[ posted to sci.stat.consult, sci.stat.edu ]
On 08 Jan 2001 08:27:17 GMT, [EMAIL PROTECTED] (Laurie Moseley) wrote:
> It might mean: "Undecided", "I don't know", "Sometimes Yes, sometimes No", "It
> all depends", and probably many other things as well. It sounds as though the
> questionnaire was not thought through or tested before being sent out.
>
> If you are really interested in the other scale points (the Agree-Disagree
> ones), you can leave the mid-point unlabelled, if it doesn't matter what it
> means other than "The Agree-Disagree does not matter in this case". If ,
> however, the exact meaning of the mid-point matters, then there is no
> alternative to doing pilot work aimed specifically at wording this mid-point.
>
> It may be that it is not just Agree-Disagree directly that you are interested
> in. You might try "In how many cases is the item true ?" with options such as
> "On average about half" or "It varies too much to tell".
>
Ah-ha! I think I figured out what has been bothering me.
If you have a "Likert-type scale", you intend to average it.
Originally, "Likert" meant, agreement on attitudes; so "missing"
or "Not applicable" was not going to occur. All the mid-scale
responses would mean the same, "mediocre" thing. But,
these days, you also expect to average (or total) the items if you
have a scale with Likert-style label.
If you are going to be technical about defining scales, "Likert"
strongly implies that the "?" in the middle will signify an
intermediate attitude. If it also means "Not Applicable" , that
creates ambiguity. Is that the scoring you want? Really?
And, Does this affect more than a tiny number of your respondents? It
is harder to be sure the distinction does not matter if "?" is the
modal response.
I am tempted to call this "bad scaling" instead of Likert
scaling, and I begin to sympathize, a tad, with the folks
who don't like treating Likert scales as interval scales.
LM>
> The other alternative is to give up simple Likert-type scales and go to, say,
> Thurstone or Magnitude Ratio scales.
That's skipping a step, isn't it? You can't do "magnitude ratio"
scaling without separating out the distinct "?" responses. But a
simple, intelligible Likert response is no problem after you make it
two questions. (Of course, there remains a problem that someone might
have been trying to hide, of assigning a scale value where the scale
is irrelevant.)
LM>
> The nature of the answer you would find informative depends both on the
> research question and on the question wording as written.
>
> The lesson is that scales are much more complex than at first sight appears,
> and you have to do more work before sending out the main instrument.
>
- right.
--
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/
=================================================================
