Summary: Here's the bottom line for the busy reader. Over time and in common usage, the term, "Likert scale" has come to be applied to things far removed from its original meaning. Most importantly:
People have gradually come to use the term, "Likert scale" in very different ways. It is variously applied it to both groups of items and to single items, and in either case there is disagreement about what specific formats apply.
This is a not good in general, since we would like mutually agreed-on definitions. Otherwise if a researcher says, "We used a Likert scale" it isn't clear what's meant.
Further, there is lack of consensus about what statistical methods are appropriate for this class of variables. This is an important matter because such variables are often used in serious applications like clinical trials. To make headway on the issue of what statistical methods are appropriate for such variables (and this is the subject of some degree of controversy), we must first agree on terms.
Likert scales were originally developed by Rensis Likert, a sociologist at the University of Michigan from 1946 to 1970. Likert was concerned with measuring psychological attitudes, and wished to do this in a "scientific" way. Specifically, he sought a method that would produce attitude measures that could reasonably be interpreted as measurements on a proper metric scale, in the same sense that we consider inches or degrees Celsius true measurement scales.
Other social scientists, such as Thurstone, had already developed sophisticated methods for measurement of psychological phenomena, but these were unsuited for Likert's attitude research. Likert, after trying various alternatives, gradually developed what we now call Likert scales.
Likert used a number of specific techniques to first generate items, and then select from among them those that were valid, unidimensional (all measuring a common trait), and well discriminating. For example, he sometimes used judges to rate items' quality or content. All of these methods collectively go into what is formally called Likert scaling. Without failing to appreciate Likert's contributions to the science of scaling, we use the term "Likert scale" in a somewhat broader sense here to include basically any scale composed of Likert or Likert-type items. That is, we distinguish between Likert scaling and Likert scales, the former term being more specific. (We thereby avoid having to introduce yet another category, Likert-type scales.)
Once constructed, Likert's scales had a format like this:
Please indicate how much you agree or disagree with each of these statements: Neither agree Strongly Somewhat nor Somewhat Strongly disagree disagree disagree agree agree -------------------------------------------------------------------- The president is 1 2 3 4 5 doing a good job. The Congress is 1 2 3 4 5 doing a good job. The Secretary of 1 2 3 4 5 Defense is doing a good job.
Here the construct being measured might be attitude towards American politics.
By Likert's method, a person's attitude is measured by combining (adding or averaging) their responses across all items. This summing or averaging across several items was essential for Likert to contribute to genuine measurement.
We note several characteristics or features that define a Likert scale:
Response levels are arranged horizontally.
Response levels are anchored with consecutive integers.
Response levels are also anchored with verbal labels which connote
more-or-less evenly-spaced gradations.
Verbal labels are bivalent and symmetrical about a neutral middle
Criterion 5 usually means there is an odd number of response levels. Typically the number is 5, though sometimes 7, 9, or 11 levels are used.
Only a scale with all these characteristics might qualify as a genuine Likert scale. We probably don't want to be that strict, however. In particular, it seems reasonable to apply Likert's methodology to domains other than attitude measurement. The view recommended here is that features 1-4 above comprise the main requirements for what can be accurately termed a Likert scale.
This much said, we now address two of the biggest and most common confusions people make. First:
A Likert scale is never an individual item; it is always a set of several items, with specific format features, the responses to which are added or averaged to produce an overall score or measurement.
A single item, even if formatted exactly as one of Likert's items, is not a Likert scale.
The confusion is understandable, however, since each item of a Likert scale itself has scale-like appearance. But these are definitely to be distinguished from the Likert scale proper, which is made up of the entire set of items.
The question then arises: so what *should* we call single items of this kind? Is there a term by which we may distinguish them from other kinds of items, such as ordinary multiple choice ones?
If features 2 through 5 above are all present, we may justifiably call them Likert items. If only 2 through 4 are present, we might call them Likert-type items instead.
How do you feel about the President's performance in domestic affairs? Strongly Somewhat Somewhat Strongly disapprove disapprove Neutral approve approve 1 2 3 4 5
Consider the example above. Here we meet all criteria 2 - 5. It seems fair to call this a Likert item, even though it doesn't refer to agreement/disagreement to a target statement.
How far can we legitimately broaden the definition? This is a judgment call on the part of the researcher. It is this writer's opinion that only criterion 5--that the anchor labels be bivalent (distinctly two-directional) and symmetrical--may be relaxed, and then this produces a Likert-type item. Without conditions 2-4 the item is basically not Likert-type in any sense.
The following, then, would be considered a Likert-type item.
How often do you go out to see a movie? Very Never Sometimes Average Often often 1 2 3 4 5
Here the response levels are not bivalent: the lower terminus is merely "Never". There is no exact opposite of "Very often". Yet the categories are reasonably interpretable as evenly spaced, especially when associated with consecutive integers in an evenly-spaced printed format. The label for level 3, "Average" clearly denotes centrality of this response category.
Similarly, we seem reasonably justified in permitting an even number of response levels, and, along with this, that there not be an exact middle or neutral category--provided the other criteria are maintained.
There are shades of gray here, so it is difficult to provide a universally applicable set of rules. But certainly the further away an item is from the criteria shown, the less inclined one should be to refer to it as a Likert item or Likert-type item. In particular, most regrettable is the tendency, not uncommon today, to refer to any ordered category item as Likert-type. We may treat this kind of error rather summarily, as follows:
This is not a Likert-scale, a Likert item, or a Likert-type item:
How often do you smoke cigarettes?
This is simply an item with ordered response levels, or an ordered-category item.
This is so even if the item is one of several that will be combined to form an aggregate scale. In this case, one simply has a summated rating scale comprised of several ordered-category variables.
|How helpful did you find this software?|
or even this:
|How helpful did you find this software?|
Note that several of our criteria are missing here. In particular, there are no verbal labels at all, except to define the poles of a continuum. Example 4 corresponds to what is called a semantic differential item.
The recommendation here is that items of either type above should be called simply discrete visual analog scale (DVAS) items. Such an item is a visual analog scale, because the printed format implies specific metric relations among the response levels. It is discrete because only pre-specified levels may be given by the respondent; this is in contrast with a more general visual analog scale format, where, for instance, the respondent may be instructed to place a mark on anywhere on a line to indicate his or her level of response.
Note that DVAS is a more generic term than "Likert-type item", and superordinate to it. All Likert-type items, at least according to the view here, are DVAS, but all DVAS are not Likert-type items.
Some might suggest that the essential innovation, and, from this, the defining feature, of a Likert scale was not so much the item format, as that the responses of several items were added or averaged to produce a composite score; f further, as this is an important and common technique, some name is needed. Here we simply note that there is an alternative name for such multi-item scales, namely summated rating scales. Multi-item scales, where individual items lack the features listed above, should be called this or something equivalent, and not Likert scales.
Last updated: 31 August 2006 (Likert scaling)