Course contentsShow
← Data Science
Lesson 812 of 2,14521. Survival AnalysisFree lesson

Handling Event Times and Censoring

How the estimator incorporates both observed events and censored observations at each time point.

Handling Event Times and Censoring

You'll learn: How the Kaplan-Meier estimator properly accounts for both completed events and censored observations when calculating survival probabilities.

The Core Challenge

When computing the Kaplan-Meier product at each time point, you need to handle two types of observations:

  1. Events — subjects who experienced the event (death, churn, failure) at that exact time
  2. Censored observations — subjects still "at risk" but whose outcome is unknown (they left the study, were lost to follow-up, or the study ended)

How the Estimator Incorporates Both

At each distinct event time t, the Kaplan-Meier estimator calculates:

  • Number at risk = all subjects who haven't yet had the event and haven't been censored before time t
  • Number of events = only those who actually experienced the event at time t

Censored observations contribute to the "at risk" count up until their censoring time, then they're removed from the calculation. They don't count as events, but they do inform us that those subjects survived at least until their censoring point.

A Simple Analogy

Think of tracking runners in a marathon. Some finish (events), some drop out mid-race (censored). When calculating the proportion still running at mile 15, you count everyone who made it to mile 15—including those who later dropped out at mile 18. The person who quit at mile 18 proved they could survive to mile 15, so their data helps estimate survival at that checkpoint.

Key Takeaway: The Kaplan-Meier estimator uses censored observations to keep the risk set accurate at each time point, extracting partial information from incomplete data rather than discarding it.