper” style
  • easy to use, hard to modify

# First create the scatter plot
plot(iris$Sepal.Length, iris$Sepal.Width,
     main = "base R: Iris Sepal Length vs Width",
     xlab = "Sepal Length",
     ylab = "Sepal Width",
     col = as.numeric(iris$Species),
     pch = 19)

# Add regression lines for each species
species_levels <- levels(iris$Species)
colors <- 1:3

for(i in 1:3) {
    subset_data <- iris[iris$Species == species_levels[i], ]
    reg <- lm(Sepal.Width ~ Sepal.Length, 
              data = subset_data)
    abline(reg, col = colors[i], lwd = 2)
}

# Add legend
legend("topright", 
       legend = levels(iris$Species), 
       col = colors, 
       pch = 19)

pen and paper -> you draw it, it’s there.

Plotting with lattice

• lattice
  • lots of customization
  • constrained structure

# Lattice plot
# uses one function with many arguments
xyplot(Sepal.Width ~ Sepal.Length, 
       data = iris,
       groups = Species,
       auto.key = TRUE, # add legend
       type = c("p", "r"), # points and regression line
       main = "lattice: Iris Sepal Length vs Width",
       xlab = "Sepal Length",
       ylab = "Sepal Width")

Plotting with ggplot2

• ggplot2
  • flexible and powerful
  • tidyverse approach to data science

# ggplot plot
# begins with a ggplot() function
# then adds layers to customize

ggplot(iris, aes(x = Sepal.Length, y = Sepal.Width, color = Species)) +
  geom_point() +
  geom_smooth(method = "lm", se = TRUE) +
  labs(title = "ggplot2: Iris Sepal Length vs Width",
       x = "Sepal Length",
       y = "Sepal Width")

ggplot2

Tidyverse’s ggplot2 package

• easy plotting system for R
• “grammar of graphics” akin to dplyr’s “grammar for data manipulation”
  • adding layers ~ adding pipes
• flows best with tidy datasets
• customizable
  • modify many visual properties on the fly
• highly flexible
  • represent the same data in dramatically different ways

What’s the difference between ggplot and ggplot2?

ggplot() is a function, ggplot2 is a package.

So ggplot2 is version 2?

No.

Customization

ggplot(iris, aes(x = Sepal.Length, y = Sepal.Width, 
                 color = Species)) +
  # Add points with custom appearance
  geom_point(size = 3, alpha = 0.7) +
  # Add regression lines with custom appearance
  geom_smooth(method = "lm", se = TRUE, alpha = 0.2, 
             linewidth = 1.2, linetype = "dashed") +
  # Customize colors using a custom palette
  scale_color_manual(values = 
                         c("#FF6B6B", "#4ECDC4", "#45B7D1")) +
  # Add labels with custom formatting
  labs(title = "Sepal Dimensions Across Iris Species",
       subtitle = "Comparing Length vs Width with Trend Lines",
       x = "Sepal Length (cm)",
       y = "Sepal Width (cm)",
       caption = "Data: Edgar Anderson's Iris Dataset") +
  # Customize theme elements: 
  # title, axis, legend, panel, borders
  theme_minimal() +
  theme(
plot.title = element_text(size = 16, face = "bold", 
                          margin = margin(b = 20)),
    plot.subtitle = element_text(size = 12, color = "grey40"),
axis.title = element_text(size = 10, face = "bold"),
    axis.text = element_text(size = 9),
legend.position = "bottom",
    legend.title = element_text(face = "bold"),
    legend.background = element_rect(
        fill = "white", color = "grey90"),
panel.grid.major = element_line(color = "grey90"),
    panel.grid.minor = element_blank(),
plot.background = element_rect(fill = "white", color = NA),
    panel.border = element_rect(color = "grey90", fill = NA)
  ) +
  # Set specific axis limits
  coord_cartesian(
    xlim = c(min(iris$Sepal.Length) - 0.2, 
             max(iris$Sepal.Length) + 0.2),
    ylim = c(min(iris$Sepal.Width) - 0.2, 
             max(iris$Sepal.Width) + 0.2)
  )

“on the fly” as in you can tweak very big and very tiny things entirely independently

Flexibility

Iris Sepal Length vs Width: Scatter Plot + Regression Line

ggplot(iris, aes(x = Sepal.Length, 
                 y = Sepal.Width, 
                 color = Species)) +
  geom_point() + # scatter plot
  geom_smooth(method = "lm", se = TRUE) + # reg. line
  labs(title = "ggplot2: Iris Sepal Length vs Width",
       x = "Sepal Length",
       y = "Sepal Width") +
  theme_minimal()

• highly flexible
• represent the same data in dramatically different ways

Flexibility

Iris Sepal Length by Species: Violin + Box Plots

ggplot(iris, aes(x=Species, y=Sepal.Length, fill=Species)) +
  geom_violin(alpha=0.5) + # violin plots
  geom_boxplot(width=0.2, alpha=0.8) + # boxplots
  theme_minimal() +
  labs(title="Distribution of Sepal Length by Species",
       y="Sepal Length (cm)")

Flexibility

Density Distributions of Iris Measurements: Faceted Histograms

pivot_longer(iris, cols=c(Sepal.Length, Sepal.Width, Petal.Length, Petal.Width),
    names_to="Measurement",
    values_to="Value") %>% 
ggplot(aes(x=Value, fill=Species)) +
  geom_histogram(alpha=0.8, bins = 15, # histogram
    position = "identity") +
  facet_wrap(~Measurement, scales="free") + # subplots
  theme_minimal() +
  labs(title="Density Distributions of Iris Measurements",
       x="Measurement Value (cm)")

Grammar of Graphics

What do you need to generate this?

The data themselves (iris) What goes on the x and y axis & what makes the colors The kinds of graphs we want (scatter + regression line) Decorative stuff, like transparency of the dots and making the lines dashed How the axes are labeled (the range of numbers on the axes) What and where all the text should be, like axis names, title, caption, etc.

Grammatical Elements

• data: ggplot(data = df)
  • Tidy dataframe to visualize
• aesthetics: aes()
  • Mapping data structures to visual properties
• geometries: geom_*()
  • Plotting structures
• statistics: stat_*() or geom_*(stat = "...")
  • Statistical transformations before plotting

• scales: scale_[aes-type]_[data-type]()
  • Relationships between mapped data and visual space
• coordinates: coord()
  • Spatial structures for visual representation
• facets: facet_wrap() or facet_grid()
  • Subplots based on data mapping
• themes: theme()
  • Unmapped, fixed visual properties

The grammar, more or less in order of importance: Data: The data you want to visualize. This is the foundation of your visualization. Can’t map data to aesthetics without data. Aesthetics: The visual properties of the data. This is how you map your data to visual elements like color, shape, size, etc. Critically, aesthetics are the visuals that are actually mapped to your data, the stuff that will change if your data change. Ironically, “aesthetics” does not refer to the parts of your plots that are purely aesthetic1, like the color of the background or the size of the axis labels (those are in the “theme” layer). Geometries (aka “geoms”): The shapes you use to represent your data, like points, lines, bars, etc. This is pretty much just defining what kind of plot you’re making – line scatterplots, line graphs, histograms, etc. Statistics (aka “stats”): Statistical transformations that you apply to your data before plotting, like calculating means, binning observations, or fitting regression lines. Scales: How data maps onto space. The axes and legends will generate based on the data, aesthetics, and geometries you’ve defined, but scales are what determine how the data is actually represented on those axes. For example, they can determine the range of values that are represented, the breaks between values, and the labels that are used. Coordinates: The space in which your data is represented. This is where you define the type of plot you’re making – Cartesian, polar, etc. Nearly anything you plot will be Cartesian, the basic X-Y space, and you can usually ignore this layer. Facets: How you divide your data into subplots. This is useful for visualizing data that has multiple categories or dimensions, especially if you are already mapping grouping aesthetics like color, shape, or size to other variables. Themes: The non-data, non-aesthetic parts of your plot. (Except the actually are aesthetic, they just aren’t ggplot’s “aesthetics.” I hate this.) This is how you change visuals that aren’t mapped to your data, the stuff that should stay the same even if your data change. Theme layers are commonly used to customize things like appearance of grid lines, font and size of axis labels, and non-data-dependent color (like if you just want all points to be blue no matter what), but they can do much more. Themes let you customize the visual properties of nearly anything in your plot.

Layers

• Layers work together to plot elements
  • Not all layers are necessary
  • Each can have a powerful effect on outcome

ggplot(data = iris,                 # Data layer
       aes(x = Sepal.Length,        # Aesthetics
           y = Sepal.Width,
           color = Species)) +
  geom_point() +                    # Geometries
  stat_smooth(method = "lm") +      # Statistics
  scale_color_viridis_d() +         # Scale
  coord_cartesian(xlim = c(4, 8)) + # Coordinates
  facet_wrap(~Species) +            # Facets
  theme_bw() +                      # Theme 
  labs(title = "Iris Plot")

Layers

• Layers work together to plot elements
  • Not all layers are necessary
  • Each can have a powerful effect on outcome

ggplot(data = iris,                 # Data layer
       aes(x = Sepal.Length,        # Aesthetics
           y = Sepal.Width,
           color = Species)) +
  geom_point() +                    # Geometries
  #stat_smooth(method = "lm") +      # Statistics
  scale_color_viridis_d() +         # Scale
  coord_cartesian(xlim = c(4, 8)) + # Coordinates
  #facet_wrap(~Species) +            # Facets
  theme_bw() +                      # Theme 
  labs(title = "Iris Plot")

Layers

• Layers work together to plot elements
  • Not all layers are necessary
  • Each can have a powerful effect on outcome

ggplot(data = iris,                 # Data layer
       aes(x = Sepal.Length,        # Aesthetics
           y = Sepal.Width,
           color = Species)) +
  geom_point() +                    # Geometries
  stat_smooth(method = "lm") +      # Statistics
  #scale_color_viridis_d() +         # Scale
  #coord_cartesian(xlim = c(4, 8)) + # Coordinates
  facet_wrap(~Species) +            # Facets
  #theme_bw()                        # Theme 
  labs(title = "Iris Plot")

Layered Layers

Grammar of data manipulation

• dplyr pipelines
• link lines of code (pipes) with |> or %>%
• each line serves 1 purpose
• add, change, remove single lines independently
• modify data sequentially (vs. embedding)

Grammar of graphics

• ggplot2 layers
• link lines of code (layers) with +
• each line assigns attributes to 1 layer
• add, change, remove single lines independently
• layer data and visuals sequentially (vs. dependently)

• “Grammatical” elements of ggplot are called layers
  • Imagine transparent sheets with images that stack on top of one another
  • Some just add to the image; others overlap to obscure what came before
  • Sequence matters

Each line serves a single purpose Add, change, or remove any line without reworking the whole pipeline Power comes from sequencing

Each line is a layer, assigning attributes to that layer Add, change, or remove any layer without reworking the whole plot Sequencing creates a data-dependent and beautiful plot

Layer Necessity

Essential

• Data
• Aesthetics
• Geometries

Common

• Scales
• Facets
• Themes

Specialized

• Statistics
• Coordinate

Layer: Data

• You need data to plot data.
• It has to be a data frame.
• Helps if it’s tidy.

Like with dplyr pipelines, the data is always the first argument of a ggplot, which means you can combine pipes and plots

Layer: Aesthetics (aes)

• Visual properties of a ggplot that are mapped to data
• Dependent on type of plot, e.g.:
  • Use the linetype aes with a line graph, but not a scatterplot
  • Use the fill aes for filled shapes like bar graphs, but not line graphs
• Dependent on data type of variable, e.g.:
  • Use shape with categorical/discrete data, because there is a fixed set of shapes to use
  • Use size with continuous data, because size itself is continuous and unbounded
  • Use color for continuous or categorical, because colors can be assigned discretely or generated across a spectrum

Terminology: aesthetics vs “aesthetic”

• fill, color, size, etc. can be aesthetic or aesthetic

• aesthetic: data-dependent visual properties

  • defined as arguments in an aes() function

• Fixed properties use the same elements without data-dependence

  • treated as “theme” even if not part of a theme layer

Color aesthetic:

geom_point(aes(x=length, y=width, color=species))

Aesthetic color:

geom_point(aes(x=length, y=width), color="blue")

In ggplot, an aesthetic is a visual property dependent on data. They are defined as arguments in an aes() function. Visual properties that are fixed and not data-dependent use the same arguments (color, fill, size, etc.) outside of an aes() function. They are not called aesthetics, and are usually considered part of a “theme” even if they aren’t actually in a theme layer.

(bubbles) Why are the things that are purely aesthetic not called “aesthetics”?

GREAT QUESTION. I DO NOT KNOW.

aes reference (1/2)

Aesthetic	Description	Mapped data	Unmapped specs
x and y (position)	x and y coordinates of the plot. Nearly every plot requires at least one of these “position” aesthetics.	continuous, categorical	n/a
group	How observations are grouped together (if not defined with another grouping aes)	categorical	n/a
color	Color of points, lines, text, and other 1D shapes. For filled shapes, this will be the outline color.	continuous, categorical	string (color name or hex code)
fill	Fill color of 2D shapes like bars, polygons, etc.	continuous, categorical	string (color name or hex code)
alpha	Opacity/transparency of any element	continuous	number between 0 and 1

aes reference (2/2)

Aesthetic	Description	Mapped data	Unmapped specs
size	Size of points and width of lines	continuous	numeric values
shape	Shape of points, out of 26 options. Default is a solid circle (19)	categorical	integers 0-25 or shape name (string) (view guide)
linetype	Type of line, out of 6 options or blank (0). Default is solid line (1).	categorical	integers 0-6 or line name (string) (view guide)
linewidth	Width of lines	continuous	numeric values
stroke	Width of shape outlines	continuous	numeric values
label	Text content	continuous, categorical	strings
fontface	Font style for text	categorical	String: “plain”, “bold”, “italic”, or “bold.italic”
family	Font family for text	categorical	String with font name

Spot the aesthetics (1/4)

Data-dependent: x & y axis, color

Static: shape and size possibly?

Spot the aesthetics (2/4)

Data-dependent: x (continuous) & y (categorical) axis

Static: color & fill

Spot the aesthetics (3/4)

this one is kind of tricky…there’s layering going on

Data-dependent: x & y axis, color, shape

Static: color, size (x2)

Spot the aesthetics (4/4)

Geometries

• define plot shape (i.e., graph type)
• required aes defined in an aes() function
• required & optional aesthetics depend on type of geom

Scatter plots:

• geom_point()
• Require 2 continuous variables mapped to x and y
• Accept continuous or categorical variables mapped to color, shape, size, etc.

Bar charts:

• geom_bar()
• Require exactly 1 categorical variable mapped to x
• Accept categorical variable mapped to fill
• Do not allow a y mapping

Define the shape of the plot, basically the “type” of graph Require aesthetics defined in an aes() function aes() can be in its own layer or an argument of the geom_() layer Required and optional aesthetics depend on type of plot, e.g.: Scatter plots – geom_point() – require 2 continuous variables mapped to x and y aesthetics (axes) accept additional continuous or categorical variables mapped to color, shape, size… Bar charts – geom_bar(): require one categorical variable mapped to x accept an additional categorical variable mapped to fill do not allow a y mapping**

Geoms

Geom reference (1/2)

Geom Layer	Variables*	Description	Data Types
geom_histogram()	1	Creates bins and counts observations within each bin	Continuous x
geom_density()	1	Creates a smoothed density estimate	Continuous x
geom_boxplot()	1-2	Shows distribution summary with quartiles and outliers	Continuous y, Optional categorical x
geom_violin()	1-2	Shows density estimate symmetrically	Continuous y, Categorical x
geom_bar()	1-2	Creates bars with heights proportional to number of cases	Categorical x

Geom reference (2/2)

Geom Layer	Variables*	Description	Data Types
geom_point()	2	Creates a scatter plot	Continuous x & y
geom_line()	2	Connects observations in order	Continuous x & y
geom_smooth()	2	Adds a smoothed conditional mean	Continuous x & y
geom_area()	2	Creates a line plot filled to the x-axis	Continuous x & y
geom_tile()	2-3	Creates rectangles based on x and y positions	Any x & y, Optional fill

Basic Construction

Pick your plot

• ggplot’s flexibility is a double edged sword!
• It makes the plot you ask for, not caring if it is confusing, messy, uninterpretable, or pointless
  • Recall the “garbage in, garbage out” principle
• Build the plot that matches your data and tells your story

ggplot’s flexibility is a double edged sword, allowing you to create many plots that are confusing, messy, uninterpretable, and pointless Selecting the best plot for your data and message is the most important part of the data visualization process, and it’s something you-the-human have to do. The “right” visualization is the one that best communicates your message to your audience.

1 Continuous Variable

geom_histogram(): binned distribution

ggplot(iris, aes(x=Sepal.Length)) +
  geom_histogram(binwidth = 0.5, fill="steelblue", color="white") +
  labs(title="Histogram of Sepal Length",
       x="Sepal Length (cm)",
       y="Count")

geom_density(): smoothed distribution

ggplot(iris, aes(x=Sepal.Length)) +
  geom_density(fill="steelblue", alpha=0.5) +
  labs(title="Density Plot of Sepal Length",
       x="Sepal Length (cm)",
       y="Density")

1 Continuous & 1(+) Categorical Variable

geom_boxplot(): distribution with quartiles and outliers

ggplot(iris, aes(x=Species, y=Sepal.Length, fill=Species)) +
  geom_boxplot() +
  labs(title="Box Plot of Sepal Length by Species",
       x="Species",
       y="Sepal Length (cm)") 

geom_violin(): distribution summary with density estimate

ggplot(iris, aes(x=Species, y=Sepal.Length, fill=Species)) +
  geom_violin(alpha=0.5) +
  labs(title="Violin Plot of Sepal Length by Species",
       x="Species",
       y="Sepal Length (cm)")

1-2 Categorical Variables

geom_bar(): count of observations in each category

ggplot(slice_sample(iris, n=50), aes(x=Species)) +
    geom_bar(fill="steelblue", color="white") +
    labs(title="Bar Chart of Iris Species",
        x="Species",
        y="Count")

geom_col(): height of bars based on pre-summarized data

iris |>
  pivot_longer(
    cols = starts_with("Sepal"),
    names_to = "Measurement",
    values_to = "Value"
  ) |>
  group_by(Species, Measurement) |>
  summarise(Value = mean(Value), 
            .groups = "drop") |>
  ggplot(aes(x = Species, y = Value, 
             fill = Measurement)) +
  geom_col(position = "dodge") +
  labs(
    title = "Mean sepal measurements by species",
    x = "Species",
    y = "Mean (cm)"
  )

For comparing quantities across categories: - geom_bar(): Use for counting occurrences in categorical data. Default behavior counts observations in each category[57]. - geom_col(): Use when you already have pre-calculated values to plot (requires y-values)[56].

THE DIFFERENCE between bar and col is in the default stat used for mapping. in bar, it’s “count”, in col it’s “identity”

BAR CHARTS ARE BAD FOR COMPARING MEANS. Use boxplots, violin plots, or dot plots instead, which show the distribution of the data and allow for more accurate comparisons of central tendency and variability across groups.

Aside from not showing distribution, bar charts are necessarily filled from the bottom up, which gives an illusion of a meaningful baseline that something is rising above or below, “filling in” the area under the curve, which is not actually meaningful in most cases. This can lead to misinterpretation of the data, especially if the bars are not starting at zero or if the differences between groups are small.

2+ Continuous Variables

geom_point(): scatter plot

ggplot(iris, aes(x=Sepal.Length, 
                 y=Sepal.Width, color=Species)) +
  geom_point() +
  labs(title="Scatter Plot of Sepal Length vs Width",
       x="Sepal Length (cm)",
       y="Sepal Width (cm)")

geom_jitter(): scatter plot with random noise

geom_smooth(): smoothed mean (e.g., regression line)

ggplot(iris, aes(x=Sepal.Length, 
                 y=Sepal.Width, color=Species)) +
  geom_jitter() +
  geom_smooth(method="lm") +
  labs(title="Jittered Scatter Plot with Regression Line",
       x="Sepal Length (cm)",
       y="Sepal Width (cm)")

the point of jittering is to reduce overlap, an alternative is using alpha to make points more transparent

geom_smooth is rarely used alone. almost always layered over a geom_point() or geom_jitter() to show the relationship between the points and the smoothed line

by default smooth uses method = “loess”, which is a non-parametric local regression. method = “lm” fits a linear regression line, which is more interpretable and often more appropriate for simple relationships

2+ Ordered Variables

geom_line(): connects observations in order

ggplot(economics_long, aes(x=date, y=value01, color=variable)) +
  geom_line() +
  labs(title="Time Series of Economic Indicators",
       x="Date",
       y="Value (normalized)")

geom_area(): line plot filled to x-axis

ggplot(economics_long, aes(x=date, y=value01, fill=variable)) +
  geom_area(alpha=0.6) +
  labs(title="Area Plot of Economic Indicators",
       x="Date",
       y="Value (normalized)")

LINE GRAPHS ARE LESS USEFUL THAT YOU THINK. The idea is that each point is connected to the next in a meaningful way, but often the points are just measurements taken at regular intervals and the line is just a visual aid to show the overall trend. In those cases, a scatter plot or a smoothed line might be more appropriate.

They are good for time because time always moves forward at a continuous rate. One point in time is always meaningfully connected to the next, and the line helps show the overall trend. But for other types of data, the line can be misleading if it implies a connection that isn’t really there.