On 9 Jan 2009 at 10:53, Mike Palij wrote:

> On Thu, 08 Jan 2009 20:04:23 -0800, Karl Wuensch wrote:
> >        I'm even less conservative than Stephen.  I would not apply the
> >Bonferroni adjustment.  After all, these are PLANNED comparisons, eh?
> 
> This is a curious point:
> 
> Why should the state of knowledge (i.e., able to predict the size
> of difference, the direction of a difference, etc.) affect the probability
> of making an error of inference?

I've come to the opinion (based on reading and struggling with this issue 
over the years) that neither requiring a significant overall ANOVA nor 
prediction are critical. What is critical is how many questions you ask 
(number of comparisons) and why you want to do them (justification) 
Unlike the planned/post hoc concern,  justification is open to inspection 
and evaluation by others. 

If you have only a few comparisons, and those tests can be justified as 
meaningful and essential to your investigation, then you can dispense 
with a Bonferroni or other correction for multiple comparisons, as Karl 
recommends above. However, as the number of tests goes up and especially 
if the justification for doing them is weak,  the need for a Bonferroni 
correction correspondingly increases.

This is admittedly subjective, but not excessively. Whether the data has 
been collected or not doesn't matter, as long as you don't let yourself 
be influenced by it (and others can see whether you have).  You start by 
listing the set of meaningful and important comparisons which you must 
test, keeping the number to a minimum because more tests mean a more 
severe correction. If none are significant uncorrected, or if all are 
significant corrected for multiple comparisons, report that. For results 
which fall in-between (significant uncorrected, but not significant when 
corrected), report both results, and be correspondingly cautious in what 
you claim. Your reader can then decide for him/herself. 

This is essentially a planned approach, which I find far preferable to 
the clutter and confusion of testing with wild abandon, but it still 
requires correction for multiple comparisons in most cases.

> 
> Consider t-tests, if we "know" that a difference is in a particular
> direction, then using a one-tailed test will give us a lower critical
> value and possibly make it easier to reject the null hypothesis that
> the two means estimate the same population mean.  But if our
> "knowledge" is faulty or incomplete and the difference is in the
> opposite direction we either (a) ignore a statistically significant
> result because it was not in "right" direction, (b) say "Ooops!",
> and claim we "planned" on doing a two-tailed test all along.
> <snip>
>   It was for reasons
> like this that Jack Cohen used to say that one-tailed tests should
> not be done (how could one distinguish delusional self-confidence
> from solid knowledge?) 

Absolutely.  This isssue drives me nuts. One-tailed tests should be 
banned (almost), and certainly raise my suspicion whenever I see them 
used. But the issue isn't one of knowledge, and just because you make a 
prediction doesn't give you the right to use them. You must be able to 
swear (on the Bible, the Koran, or _The Design of Experiments_) that even 
if, contrary to all that you hold reasonable, the result falls in the 
"wrong" tail, you would have not the slightest interest in such an 
outcome, which you would declare to be totally meaningless. 

It's not easy to find such examples which fit this criterion allowing a 
one-tailed test. The only one I can recall is, say, that you're testing 
whether a therapy is better than a placebo. You predict that it is but, 
more important, you have absolutely no interest in an outcome which shows 
the therapy causes harm. Because harm and merely useless lead to the same 
conclusion: don't use the therapy.

Even this isn't a good example, because a finding that a therapy causes 
deterioration could well have important implications. But a clinician 
might not care.

Stephen

-----------------------------------------------------------------
Stephen L. Black, Ph.D.          
Professor of Psychology, Emeritus   
Bishop's University      e-mail:  [email protected]
2600 College St.
Sherbrooke QC  J1M 1Z7
Canada

Subscribe to discussion list (TIPS) for the teaching of
psychology at http://flightline.highline.edu/sfrantz/tips/
-----------------------------------------------------------------------

---
To make changes to your subscription contact:

Bill Southerly ([email protected])

Reply via email to