The Likert scale is arguably the most widely used type of attitude scale in the social sciences. The typical Likert scale appears as a collection of statements about an attitude object (person, group, institution, idea, etc.) reflecting favorable or unfavorable attitudes toward the object. Each statement is accompanied by a graded-response rating scale, typically with five response choices, the most commonly used being: “Strongly Agree,” “Agree,” “Undecided,” “Disagree,” and “Strongly Disagree.” The respondent is instructed to select one of the response choices for each statement. To score the scale, the response choices are given weights to reflect the attitude continuum, typically weights of 1 to 5, 1 for the most unfavorable (or least favorable) attitude and 5 for the most favorable. The weights for the respondent’s choices are then summed across all statements. The resulting total score may be interpreted normatively, with reference to some comparison group, or absolutely, with reference to theoretically or empirically chosen cut-off scores.
As examples, here are two contrasting items from a Likert scale on attitudes toward work:
1. Work is an activity to be avoided.
__Strongly Agree __Agree __Undecided __Disagree __Strongly Disagree
2. Work brings out the best of my abilities.
__Strongly Agree __Agree __Undecided __Disagree __Strongly Disagree
The first item reflects an unfavorable attitude toward work and therefore “Strongly Agree” is scored 1, “Agree” is scored 2, and so on, up to 5 for “Strongly Disagree.” The second item, reflecting a favorable attitude, is scored 5 for “Strongly Agree,” 4 for “Agree,” and so on, up to 1 for “Strongly Disagree.”
The format of five response choices per statement is so widely used that some have mistakenly called any instrument using this format a Likert scale. But it is not the five-choice format that characterizes a Likert scale. Rather, it is the method used in developing the scale—Rensis Likert’s method—that makes an instrument a Likert scale, regardless of the number of response choices used.
Likert’s method is essentially the construction of an internally consistent scale. Likert’s idea, innovative for its time in the 1930s, was to use total score as the criterion for item selection, on the assumption that total score was a “best estimate” of a respondent’s attitude. Likert proposed two different methods of item selection: (1) item analysis, in which selection is based on the correlation of item score with total score; and (2) the employment of a criterion of internal consistency, which is used to examine, for each statement, the difference in average item score between high-scoring and low-scoring groups defined on the basis of total score. Both methods turned out to be variants of the same general method for constructing internally consistent scales.
Likert’s insight was that psychometric methods (at that time used only in ability and achievement test construction) could be applied to attitude measurement, just as Louis Thurstone’s insight was that psychophysical methods, such as the method of equal-appearing intervals, could be used in attitude measurement. Louis Guttman, the third pioneering great of attitude measurement, invented a unique scaling method based on a rigorous definition of unidimensionality and its implications for the rank-ordering of attitude statements and respondents. Whereas Thurstone’s method scaled stimuli (attitude statements) and Likert’s method scaled responses and respondents, Guttman’s method scaled both stimuli and respondents simultaneously. However, Guttman’s method was rather difficult to apply, and Thurstone’s method was quite labor-intensive. Consequently, Likert’s method became the method of choice for constructing attitude measures and other self-report instruments, especially with the advent of computers and statistical computer programs. Nowadays, the application of Likert’s method often begins with a factor analysis of a large pool of statements, followed by the (Likert) scaling of high-loading statements for each factor. The usual criterion for item retention is contribution to internal consistency reliability. All the techniques of psychometrics, including—more recently—item response theory, can be applied to the construction of Likert scales.
Likert scales are essentially rating scales, hence they are susceptible to the problems of rating scales. Many of these problems involve response sets, which comprise response tendencies that are independent of the semantic content of the attitude statements. Response sets may result in errors of three kinds: (1) mean error, the tendency to center responses around some preferred mean value (leniency error, when the preferred mean is high, and strictness error, when the preferred mean is low); (2) variance error, the tendency to restrict or to expand the rating range regardless of correspondence to reality; and (3) covariance error, the tendency to rate all statements similarly (also called the halo effect ). These errors may be avoided or minimized by various procedures described in psychometric texts.
Likert scales are also self-report scales, hence they are susceptible to the biases implicit in self-disclosure. Clerical errors aside, responses to a Likert scale reflect the extent to which a respondent decides to “self-disclose,” which, in turn, depends on the ability and the motivation of the respondent to self-disclose. Ability to self-disclose is affected by such factors as reading comprehension (of the scale’s statements, instructions, etc.), experience, and familiarity with the subject matter of the scale. Motivation to self-disclose is more difficult to parse; respondents may be motivated by any number of reasons to understate or to overstate their attitudes. Such distortion may be avoided or minimized if the scale administrator has credibility and rapport with the respondents.
Likert scales can be measures of either stable or changeable attitudes. It behooves scale constructors to determine which kind of attitude they are measuring by conducting test-retest and longitudinal studies. These studies are especially important when the scale constructor intends (for theoretical or other purposes) to measure one kind of attitude rather than the other.
Finally, unless it is actual self-report that is of interest (as, for example, in opinion surveys), Likert scales should be validated against other independent indicators of the attitude(s) being measured. As psychometric instruments, Likert scales have to meet the basic psychometric requirements of reliability and validity.
SEE ALSO Guttman Scale; Reliability, Statistical; Scales; Surveys, Sample; Validity, Statistical
Likert, Rensis. 1932. A Technique for the Measurement of Attitudes. Archives of Psychology 22 (140): 5–54.
Rene V. Dawis