A set of statistical techniques for modeling the timing until the occurrence of discrete events (time‑to‑event), explicitly accounting for censoring, varying risk sets, and time‑dependent covariates; commonly used to estimate hazard (instantaneous risk) or survival functions and to test how covariates affect the rate or risk of event occurrence.

Principle

Correct inference about temporal risk requires modeling both the duration process and censoring mechanism; analyses typically construct risk sets at each observed time and estimate how covariates shift hazard rates, distinguishing between time‑invariant and time‑varying predictors.

Demonstration

Illustrative scenario: Modeling time from government formation to cabinet collapse for a set of countries observed over a decade. Some cabinets remain in office at the study's end (right‑censored); a Cox proportional hazards model or a parametric duration model can estimate how variables (e.g., coalition size, economic shocks) are associated with the instantaneous risk of collapse while properly handling censored observations.

Misapplication

Applying ordinary least squares or naive duration averages to time‑to‑event data without accounting for censoring or changing risk sets, or interpreting hazard ratios as direct probabilities of event occurrence over fixed intervals without clarifying the time scale and baseline hazard assumptions.

Consequence

When applied correctly, EHA yields unbiased estimates of how covariates influence timing and risk, enabling more accurate policy and theoretical inference about event dynamics; misapplication leads to biased estimates, mistaken inference about effects, and incorrect policy prescriptions.

Reversal

Specialized alternatives are required when events are recurrent (repeated spells), when competing risks (mutually exclusive event types) are present, when interval or left censoring dominates, or when hazards are driven by unobserved heterogeneity requiring frailty or multilevel specifications.

Boundary

Covers continuous‑ or discrete‑time models for single or recurrent event timing with censoring. Excludes cross‑sectional analyses where timing is irrelevant, naive regressions ignoring censoring, and causal identification tasks that require experimental designs unless combined with appropriate identification strategies.

Semantic Tension

Hazard‑rate (instantaneous risk) interpretation ↔ cumulative probability interpretation: hazard ratios describe relative instantaneous risk conditional on survival to a time, which is distinct from the probability of event occurrence over a specified interval; conflating them produces misinterpretation.

Synthesis

Time contains information beyond event occurrence; treating timing appropriately—through risk sets, censoring adjustments, and time‑varying covariates—yields qualitatively different and often more informative inferences than analyses that ignore temporal structure.