VMC: A Grammar for Visualizing Statistical Model Checks

Area of Science:

• Statistics
• Data Visualization
• Computational Statistics

Background:

• Visualizations are crucial for validating and refining statistical models.
• The design principles for effective graphical model checks are not well-defined, hindering exploration.
• Existing tools offer limited support for systematic creation of model check visualizations.

Purpose of the Study:

• To define a structured framework for creating model check visualizations.
• To develop an R package implementation of the Visualization for Model Checking (VMC) framework.
• To evaluate the effectiveness of VMC in generating accurate and exploratory model check visualizations.

Main Methods:

• Defined Visualization for Model Checking (VMC) based on four components: data samples, transformations, visual representations, and layouts.
• Implemented VMC as an R package.
• Validated VMC by reproducing canonical model check examples and comparing visualization edit distances with existing toolkits.
• Conducted an interview study with expert modelers on VMC usability.

Main Results:

• VMC provides a systematic approach to designing model check visualizations.
• The VMC R package facilitates the generation of model check visualizations.
• Using VMC reduced the edit distance between visualizations compared to existing toolkits.
• Expert modelers identified challenges and opportunities for VMC in promoting effective visualization exploration.

Conclusions:

• VMC offers a structured method for developing effective graphical model checks.
• The VMC R package can improve the process of statistical model validation.
• Further development of VMC can enhance the exploration and creation of accurate model check visualizations.