Factor analysis of semantic differential data

Abstract

Since the semantic differential was first introduced by Osgood et al. (1957) in The Measurement of Meaning, its use as a psychological research instrument has become firmly established. Heise, reviewing the whole field of semantic differential methodology in 1969, noted a thousand books or articles concerned with the technique. In discussions of the method it is rarely appreciated that many applications depart quite radically from the purposes for which it was originally devised, and that many of these pose special problems for data analysis. Osgood et al. developed the method originally for the investigation of connotative or metaphorical meaning. In the basic procedure, subjects rated each of a large number of concepts (e.g., Home, Me, Mother) on a set of bipolar adjectival scales (e.g., ho t -co ld , good-bad, qu ickslow). Each scale contained seven divisions to which scores could be allocated, although in subsequent studies, in line with the arguments of such writers as Peabody (1962), the number of scale divisions has often been reduced to five or even three. By forming correlations between the scales in various ways and then factor-analysing these correlations, Osgood et al. were able to reduce the scales to the three well-known primary dimensions of connotative meaning, Evaluation, Potency, and Activity (E, P and A). These three dimensions held up over a wide range of scales and concepts and have frequently been replicated in different populations (Tanaka, 1967; Osgood, 1975), An essential feature of Osgood's procedure is that both the concepts and the scales should be as diverse and numerous as possible. Subsequent work with the method has departed from this principle in a