RE: st: OT: how to report statistics in (medical) journals

[Nick Cox <n.j.cox@durham.ac.uk>]

I agree very much with the general attitude here. For once, I was just answering the question.  

On one detail: rank-based tests can often be thought of very fruitfully as based on estimating the probability that some A > some B and several of Roger Newson's commands have this strong flavour. 

Nick 
n.j.cox@durham.ac.uk 

Seed, Paul

- --- On Tue, 16/11/10, Kaulisch, Marc wrote:
> I was asked to provide some tests for analyses in an
> article for a medical journal. <snip> Are there any
> guidelines for reporting statistics?

Nick Cox & Maarten Buis both suggest Marc looks at the 
journal's practice & policy.  This is fine as far as it gues, but current 
practice is not always best practice.  (Altman DG (2002) Poor Quality 
Medical Research. What Can Journals Do? JAMA 287 2765-2767 ) 

There are a number of guidelines for reporting clinical studies
(CONSORT - randomised comtrolled trials, STARD  - diagnostic 
tests, PRISMA - meta-analyses) and others.  
Marc should google these.
(Altman DG, Schulz KF, Moher D, Egger M, Davidoff F, Elbourne D, 
Getzsche PC & Lang T (2001). The revised CONSORT statement 
for reporting randomized trials: explanation and elaboration. Annals of 
Internal Medicine 134:663-694.)

These stress producing estimates with confidence intervals as well as 
(and in preference to) p-values.  Rank-based tests such as 
Wilcoxon & Kruskal-Wallis are unhelpful in that 
they give no such estimates.  If Marc has the time, he may investigate 
the possibility of parametric tests, posibly after log-transformation, or 
dropping outliers, so that he can report meaningful estimates with 
confidence intervals.


*
*   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/