How Data Analysts Use a Visualization Grammar in Practice

Title of the Paper

How Data Analysts Use a Visualization Grammar in Practice

Paper Information

• Subject Area: Data Visualization and Human-Computer Interaction
• Keywords: Visualization Grammar, TidyTuesday, Computational Notebooks, Data Analysis, ggplot2, Data Wrangling, Data Visualization, Layered Grammar, Interactive Visualization, Visualization Errors

Research Background and Problem

• Identified Problems or Challenges:

  • Although visualization tools based on the Grammar of Graphics (GoG) have theoretical advantages, there is limited understanding of how analysts actually use these tools and the difficulties they may encounter.
  • The design of visualization grammars may not fully align with analysts' actual practices and needs, potentially leading to hard-to-detect "silent errors," i.e., errors that produce reasonable outputs but contain underlying issues.
  • There is a lack of research on how visualization grammars integrate with real-world data analysis workflows.

• Significance:

  • Understanding how analysts genuinely use visualization tools can help improve grammar design, making them more practical and user-friendly, thereby helping analysts avoid errors.

• Research Motivation and Related Work:

  • GoG-based visualization tools (e.g., ggplot2 in R and Vega in JavaScript) have been widely adopted in recent years.
  • Existing studies often rely on experimental methods, such as asking participants to replicate given visualizations, which fail to reveal how analysts iteratively explore data through visualization.
  • There is a need to deeply understand how analysts use visualization grammars in real-world programming environments like computational notebooks to inform tool design improvements.

Solution

• Research Methods:

  • Conducted a qualitative study with six intermediate to advanced users from the TidyTuesday community.
  • Collected screen recordings of users performing data analysis and visualization with ggplot2, followed by retrospective interviews.
  • Applied "reflexive thematic analysis" to analyze user data and extract patterns and usage behaviors.

• Key Findings and Innovations:

  1. Alignment of GoG Grammar Components with Analysts' Tasks:
     • The intended design of the grammar (e.g., modularity) does not always align directly with analysts' customization needs or practical requirements.
  2. Feedback Loop Between Analysis and Visualization:
     • A "feedback loop" exists between data wrangling and visualization operations, where data outputs influence visualization design, and visualization results, in turn, affect data wrangling.
     • This highly coupled process can improve efficiency but may also lead to consistency issues and potential errors.
  3. Silent Errors and Challenges in Evaluating Visualizations:
     • Common silent errors include mismatches in data semantics, which can occur between data wrangling and visualization.
  4. Analysts' Iterative Patterns:
     • Two iterative patterns were observed: "incremental" and "experimental," which support the original design intent of the GoG grammar.
  5. Use and Need for Visualization Templates:
     • Analysts tend to rely on "visualization templates," which help them quickly create similar charts.

• Implementation Steps and Key Techniques:

  • Used structured observation and interview techniques to document analysts' visualization and wrangling processes in TidyTuesday projects.
  • Extracted code logs and behavioral patterns from screen recordings and interview data using reflexive thematic analysis.
  • Analyzed the differences between user-customized grammars and the standard GoG design from statistical and graphical perspectives.

Research Outcomes

• Specific Outcomes:

  • Identified typical patterns in analysts' use of GoG grammar and common difficulties encountered during grammar execution and iterative processes.
  • Conducted quantitative analysis of usage habits in the TidyTuesday community codebase, finding that participants' usage patterns were representative.
  • Proposed key design recommendations for improving visualization grammars, including support for custom templates and enhanced consistency validation between data and visualizations.

• Advantages:

  • The study is based on real user data, covering various challenges encountered in practical analysis applications rather than theoretical or lab-based explorations.
  • Implemented an integrated research framework combining analysis and feedback, capable of revealing complex iterative processes.

• Experimental or Evaluation Results:

  • Participants' average code files involved reasonable amounts of data wrangling and visualization editing, consistent with typical TidyTuesday practices.
  • Identified two main types of silent errors during the experiments and provided specific case analyses.

• Limitations and Future Directions:

  • The participant group may be biased, as all users were from the active TidyTuesday community and likely more familiar with ggplot2 and related tools.
  • Future research could expand to other GoG-style visualization grammar tools (e.g., Vega-Lite).
  • Further exploration is needed on improving visualization grammars in programming environments (e.g., computational notebooks), focusing on enhancing transparency and consistency in analysis and visualization tasks.