By Glenn O’Neil at http://www.isixsigma.com/offsite.asp?A=Fr&Url=http://intelligentmeasurement.wordpress.com/2007/11/20/likert-scale-surveys-best-practices/ (November 20, 2007)
I’ve been looking into the best practices for using the Likert scale type of question, probably the most widely used response scale featured in surveys – often used to measure attitudes and other factors (e.g. “Excellent” to “Poor”). Created by Rensis Likert (pictured above) in the 1930s, his original scale featured five points. Over time, there has been many discussions and disagreements focused on one central question: What works best with the Likert scale to give you the most accurate responses?
I have read a number of studies on this question (sorry, I don’t link to them as they are all books or academic journals (that require a fee) but if you are interested write to me and I’ll give you the references) and the following are the points that most (but not all) scholars agree on:
More than seven points on a scale are too much. Studies show that people are not able to place their point of view on a scale greater than seven. So go for seven or less. What is the perfect number? Studies are not conclusive on this, most commonly mentioned are five, four or three point scales.
Numbered scales are difficult for people. For example, scales that are marked “1 to 5, with 5 being the highest” result in less accurate results than scales with labels such as “good” or “poor”. If numbered scales are used, signposts are recommended (e.g. put “poor” above 1, “satisfactory” above 3 and “excellent” above 5).
Labelled scales need to be as accurate as possible. Commonly uses labels such as “often” or “sometimes” often result in inaccurate responses. As these terms mean different notions of engagement from person to person, culture to culture (not to add the complexity of translating these terms). Scholars recommend using time-bound labels for frequency measures such as “once a week” (although problems of correct recall are also an issue). In addition, studies show that people find it difficult to differentiate between “very good” and “good” – better to use “good” and “excellent”.
And that’s it! Basically, there are inconclusive results on the use of a middle or neutral point (e.g. four point vs. a five point scale). Some scholars advocate a five point scale where respondents can have a “neutral” middle point whereas others prefer to “force” people to select a negative or positive position with a four point scale. In addition, the use of a “don’t know” option is inconclusive. I personally believe that a “don’t know” option is essential on some scales where people may simply not have an opinion. However, studies are inconclusive on if a “don’t know” option increases accuracy of responses.