Correspondence Analysis Example 2025

So this is a toy repo to illustrate a cheap hack - taking tables that can be interpreted as some measure of closeness/distance between entities such as brands - and reducing that to a two dimensional map.

What This Demonstrates

Correspondence Analysis (CA) reveals hidden relationships in contingency tables by mapping categories into a low-dimensional space that preserves chi-square distances. Perfect for market research, survey analysis, and categorical data exploration.

Examplsa

There are three examples here - two from real data and one synthetic. The code to generate the synthetic data is also is the repo to give you a sense of the end to end process.

1. Energy Company Cross-Visitation - Website overlap patterns (42.5% + 39.5% variance)
2. Categorical Supplier Analysis - Business relationship mapping (44.9% + 34.1% variance)
3. 🆕 Synthetic Airline Market Segmentation - 31.19% + 8.55% = 39.74% variance

Market Insights baked in

In the simulated airline data we bake in...

• Budget vs Premium: Clear separation between Ryanair and traditional carriers
• Geographic Preferences: UK business (BA/Virgin) vs European business (Lufthansa/Air France)
• User Behavior: Realistic segmentation with zero crossover between budget-conscious and business travelers

Quick Start

# Clone and run
git clone https://github.com/spm1001/correspondence-example-2025.git
cd correspondence-example-2025
uv sync
uv run correspondence_demo.py  # Runs all three analyses

Individual analyses:

uv run correspondence_demo.py data.csv                           # Energy companies
uv run correspondence_demo.py co_occurrence_cat.csv             # Suppliers  
uv run correspondence_demo.py airline_usertype_contingency.csv  # Airlines (39.74% variance!)

Example output

• data_correspondence_analysis.png - Energy market analysis
• co_occurrence_cat_correspondence_analysis.png - B2B supplier relationships
• airline_usertype_contingency_correspondence_analysis.png - Market segmentation showcase

Synthetic Data Pipeline:

uv run generate_airline_data.py          # Create 100k realistic user sessions
uv run create_proper_contingency_table.py # Transform to CA-ready format
uv run correspondence_demo.py airline_usertype_contingency.csv # Analyze

Mathematical stuff:

• SVD decomposition of standardized residuals
• Chi-square distance preservation
• Proper eigenvalue/coordinate calculations
• Equivalent to R's FactoMineR but optimized for Python