s-news
[Top] [All Lists]

adding and averaging likert scale or similar variables

To: s-news@lists.biostat.wustl.edu
Subject: adding and averaging likert scale or similar variables
From: ellisp@netspeed.com.au
Date: Fri, 10 May 2002 15:11:45 +1000
User-agent: Internet Messaging Program (IMP) 3.0

All

I apologise for the slightly off topic nature of this question and will keep it 
brief (I am prepared to wear any flames and humbly apologise for the blatantly 
non S focus).

In many fields (eg psychology, marketing, political science) it is common to 
add and average ordinal ratings as though they are numbers in a way that is on 
the face of it completely inappropriate.  Similarly, people use ordinary 
correlation coefficients when they should use polychoric correlations; bog 
standard continuous response linear models when they should have some sort of 
more complex GLM; etc.  However, my hunch and to a certain degree experience is 
that at least sometimes this approach does not do too much damage to the data 
(eg more complex appropriate responses come up with similar results).

1) What are people's views?  Is it unfair, for example, to say "such scales 
should never be added and averaged"?

2) Where should I look for the literature on this? (what literature I have 
found concentrates on pointing out the flaws of common techniques and the 
importance of using some more sophisticated methods, usually one developed by 
the author; but I have not seen a comprehensive conceptual, theoretical and 
empirical comparison of the cheats' methods versus the "proper" methods)

cheers, ta, and sorry again
PE

<Prev in Thread] Current Thread [Next in Thread>