Cox Proportional Hazards Survival Analysis

Description

Models when users abandon or complete a task, estimating hazard ratios for design variants and covariates such as SNR and mistake rate rather than only whether they finish.

Survival analysis models time to an event while handling censoring. The Cox model, h(t|X) = h0(t)·exp(βX), leaves the baseline hazard unspecified and estimates hazard ratios (HR): the relative risk of the event per unit time. In Vivek Gopinath's playbook the event is abandonment of a task, and the model includes design variant, SNR and mistake rate: in a 600-participant prototype A/B, the new IA had HR 0.68 [0.59, 0.77] (32% lower abandonment risk), each +0.10 SNR HR 0.92, and each extra mistake per minute HR 1.28. The playbook stresses that reporting only completion rates ignores when users abandon, and that the proportional hazards assumption must be tested with Schoenfeld residuals. HRs become decision bars (HR ≤ 0.80) and are re-checked in production during validation (HR 0.72, 95% CI 0.63–0.82).

Attributes

Loss type
Latent
Effort
Days

Steps

  1. Define the event (abandonment) and time origin (task start), and log time to completion or abandonment per participant with censoring.
  2. Choose covariates from earlier stages (variant, SNR, mistake rate, segment).
  3. Fit h(t|X) = h0(t)·exp(β1·Variant + β2·SNR + β3·Mistake rate).
  4. Report HRs with 95% CIs and p-values, interpreting HR < 1 as reduced risk.
  5. Test the proportional hazards assumption with Schoenfeld residuals.
  6. Compare HRs with the decision bar (e.g. HR ≤ 0.80) and re-estimate in production.

Sources

  • Vivek Gopinath (2025), Forward-Thinking UX Research Playbook: A Mixed-Methods, Decision-Ready Approach from Prediction to Validation
  • Statistics for the Behavioral Sciences
  • lifelines (survival)
  • scikit-learn (GMM, LDA, UMAP)
  • pandas (log parsing)
  • matplotlib
  • plotly
  • Applied Survival Analysis (Hosmer, Lemeshow & May)
  • The Inquiry Series, Vol III, Ch 44
  • The Inquiry Series, C4 — Behavioural Log Analysis — From Clickstreams to Personas
  • Causal Inference: What If (Miguel Hernán & James Robins)
  • Forward-Thinking Playbook (Vivek Gopinath)
  • Behavioral Drift Playbook (Vivek Gopinath)
  • Cox (1972) Regression models and life-tables
  • Causal Inference (Hernán & Robins)
  • Forward-Thinking Playbook (Gopinath)
  • Behavioral Drift Playbook (Gopinath)
  • The Inquiry Series, Ch 44 — Survival Analysis, Longitudinal Design & Behavioral Drift

Also known as

  • Cox regression
  • Time-to-abandonment modelling
  • Survival analysis
  • Time-to-event analysis
  • Kaplan-Meier analysis
  • Cox proportional hazards model
  • Hazard modelling
  • Proportional hazards regression
  • Stratified Cox model

Categories

  • Analytics & Behaviour
  • Statistics & Modelling