Yes, Alpha should always be set a priori, but too many people simply set
it arbitrarily at .05 without a priori examination of  the important
(oops--I almost made the Freudian slip of spelling that last word
i-m-p-o-t-e-n-t) factors related to the choice of the alpha level
to-be-used.
Cheers,
Hank

-----------------------------------------------------------------------
Hank Goldstein,                     |   HOME:   (563) 556-2115
Department of Psychology     |   FAX:      (563) 588-6789
Clarke College                       |   EMAIL:  
[EMAIL PROTECTED]
Dubuque, IA  52001              |   HOME:   1835 Cannon St.
Office: (563) 588-8111          |                  Dubuque, IA
52003-7904
-----------------------------------------------------------------------
"There is no cure for birth and death save to enjoy the interval." -
George Santayana
"The most wasted of all days is one without laughter." - e.e. cummings
-----------------------------------------------------------------------
>>> [EMAIL PROTECTED] 11/11/02 12:40 PM >>>
Not to really disagree, but there are many ways of increasing power
besides increasing sample size. Sample size is often the most
expensive/difficult one. Others include more reliable measures, use of
more homogenous samples, better control of other (non-subject)
extraneous variables (e.g. background noise levels), and measurement and
"covarying out" of extraneous variables. On top of that, there's the
method of using a strong treatment rather than a weak one. Sometimes
students use weak treatments to "be fair", even in what should be
exploratory research. There's nothing wrong with using an
unrealistically strong treatment to establish an effect, and then
weakening the treatment to more realistic levels in later research. 

I couldn't agree more that setting the alpha level should be done a
priori. I don't understand why one would use an alpha level at all if
one were going to change it after the data were in. That seems to
entirely defeat the purpose of setting a standard. 

Paul Smith
Alverno College
Milwaukee

-----Original Message-----
From: Rick Froman [mailto:RFroman@;jbu.edu]
Sent: lundi 11 novembre 2002 11:53
To: Teaching in the Psychological Sciences
Subject: RE: Marginally Significant?


Although it is true that alpha is somewhat arbitrary, I think to avoid
all
kinds of mischief, if there is a justification for changing the alpha
level,
it should be set before the analysis, not post hoc. If you are concerned
about making a Type II error, your best choice is to increase the sample
size, if possible, instead of raising alpha.

Rick

Dr. Richard L. Froman
Psychology Department
John Brown University
Siloam Springs, AR 72761
e-mail: [EMAIL PROTECTED]
phone and voice mail: (479)524-7295
http://www.jbu.edu/sbs/rfroman.html

-----Original Message-----
From: Hank Goldstein [mailto:Hank.Goldstein@;clarke.edu]
Sent: Monday, November 11, 2002 10:08 AM
To: Teaching in the Psychological Sciences
Subject: Re: Marginally Significant?


Since the choice of alpha is somewhat arbitrary and should depend, to a
large extent, on the relative importance (i.e., practical consequences)
of Type I and Type II errors, I don't agree that significance is an
either-or decision. It may seem that it should be an either-or decision,
depending on how simplistic an approach one wants to take to the whole
complex concept of hypothesis testing.

"Them there" is my 3 cents worth!

Warm regards,
Hank


-----------------------------------------------------------------------
Hank Goldstein,                     |   HOME:   (563) 556-2115
Department of Psychology     |   FAX:      (563) 588-6789
Clarke College                       |   EMAIL:  
[EMAIL PROTECTED]
Dubuque, IA  52001              |   HOME:   1835 Cannon St.
Office: (563) 588-8111          |                  Dubuque, IA
52003-7904
-----------------------------------------------------------------------
"There is no cure for birth and death save to enjoy the interval." -
George Santayana
"The most wasted of all days is one without laughter." - e.e. cummings
-----------------------------------------------------------------------
>>> [EMAIL PROTECTED] 11/11/02 06:45 AM >>>
One of my students doing her senior thesis ran her stats and got results
of
.056 and .08 for two different ANOVAs. In the past I have seen published
..056 and .08 for two different ANOVAs. In the past I have seen
published
..056 and .08 for two different ANOVAs. In the past I have seen ...056
and .08 for two different ANOVAs. In the past I have seen published
studies indicating that these are "marginally significant." How do you
deal
with results of this nature? More importantly, do you have any citations
(journals or books) that discuss the value of including/discussing
results
that seem to "approach significance"?

Thanks,

Rob Flint
-------------------------------------------------------------
Robert W. Flint, Jr., Ph.D.
The College of Saint Rose
Department of Psychology
432 Western Avenue
Albany, NY  12203-1490

Office: 518-458-5379
Lab: 518-454-2102
Fax: 518-458-5446

Behavioral Neuroscience Homepage:
http://academic.strose.edu/academic/flintr/
Department of Psychology Homepage:
http://academic.strose.edu/academic/psychology/index.htm


---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to
[EMAIL PROTECTED]


---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to
[EMAIL PROTECTED]

---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to
[EMAIL PROTECTED]

---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to
[EMAIL PROTECTED]


---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to [EMAIL PROTECTED]

Reply via email to