Originally Posted by
MoAlpha
So time to occurrence, rather than number of occurrence over time? Is that what they did here?
Close. Ordered rank to occurrence (in this case, to death). Proportional hazards is, um, proportional, and Cox realized that proportionality is a strong enough restriction that you can get the risk ratios even without needing the underlying absolute risk. I *think* I might be one of the few professors who forces students to sit through a lecture where I do a proportional hazards estimation with two covariates on the blackboard by hand. Like most survival analysis techniques, it handles censored observations pretty well.