Proportions are commonly reported as outcomes from clinical trials.
Naturally, we want to know what the confidence intervals around the mean
proportions are. One common way to obtain them is to use binomial
distribution, or chi-square if the samples are large. If the proportions
are not close to 0 or 1, there should generally be no problems.
The appears to be no reason, however, not to use bootstraping to obtain
CIs. I am analysing output from a simulation model based on an RCT. The
sample size in the trial was about 200 and I am sampling in blocks of
size 200 - 10,000 times to get a reasonably good estimate.
The problem is that, despite the results converging even after 1,000
resamplings, the CIs are much different from those obtained based on
binomial distribution, e.g., the bootstrap intervals are much narrower.
In many cases that could affect the final conclusions re significance of
results.
Nonparametric methods, such as bootstrapping and permutation tests are
supposed to be much more accurate than parametric chi-square or t-test
approximations. They are starting to be popular in medical statistics,
as computer power makes computation feasible. But there still is a
chance of making wrong assumptions, which is perhaps what I am doing.
I realize that there is a whole rocket science to bootstrap CIs, but
perhaps, the solution to my problem is relatively simple?
Thanks for any tips.
Jaro
-------------------------------------------------------------------
Jaroslaw Wechowski, MB BS, PhD (econ)
Principal Health Economist
Cardiff Research Consortium Ltd
The Medicentre
Heath Park
Cardiff, CF14 4UJ
UK
www.crc-limited.co.uk
http://www.linkedin.com/in/wechowski
-------------------------------------------------------------------
Phone +44 (0) 2920 682049
Fax + 44 (0) 2920 750239
Mobile 0044 (0)77 16 29 98 35
Skype : jarowech
-------------------------------------------------------------------
----------------------------------------------------
Wygraj wyścig o swoją przyszłość. Wybierz ING N-NP OFE,
który od lat bardzo dobrze zarabia dla swoich Klientów.
Przystąp on-line>>
http://klik.wp.pl/?adr=http%3A%2F%2Fadv.reklama.wp.pl%2Fas%2Fing-2.html&sid=1204
|