Re[2]: Likert-type scales: interval or ordinal?
P.Ja...@lse.ac.uk
This is a hoary old topic, I know, but here is my two penny worth.
A crucial (if not *the* crucial) issue here is surely how one attaches meaning
to the scales used. In particular, howe the respondant "interprets" such scales
and in turn what the researcher assumes about the respondent's interpretation.
If such scales were *really* being regarded as purely ordinal, then attaching
different numbers would make no analytic difference (eg use 10, 324 and 5000
instead of 0, 1 and 2) However, ask a researcher if they would mind actually
printing these values next to the response boxes on the questionnaire and they
would yell "no" ! Why? Because , rightly in my view, they would suspect this
would lead the respondent into thinking that the intervals on the scale were of
particular magnitudes. I would maintain that respondents do, *at least to some
degree*, respond as if there is an underlying scale and that, unless otherwise
led astray, they see the response alternatives on such scales as equally spaced,
or see the intervals as having some magnitude which influences their response.
We may be wrong (sometimes? often?) in assuming they are treating the intervals
as equal, but I believe we are also wrong in assuming that the repondent is
responding merely on the basis of ordinality.
Paul Jackson
Dept Social Psychology
London School of Economics
______________________________ Reply Separator _________________________________
Subject: Re: Likert-type scales: interval or ordinal?
Author: win...@commerce.murdoch.edu.au at :external_mail
Date: 18/05/95 07:36
>From edst...@jse.stat.ncsu.edu
X-Envelope-From: edst...@jse.stat.ncsu.edu
Received: from jse.stat.ncsu.edu by res.lse.ac.uk with SMTP (PP);
Thu, 18 May 1995 07:00:41 +0100
Received: from by jse.stat.ncsu.edu (8.6.11/SL31jan95) id CAA28228;
Thu, 18 May 1995 02:06:54 -0400
Date: Thu, 18 May 1995 02:06:54 -0400
Message-Id: <winzar.5...@commerce.murdoch.edu.au>
Reply-To: win...@commerce.murdoch.edu.au
Originator: edst...@jse.stat.ncsu.edu
Sender: edst...@jse.stat.ncsu.edu
Precedence: bulk
From: win...@commerce.murdoch.edu.au (Hume Winzar)
To: Multiple recipients of list <edst...@jse.stat.ncsu.edu>
Subject: Re: Likert-type scales: interval or ordinal?
X-Listserver-Version: 6.0 -- UNIX ListServer by Anastasios Kotsikonas
X-Comment: Statistics Education Discussion
In article <D8nI7...@news.uwindsor.ca> g...@uwindsor.ca (Rick Holigrocki)
writes:
>I've been having an argument with a friend of mine about interval and
>ordinal data. I am using a three-point scale that judges use to rate
>whether participants have learned a set of concepts. Judges score a
>0 for wrong
>1 for somewhat right
>2 for right
>As far as I am concerned, this is ordinal data. The interval between
>...
>My friend says that because there is an equal difference between
>0 and 1 and 1 and 2 the data is therefore interval data.
Technically, you are right. However, in practice, it seems to make very
little difference if you do treat likert-type variables as interval scaled.
With just two "intervals" in your three level scale, that could be very
arguable. With more levels, say with five-point or seven-point "agree /
disagree" scales, the data have to be extremely skewed before you get
appreciably different results between parametric and non-parametric
statistical procedures.