Item Construction in Attitude Measurement

Abstract

Generalizations based on verbal attitude measures are dependent on the universe of content from which the item sample is drawn. Inadequate recognition of the need to coordinate item content to the intended realm of generalization has led traditional and current discussions of attitude measurement to an inappropriate emphasis on the assessment of global attitudinal responses. The implications of restricting the domain of content either to a specific persuasive event or to the general attitude object are discussed, and corresponding procedures for preparation of an initial item pool are outlined. Thomas M. Ostrom is Associate Professor of Psychology at The Ohio State University. T H HE MEASUREMENT of attitude involves locating the individual's typical response toward the attitude object on an evaluative continuum. The population of responses to which verbal attitude scores can be generalized is, therefore, determined by the kinds of feelings, beliefs, and stands expressed in the item pool. The variety of themes represented in the pool is potentially quite broad; investigations employing factor analysis, facet analysis, and content analysis have consistently found a variety of such themes relating to single attitude objects. When no attention is given to item composition, the very meaning of an attitude index (its realm of generalization) can differ from scale to scale. The preparation of an initial item pool is, consequently, a basic concern in the construction of any verbal attitude measure. However, while the statistical bases of test construction have become increasingly sophisticated,' little progress has been made on the problems surrounding item construction. Problems of item construction were recognized by the earliest workers in attitude measurement. At least five papers on this subject had appeared prior to 1940.2 Unfortunately, the guidance offered by these