survival-analysislisted

# Survival Analysis ## When to use this skill Use when the outcome is "**how long until X happens?**" and some observations haven't experienced X yet (censored). Triggers: - "Time to churn / cancel / conversion" - "Survival curve for…" - "Hazard ratio of feature X" - "How long do users stay before churning?" - "Compare retention across segments over time" If you only care about *whether* an event happens within a fixed window, use `logistic-regression`. If you need fixed-time-point retention rates without timing, use `cohort-analysis`. ## Why not just use logistic regression? | Question | Right tool | |---|---| | Did the user churn within 90 days? | logistic | | When did the user churn? Distribution? | survival | | How does plan tier affect churn timing? | survival | | Users still active — do we throw them out? | survival (treats as censored, not missing) | Throwing out censored observations biases logistic regression toward early events. ## Required inputs | Input | Why it matters | |---|---| | Subject ID | One row per subject | | Duration | Time from start to event OR last observation | | Event indicator (1/0) | 1 = event happened, 0 = censored (still observed without event) | | Covariates | Features that may affect timing | | Start time | When observation begins (often signup_at) | ## Workflow 1. **Define event precisely.** "Churn" must have an exact definition: - Subscription cancellation date - No-activity threshold (e.g., 30 consecutive inactive days)