Description
Graphs of variables and one-way causal arrows that make assumptions explicit and show which confounders to adjust for, which colliders to avoid and which mediators to estimate.
A DAG encodes assumed causal relations as arrows with no cycles. It does not estimate anything; it determines identification: which variables must be measured and adjusted to estimate an effect without bias, and which must not be controlled (colliders) to avoid inducing bias. In Vivek Gopinath's payments protocol the DAG is drawn from qualitative themes before estimation: users double-check because they do not fully trust tap-to-pay, double-checking makes paying feel slow, and slowness suppresses adoption. The DAG links Prior Fraud Exposure and Support Responsiveness to Trust, Trust to Double-Checking and Adoption, Double-Checking to Perceived Slowness, and Slowness to Adoption, with confounders such as risk tolerance, device type and prior bank relationship. Benefits named are transparency for peer critique, identification, and an analysis plan pointing to mediators. The DAG and qual arrays are kept under version control so every assumption change is traceable before re-estimation.
Attributes
- Discipline
- Experimentation & Causal Inference
- Loss type
- Unmigrated
- Goal
- Prove causation
- Stage
- Definition, Post-launch
- Data
- Analytical / expert
- Effort
- Hours
Steps
- Extract candidate causes, mediators and outcomes from qualitative themes.
- Draw arrows for assumed causal directions, ensuring no cycles.
- Add confounders that affect both exposure and outcome (e.g. past fraud experience, device, tenure).
- Identify colliders and avoid adjusting for them, and mark mediators whose indirect effects you will estimate.
- Derive the measurement and adjustment plan from the graph.
- Version-control the DAG and update it before any re-estimation.
Sources
- Vivek Gopinath (2025), DAGs, SEM & Mixed Methods in UX: Turning Insight into Evidence in Banking & Payments
- Causal Inference
- Causality (Judea Pearl)
- Causal diagrams for epidemiologic research (Greenland, Pearl & Robins)
- causalgraphicalmodels
- dowhy
- causalml
- econml
- statsmodels
- Causality (Pearl)
- The Inquiry Series, Vol III, Ch 46
- The Inquiry Series, C5 — Causal Inference in Python — DAGs, Target Trial Emulation & Sensitivity Analysis
- Statistical Rethinking — Richard McElreath
- Lost Methods in UX — A Complete Treatise (Gopinath)
- Pearl J (2000; 2nd ed 2009) Causality: Models, Reasoning and Inference
- Hernán MA & Robins JM (2020) Causal Inference: What If
- Greenland S, Pearl J & Robins JM (1999) Causal diagrams for epidemiologic research. Epidemiology 10(1)
- Cinelli, C., Forney, A. & Pearl, J., A Crash Course in Good and Bad Controls (2022)
Also known as
- Causal diagrams
- Causal DAGs
- Structural causal models
- Directed acyclic graphs
- Causal graph specification
- DAG specification
- Directed acyclic graphs for UX
- Causal Model Specification Before Analysis
- Causal graphs
- Do-calculus
- Directed Acyclic Graphs (DAGs) for Causal Reasoning
Categories
- Analytics & Behaviour
- Experimentation & Causal Inference