Tuesday, January 31, 2012

How to bake the perfect Likert Scale

I've been building a survey and needed to argue a point about some "Likert Scale" questions in the design... so I started reading a little and all the bits that I thought turn out to be crap..... damit.

What's a Likert scale?(First Error...) A tool(Second Error) to measure attitude strength and direction. (And maybe also "Conviction" or the firmness with which the attitude is held) (Third Error)

* Error 1 - "Scale" can mean a number of things.  The intended meaning of this would more correctly be a "Likert response format item"; probably as part of a scale, composed of many of these response items... which can then be called a "Likert Scale". 

* Error 2 - It's not a tool, its simply a mechanism by which to filter the possible responses of the participant into a very limited subset that are acceptable to the researcher. 

* Error 3 - One answer to one question does not measure anything.  Statistical measurement requires multiple tests and is based on a summative result.  Its pattern recognition. One data point does not make a trend line.

This post is simply looking at the variety of flavours of Likert response items in an effor to make sense of it all.

Likert item Stylin'

How many ways can you present a Likert response format?  Seems simple enough....but.... maybe not.  Some of the options I have been asked to use are...

1. Text anchors  vs text labels on all points
2. Numeric scale vs "dots" vs line vs line with dots.
3. Midpoint or no midpoint points ("Forced Choice" model)
4. NA column, "dont know" and "No opinion" choice.
5. Reverse scale or not
6. Number of points
7. One stage or two (Split the direction and the strength into seperate question items)

Seriously, can all these thing actually end up being used in the same way are they real options?  Probably not.  What you may be calling a likert scale/item may infact be a...

Discrete visual analog scale(DVAS)
Ordered-category item
Summated rating scale
Random crap

This article breaks down the above similar constructs and explains the differences (except the random crap). Essentially if your items do not have a agree-to-disagree response format... your doing something else. 

So are you even talking about anything Likert "like"? Here's an article on constructing a Likert style Scale with some rigor.  Chances are, if your's is not being constructed and used in a very similar way... its something else. So go find another name.


Questions of Analysis

There are various arguments about what can and cannot be done with data collected using a Likert style scale.  Parametric or not?  F-Test or not? Ordinal, Interval or Norminal?  Most of these questions illustrate a more fundamental flaw in understanding the context and the instrument that has collected the data than in any choices in analysis.  Go read a good text, define the terms clearly and re-examine the research questions... it should solve itself.

And in conclusion... 

I vow never to misuse the term "Likert scale" again.... bad me.

Further Reading

Stopping the abuse... with the one true article
(The writing style in this article is worth a look... talk about venom and breathless, hypobolic writing and run on sentences and .... you get the idea.)
(Carifio, J. & Perla, R. (2007). Ten Common Misunderstandings, Misconceptions, Persistent Myths and Urban Legends about Likert Scales and Likert Response Formats and their Antidotes. Journal of Social Sciences, 2, 106-116. )

This is the Jamison article that seems to have pissed off the writer of the above article.
(Jamieson, S., 2004. Likert scales: how to (ab)use
them. Medical Education, 38: 1212-1218.)

To Mispoint or not to midpoint...

Breaking it up into bits and re mixing it

No comments:

Post a Comment