MediVizor: Visual Mediation Analysis of Nominal Variables

Area of Science:

• Social Sciences
• Communication Science
• Sports Science

Background:

• Mediation analysis is vital for understanding indirect causal relationships across scientific disciplines.
• Analyzing nominal variables in mediation is complex, often requiring tedious comparisons of multiple effects using classical tools like Excel.
• Existing methods present challenges for efficiently examining direct and indirect causal effects within mediation models.

Purpose of the Study:

• To design and evaluate MediVizor, a novel visualization system for conducting visual mediation analysis of nominal variables.
• To enable researchers to easily browse, compare, and understand the composition of total effects from direct and indirect causal effects.
• To facilitate the examination of how positive and negative effects contribute to or reduce total effects.

Main Methods:

• Collaborative design of MediVizor with domain experts in sports and communication science.
• Development of a visualization system enabling users to compare multiple total effects and their constituent direct/indirect effects.
• Evaluation through two domain-specific case studies and a user study with general users.

Main Results:

• MediVizor allows users to visually explore and compare multiple total effects and their underlying direct/indirect effects.
• The system effectively illustrates the contribution of positive and negative effects to total effects.
• Case studies and user studies yielded positive feedback, confirming the system's effectiveness and generalizability.

Conclusions:

• MediVizor offers an effective visual approach to mediation analysis for nominal variables, overcoming limitations of traditional tools.
• The system enhances the ability of researchers to interpret complex causal relationships.
• The design is versatile and applicable across different scientific domains, as demonstrated by its successful application in sports and communication science.