Likert scaling

From Citizendium, the Citizens' Compendium
Jump to: navigation, search
This article is developing and not approved.
Main Article
Talk
Related Articles  [?]
Bibliography  [?]
External Links  [?]
Citable Version  [?]
 
This editable Main Article is under development and not meant to be cited; by editing it you can help to improve it towards a future approved, citable version. These unapproved articles are subject to a disclaimer.

Likert scaling is the most common technique for constructing questions in survey research, such as political opinion polling. In opinion research, the question asks the respondent to state the opinion of "strong agreement" to "strong disagreement" with a statement, giving a numerical response. The responses are often 1 through 5, but may be 1 to 7 or 1 to 10.

It is also used in health assessment, with an assumption that the respondent would agree with all activities that are less challenging than the ones below the ones select. For example, if the highest value is "runs marathons" and the next highest is "runs one mile", it would be assumed that the lower-intensity activities such as "walks one mile" and "can climb one flight of stairs" would also be true. If there is uncertainty that this cumulative agreement is true, the question may need to be validated with a technique such as Guttman scaling.

In these applications of Likert scaling, care must be taken to have all responses use common assumptions: riding a bicycle is less demanding than running, but a question of "cycles 26 miles" is not a reasonable second level to "runs 26 miles", for a very physically fit person that does not know how to ride a bicycle.

The underlying nonparametric statistical methods assume that the range of responses will have a roughly uniform, or perhaps normal, distribution across the range. Using the example of 1 to 5, while it would not be expected that there would be an exact 20% in each "bucket", neither would it be expected that there might be 40% each in 1 and 5, or 70% in 5. If the survey designer, with knowledge of the population, expects a more clustered response pattern such as the latter two, a different statistical basis, such as Mokken scaling should be used for the design of questions and the analysis of results. Mokken and Likert questions look alike but are designed and evaluated differently.