Description
Turns qualitative themes into a causal DAG, tests it with CFA and SEM, and binds each estimated path to a quote and a design prescription.
Reports usually end with themes or dashboards; what is missing is a mechanism showing how a perception change propagates through behaviour to outcomes. In this protocol the DAG expresses the mechanism and SEM estimates its strength. Qualitative discovery yields a codebook of constructs; the DAG makes causal assumptions and confounders explicit and identifies mediators and colliders; a CFA measurement model validates latent constructs; the structural model estimates DAG paths with direct, indirect and total effects; and a joint display maps qual mechanism → quant evidence → prescribed change. In the tap-to-pay case (n = 520, MLR), FR → TR −0.45, SR → TR +0.31, TR → DC −0.50, DC → SL +0.42, SL → AD −0.28 and TR → AD +0.26, with indirect effect ≈ +0.059 and total TR → AD ≈ +0.32. Robustness checks include multi-group invariance, common method bias, sensitivity to confounders and out-of-sample validation. Qual arrays and the DAG are kept under version control.
Attributes
- Discipline
- Community-Proposed Frameworks
- Loss type
- Community-proposed
- Tested
- Not yet tested
- Stage
- Definition, Evaluation
- Data
- Mixed
- Effort
- Months
Steps
- Run qualitative discovery (interviews, think-aloud) and build a codebook of constructs, marking each as latent or observed (e.g. Trust, Fraud Risk, Double-Checking, Slowness, Support, Adoption).
- Draw the DAG from qual: write causal assumptions, confounders to adjust (past fraud, device type, tenure) and colliders not to control.
- Design 3-item Likert batteries per latent construct; screen missingness < 10% (FIML), check Cronbach's α ≥ 0.70 and run CFA, dropping loadings < 0.60 unless theory demands.
- Check convergent validity (AVE ≥ 0.50, CR) and discriminant validity (Fornell-Larcker or HTMT < 0.85).
- Estimate the structural model per the DAG with robust ML (MLR) and report CFI, TLI (≥ 0.95), RMSEA (≤ 0.06) and SRMR (≤ 0.08).
- Decompose direct, indirect and total effects (|β| ≈ 0.10 small, 0.30 medium, 0.50 large) and never interpret paths without quotes beside them.
- Run robustness checks: multi-group invariance (iOS/Android, new/veteran), common method bias via behavioural covariates, confounder sensitivity and split-sample or time-based validation.
- Bind paths to prescriptions in a joint display and validate with A/B tests, guardrails and brief qual follow-ups (e.g. decision bar +5–8 points adoption in 4 weeks; 5 post-launch sessions).
Sources
- Vivek Gopinath (2025), DAGs, SEM & Mixed Methods in UX: Turning Insight into Evidence in Banking & Payments
Also known as
- Quotes to Coefficients stitching flow
- DAGs, SEM & Mixed Methods in UX
Categories
- Community-Proposed Frameworks