Attitude scales
Thurstone scale
A Thurstone scale is an attitude scale consisting of items (in the form of statements) with which the respondent has either to agree or disagree.
Only those items with which they agree are scored. Each item has a value and the respondent’s score on the scale corresponds to the median score of the items with which the respondent agrees.
The item scores are usually derived from asking a number of judges to rank each item on the scale using an eleven-point scale reflecting the attitude that is being measured. The final score of each item is the median of the judges’ individual scores. Usually, more items are judged than are used and the final selection is based on two criteria: first, that items covering the whole eleven-point range are included; second items should have a small variation (between judges).
Likert scale
Likert scaling is an attitude scaling method in which respondents indicate the extent of their agreement with each item on a scale (e.g., a five- or-seven point scale). Their score on the scale is the sum of the scores for each item.
Likert scales were devised in 1932 as a development of Thurstone scales, the aim was to eliminate the unreliability of using intermediary judges in scale construction. Original Likert scales had no neutral or middle point and respondents were ‘forced’ to some degree of agreement or disagreement with the scale item. This specification is not enforced by most current users of Likert-type scales.
Likert scales are relatively easy to construct. At the pilot stage, each test item is analysed to see to what extent it contributes consistently to the scale. This can be done by correlating each item score with the overall scale score. Alternatively, the sample can be split into quartiles on the basis of their scale score. The mean score on each item for the upper quartile is compared with the mean score on the same item for the lower quartile. The difference in mean scores for each item is called the discriminatory power of the item. Those with larger discriminatory power are preferable, especially if they have overall item score means approximately equal to the expected mean (i.e., the mean of the possible scores for the item, or mid-point of the range of possible item scores).
Likert scales may not always be unidimensional but the approach basically assumes a single dimension. It has been argued that analysis of a Likert scale could identify clusters that are indicative of a number of dimensions, although this is much weaker than the multi-dimension identification claimed for Guttman scaling.
The Likert method (like Thurstone and latent structure analysis) involves making inferences about the latent classes into which the manifest data can be made to fit. Unlike the Thurstone scale (where judges have to rationally assess an item’s relation to others) Likert scale items can be included that need not be overtly related to the attitude being tested. What is included is determined by the item’s correlation with the entire scale score. Thus items that show ‘underlying’ relationships can be included. For some critics, however, this raises issues of objectivity.
The Likert method attains only index measurement.
Semantic differential
Semantic differential scaling is a flexible method of attitude scaling in which subjects rate the concepts, in which the researcher is interested, on a bipolar (usually) seven-point scale. The two ends of the scale are defined by pairs of adjectives with supposedly opposite meanings (e.g. good/bad, etc.)