Reinventing the Contingency Wheel: Scalable Visual Analytics of Large Categorical Data.

Area of Science:

• Data Visualization
• Visual Analytics
• Statistical Computing

Background:

• Contingency tables are crucial for analyzing relationships between categorical variables in science and business.
• Large, asymmetric two-way contingency tables present challenges for existing visualization techniques.
• The current Contingency Wheel method has limitations in scalability and readability for large, dense tables.

Purpose of the Study:

• To introduce Contingent Wheel++, an advanced visual analytics method.
• To overcome the scalability and readability limitations of existing methods for large contingency tables.
• To enhance the analysis of associations within large categorical datasets.

Main Methods:

• Developed an improved measure of association using Pearson's residuals to address bias in raw residuals.
• Implemented a frequency-based abstraction for visual elements to eliminate overlap and enable analysis of positive/negative associations.
• Created a multi-level overview+detail interface for interactive exploration of aggregated data.

Main Results:

• Contingency Wheel++ effectively handles large and dense contingency tables.
• The new methods facilitate the discovery of nontrivial patterns and associations.
• The approach enables detailed analysis of individual data items through coordinated views.

Conclusions:

• Contingency Wheel++ significantly advances the visual analytics of large contingency tables.
• The enhanced methods provide a more scalable and readable approach to exploring complex categorical data.
• This work offers powerful tools for uncovering hidden relationships in large datasets.