Log-Rank Test vs Other Tests
What you'll learn: How the log-rank test compares to alternative tests like Wilcoxon and Peto tests when survival curves differ in specific ways.
Why Multiple Tests Exist
You've learned the log-rank test for comparing survival curves, but it's not the only option. Different tests have different sensitivities to when and how survival curves diverge. Think of it like having different thermometers optimized for different temperature ranges—each works best under certain conditions.
The Log-Rank Test's Strength
The log-rank test gives equal weight to all time points. It's most powerful when:
- Hazard ratios are constant over time (proportional hazards)
- Differences occur throughout the study period
- Late survival differences matter as much as early ones
Wilcoxon Test: Early Differences Matter More
The Wilcoxon test (also called Breslow test) weights earlier time points more heavily because more subjects are at risk early on. It's more sensitive when:
- Survival curves diverge early then stay parallel
- You care more about short-term differences
- The risk set is larger in early periods
Peto Test: A Middle Ground
The Peto test (Peto-Peto modification) uses a weighting scheme between log-rank and Wilcoxon. It's useful when:
- You want balance between early and late differences
- Sample sizes are unequal between groups
- You're uncertain about the pattern of divergence
Choosing the Right Test
Match your test to your survival curve pattern. If curves cross or proportional hazards don't hold, all three tests may give conflicting results. In practice, if you have strong proportional hazards and consistent separation, all tests typically agree.
Key Takeaway: The log-rank test is optimal for constant hazards; Wilcoxon emphasizes early events; Peto balances both—choose based on when survival curves actually diverge.