Michelle,
I think you have your standards set higher that it is
possible to meet. Or than it is necessary. And you
haven't considered how easy it is to FAIL the tough
standards ....
On Sun, 9 Sep 2001 13:25:24 +1200, "Magenta"
<[EMAIL PROTECTED]> wrote:
[ snip ]
>
> My biggest problem with Likert (and similar) scales is the assumption that
> there is within-subject and between-subject consistency in responding to the
> items. For example, if I moderately agree with an item, then I have to
> select between "agree" and "strongly agree" on a typical 5-point scale.
> Sometimes I may choose "agree" and other times I may choose "strongly
> agree", but at all times my "true" response would be "moderately agree".
> Another respondent may choose "agree" for the same strength of feeling on
> all items. (This is just like people deciding on "reasonable doubt" if they
> are on a jury. One person's level may be 98% confidence, another's may be
> only 80%. It's the lack of consistency that is the problem.)
>
And by a similar logic, does Reaction Time in milliseconds
represent how fast a person CAN react? or how fast the person
was WILLING to react, to avoid making mistakes?
"Simple reaction time", with obvious measures, was collected
more than a century ago. It serves purposes, both between
and within subjects. But the standard between subjects needs
to be adjusted for "equal error rate-when-trying-to-go-fast" --
if you want to test *abilities*. That leads to 'discriminability'
and something called d-prime.
> One way to try to remove this problem is to behaviourally anchor the
> options. I have not often seen this method used.
>
- or, you can run some analyses and see if the accuracy
is good enough to WORK.
You can do one-way ANOVAs on Ranks and get
excellent p-levels, despite how far Uniform is from
Normal.
> This is the reason I prefer visual analog scales, then the person decides
> what their strength of feeling is. And it makes it easier to be consistent
> between questions.
Now, this puzzles me. Are you assuming that a visual analog
scale is always equipped with precise instructions about how
the ratings are supposed to relate?
I think I learned something about scaling when I read
about "magnitude estimation". If you let people use
numbers that are arbitrarily small fractions or huge numbers,
then you can let them relate a new measure to whatever
number came before. And (presumably) using the log of
that gives a fairly decent representation, of something.
--
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/
=================================================================
