Generating Coherent Visualization Sequences for Multivariate Data by Causal Graph Traversal.

Area of Science:

• Data Visualization
• Causal Inference
• Human-Computer Interaction

Background:

• Multivariate data analysis presents challenges in information navigation.
• Existing visualization techniques often rely on statistical metrics like correlation.
• Ordering bivariate scatterplots or parallel coordinates axes is crucial for effective data interpretation.

Purpose of the Study:

• To develop and evaluate novel visualization sequences for multivariate data.
• To incorporate causal relationships into the ordering of visualization elements.
• To enhance user comprehension of underlying causal structures within data.

Main Methods:

• Deriving causal graphs from multivariate data.
• Implementing semantic traversal schemes based on causal relationships.
• Utilizing bivariate scatterplots and parallel coordinates plots for data representation.
• Conducting crowd-sourced user studies and interviews for evaluation.

Main Results:

• Causality-informed visualization sequences significantly improve user understanding of causal relationships.
• The proposed method outperforms sequences based solely on correlation or randomization.
• User studies confirm enhanced grasp of data causality.

Conclusions:

• Incorporating causal relationships into visualization sequences offers a superior approach for exploring multivariate data.
• This method aids users in identifying and understanding complex causal structures.
• Future work can explore diverse causal discovery algorithms and traversal strategies.