Conference proceeding
Interpreting Latent Variables in Causal DAGs: A Study of Non-Expert Understanding
EuroVA 2026 - EuroVis Workshop on Visual Analytics, 20261006
2026
DOI: 10.2312/eurova.20261006
Abstract
Causal Directed Acyclic Graphs (DAGs) are a common way to represent cause-and-effect relationships between variables. While simple DAGs are often easy to read, graphs that include latent variables—causes that cannot be directly observed—may be harder for non-experts to interpret. We conducted a controlled user study with 60 non-expert participants using two real-world scenarios: college admissions and police funding, both based on structurally identical DAGs containing latent variables. After a short training phase, participants answered comprehension questions about the graphs. Overall accuracy was high (89.7%), but participants struggled most when reasoning about the relationship between latent variables and the observable signals derived from them. An error analysis shows that the most common misconception was treating latent variables as directly observable. These findings highlight a gap in how non-experts interpret latent variables in causal DAGs and suggest that visual analytics tools should make the distinction between true and inferred values more explicit.
Details
- Title: Subtitle
- Interpreting Latent Variables in Causal DAGs: A Study of Non-Expert Understanding
- Creators
- Amit Kumar Das - Stony Brook UniversityShahreen Salim - Stony Brook UniversityNaimul Hoque - University of IowaKlaus Mueller - Stony Brook University
- Resource Type
- Conference proceeding
- Publication Details
- EuroVA 2026 - EuroVis Workshop on Visual Analytics, 20261006
- DOI
- 10.2312/eurova.20261006
- ISSN
- 2664-4487
- eISSN
- 2664-4487
- Language
- English
- Date published
- 2026
- Academic Unit
- Computer Science
- Record Identifier
- 9985182471302771
Metrics
1 Record Views