Week 9 — Color & Accessibility

Understanding color theory, palette selection, and accessibility best practices in data visualization.

---

📖 Background & Motivation

Color is one of the most powerful and expressive tools in visualization. It can highlight differences, show magnitude, encode categories, or evoke emotion—but it can also distort, confuse, or exclude if not used carefully. This module explores color theory, perceptual uniformity, and accessibility guidelines to ensure your visuals communicate effectively to all audiences.

We will discuss the use of color in sequential, diverging, and categorical scales, and how to design visualizations that remain interpretable for viewers with color vision deficiencies.

---

🔎 Learning Objectives

• Distinguish between sequential, diverging, and categorical color schemes.
• Select appropriate palettes based on data type and context.
• Apply color theory concepts (hue, saturation, luminance) to improve legibility.
• Use color tools to ensure accessibility for all users.
• Test visualizations for color blindness and accessibility compliance.

---

📚 Readings & Resources

• Fundamentals of Data Visualization — chapters on color and perception.
• ColorBrewer2 — tool for choosing effective color schemes.
• Colorcet - Perceptually accurate 256-color colormaps
• Adobe Color — create custom palettes.
• Coblis — Color Blindness Simulator
• Accessible Data Visualization Guide (W3C)
• Seaborn color palettes documentation

Sample Data Sources:

• Gapminder dataset (for categorical and quantitative mapping)
• U.S. demographic data (for diverging and sequential maps)
• Simulated survey data (for categorical comparison)

---

🛠️ Setup Checklist

Ensure your environment includes everything the notebook imports:

pip install numpy pandas matplotlib seaborn colorcet plotly altair vega_datasets geopandas mapclassify

Don't skip mapclassify. The notebook never imports it by name, but GeoPandas needs it the moment you pass scheme="Quantiles" or scheme="FisherJenks" to .plot(). Without it those calls fail with an error that does not mention the missing package.

Confirm you can load the sample datasets (seaborn, vega_datasets, plotly.express all download theirs on first use, so you need a network connection) and that data/map_us_only.geojson and data/unemployment.csv sit next to the notebook.

---

🧭 Lecture Outline

Session 1 (Theory + Hands-on)

1. Introduction: Why color matters in visualization
2. Color theory basics — hue, saturation, brightness
3. Types of color schemes — sequential, diverging, categorical
4. Color perception & accessibility — common pitfalls
5. Experiment with Seaborn and Matplotlib color palettes
6. Apply ColorBrewer and Colorcet palettes to the previous week’s map visualizations
7. Classification and class breaks — quantile vs equal-interval vs Fisher-Jenks
8. Download the Jupyter Notebook and Data

---

💻 Starter Notebook Snippets

These are taken from the course notebook (Week09_Color_Accessibility_Demos.ipynb in the download above), so you can follow along or run them on your own data.

1. Match the palette to the data type

import numpy as np
import matplotlib.pyplot as plt

# Sequential: one direction, low to high
data_seq = np.outer(np.linspace(0, 1, 120), np.ones(120))
plt.imshow(data_seq, aspect="auto", cmap="viridis")
plt.title("Sequential colormap: viridis (ordered data)")
plt.colorbar()
plt.show()

# Diverging: a meaningful midpoint at 0
data_div = np.outer(np.linspace(-1, 1, 120), np.ones(120))
plt.imshow(data_div, aspect="auto", cmap="coolwarm")
plt.title("Diverging colormap: coolwarm (deviation around 0)")
plt.colorbar()
plt.show()

2. Check text contrast against WCAG

def srgb_to_linear(c):
    """One sRGB channel (0-255) -> linear light (0-1).

    WCAG defines luminance on linear light, so the gamma has to come off first.
    """
    c = c / 255.0
    return c / 12.92 if c <= 0.04045 else ((c + 0.055) / 1.055) ** 2.4


def rel_luminance(rgb):
    r, g, b = (srgb_to_linear(v) for v in rgb)
    return 0.2126 * r + 0.7152 * g + 0.0722 * b


def contrast_ratio(fg, bg):
    L1, L2 = rel_luminance(fg), rel_luminance(bg)
    hi, lo = max(L1, L2), min(L1, L2)
    return (hi + 0.05) / (lo + 0.05)


contrast_ratio((33, 33, 33), (255, 255, 255))     # 16.10 -> passes
contrast_ratio((150, 150, 150), (255, 255, 255))  #  2.96 -> fails (needs 4.5)

3. Simulate color vision deficiency

import numpy as np

def srgb_to_linear_arr(c):
    return np.where(c <= 0.04045, c / 12.92, ((c + 0.055) / 1.055) ** 2.4)


def linear_to_srgb_arr(c):
    c = np.clip(c, 0, 1)
    return np.where(c <= 0.0031308, c * 12.92, 1.055 * c ** (1 / 2.4) - 0.055)


# Vienot, Brettel & Mollon (1999). The matrix is only valid in LINEAR light,
# so undo the sRGB gamma first -- skipping that step overstates the damage and
# can make a safe palette look as bad as a risky one.
DEUTERANOPIA = np.array([[ 0.2920, 0.7054, -0.0003],
                         [ 0.2934, 0.7089,  0.0000],
                         [-0.0209, 0.0272,  0.9922]])


def simulate_cvd(rgb, M):
    """rgb: array of floats 0-1, shape (..., 3)."""
    lin = srgb_to_linear_arr(np.asarray(rgb, float))
    out = np.clip(lin.reshape(-1, 3) @ M.T, 0, 1).reshape(np.shape(rgb))
    return np.clip(linear_to_srgb_arr(out), 0, 1)

Run it on a red/green pair and a blue/orange pair and compare how far apart the two colors stay. Red and green collapse into two olives; blue and orange do not.

4. Class breaks change the map

import pandas as pd
import geopandas as gpd

polys = gpd.read_file("data/map_us_only.geojson")
attrs = pd.read_csv("data/unemployment.csv")
g = pd.merge(polys, attrs, left_on="GEO_ID", right_on="Geo_GEO_ID", how="left")
g = g.to_crs("ESRI:102003")          # equal-area projection

# Social Explorer table A17002. The unemployment rate is
# unemployed / civilian labor force -- NOT a share of total population.
g["unemp_rate"] = g["SE_A17002_006"] / g["SE_A17002_004"] * 100

for scheme in ("EqualInterval", "Quantiles", "FisherJenks"):
    ax = g.plot(column="unemp_rate", cmap="OrRd", scheme=scheme, k=5,
                legend=True, edgecolor="white", linewidth=0.2)
    ax.set_title(scheme)
    ax.set_axis_off()

Same data, same palette, three different maps. Say in your caption which classification you used and why.

---

🧪 In-Class Activity

• Create 3 mini visualizations (one sequential, one diverging, one categorical) using Seaborn.
• Use ColorBrewer to test and adjust color schemes.
• Simulate color blindness and discuss readability.
• Redesign a prior week’s map or chart with improved accessibility.

---

🏠 Homework

1. Choose one of your visualizations from a previous week and redesign it to improve color and accessibility.
2. Include:
   • One before-and-after comparison of color schemes.
   • Discussion (150–250 words) of how your new design improves clarity and inclusivity.
   • Verification of color accessibility (using simulator or contrast tool).
3. Submit .ipynb and .html.

Rubric (10 pts)

• Appropriate use of color schemes (3)
• Accessibility testing and discussion (3)
• Clarity and improvement of redesigned visual (2)
• Reproducibility and documentation (2)

---

🧩 Optional Extensions

• Explore colorcet for perceptually uniform colormaps.
• Create a color legend that works both in grayscale and color.
• Compare the performance of different palettes in color‑blind simulation.

---

✅ Submission Checklist

Before submitting, make sure:

• Your assignment has fulfilled all the basic requirements listed above.

• Use Quarto to render the notebook into HTML and zip the files for submission.

• Double-check the visualizations and your reflections in the HTML are properly organized and displayed.