On Thu, 27 Nov 2008, [EMAIL PROTECTED] went:

> But neither control vs therapy comparison is even close to
> significance by a chi-square test (e.g. Fisher´s exact), which means
> either there´s nothing there, or not enough subjects were studied to
> show it.
>
> So, what´s going on? Can Cox proportional hazards analysis
> demonstrate something not evident by simple statistics?

My mostly intuitive response is that survival analysis is more
sensitive than endpoint analysis because it accounts for every time
point.  In fact, when you do a survival analysis, you can use
different types of significance tests to emphasize early differences
in survival (that would be a Wilcoxon chi-square) or emphasize all
time points equally (that would be a log-rank chi-square).  I think
the Cox model is more similar to the latter, but I'm a bit fuzzy on
that.  Either way, you're testing the whole pair of curves, not just
the difference at the endpoint.

So I think it makes sense for Anderson et al. to equate longer
survival times with delayed mortality.  As you noted, the Gigerenzer
objection doesn't seem to apply here.

--David Epstein
   [EMAIL PROTECTED]

---
To make changes to your subscription contact:

Bill Southerly ([EMAIL PROTECTED])

Reply via email to