OT: t-tests, non-parametric tests, and large studies—a paradox of statistical practice?

[Bruce Weaver]

Here's an (open access) article you may find interesting.

   http://www.biomedcentral.com/1471-2288/12/78

Cheers,
Bruce

[Rich Ulrich]

[re-posted; first try seems to have failed.]
Oh, thanks!  That looks important. 

I have disliked the rank-based tests for the simpler reasons, for a
long time.  I would not have guessed that large Ns made them
so strong as tests against  "shape"  instead of location. 

 - I wonder, though, if Fagerland told the whole story, about the increased
use of non-parametric tests in studies with large Ns.  Since there are
some people who are (overly) concerned with non-normality, I did
eventually become willing to do my t-tests and report them in detail,
while tossing in the assurance that non-parametric tests showed the
same conclusions.  By nominal tabulation, my use increased over time. 
You would have to separate out that sort of back-up confirmation before
you conclude that there is really a lot of cherry-picking of results for their
p-values.

--
Rich Ulrich

> Date: Thu, 10 Jan 2013 12:56:11 -0800

> From: [hidden email]
> Subject: OT: t-tests, non-parametric tests, and large studies—a paradox of statistical practice?
> To: [hidden email]
>
> Here's an (open access) article you may find interesting.
>
> http://www.biomedcentral.com/1471-2288/12/78
>
> Cheers,
> Bruce
> ...

[Kornbrot, Diana]

Hate the use of non-parametric when people actually mean rank based.
In my view, there is great advantage in using generalizED linear methods, with either a logistic or normal transformed proportion as the DV.
Often this gives same results as normal based methods. However, there are differences when distributions have floor or ceiling effects, or are J shaped.
J shapes often occur for sympton counts for abnormality  - most controls score 0-2 while even the exptals may have more in 0-2 than in any 1 of 3 –N

Heartily agree that both normal based and other test results should be reported
Best

Diana


On 11/01/2013 19:41, "Rich Ulrich" <rich-ulrich@...> wrote:

[re-posted; first try seems to have failed.]
Oh, thanks!  That looks important.  

I have disliked the rank-based tests for the simpler reasons, for a
long time.  I would not have guessed that large Ns made them
so strong as tests against  "shape"  instead of location.  

 - I wonder, though, if Fagerland told the whole story, about the increased
use of non-parametric tests in studies with large Ns.  Since there are
some people who are (overly) concerned with non-normality, I did
eventually become willing to do my t-tests and report them in detail,
while tossing in the assurance that non-parametric tests showed the
same conclusions.  By nominal tabulation, my use increased over time.  
You would have to separate out that sort of back-up confirmation before
you conclude that there is really a lot of cherry-picking of results for their
p-values.

---

Emeritus Professor Diana Kornbrot
email:  d.e.kornbrot@...    
 web:    http://dianakornbrot.wordpress.com/
Work
Department of Psychology
School of Life and Medical Sciences
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
voice:   [phone]
Home
19 Elmhurst Avenue
London N2 0LT, UK
voice:   [phone]
mobile: [phone]