home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!news.funet.fi!hydra!klaava!tarkkone
- From: tarkkone@klaava.Helsinki.FI (Lauri Tarkkonen)
- Newsgroups: bit.listserv.stat-l
- Subject: Re: guttmann scales
- Message-ID: <1992Nov7.070942.28258@klaava.Helsinki.FI>
- Date: 7 Nov 92 07:09:42 GMT
- References: <STAT-L%92110617193393@VM1.MCGILL.CA>
- Organization: University of Helsinki
- Lines: 54
-
- In <STAT-L%92110617193393@VM1.MCGILL.CA> ZUMBO@ACADVM1.UOTTAWA.CA (Bruno D. Zumbo) writes:
-
- > We have developed a clinical questionnaire for head injured
- >patients on their social and adaptive functioning. The
- >questionnaire consists of 6 or 7 major scales, each represented by
- >8 to 12 questions. Each scale, e.g. "taking intiative" or "being
- >organized", is Guttman scaled more or less so that there is some
- >sort of threshold for problem behaviour depending on how impaired
- >the patient is. But instead of YES/NO responses, we supply a
- >Likert scale. We then adjust the polarity appropriately and simply
- >add up the scores on the 1 to 5 ratings. We have found the results
- >useful and have externally validity in appropriate places. Now, I
- >have an honours student doing a project on checking some
- >psychometric features (something new for me), such as doing a
- >proper item analysis, seeing whether the Guttman scaling is as we
- >hoped, factor analysis, etc. The reason I am writing is that I
- >have not been able to find any relevant discussant in psychometry
- >books on such a type of test with Likert ratings on Guttman scales.
- >Do you know of any? and is there anything I should be aware of?
-
- There is a problem in trying to fit a Guttman scale and factor analysis
- in the same package. In Factor analysis you think that there are one
- or more factors affecting the variables observed, the strong correla-
- tions are caused by common factors. In a Guttman scale the items by
- structure cannot correlate very well:
-
- items
- observations i1 i2 i3 .... ip
- o1 0 0 0 .... 0
- o2 0 0 0 1
- . . . . .... 1
- . . . . .... 1
- . . . 1 .... 1
- on-1 0 1 1 1
- on 1 1 1 1
-
- The correlation of items i1 and ip is -1 but the variance is about 0.
- The correlation of first and last items to the middle items are about 0.
-
- If i draw a picture of the situation:
-
- median item
- ^
- !
- !
- !
- i1 <-!-> in
-
- The variance of the median iten is largest and the variance of the first
- and last item is smallest, aproaching nill. The direction you are supposed
- to separate your objets has almost no variance, the main variance is ortho-
- gonal to this. So Guttman leads you into trouble.
-
- - Lauri Tarkkonen
-