Course contentsShow
← Data Science
Lesson 805 of 2,14521. Survival AnalysisFree lesson

Left and Interval Censoring

Recognizing left censoring (event before observation) and interval censoring (event within a time window).

Left and Interval Censoring

What you'll learn: How to recognize and handle situations where you don't know the exact event time—either because it happened before you started watching, or somewhere within a time window.

In Lesson 804, you learned about right censoring (when someone drops out before the event happens). Now let's look at two other incomplete-data scenarios that appear in survival analysis.

Left Censoring: The Event Already Happened

Left censoring occurs when you know an event has already occurred before your observation period began, but you don't know exactly when.

Analogy: You're studying when employees first received a promotion. A new employee tells you they were promoted at their previous company, but they don't remember the exact date—only that it was before they joined your firm. You know the event happened, but the exact timing is censored on the left side of your timeline.

Real-world examples:

  • Disease detection where infection occurred before diagnosis
  • Age at which a skill was first learned (self-reported retrospectively)

Interval Censoring: Somewhere In Between

Interval censoring means you know the event occurred within a specific time window, but not the precise moment.

Analogy: You check on a patient every 6 months. At Month 0, no disease. At Month 6, disease detected. You know it happened sometime between those checkups, but not exactly when.

Real-world examples:

  • Clinical visits with gaps between appointments
  • Equipment inspections revealing failure between checks
  • Survey data collected at discrete intervals

Both types require specialized survival analysis methods that account for this uncertainty rather than ignoring it or making false assumptions about exact timing.

Key Takeaway: Left censoring means the event happened before observation started; interval censoring means it happened within a known time window—both capture real-world measurement limitations.