Assumptions of the Log-Rank Test
What you'll learn: The key assumptions that must hold for the log-rank test to validly compare survival curves between groups.
Why Assumptions Matter
You've learned how the log-rank test compares survival between groups using a statistical hypothesis test. But like any statistical method, it only works correctly when certain conditions are met. Violating these assumptions can lead to misleading p-values and wrong conclusions.
The Two Critical Assumptions
1. Proportional Hazards
The most important assumption is that the hazard ratio between groups remains constant over time. Think of it like this: if Group A has twice the risk of an event compared to Group B at month 1, that ratio should stay roughly 2× throughout the entire study period.
Analogy: Imagine two runners racing. The proportional hazards assumption is like saying the faster runner maintains a constant speed advantage—not pulling ahead dramatically at the start then falling behind later.
When hazards cross or diverge over time, the log-rank test loses power and may miss real differences.
2. Independent Censoring
Censoring must occur independently of the event risk. This means that why someone drops out of your study (gets censored) should have nothing to do with their likelihood of experiencing the event.
Example violation: In a customer churn study, if high-risk customers are more likely to stop using your product and more likely to unsubscribe from your tracking emails (causing censoring), your results will be biased.
The censoring pattern must be similar across all groups you're comparing.
Checking the Assumptions
Before trusting your log-rank test p-value, visually inspect whether survival curves maintain roughly parallel spacing (proportional hazards) and verify that censoring rates and patterns are comparable between groups.
Key Takeaway: The log-rank test requires proportional hazards across time and censoring that's independent of event risk—violating these assumptions invalidates your comparison.