Effects of Point Size and Opacity Adjustments in Scatterplots

Title of the Paper

Effects of Point Size and Opacity Adjustments in Scatterplots

Paper Information

• Research Area: Data Visualization and Human-Computer Vision Interaction
• Keywords: Scatterplots, Correlation, Perception, Point Size, Opacity Adjustment, Visual Design, User Study, Experimental Analysis

Research Background and Problem

• Problem or Challenge:

  • People generally exhibit low accuracy in estimating positive correlations in scatterplots, particularly within the correlation range of 0.2 to 0.6.
  • Current designs may fail to effectively support accurate interpretation of data visualizations by the general public.
  • Adjusting point size and opacity is believed to enhance the accuracy of correlation estimation in scatterplots, but the optimal combination of these adjustments remains unclear.

• Significance:

  • The COVID-19 pandemic highlighted the public's need to accurately interpret data visualizations daily to support decision-making.
  • This research can contribute to designing data visualizations that are more interpretable for users lacking statistical or graphical training.

• Research Motivation:

  • To explore the impact of visual features (e.g., point size and opacity) and their combinations on users' perception of correlation.
  • To propose a data visualization design framework based on human perceptual mechanisms.

Solution

• Proposed Method:

  • The authors conducted experiments to separately adjust point size and opacity in scatterplots, testing the effects of their combination on correlation estimation.
  • Specific configurations included: combinations of typical and reversed decay functions for point size and opacity, resulting in four experimental conditions.

• Innovations:

  • Proposed and validated the nonlinear interaction effects of point size and opacity adjustments, rather than simple additive effects.
  • Built on prior research to propose a visual optimization framework based on human perceptual mechanisms.

• Implementation Steps and Techniques:

  • Experimental Methodology:
    1. Randomly generated scatterplot datasets with defined correlations (r ranging from 0.2 to 0.99).
    2. Designed two styles of decay functions for point size and opacity (typical decay and reversed decay) and applied them to scatterplots.
    3. Conducted correlation estimation experiments with online crowdsourced participants, recording errors and accuracy.
  • Key Techniques:
    • Used linear mixed-effects models to analyze participants' estimation errors.
    • Controlled point opacity through graphical parameter α values; controlled point size using a mapping formula based on residuals and regression line distance.

Research Findings

• Specific Findings:

  • Discovered that the combination of point size and opacity exhibits nonlinear interaction effects, significantly influencing correlation estimation accuracy.
  • Correlation estimation accuracy was highest when combining typical decay for both point size and opacity, though it led to overcorrection effects, causing overestimation in most cases.
  • Correlation estimation accuracy was lowest, with the largest errors, when combining reversed decay for both point size and opacity.

• Advantages Compared to Existing Solutions:

  • Unlike prior studies that focused on a single adjustment method, this research provides the first exploration of the combined effects of two adjustment methods.
  • Demonstrated that point size adjustments have a stronger effect, with typical decay significantly outperforming reversed decay.

• Experimental or Evaluation Results:

  • In most cases, the typical decay direction positively improved perceptual accuracy, while the reversed decay direction was unhelpful and deviated from participants' accurate estimations.
  • The proposed typical combination was significantly associated with correlation estimation, explaining approximately 10.4% of the error variance.

• Limitations and Future Directions:

  • Limitations:
    • The smallest point size and opacity might not be visible on some participants' devices, affecting data perception.
    • The current experiment did not isolate the individual contributions of size and opacity adjustments.
    • The online crowdsourced experiment environment may have introduced device configuration variability, affecting data homogeneity.
  • Future Directions:
    • Conduct experiments on negatively correlated scatterplots to observe the symmetry of size and opacity adjustments.
    • Further optimize adjustment parameters to identify the best combination methods (including new mathematical formulas or multi-dimensional channel encoding).
    • Extend the experimental framework to other chart types (e.g., bar charts, heatmaps) or other tasks (e.g., cluster detection, outlier identification).

Summary and Contributions

• Provided baseline experimental data on point size and opacity adjustment methods for optimizing correlation estimation in scatterplots.
• Enriched theoretical research on how visual features influence numerical perception.
• Offered an open experimental framework to support further empirical research on data visualization design.