home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.stat-l
- Path: sparky!uunet!utcsri!torn!cunews!sjones
- From: sjones@alfred.carleton.ca (Stan Jones)
- Subject: Re: likert scales
- Message-ID: <sjones.716044733@cunews>
- Sender: news@cunews.carleton.ca (News Administrator)
- Organization: Carleton University
- References: <STAT-L%92090907195568@VM1.MCGILL.CA>
- Date: Wed, 9 Sep 1992 13:18:53 GMT
- Lines: 28
-
- In <STAT-L%92090907195568@VM1.MCGILL.CA> A716HOX@HASARA11.BITNET (Joop Hox) writes:
-
- >uses the simple summated score. Still, there is a psychometric theory behind
- >all this which resembles the Thurstone scale. One obvious assumption is that
- >all items measure one single underlying factor. Since the reliability
- >coefficient alpha measures the common factor saturation some reliability
- >analysis is often used in connection with Likert scales. If you do not scale
- >the item scores, you cannot assume they are interval scores, and the sum score
- >can be considered an interval score only by virtue of its close approximation
- >of the more complex original Likert score.
-
- From a data theory perspective (Coombs, Young) there is little
- difference between the Likert scales and equal-appearing-interval
- (Thurstone) scales except the former focuses on subjects and the
- latter on items (but both necesarily scale both items and subjects).
-
-
- >BTW, as summated score is a sufficient statistic for all the information in
- >the item scores only if the Rasch model holds. Since I have never seen real
- >data that totally support the Rasch model, I mention this only as a curious
- >fact.
-
- Likert scales are 'good' for getting a score for each subject - they
- are terrible for determining whether they are a good fit to the data.
- hence something (such as Rasch, though there are other ways as well)
- are necessary. It is also important, of course, to justify that the
- interpretation of the scale (its latent equivalent) is interval,
- ordibal, or what.
-