Robert Abelson, in his excellent book Statistics as Principled Argument, 
advocates the use of what he calls a one-and-a-half tailed test for directional 
predictions; for example, you could set the alpha level at .04 in the tail of 
the predicted direction and .01 in the unpredicted direction. I always thought 
that was a neat idea, but I've never seen anyone use it.

________________________________________
From: [email protected] [[email protected]]
Sent: Friday, January 07, 2011 11:04 AM
To: Teaching in the Psychological Sciences (TIPS)
Subject: Re: [tips] Don't Be Surprised If Your Physics Colleagues Snicker When 
They Pass You In the Hall

On 6 Jan 2011, I wrote in response to Martin Bourgeois pointing out
that Bem used one-tailed tests:

> That's interesting, disturbing, in fact. I've long argued that one-
> tailed tests are almost uniformly misused in psychology and should be
> banned. Making a one-sided prediction is insufficient justification.
> One should only be allowed to use a one-tailed test if one can
> plausibly argue that not only do I not predict a result in the
> "wrong" tail, but that if such a perverse result occurred, it would
> either be meaningless or of no interest at all.

I am happy to report that a paper hot off the press (or do they not
do that any more?), in what appears to be a brand-new journal, has an
excellent opinion piece which fully supports my view above. How they
got it into print so quickly after reading my post, I'll never know.

The paper is this:

Ruxton, G, and Neuhauser, M. (2010). When should we use one-tailed
hypothesis testing? Methods in Ecology and Evolution, 1, 114-117.

After giving a great example of the tangled mess an unwary researcher
is faced with when a one-tailed test goes rogue, they say this:

"On the basis of their statistical test, the scientist has no grounds
for treating an experiment where the birds having a spectacular
adverse reaction to the supplement any differently from the birds
having no reaction. This philosophical lack of ability to act in
response to unexpected results is the cost of one-tailed testing".

Amen! Cut your costs! Help stamp out one-tailed tests!

They also say "We rarely find ourselves in a position where we are
comfortable with using  a one-tailed test", and later,  "Use of one-
tailed testing is more common than we would expect" [in ecology
journals], They report that none of these uses involved a
satisfactory explanation.

The authors' summary, in modest and restrained language, offers this
good guide for the perplexed:
----------------------------
Summary

1. Although one-tailed hypothesis tests are commonly used, clear
justification for why this approach is used is often missing
frompublished papers.

2. Here we suggest explicit questions authors should ask of
themselves when deciding whether or not to adopt one-tailed tests.

3. First, we suggest that authors should only use a one-tailed test
if they can explain why they are more interested in an effect in one
direction and not the other.

4. We suggest a further requirement that adoption of one-tailed
testing requires an explanation why the authors would treat a large
observed difference in the unexpected direction no differently
from a difference in the expected direction that was not strong
enough to justify rejection of the null hypothesis.

5. These justifications should be included in published works that
use one-tailed tests, allowing editors, reviewers and readers the
ability to evaluate the appropriateness of the adoption of one-tailed
testing.

6. We feel that adherence to our suggestions will allow authors to
use one-tailed tests more appropriately, and readers to form their
own opinion about such appropriateness when one-tailed tests
are used.
-------------------------


Stephen

--------------------------------------------
Stephen L. Black, Ph.D.
Professor of Psychology, Emeritus
Bishop's University
Sherbrooke, Quebec, Canada
e-mail:  sblack at ubishops.ca
---------------------------------------------

---
You are currently subscribed to tips as: [email protected].
To unsubscribe click here: 
http://fsulist.frostburg.edu/u?id=13390.2bbc1cc8fd0e5f9e0b91f01828c87814&n=T&l=tips&o=7744
or send a blank email to 
leave-7744-13390.2bbc1cc8fd0e5f9e0b91f01828c87...@fsulist.frostburg.edu
---
You are currently subscribed to tips as: [email protected].
To unsubscribe click here: 
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=7746
or send a blank email to 
leave-7746-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu

Reply via email to