In article <[EMAIL PROTECTED]>,
Rich Ulrich <[EMAIL PROTECTED]> wrote:
>On Thu, 03 Aug 2000 10:34:24 GMT, [EMAIL PROTECTED] wrote:
>> Hello from Germany,
>> as a part of my dissertation in medicine, I have
>> to summarize some results of clinical trials.
>> My question: By summarizing the results
>> (percentage differences of certain parameters),
>> how can I regard for the different p-values
>> (which are calculated with different tests in the
>> trials). Is it possible to form something like a
>> weighet mean with the p-values and the sample
>> sizes in the trials to generate an common effect
>> size of the different results in the trials?
>You might try some textbooks or articles concerning meta-analysis, and
>read what you find online. There are some comments in my stats-FAQ.
>You could look at examples of actual meta-analyses -- ones that were
>not carried out by statisticians as models to be followed -- but be
>aware that the majority of those are poorly done and misleading, both
>as to style and conclusions.
I have deleted the rest of this article; it does describe
what is often done, but it does not give the important
warnings about what is done incorrectly, and even more
importantly, what is ignored.
I will address this point first, as it is equally valid no
matter what one's view of statistical inference happens to
be. The point is that the probability of publication
depends on the p-value. If you have 100 studies published,
and the distribution of p-values is roughly uniform between
0 and .05, this should be looked upon as selection bias,
and not as meaning anything. On the other hand, if one
had 100 studies with the p-values coming from a normal
distribution with mean 1 and variance 1, while these
p-values would be considerably larger, the evidence of
the effect would be overwhelming.
The other point is that the p-value gives no information
whatever about the magnitude of the effect; p-values are
not what should be combined, but likelihood functions.
At least in parametric models, these are very easy to
combine, and they give the full information. However,
the publication problem above still remains.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
[EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
