Preliminary Guidelines for Combining Data Integration and Visual Data Analysis

Ashley Suh, Adam Coscia, Ashley Suh 0001, Remco Chang, Alex Endert

Introduction

In modern data analysis, the processes of data preparation and visual exploration are often treated as distinct, sequential phases. Analysts frequently encounter a "context-switching tax" when they realize mid-analysis that essential data resides in an external file. This paper investigates the integration of data joining directly into the visual analysis workflow—a concept termed in-situ integration. By merging these traditionally separate tasks, the researchers aim to reduce cognitive hurdles and allow analysts to maintain their "flow" during discovery.

Methodology

The study employed a within-subjects design with 16 participants to compare two interface archetypes:

  1. The Separated Interface: This baseline represents the traditional workflow. Participants used Microsoft Excel to manually join CSV files before importing the data into a visualization tool.
  2. The Combined Interface: This experimental tool allowed users to perform joins "on the fly." By selecting a join key, users could instantly access and integrate columns from external files directly within the visualization interface, such as in axis selection or filtering menus.
The Combined Interface design allows for seamless attribute integration across various panels.

Participants were tasked with both specific data retrieval and free-form insight generation using three multi-file datasets: Colleges (34 attributes across six files), Loans (27 attributes across six files), and a smaller Movies dataset for training.

Key Insights and User Behavior

The study revealed a significant shift in how users allocated their time. In the Separated interface, data preparation consumed 25–65% of the session. Conversely, the Combined interface reduced this to 10–30%, allowing the "saved" time to be reinvested into actual analysis and sensemaking.

The researchers identified four distinct integration strategies:

  • S1 (Integrate First): Proactively shortlisting all potentially useful attributes.
  • S2 (Direct Use): Integrating attributes directly within functional panels without adding them to a master list.
  • S3 (Demand-Driven): Integrating only when specific analysis questions required it.
  • S4 (Incremental): A meticulous, one-at-a-time approach.
A live view of the Combined Interface during a task involving the Colleges dataset.

Despite the efficiency gains, the study uncovered a "Trust Paradox." Some participants preferred the manual Excel method because it offered transparency; they could verify the raw data rows. Automated joins in the Combined interface were sometimes perceived as a "black box," suggesting that future tools must prioritize provenance and transparency. Additionally, an "anchoring bias" was observed: even with easy integration, some users stuck to existing attributes rather than exploring more relevant ones in external files.

Results and Discussion

While there was no statistically significant difference in total task completion time, the quality of time spent differed. The Combined interface facilitated more meaningful exploration and hypothesis testing without the fear of breaking the analytical train of thought. Interaction volumes remained stable, indicating that ease of integration does not necessarily lead to an explosion of trivial interactions.

Conclusion and Future Work

The paper concludes with preliminary guidelines for designers: prioritize in-situ integration for exploratory tasks, maintain clear data provenance to build trust, and balance automation with manual control. Future research should address more complex data relationships (e.g., many-to-many joins), entity resolution for messy data, and the scalability of UI elements like dropdown menus when dealing with hundreds of potential attributes.

Summary updated

Abstract

Data integration is often performed to consolidate information from multiple disparate data sources during visual data analysis. However, integration operations are usually separate from visual analytics operations such as encode and filter in both interface design and empirical research. We conducted a preliminary user study to investigate whether and how data integration should be incorporated directly into the visual analytics process. We used two interface alternatives featuring contrasting approaches to the data preparation and analysis workflow: manual file-based ex-situ integration as a separate step from visual analytics operations; and automatic UI-based in-situ integration merged with visual analytics operations. Participants were asked to complete specific and free-form tasks with each interface, browsing for patterns, generating insights, and summarizing relationships between attributes distributed across multiple files. Analyzing participants' interactions and feedback, we found both task completion time and total interactions to be similar across interfaces and tasks, as well as unique integration strategies between interfaces and emergent behaviors related to satisficing and cognitive bias. Participants' time spent and interactions revealed that in-situ integration enabled users to spend more time on analysis tasks compared with ex-situ integration. Participants' integration strategies and analytical behaviors revealed differences in interface usage for generating and tracking hypotheses and insights. With these results, we synthesized preliminary guidelines for designing future visual analytics interfaces that can support integrating attributes throughout an active analysis process.
BibTeX
@article{suh2024preliminary,
  author = {Ashley Suh and Adam Coscia and Ashley Suh and Remco Chang and Alex Endert},
  title = {{Preliminary Guidelines for Combining Data Integration and Visual Data Analysis}},
  journal = {IEEE TVCG},
  year = {2024},
  doi = {10.1109/tvcg.2023.3334513},
}
Figure