Discriminability Tests for Visualization Effectiveness and Scalability.

Area of Science:

• Information Visualization
• Human-Computer Interaction
• Computer Vision

Background:

• Visualization scalability depends on users discerning plot differences across datasets.
• Current visualization design often lacks evaluation across diverse datasets.
• Mismatch between data characteristics and visual encoding hinders effectiveness.

Purpose of the Study:

• To explore the use of Multi-Scale Structural Similarity Index (MS-SSIM) for assessing data visualization discriminability.
• To evaluate MS-SSIM's ability to capture perceptible changes in visualizations across various datasets.
• To provide a computational method for ranking visualization encodings based on discriminability.

Main Methods:

• Utilized MS-SSIM, an image similarity measure, to quantify visualization differences across scales.
• Conducted empirical studies comparing human similarity judgments with MS-SSIM scores for scatterplots.
• Calculated discriminability values for basic visualizations and compared them with empirical effectiveness measurements.

Main Results:

• MS-SSIM effectively captures both granular and high-level pattern similarities between visualizations.
• Significant data changes missed by MS-SSIM correlate with low visualization discriminability.
• The computational MS-SSIM measure closely approximates empirical human judgments and effectiveness.

Conclusions:

• MS-SSIM serves as a valuable computational tool for evaluating data visualization discriminability.
• This approach aids in selecting appropriate visualizations for specific data distributions.
• The method can objectively rank visualization encodings, enhancing design and evaluation processes.