 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: Wald interval and the WSJ
Your point is well-taken. The confidence interval approach for the
proportion of 'successes' would be appropriate.
The Center for Devices and Radiological Health (CDRH) is keen on
Bayesian intervals.
The limitations I mentioned in my earlier e-mail (which may arrive after
this on) are far more problematic than the difference between 0.049 and
0.051.
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of David Airey
Sent: Thursday, August 14, 2008 7:12 AM
To: [email protected]
Subject: Re: st: Wald interval and the WSJ
.
Another point of view is that for something like a stent that goes in
a human body, I'd want my probability to be quite a bit better than .
05 anyway, like .01 or .001. I've seen a lot of .05 results evaporate
on additional scrutiny. And also the article doesn't emphasize effect
size, which might make the quibbling over p values moot too.
-Dave
On Aug 14, 2008, at 8:56 AM, Maarten buis wrote:
> It looks to me (though the article doesn't say) that this deals with a
> test of multinomial proportions. This is a difficult problem, the
> exact
> test statistic exists, but is actually not exact at all but
> conservative, various other/better approximations have been proposed,
> but now you have to decide how to choose between them. In this case
> all
> the different p-values are very close to .05 (from .049 to .052
> whith a
> spike at .55) The articles puts great emphasis on the fact that all
> but
> one test results in p-values more then .05, and that the company
> conveniently choose the one test that supported their claim. That may
> or may not be suspicious, but my reading of a study like this is that
> these numbers are so close to .05 (and there is nothing magical or
> `scientific' about the number .05) that making a big deal about a
> black
> and white distinction between significant and non-significant is
> pretty
> ridiculous. I know that this doesn't help an agency like the FDA who
> have to make a approve or disaprove decision (and they certainly don't
> ask my advise), but sometimes a study just ends up in a grey area
> between significant and insignificant.
>
> -- Maarten
>
> --- Scott Merryman <[email protected]> wrote:
>
>> There is an interesting article in today's Wall Street Journal on the
>> use of the Wald interval in clinical trials.
>>
>> "Boston Scientific Stent Study Flawed" by Keith Winstein, August 14,
>> 2008; Page B1.
>>
>> "But Boston Scientific's claim was based on a flawed statistical
>> equation that favored the Liberte stent, a Journal analysis has
>> found.
>> Using a number of other methods of calculation -- including 14
>> available in off-the-shelf software programs -- the Liberte study
>> would have been a failure by the common standards of statistical
>> significance in research.
>> Boston Scientific isn't the only company to use the equation, known
>> as
>> a Wald interval, which has long been criticized by statisticians for
>> exaggerating the certainty of research results. "
>>
>>
>
http://digg.com/business_finance/WSJ_com_Boston_Scientific_Stent_Study_F
lawed
>>
>> Scott
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/statalist/faq
>> * http://www.ats.ucla.edu/stat/stata/
>>
>
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting address:
> Buitenveldertselaan 3 (Metropolitan), room Z434
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
> Send instant messages to your online friends
http://uk.messenger.yahoo.com
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/