The same old question- Likert scale if Ordinal measurement (Non Metric) or Interval (Metric)

[Dr Neeraj Kaushik]

Taken from Yahoo Groups
http://in.groups.yahoo.com/group/Statisticians_group/messages/4817?threaded=1&m=e&tidx=1
======================================================================

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Date:: Mon Dec 20, 2010 4:31 am
Subject:: Re: info query kabonline07
Offline Offline
Send Email Send Email

Hi Jay & everyone ,

Is there any yes /no answer to the question " Whether the measurement
obtained from Likert scale is in ordinal/interval scale?"
If the answer is something like it depends (e.g."If you assert that
the 'distance' between each level of the Likert scale is equal, then
you can use the rating as a continuous response.") I wonder it may
also be in a ratio sacle?(In fact I have seen people saying/doing
that.)

But then what about typical text book like "Zikmund" clearly stating
that Likert scale is an ordinal scale? A MBA student will only read
those books for 'Resaerch Methodology' , right?

Kaushik Bhattacharjee

=================================================================================

Short answer: no., a Likert scale is _not_ an interval or ratio
scale. Check the definitions of these two scales, and see if your
scale fits.

Long answer (and the one I prefer:) : How did you devise your scale?
A "Likert" scale by original definition involved a range of like/
dislike attitudes or emotions, in response to a statement. It also
works very well, thank you very much, where there is no scale that can
be measured (as with a ruler), and the respondent is asked to respond
with something more detailed than 'yes' or 'no.' For example, I have
used it very well to establish a rating on a surface finish, and also
on a rating of 'quality' of a braze joint. Both of these are very
hard, and expensive, to measure in such a way that a dial display will
yield a single useful number - a measurement. But a visual
assessment, backed up with a 'standard' sample or three, gives a
remarkably precise and objective result.

Now, look at your levels on the scale, not as numbers, but as the
original words that you used to define them. What is the 'distance'
between each one? Is it the same in every case? Of course, you can't
tell - you're only discussing words, whose meanings are not
necessarily the same for you, the analyst, as for the respondent. So
on its face, a Likert scale is not interval. It must be ordinal.
Check those definitions again to see why the scale can be ordinal, and
not categorial. (If you put 'no opinion' or 'no answer' in your
scale, then the scale is not even ordinal.)

IF and ONLY IF you declare/assert that the intervals are equal, then
you can treat the scale as interval, and you can calculate averages &
stdev's. Can't do that for an ordinal scale, by the way. It still
won't be a ratio scale unless a rating of 1 is one-third the 'value'
of a rating of 3.


Now, suppose you do claim the scale is interval. By its nature, a
scale that runs only from 1 to 5, or to 7 or 10 for that matter, is
NOT Normal. It cannot in principle go below 1, nor over the highest
scale point. What I did, was to add the points for 0 (pitiful
outcome) and 6 (better than we ever anticipated), and crunched the
results. So it wasn't Normal; so it was only mostly interval. Didn't
matter; I was looking for which direction to go for improvement, and I
got it. IN your work, you may want more statistically confident
predictions than I used.

Cheers,
Jay
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA