NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/bit/listserv/statl/1475

Text File | 1992-09-03 | 1.7 KB | 35 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!paladin.american.edu!auvm!IRLEARN.UCD.IE!RCONROY
Message-ID: <[e-mail]>
Newsgroups: bit.listserv.stat-l
Date:         Thu, 3 Sep 1992 10:51:10 GMT
Sender:       "STATISTICAL CONSULTING" <[e-mail]>
From:         Ronan M Conroy <[e-mail]>
Subject:      ordinal, interval, cautions
Lines: 24

The classic 5-point scale rangine from "agree strongly" to
"disagree strongly" does not meet the assumptions of equal-interval
scales, since the scale steps do not correspond to increases in
a defined unit of anything. More important, however, is that
the use of these scales correlates with things like education
and social class - two tendencies, in particular, are worth
noting: one is for better-educated people to make more use of
the extremes of the scale, and the other is for a response set
to set in, so that responses to later items are patterened by
responses to earlier ones (if you start with a few "don't know"
items, you'll be more likely to use the "don't know" box in later
items, for instance). For these reasons, I usually recommend
rescaling bunches of 5-point items as 2-point, by grouping
together all the responses that represent what you are interested
in and scoring them as '1', and putting everything else as '0'.
If you are intested in the presence of significant levels of
depression, for example, you could score a 1 for any item agreeing
with a 'depressed' statement or disagreeing with a 'cheerful' one.

I am very much against adding responses on the 5-point scale, since,
if it isn't an interval scale, you can't do addition on it. The 1/0
system allows you to add items and explain what you've done.

ronan