( Posted to sci.stat.edu and sci.stat.consult. I notice that there
are several replies in sci.stat.edu that came from that mailing
list, and were not cross-posted.)
On 7 Sep 2001 07:38:00 -0700, Michael Lacy
<[EMAIL PROTECTED]> wrote:
> Here's the locus classicus, I think:
> Likert, Rensis. 1932. "A Technique for the Measurement of Attitudes."
> Archives of Psychology 140: pp. 44-53.
>
> This and numerous other classic articles in scaling are reprinted in
> Maranell, Gary M.(ed.) 1974. Scaling: A Sourcebook for Behavioral
> Scientists. Chicago: Aldine.
>
> FWIW, I would argee that the term "Likert Scale" should only
> be applied to a summated ratings scale based on several items
> and subjected to some sort of item analysis to demonstrate
> unidimensionality, etc. I don't think the terminology needs
> necessarily be restricted to the measurement of attitudes,
> however.
My local library has the Maranell volume, so I will be looking for
it - thanks.
I agree with Mike's opinion, above, that "Likert Scale" does
not need to refer to attitudes, and that it still ought to imply
that some amount of reliability testing has been performed.
When I looked for textbook opinions a couple of years ago,
the handful of descriptions were not entirely explicit. However,
(if I remember right) all the old textbook *examples* used agree/
disagree, on attitude questions.
No one seemed to consider that a single item would be
called "a Likert scale" -- I think that is a recent formulation:
Nowadays there are a lot of "scales"; they no longer belong to
the psychologists alone, many of whom used "Likert scale"
as practically synonymous with "additive scale." I do think
the notion of multiple items, added, should be considered
as essential, since a Likert item can be called an "item."
Roderick McDonald's 1999 text, "Test Theory, a unified
treatment," has a few paragraphs of interesting commentary.
Page 22 offers an opinion which is more forgiving than my own:
"Scoring the items using integers is so commonly referred to
as /Likert scaling/ that it is virtually impossible to correct the
misunderstanding. It seems inevitable that we follow the
accepted usage and refer to an integer score for an ordered-
category item as a /Likert score/.
"Ordered-category items can also be described as /graded
response items/. ..."
What do we insist on as reviewers? What do we tell
editors if they ask our advice? I think that the label of
"Likert scale" should mean, at the least,
a) a summed score,
b) based on one set of items and ordered categories, and which
c) has already passed at least *some* testing. (Face validity
may be enough, if that implies the opinion of an expert.)
I would try to insist on (a) and (b) as a reviewer, but (c)?
- That requirement is an ordinary requirement for publication.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
