The Grammar of Graphics

Understand the grammar of graphics framework. Learn how to map data to visual properties. Explore different aesthetic mappings beyond x and y coordinates.

NoteLearning Objectives
  • Understand the grammar of graphics framework
  • Map data variables to different visual properties (aesthetics)
  • Use color, size, and shape to encode information
  • Create multi-dimensional visualizations
  • Choose appropriate aesthetic mappings for different data types
TipKey Questions
  • What is the grammar of graphics?
  • How do I map data to visual properties?
  • What aesthetic properties can I use?
  • How do I choose the right mappings for my data?

The Grammar of Graphics Framework

The grammar of graphics is a systematic approach to creating data visualizations. Instead of thinking about “chart types” (bar chart, scatter plot, etc.), we think about the components that make up any visualization.

Think of it like building with LEGO blocks:

  • You don’t have pre-made houses or cars
  • You have basic blocks that you can combine in infinite ways
  • Different combinations create different results

Similarly, data visualizations are built by combining:

  1. Data - What you want to visualize
  2. Aesthetic mappings - How data variables map to visual properties
  3. Geometric objects (marks) - The shapes that represent data
  4. Scales - How data values translate to visual values
  5. Coordinate systems - The space where the plot is drawn
  6. Facets - Subplots for different data subsets

Visual Properties (Aesthetics)

When we create a visualization, we’re encoding data as visual properties. The main aesthetic properties we can use are:

Aesthetic What it controls Best for
x Horizontal position Continuous or categorical data
y Vertical position Continuous or categorical data
color Color of marks Categorical groups or continuous gradients
pointsize Size of marks Continuous values (showing magnitude)
marker Shape of marks Categorical groups (limited categories)
alpha Transparency Emphasis or showing density
stroke Border thickness Emphasis or additional grouping
fill Fill color Similar to color, for filled shapes

Setting Up

Let’s load our libraries and data:

import seaborn as sns
import seaborn.objects as so
import pandas as pd

# Load the penguins dataset
penguins = sns.load_dataset("penguins")

# Remove rows with missing values for cleaner examples
penguins_clean = penguins.dropna()

Mapping Color

Color is one of the most powerful ways to encode information. Let’s see how it works:

# Color by categorical variable (species)
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        color="species"
    )
    .add(so.Dot())
)

Here, each species gets a different color automatically. Seaborn:

  • Assigns distinct colors to each category
  • Creates a legend
  • Maintains consistent colors across plots

Color for Continuous Variables

Color can also represent continuous data:

# Color by continuous variable (body mass)
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        color="body_mass_g"
    )
    .add(so.Dot())
)

Notice how the color gradient shows the range of body mass values. Lighter or darker colors indicate higher or lower values.

NoteChoosing Colors for Data Types
  • Categorical data (species, gender, treatment group) → Use distinct colors
  • Continuous data (income, temperature, age) → Use color gradients
  • Diverging data (change from baseline, profit/loss) → Use diverging palettes (e.g., blue-white-red)

Mapping Size

Size is excellent for showing magnitude or importance:

# Size by continuous variable
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        pointsize="body_mass_g"
    )
    .add(so.Dot())
)

Larger dots represent penguins with greater body mass. This creates a “bubble plot” where we’re visualizing three variables simultaneously!

Combining Multiple Aesthetics

Let’s get more sophisticated by combining color AND size:

# Multi-dimensional visualization
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        color="species",
        pointsize="body_mass_g"
    )
    .add(so.Dot(alpha=0.6))  # Added transparency to see overlapping points
)

Now we’re showing four variables at once:

  • Bill length (x-axis)
  • Bill depth (y-axis)
  • Species (color)
  • Body mass (size)

This is powerful for exploratory data analysis!

Mapping Shape (Marker)

Shape (called marker in seaborn.objects) is useful for categorical variables, especially when printing in black and white:

# Different shapes for different categories
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        marker="species"
    )
    .add(so.Dot())
)
WarningLimit Shape Categories

Don’t use shape for variables with many categories (more than 6-7). It becomes hard to distinguish between shapes. For many categories, use color instead.

Combining Color and Shape

For maximum clarity and accessibility (including for colorblind viewers), combine color and shape:

# Both color and shape for species
(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        color="species",
        marker="species"
    )
    .add(so.Dot())
)

Now species are distinguished by both color AND shape - making the plot accessible to everyone!

Using Alpha (Transparency)

Alpha controls transparency and is useful when points overlap:

# Without alpha - hard to see overlapping points
(
    so.Plot(
        penguins_clean,
        x="flipper_length_mm",
        y="body_mass_g"
    )
    .add(so.Dot())
)

# With alpha - can see density of overlapping points
(
    so.Plot(
        penguins_clean,
        x="flipper_length_mm",
        y="body_mass_g"
    )
    .add(so.Dot(alpha=0.5))
)

Where many points overlap, the color becomes darker, showing areas of high density.

Research Example: Education and Income Data

Let’s imagine we have data from a household survey in Kenya. Here’s how we might use different aesthetics:

# Create sample research data
import numpy as np

np.random.seed(42)
n = 200

research_data = pd.DataFrame({
    'household_income': np.random.lognormal(10, 1, n),
    'years_education': np.random.normal(8, 3, n),
    'household_size': np.random.randint(1, 10, n),
    'county': np.random.choice(['Nairobi', 'Kisumu', 'Mombasa', 'Nakuru'], n),
    'program_participant': np.random.choice(['Yes', 'No'], n)
})

# Ensure positive values
research_data['years_education'] = research_data['years_education'].clip(lower=0)

Now let’s visualize this data with multiple aesthetics:

# Rich multi-dimensional visualization
(
    so.Plot(
        research_data,
        x="years_education",
        y="household_income",
        color="county",
        pointsize="household_size",
        marker="program_participant"
    )
    .add(so.Dot(alpha=0.6))
)

This single plot shows:

  • Education level (x-axis)
  • Household income (y-axis)
  • County (color)
  • Household size (point size)
  • Program participation (shape)

That’s five variables in one visualization!

Exercises

NoteExercise 1: Experiment with Aesthetics

Using the penguins dataset, create a plot that shows:

  • flipper_length_mm on the x-axis
  • body_mass_g on the y-axis
  • island as color
  • sex as marker shape
# Your code here
# Clean data to remove missing sex values
penguins_complete = penguins.dropna(subset=['sex'])

(
    so.Plot(
        penguins_complete,
        x="flipper_length_mm",
        y="body_mass_g",
        color="island",
        marker="sex"
    )
    .add(so.Dot(alpha=0.7))
)

What patterns do you see? Do males and females differ in body mass? Do penguins from different islands cluster differently?

NoteExercise 2: Too Many Aesthetics?

Create a plot with the penguins data that uses all of these aesthetics:

  • x, y, color, pointsize, marker, alpha

Then discuss: Is this plot easy to understand? When does adding more aesthetics become confusing rather than helpful?

(
    so.Plot(
        penguins_clean,
        x="bill_length_mm",
        y="bill_depth_mm",
        color="species",
        pointsize="body_mass_g",
        marker="sex"
    )
    .add(so.Dot(alpha=0.5))
)

Discussion: While technically you can encode many variables, there’s a limit to how much information people can process visually. Generally:

  • 3-4 variables is very manageable
  • 5 variables can work if the relationships are clear
  • 6+ variables usually creates confusion

Consider whether faceting (small multiples) might be clearer than packing everything into one plot.

NoteExercise 3: Research Application

Think about a dataset from your own research work. Sketch out (on paper or describe in words) a visualization that would show:

  1. What would be on your x and y axes?
  2. What categorical variable would you show with color?
  3. Is there a continuous variable you could show with size?
  4. Would you need shape for an additional categorical variable?

Discuss with a partner: Would stakeholders understand this visualization? What would make it clearer?

Choosing the Right Aesthetics

Here are some guidelines for choosing aesthetic mappings:

For Position (x, y)

  • The most important variables should go on axes
  • Position is the most accurate way humans perceive quantitative data
  • Typically: independent variable on x, dependent variable on y

For Color

  • Categorical: Use for up to 8-10 categories maximum
  • Continuous: Use for highlighting patterns or gradients
  • Ensure sufficient contrast between colors
  • Consider colorblind-friendly palettes

For Size

  • Best for continuous positive values
  • Don’t use size for negative values or data that includes zero
  • Humans aren’t great at comparing areas - use size as a supplementary encoding

For Shape

  • Only for categorical data
  • Limit to 3-6 categories maximum
  • Essential for black-and-white printing
  • Combine with color for accessibility

For Alpha (Transparency)

  • Use to show density when points overlap
  • Use to de-emphasize less important data
  • Usually set as a constant (like alpha=0.5), not mapped to data

Common Pitfalls

WarningThings to Avoid

  1. Rainbow color schemes for continuous data - they’re not perceptually uniform and are problematic for colorblind viewers

  2. Too many aesthetics - More isn’t always better. Keep it simple enough to understand.

  3. Inappropriate mappings - Don’t use size for categorical data or shape for continuous data

  4. No legend - Always ensure viewers can decode your aesthetic mappings

  5. Inconsistent mappings - If you use color for species in one plot, use the same mapping in related plots

ImportantKey Points
  • The grammar of graphics breaks visualizations into reusable components
  • Aesthetic mappings connect data variables to visual properties
  • Main aesthetics: position (x, y), color, size (pointsize), shape (marker), and alpha
  • Color works for both categorical and continuous data
  • Size is best for continuous positive values
  • Shape works for small numbers of categories
  • Combining multiple aesthetics creates multi-dimensional visualizations
  • Choose aesthetics based on your data type and what you want to emphasize
  • Always consider accessibility and clarity for your audience
TipLooking Ahead

In the next lesson, we’ll explore different types of marks (geometric objects) beyond dots - including lines, bars, areas, and bands. We’ll learn how to choose the right mark for your data and research questions.

Back to top