Re: power for recurrent events
- From: Richard Ulrich <Rich.Ulrich@xxxxxxxxxxx>
- Date: Thu, 07 Sep 2006 18:59:58 -0400
On Thu, 7 Sep 2006 14:06:20 +0000 (UTC), naught@xxxxxxx wrote:
Hi all,
Has anyone had experience in planning a study that will use recurrent
events as an outcome? I imagine that the data will ultimately have a fair
number of zeros on the left, be fairly overdispersed as well as
right-censored. I have a bit of experience modeling this sort of data,
but none whatsoever in terms of planning sample size/power in this
What do you do when modeling it? If you use the means
and standard deviations, why not analyze the power in those
terms? - Or, in the simple version of using the terms, look for the
power for a simple t-test.
If there are a large enough number of zeros, most of the power
of the comparison will be in zero-versus-other. In that case,
you can find a conservative estimate of power by using
the 2x2 table alone.
context. I have to admit that I always find myself a bit cynical about
anyting but ballpark power analyses in the first place, but if there is
something reasonable to do here, I'd certianly consider it. Given all the
assumptions I'd need to make for estimating sample size for a fairly
sophisticated model, would it be just as acceptable to just use, say, a
simple poisson model and assume that it's likely optimistic?
Any pointers to literature, software, or, of course, your own thoughts
would be greatly appreciated.
--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.
- References:
- power for recurrent events
- From: naught
- power for recurrent events
- Prev by Date: Re: Correlation ratio = eta-squared, or eta?
- Next by Date: Concurrent Segmentation
- Previous by thread: power for recurrent events
- Next by thread: Re: power for recurrent events
- Index(es):
Relevant Pages
|
