Scale

A scale is a composite measure of some underlying concept. Many phenomena in social science are theoretical constructs that cannot be directly measured by a single variable. To understand the causes, effects, and implications of these phenomena, the researcher must develop a valid and reliable empirical indicator of these constructs. This empirical indicator is called a scale. In this sense, the scale is composed of a set of measurable items that empirically captures the essential meaning of the theoretical construct.

Good scales are data reduction devices that simplify the information and, in one composite, measure the direction and intensity of a construct. They are usually constructed at the ordinal level of measurement, although some can be at the interval level. Because they are constructed of more than one indicator of a particular phenomenon, they increase both the reliability and the validity that would be attainable if the individual indicators were used in analysis.

Several possible models can be used to combine items into a scale. These include Thurstone scaling, Likert scaling, and Guttman scaling. The method chosen depends on the purpose that the scale is intended to serve.

Scaling has three related but distinct purposes. In the first case, scaling may be intended to test a specific hypothesis (e.g., that a single dimension, party identification, structures voters' choice of presidential candidate). In this case, the scaling model is used as a criterion to evaluate the relative fit of a given set of observed data to a specific model.

Another purpose for scaling is to describe a data structure. If a political scientist attempted to discover the underlying dimensions (e.g., social, economic, cultural, etc.) of the United States' electoral system, this would be an exploratory analysis rather than a hypothesis-testing approach.

The third purpose for scaling is to construct a measure on which individuals can be placed and then relate their scores on that scale to other measures of interest.

Scaling models may be used to scale people, stimuli, or both people and stimuli. The Likert scale is a scaling model that scales only subjects. Broadly, any scale obtained by summing the response scores of its constituent items is called a Likert or “summative” scale, or a linear composite. An examination in a mathematics class is an example of a linear composite. The scale (test) score is found by adding the number of correct answers to the individual items (questions). The composite score on the scale (test) is a better indicator of the student's knowledge of the material than is any single item (question).

Thurstone's interest lay in measuring and comparing stimuli when there is no evident logical structure. The Thurstone scaling model is an attempt to identify an empirical structure among the stimuli. To do this, Thurstone scaling uses human judgments. Individuals are given statements, maybe as many as 100. These “judges” order the statements along an underlying dimension. Those statements that produced the most agreement among the judges are selected as the items to be included in the scale. These remaining items (say, 20) should cover the entire latent continuum. These items are then presented to subjects, who are asked to identify those they accept or with which they agree. The average scale value of those items chosen represents the person's attitude toward the object in question.

...