Designing Computational Tools for Exploring Causal Relationships in Qualitative Data
This is a credible CHI systems paper with a clear design contribution: it reframes qualitative causal analysis as an interactive, traceable workflow rather than a purely manual reading task. The strongest contribution is the combination of causal-network construction and multi-view visualization, but the empirical support is still scoped to a small formative/feedback loop and a technical evaluation.
Axes Lens
Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.
Contribution shape
- Knowledge form
- technical knowledge typical · 50/268
- Novelty type
- system architecture typical · 35/268
- Abstraction level
- system typical · 61/268
- Generalization target
- task class typical · 63/268
- Validation mode
- mixed methods typical · 136/268
Evidence profile
- Evidence strength
- moderate typical · 105/268
- Claim alignment
- medium typical · 32/268
- Overclaim risk
- medium typical · 210/268
Review Summary
QualCausal’s contribution is best understood as a system-level rethinking of how computational tools can support qualitative causal analysis. The paper does not merely automate extraction; it proposes an interactive workflow in which analysts can move between indicators, concepts, and causal relationships while retaining traceability to source sentences. That is a meaningful design move because it addresses a real tension in qualitative work: analysts want computational leverage, but they also need credibility, context, and the ability to inspect why a relationship was proposed. The paper’s discussion of users shifting from source-first analysis to a more scaffolded, output-first workflow is especially interesting because it suggests the tool can change analytic practice, not just speed it up. At the same time, the evidence base is bounded. The technical evaluation covers two datasets, and the user study is a feedback study with 15 participants in a single scenario. That is enough to support a CHI design/system contribution, but not enough to claim broad effectiveness across qualitative domains or to settle questions about long-term adoption in real research settings. The limitation that the system only handles within-sentence causal relationships is important: it means the tool may miss discourse-level or theory-relevant causal patterns that emerge across sentences or larger stretches of text. So the paper is strongest as a design and prototype contribution with promising user feedback, rather than as a definitive solution for causal qualitative analysis.
What Changed
Canon before
Prior CHI and HCI work on qualitative analysis tools has largely emphasized coding, categorization, summarization, and other forms of text characterization. Causal relationship exploration in qualitative data is less established, and existing systems are described as limited by context handling, credibility, or complexity.
Departure from common sense
The paper’s most interesting behavioral claim is that the tool can invert the usual qualitative-analysis workflow: instead of starting from raw source text and moving upward, users may begin with extracted indicators and concepts, then only return to source data when outputs conflict with expectations or theory. That is a non-obvious shift in how computational support can scaffold analysis.
Actual novelty
QualCausal is presented as a system that combines interactive causal network construction with multi-view visualization for qualitative data analysis. The novelty is not just extraction, but the coupling of indicator extraction, user-controlled concept mapping, and LLM-based causal edge generation with traceability back to source sentences.
Evidence
The paper supports its claims with a technical evaluation of the causal-network construction pipeline on two datasets and a feedback study with 15 participants. The evidence shows both algorithmic behavior on benchmark-like data and user-perceived value in a qualitative workflow, but the scope remains bounded to a single system and a small participant sample.
“ Based on the findings, in Phase two, we developed QualCausal , a system that combines causal network visualization with a user-controlled workflow for categorizing and causally connecting qualitative data”
actual novelty · System Design (4) and Figure 1/4/5 descriptions · confidence 0.62
“ This analytical shift, combined with the system’s presentation format that had users “ labeling ” indicators via pre-structured tables, reinforced a “ task-like ” mentality that led users to habitually move from one indicator to the next as if they were “ data labelers ,” potentially reframing qualitative data analysis as a process of validating system outputs rather than one of discovery, abstraction, and theorizing [ 78 ]”
departure from common sense · Discussion 7.1.2 (From passive acceptance, confirmation to proactive cognitive forcing, exploration) · confidence 0.66
“ Second, our system only focuses on within-sentence causal relationships, potentially missing those that span multiple sentences, omitting broader contextual patterns across discourse, and thus bounding the utility for theory buildin”
limitation · Limitations and Future Research (7.2) - Second limitation · confidence 0.70
“ (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)”
validation scope · Technical Evaluation (4.3) and Feedback Study Design/Procedure (5.1-5.2) · confidence 0.55
Limits
Method limits
Validation is limited to a technical evaluation on two datasets and a feedback study with 15 participants. The evidence does not establish broad comparative superiority over alternative qualitative-analysis tools or workflows, nor does it show long-term adoption.
Deployment limits
The system is positioned for exploratory qualitative analysis of causal relationships, but the paper’s own discussion indicates that fit with established research paradigms, practices, and habits remains an open issue. Practical deployment will depend on how analysts trust, inspect, and integrate generated causal structures.
Boundary conditions
The paper explicitly notes that the system only focuses on within-sentence causal relationships, which bounds its usefulness when causal meaning spans multiple sentences or broader discourse. The workflow also appears most relevant when users are willing to work iteratively between extracted structure and source text.
Position in field
This work sits at the intersection of qualitative analysis support tools, causal inference assistance, and LLM-mediated sensemaking. It extends beyond conventional coding or summarization tools by targeting causal relationship exploration, while still relying on human interpretation and source-grounded inspection.