Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Designing Computational Tools for Exploring Causal Relationships in Qualitative Data

Han Meng, Qiuyuan Lyu, Peinuan Qin, Yitian Yang, Renwen Zhang, Wen-Chieh Lin, Yi-Chieh Lee

TL;DR

This paper tackles the challenge of exploring causal relationships in qualitative data by designing QualCausal, a system that combines automatic indicator extraction, concept creation, and causal-network construction with interactive, multi-view visualizations. Through a formative study (n=15) and a follow-up feedback study (n=15), the authors derive design goals that balance automation with human agency and introduce a dual-level visualization approach to support theory building and data traceability. Technical evaluations against established baselines show improved precision, recall, and directionality for causal extraction, while user studies reveal benefits in cognitive scaffolding and analytical efficiency alongside concerns about paradigmatic alignment and potential over-reliance. The work contributes design principles for computational tools in qualitative analysis, demonstrates a practical, open-source implementation, and discusses the broader epistemological implications for computational social science and interdisciplinary research.

Abstract

Exploring causal relationships for qualitative data analysis in HCI and social science research enables the understanding of user needs and theory building. However, current computational tools primarily characterize and categorize qualitative data; the few systems that analyze causal relationships either inadequately consider context, lack credibility, or produce overly complex outputs. We first conducted a formative study with 15 participants interested in using computational tools for exploring causal relationships in qualitative data to understand their needs and derive design guidelines. Based on these findings, we designed and implemented QualCausal, a system that extracts and illustrates causal relationships through interactive causal network construction and multi-view visualization. A feedback study (n = 15) revealed that participants valued our system for reducing the analytical burden and providing cognitive scaffolding, yet navigated how such systems fit within their established research paradigms, practices, and habits. We discuss broader implications for designing computational tools that support qualitative data analysis.

Designing Computational Tools for Exploring Causal Relationships in Qualitative Data

TL;DR

This paper tackles the challenge of exploring causal relationships in qualitative data by designing QualCausal, a system that combines automatic indicator extraction, concept creation, and causal-network construction with interactive, multi-view visualizations. Through a formative study (n=15) and a follow-up feedback study (n=15), the authors derive design goals that balance automation with human agency and introduce a dual-level visualization approach to support theory building and data traceability. Technical evaluations against established baselines show improved precision, recall, and directionality for causal extraction, while user studies reveal benefits in cognitive scaffolding and analytical efficiency alongside concerns about paradigmatic alignment and potential over-reliance. The work contributes design principles for computational tools in qualitative analysis, demonstrates a practical, open-source implementation, and discusses the broader epistemological implications for computational social science and interdisciplinary research.

Abstract

Exploring causal relationships for qualitative data analysis in HCI and social science research enables the understanding of user needs and theory building. However, current computational tools primarily characterize and categorize qualitative data; the few systems that analyze causal relationships either inadequately consider context, lack credibility, or produce overly complex outputs. We first conducted a formative study with 15 participants interested in using computational tools for exploring causal relationships in qualitative data to understand their needs and derive design guidelines. Based on these findings, we designed and implemented QualCausal, a system that extracts and illustrates causal relationships through interactive causal network construction and multi-view visualization. A feedback study (n = 15) revealed that participants valued our system for reducing the analytical burden and providing cognitive scaffolding, yet navigated how such systems fit within their established research paradigms, practices, and habits. We discuss broader implications for designing computational tools that support qualitative data analysis.
Paper Structure (65 sections, 4 figures, 2 tables)

This paper contains 65 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Qualitative Data Analysis and Causal Relationships
  4. Computational Support for Qualitative Data Analysis
  5. Computational Support for Characterizing and Categorizing Qualitative Data
  6. Computational Support for Causally Connecting and Theorizing Qualitative Data
  7. Visualizations of Causal Relationships
  8. Formative Study
  9. Participants
  10. Procedure
  11. Findings
  12. Current Practices for Exploring Causal Relationships and Generating Hypotheses
  13. Challenges in Causally Connecting Qualitative Data
  14. Desired Features for Computational Causal Discovery Tools
  15. Design Goals
  16. ...and 50 more sections

Figures (4)

  • Figure 2: Processes of formative study and feedback study. (a) The formative study was a two-hour workshop consisting of an opening (10 minutes), introduction of common methods (20 minutes), group discussions on three questions (15 minutes each), and a co-design activity for computational tools (45 minutes). (b) The feedback study was a one-hour session consisting of an introduction and system tutorial (10 minutes), followed by two hands-on evaluation sessions (15 minutes each) with QualCausal, and concluding with an experience discussion (15 minutes).
  • Figure 3: Paper prototypes created by participants during the formative study co-design activity. (a) A system that processes uploaded files through automated labeling and displays results as directed graphs. (b) A workflow where input text files are chunked into segments, tags are generated and highlighted for each segment, and the results are then organized into network structures that allow users to trace back to source data. (c) Drag-and-drop functionality with different node shapes representing different types of labels, where clicking nodes enables traceability to source data.
  • Figure 4: Interface for causal network construction. (a) Indicator Extraction: The system automatically extracts indicators from the uploaded data (a.1) based on the user-provided research overview. Users can select text to edit indicators (a.2), type to modify them (a.3), or delete indicators (a.4). (b) Concept Creation: Users can create concepts to abstract indicators into meaningful theoretical constructs (b.1) and assign colors to them (b.2). Additionally, the system allows users to map indicators to the appropriate concepts (b.3) and save these mappings (b.4).
  • Figure 5: Interface for causal network exploration with multiple coordinated views. Indicator View: causal network display where nodes represent indicators and edges denote causal relationships. Interactions include connectivity-based node sizing, slider control, semantic search, and legend-based filtering (①-⑤). Concept View: Abstracted conceptual model showing user-created concepts as colored blocks connected by weighted edges. Interactions include hovering for percentages and clicking edges for supporting indicator pairs (⑥-⑧). Node View: Localized subnetwork centered on selected indicators (⑨). Details and Concepts Panel: Sidebar for viewing context and editing concepts with re-visualization (①0-①1).