The Likert scale is the single most common rating scale in quantitative market research. In its normal form it attempts to record a respondent’s level of agreement with statements. For example a statement might be “I am completely satisfied with the service provided by ACME” and the levels might be:
- Strongly disagree
- Disagree
- Neither agree nor disagree
- Agree
- Strongly agree
In formal terms there is a distinction between a Likert scale and a Likert item. The question in the example above is an example of a Likert item. If several Likert items are used and the results summed across those items, then the result is a Likert scale.
When the words are used correctly a Likert scale relates to the sum of results across several Likert items – but this is almost never done in commercial market research. Many/most people, in commercial market research, say Likert scale when they mean Likert item.
The Likert scale was developed by US educator and psychologist Rensis Likert (1903 – 1981) in 1932.
Amongst methodologists the big debate about Likert items is whether they can be treated as providing interval data (e.g. 1, 2, 3) or whether the numbers produced are simply ordinal (e.g. 1st, 2nd, and 3rd).
The case for ordinal data
The case for treating data from a Likert item as ordinal (i.e. 1st, 2nd, 3rd as opposed to 1, 2, 3) is based on the fact that there is no reason to believe that the difference between, say, ‘Strongly disagree’ and ‘Disagree’ is numerically the same as the difference between ‘Agree’ and ‘Neither agree nor disagree’. Similarly, if the scale is an interval scale the researcher needs to believe that the difference between ‘Agree’ and ‘Disagree’ is exactly twice as big as the difference between ‘Strongly disagree’ and ‘Disagree’.
If the data is not interval data, then it is ordinal data. If it is ordinal data then we can report the median, the mode, the percentage picking each level, the percentage picking the top two boxes, the bottom two boxes, the percentage agreeing or disagreeing, etc. However, we cannot and should not calculate a mean or standard deviation.
The case for interval data
Some researchers argue that to treat Likert numbers as ordinal loses information. They argue that the symmetry of the language of most Likert questions means that respondents understand they are supposed to respond in an interval scale fashion. To supplement this many researchers layout their questions graphically to help emphasise the interval nature of the data, and include placing numbers next to the text to help reinforce the interval scaling.
If a researcher makes the wild and unjustified decision (IMHO) to treat Likert numbers as an interval scale they are rewarded by being able to use means, standard deviations, and by being able to use the regular forms of techniques such as Factor Analysis and Regression (as opposed to the less familiar non-parametric alternatives).
In a commercial environment a researcher who refuses to pretend that Likert items produce interval data will probably face commercial difficulties (either from their clients or their employer).
The number of levels
Researchers disagree about the best number of levels to use. Probably the most common Likert item is the one with 5 levels, but 4, 7, 8, 9 and 10 are all fairly common. When different levels have been tested they have been shown to produce different results, for example different means and standard deviations. The most frequent results seem to be that more levels results in higher means and more dispersion (in the 5 point scale some researchers have reported that respondents are reluctant to use the 1 and 5, resulting in very limited variation). Most researchers have found that going beyond 8 points changes the results very little.
My own feeling is that with a larger number of levels a question changes from being a Likert item to more of a numerical scale, where there may be a stronger argument for treating it as an interval scale rather than just an ordinal scale.
Comments